Voting for Money, Whips for Free
Andreea Ioana Alecu
Master’s Thesis
Department of Political Science
Faculty of Social Sciences
University of Oslo
May 2013
ii
Votes for Money, Whips for Free
Andreea Ioana Alecu
May 21, 2013
ii
© Andreea Ioana Alecu
2013
Voting for Money, Whips for Free
Andreea Ioana Alecu
http://www.duo.uio.no
Printed by: Reprosentralen, Universitetet i Oslo
Abstract
Polarization in the European Parliament has mainly been attributed to the ideological
dimension. The economic implications of bills have been only studied in relation to the
Council. However, after the ratification of the Lisbon treaty such questions have become
paramount for the Parliament. The aim of this thesis is to investigate whether legislative unity is lower for bills that would affect the budget in a way that, if passed, would
change the status quo allocation of resources. I propose that the impact of such budgetary
implications is dependent on the initial level of polarization within the European Parliament. In the absence of other polarizing factors, such as defections or splits within the
party of the rapporteur, budgetary implications are likely to decrease legislative unity.
Nonetheless, the negative effect of budgetary implications on legislative unity is mitigated
by defections on behalf of the party of the rapporteur.
In order to evaluate these propositions empirically I have collected data on final votes
in the 7th European Parliament. In contrast to datasets that the previous literature has
relied on, this dataset contains information on the budgetary implications of the bills.
Estimating a series of tobit regression models, I show that budgetary implications affect
the level of cohesion in the European Parliament, but this effect is conditional upon the
level of initial polarization as captured by defection from the party of the rapporteur. This
finding is relatively robust, thus shedding new light on voting patterns in the European
Parliament. Yet, some shortcomings remain, namely the differentiation at a theoretical
level between partisan and national interests, the small magnitude of the coefficients and
a potential endogenity problem. These caveats notwithstanding, it can be concluded that
the behaviour of legislators is affected by economic concerns, even if to a certain extent
such concerns are masked by ideological variables. It is likely that there is lower legislative
unity when bills aim to alter the existing budgetary status quo. Yet, conversely, when
both defection by the party of the rapporteur and budgetary implications are observed
this effect is alleviated.
iii
iv
Acknowledgements
First of all, I would like to thank my supervisor Professor Bjørn Høyland who gave me
the knowledge, guidance and skills that made the writing of this thesis possible.
All my gratitude goes to my aunt Anica Lacau, and my mother, Aftina Alecu, for
keeping me motivated during this period. I really want to thank you for your support,
love and care.
I would also like to thank all those wonderful people who lived their lives during the
past month in the 9th and with whom I have enjoyed many lovely coffee break and laughs.
Especially, I would like to thank Øyvin, Ida, Ingrid, Jørn, Lars Peter, and Øyvind for
their comments and suggestions. A special thanks goes to Jonas Nordkvelle for answering
all my questions.
Lastly, but not least, I would like to thank that special person, for being there in my
life and giving me more love and support than I ever hoped for.
This thesis is dedicated to the most important person in my life, my father.
I am solely responsible for any errors and omissions in this thesis.
Word Count: 22 698
v
vi
Contents
List of Figures
3
List of Tables
4
1 Introduction
5
1.1
Money makes the EU go round? . . . . . . . . . . . . . . . . . . . . . . .
6
1.2
Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
2 Theoretical framework
8
2.1
European Integration and national interests . . . . . . . . . . . . . . . .
8
2.2
Understanding the European Parliament today . . . . . . . . . . . . . . .
10
2.2.1
Negotiations at the committee level . . . . . . . . . . . . . . . . .
12
2.2.2
Voting in the European Parliament . . . . . . . . . . . . . . . . .
15
2.3
Cohesion for free? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
2.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
3 Data and Research Design
25
3.1
Data collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
3.2
Data gathering process and coding . . . . . . . . . . . . . . . . . . . . .
27
3.2.1
When do bills have budgetary implications? . . . . . . . . . . . .
30
3.3
Operationalization of legislative unity . . . . . . . . . . . . . . . . . . . .
32
3.4
What about ideology? . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
3.5
Choice of model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
3.6
How does the Tobit model work? . . . . . . . . . . . . . . . . . . . . . .
37
3.7
Linearity? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
3.8
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
4 Results
42
4.1
Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
4.2
Bivariate Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
4.3
OLS Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
4.4
Cohesion, not rejection . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
1
2
CONTENTS
4.5
4.6
4.7
4.8
Tobit Regression Results . . . . . . . . . . . . . . . . .
Budgetary implications and defections . . . . . . . . .
4.6.1 Simulated Probabilities: Substantial difference?
Model diagnostics . . . . . . . . . . . . . . . . . . . . .
4.7.1 What drives the estimates? . . . . . . . . . . .
Summary . . . . . . . . . . . . . . . . . . . . . . . . .
5 Robustness
5.1 Shortcomings of the empirical approach . . . . .
5.2 What else is there? Controlling for ideology and
5.3 Accounting for policy areas . . . . . . . . . . .
5.4 Summary . . . . . . . . . . . . . . . . . . . . .
. . .
time.
. . .
. . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
51
57
59
63
63
67
.
.
.
.
68
68
70
73
78
6 Conclusion
79
7 Bibliography
82
Appendices
90
Appendix A Influential observations
90
Appendix B Tobit models without interaction
94
Appendix C Heteroskedasticity
95
Appendix D R code
96
List of Figures
2.1
Aggregate voting cohesion per European Party Group . . . . . . . . . . .
13
3.1
3.2
Standardized residuals vs. predicted values for the AI tobit model . . . .
Test of normal distribution for residuals . . . . . . . . . . . . . . . . . .
39
40
4.1
4.2
Scatter plot between AI and RICE index . . . . . . . . . . . . . . . . . .
Party unity by budgetary implication. The green lines illustrate individual
observations, while the purple area shows the distribution. . . . . . . . .
Interaction plot based on Model (1) . . . . . . . . . . . . . . . . . . . . .
Simulated effect of budgetary implication for bills with defection . . . . .
Simulated effect of budgetary implication for bills without defection . . .
Most influential observations for interaction term . . . . . . . . . . . . .
45
4.3
4.4
4.5
4.6
A.1 Influential observations for the tobit model with AIn as the dependent
variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Influential observations for the tobit model with RICE as the dependent
variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.3 Influential observations for the tobit model with RICEabs as the dependent
variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.1 Test of heteroskedasticity for the AI tobit model . . . . . . . . . . . . . .
3
46
57
60
61
64
91
92
93
95
List of Tables
3.1
Overview of dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
4.1
4.2
4.3
4.4
4.5
4.6
Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . .
Bivariate correlations for the variables of substantial interest . . . .
OLS Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Logit models with outcome of the vote as the dependent variable . .
Tobit models with different specifications of the dependent variable
Tobit models without influential observations . . . . . . . . . . . . .
.
.
.
.
.
.
43
43
48
50
53
65
5.1
5.2
5.3
Tobit models with controls for ideology and number of sessions . . . . . .
Fixed Effects Multi-level linear models . . . . . . . . . . . . . . . . . . .
Random intercept and effect of budget controlled for party positions . .
71
74
77
B.1 Tobit models with only budgetary implication . . . . . . . . . . . . . . .
94
4
.
.
.
.
.
.
.
.
.
.
.
.
Chapter 1
Introduction
After the ratification of the Lisbon treaty the European Parliament has become a vital
player in the European Union. Understanding the factors that lie behind the legislator’s
voting decisions has become highly important in this context. In this thesis I choose to
focus on the economic determinants of voting. In many respects, the European Parliament resembles national legislatures. Much like national parliaments it is composed of
legislators representing different constituencies. In parliamentary democracies one observes a high overall level of cohesion and overall agreement (Diermeier and Feddersen
1998, Hix, Noury and Roland 2005, Persson and Tabellini 2005, Shugart 1998). Hence, in
many respects the European Parliament resembles national parliaments: it is composed
of legislators with divergent preferences, who represent different constituencies and who
are able to forge and maintain a high level of agreement in the plenary.
Despite the weak electoral connection between Members of the European Parliament
(hereon, MEPs) and national constituencies (Hix and Høyland 2011, Carey 2007), legislators have strong incentives to satisfy the desires of their national constituencies. As
proposed by Fenno (1978) legislators have three main goals: re-election, power while in
office, and good policy.
The resemblance of the European Parliament (hereon, EP) to its national counterparts
has been supported by studies of the voting behaviour (McElroy and Benoit 2007; 2012,
Hix, Raunio and Scully 1999, Hix 2002, Hix and Noury 2009). The conclusion converged
upon by both quantitative and qualitative research is that the main determinant of conflict
has been the left-right ideological dimension, and national interests mainly play a role in
high-profile salient cases (McElroy and Benoit 2007; 2012, Hix, Raunio and Scully 1999,
Hix 2002, Hix and Noury 2009). Yet, is ideology everything?
The intergovernmentalist approach highlights the interests of member states in the
bargaining process (Moravcsik 1998). Recent studies focused on the Council have shown
the importance of the economic determinants of voting. Zimmer, Schneider and Dobbins
(2005), and Mattila (2009) were among the first to focus solely on redistributive interests
in the Council. Recently, Bailer, Mattila and Schneider (2010) have shown that the
5
6
CHAPTER 1. INTRODUCTION
interactions in the Council are influenced by economic factors, such as re-distributive
outcomes. Peltzman (1984) shows that ideological variables are likely to mask economic
ones, and that constituent interests affect the behaviour of legislators. For the European
Parliament the economic aspects of voting have however, not received much attention
in the existing literature. This is puzzling, as there is a large level of agreement in
recent scholarship that the Europan Parliament has become more similar to national
parliaments. This motivates the following research question: Do budgetary implications
impact cohesiveness in the European Parliament?.
1.1
Money makes the EU go round?
The findings of this thesis suggest that budgetary implications have an impact on the
overall legislative cohesion. The effects of budgetary implications have proven relatively
robust across model specifications. In the absence of other polarizing factors, budgetary
implications decrease legislative unity. When accounting for party positions, the results
suggest that legislators prefer to abstain or not to vote, instead of directly defecting.
However, in cases where there are other polarizing factors, for example defection by the
party of one of the rapporteurs, legislative cohesion is likely to increase, despite budgetary
implications. In other words, a high level of polarization in the European Parliament
before the final vote mitigates the effects of budgetary implications.
These findings hold across various model specifications, even if the effect of budgetary
implications is somewhat weaker when including dummies for party positions. In these
cases it becomes important to account for legislators that choose to abstain or not to
vote. It seems that ideology remains an important determinant of voting within the
European Parliament, even when controlling for economic interests. This is expected, as
Peltzman (1984) shows that ideological variables mask economic interests. One finding
that remains robust in all the models estimated in this thesis is that legislative unity is
likely to be higher in cases where the bill has both budgetary implications and where the
party of the rapporteur also defected. There is little doubt that European Party Groups
are strategic actors, thus they are likely to have strong incentives to use party pressure
in order to establish and maintain a cohesive behaviour in the plenary. Controlling for
policy areas or in other words, accommodating the model to the hierarchical structure of
the data, does not change the substantial relationship.
Nevertheless, caution is advised when interpreting the results, as the magnitude of the
coefficients is small. Furthermore, as defection is observed at the same time as legislative
unity, even if chronologically it is known before, instruments should have been employed,
in order to remedy this potential problem. Unfortunately, as discussed in section 5.1 such
an approach was not possible given the weak correlation between the instruments and
the variable to be instrumented.
1.2. STRUCTURE
1.2
7
Structure
In the second chapter I will present a short overview of the literature on the European
Parliament and develop the theoretical framework and testable hypothesis. The third
chapter presents the data collection process and discusses several methodological choices
made, such as operationalization and the choice of model. The findings are presented
in the fourth chapter; the chapter is concluded by a discussion of model fit. In the
robustness chapter several other model specifications are presented and the shortcomings
of the empirical approach are discussed. The main conclusions are summarized in the
Conclusions chapter.
Chapter 2
Theoretical framework
In this chapter I develop the theoretical framework and derive a set of testable hypotheses.
The core argument is that the budgetary implications of bills have an implication for
the level of cohesion, yet these implications are contingent upon the presence of other
polarizing factors before the final vote. The mere presence of budgetary implications is
likely to reduce legislative unity. Nonetheless, Party Groups are likely to increase the
level of pressure on legislators when,for example the party of the rapporteur is split or is
going to defect on that set legislators. In these cases the effects of budgetary implications
are alleviated.
2.1
European Integration and national interests
The European Parliament is part of a complex network of institutions. When looking
at the main theoretical approaches that aim to explain the formation and perpetuation
of the European Union, such as neofunctionalism (Aspinwall and Schneider 2000), intergovernmentalism, or neointergovernmentalism (Moravcsik 1998) there is some ground to
expect that monetary gains, or concerns will affect the payoff and implicitly the behaviour
of legislators. For example, Moravcsik’s neo-intergovernmentalist approach highlights the
importance of member state’s interests in the bargaining process (Moravcsik 1998).
At a first glance this appears to apply less to the European Parliament given it’s weak
connection to the electorate (Carey 2007) and the rarity of national party defections
(Nordkvelle 2012). Nonetheless, the loyalty of MEPs is split, on the one hand they are
accountable to their national parties, which insure re-election, while on the other to their
European Party groups. It has been shown that national interests also play a role, especially in high profile cases (Hix, Noury and Roland 2006, Hix and Noury 2009), therefore
it can be argued that legislators are also interested in what their constituencies will gain
or loose after a legislation is implemented. This is further discussed in Section 2.2. Fenno
(1978) proposes that legislators will aim to satisfy the desires of their constituency in
order to insure their other goals. Even if Fenno’s (1978) theory was proposed for the US
8
2.1. EUROPEAN INTEGRATION AND NATIONAL INTERESTS
9
Congress there are limited reasons to doubt its applicability to the European Parliament,
as discussed in Section 2.3. One implication is that legislators are more willing to reject
proposals that are detrimental to their national constituencies. This is further discussed
in Section 2.3.
Furthermore, the importance of economic gains is given some attention in the neoinsitutionalist approach (Schneider and Cederman 1994, Hug and König 2002). In simple
terms, the neo-insitutionalist approach highlights that intergovernmental negotiations are
a matter of who wins and who looses. This can be extrapolated to legislative politics as
well. (Bailer 2004), and Selck and Steunenberg (2004) highlight this. Bailer (2004) focuses
on the Council of Ministers and investigates how factors such as votes, economic strength,
position on policy area and agenda setting lead to success in bargaining situations. Her
findings highlight that exogenous factors, such as resources of power and number of votes
do not always bring about bargaining success, while patient negotiations seem to be more
important Bailer (2004). Her focus does not lie on the intrinsic characteristic of the bills,
but on the characteristics of those who vote. By focusing on the intrinsic characteristics of
a bill, namely its budgetary implications, this thesis hopes to make a minor contribution
to the existing literature.
The aim of this thesis is to investigate how legislative unity is affected by the structural
characteristics of a bill controlled for other exogenous factors such as party of rapporteur
and her seniority. Arguably, a natural starting point for the theoretical framework of
this thesis are several contributions contributions addressing the factors that impact the
behaviour of legislators (Hillman 1989, Peltzman 1984, Grossman and Helpman 1996,
Stigler 1971). Peltzman (1984), for example, shows empirically that in many cases economic interests are masked by ideological variables. Peltzman (1984) mainly focused on
congressional voting, furthermore, given the similarities between the congressional voting
and the European Parliament it might be argued that his main finding can be translated
to the setting of the European Parliament. In the US Congress it has been shown that
economic interests are important determinants of the voting behaviour (Magee, Brock
and Young 1989).
Extrapolating theoretical expectations has some inherent advantages, first it allows
one to get more substantial insight in one area. Secondly, it has a methodological advantage, it allows one to make a contribution to the existing literature by employing at
least part of a design “designed for some purpose in one literature could be applied in
another literature to solve an existing but apparently unrelated problem” (King, Keohane
and Verba 1994: 17). Given that similar approaches have been employed in the Council (Bailer, Mattila and Schneider 2010, Aksoy 2010, Kandogan 2000, Carrubba 1997) it
might be more interesting to investigate how re-distributive legislation with budgetary
implication affects legislative unity. Before proceeding to developing this framework, I will
draw upon several literatures aimed at explaining legislative cohesion and the behaviour
10
CHAPTER 2. THEORETICAL FRAMEWORK
of legislators in the plenary.
2.2
Understanding the European Parliament today
The European Parliament has become an important decision maker within the European
Union. Its powers have been further increased by the adoption and ratification of the
Lisbon treaty. It can be argued that the European Parliament is almost as powerful
as the Council in all policy areas. The changes in the rules of procedure did not only
strengthen the main political groups (Kreppel 2002), but also aimed at making the EP
a stronger actor in relation to the other institutions (Hix and Høyland 2013: 3), while at
the same time it gained budgetary power. In this context it becomes even more important
to understand the dynamics of the European Parliament.
During the current session there were 1201 individual cases voted upon and 1093
of these cases were passed, while only 103 bills failed. This is a small indication that
the European Parliament has become more cohesive.1 Hix, Noury and Roland (2005)
show that there is increased cohesion within the European Parliament, despite that the
European Party Groups (EPGs) and the national ones have become more ideologically
diverse. The level of cohesion within the institution is well documented. It has been
shown numerous times that the main dimension of conflict in the European Parliament
is the ideological one (Hix 2004, Hix, Noury and Roland 2005; 2006, Hix and Høyland
2013).
Legislators are conceptualized as agents of two principals, as they have to respond to
both the demands of the European Party Groups and to those of the national parties
(Hix 2002). Transnational party groups are formed based on policy interests (Kreppel
and Tsebelis 1999, McElroy and Benoit 2012). An individual level of explanation for the
existence of transnational parties is given by Hix and Noury (2009). They posit that
division of labour combined with sharing of information are the main reasons why MEPs
with similar preferences choose to organize themselves in European Party Groups (Hix,
Noury and Roland 2007).
Following the rational presented above it is implicit that almost all national parties
have strong incentives to join EPGs. A similar approach is also presented by Hix and
Høyland (2011: Chapter 3). National parties on their own have almost no chances of
securing office or their policy goals unless they join European Party Groups (Hix and
Høyland 2011: 54). For example in 2010 there were only 3.1% of the legislators that were
Non-affiliated (McElroy and Benoit 2012: 154). Thus, it can be argued that individual
national parties have little, or no influence on their own within the European Parliament.
Yet, this does not imply that national parties are unable to influence their representatives.
One potential implications is that the effects of national and transnational parties
1
This paper focuses only on final votes. The final date of the coding is 2013-03-14.
2.2. UNDERSTANDING THE EUROPEAN PARLIAMENT TODAY
11
are different on legislators, as they can have different leverage (Lindberg, Rasmussen and
Warntjen 2008: 1109). Conventionally, legislators in the European Parliament have been
conceptualized as agents of two principals. It has been shown that MEPs are to respond
to the demands of the national parties, which insure their re-election, while at the same
time they try to accommodate the preferences of the European Party Groups (Hix 2002;
2004). Unlike national parties, transnational ones do not have the incentive to establish
strong brand names and maintain these (Lindberg, Rasmussen and Warntjen 2008: 1113).
Transnational party groups have strong incentives to form a centralized leadership
which monitors the compliance of legislators and sanctions them accordingly (Lindberg,
Rasmussen and Warntjen 2008: 1113), much like national parties. Furthermore, there is
a consensus in the field that national interest are mainly present high-profile cases (Hix
and Noury 2009).
Recently, it has been shown that European legislators are mainly influenced by their
European Party Groups (Mühlböck 2012). Following the recommendations set out by
Lindberg, Rasmussen and Warntjen (2008), Mühlböck (2012: 7) finds that there is a
culture of consensus between the institutions and that transnational groups play a strong
role. She highlights that in a very large majority of the cases national parties do not
play a role, as they cannot afford to care (Mühlböck 2012). A weakness of the study
is that it is solely based on the final stage of the decision making. Therefore, it cannot
be concluded that national parties do not have an influence during the decision making
process, for example at the committee stage. Yet, it can be argued that as they cannot
afford to care about final votes, it is doubtful that they have the resources needed to
invest in the decision making process.
McElroy and Benoit (2012: 152) argue that the European level party system is relatively fluid, as national parties switch their European level affiliation, while at the same
time some several transnational parties cease to exist, while others are established (McElroy and Benoit 2012). For example when focusing on the party groups elected in the European Parliament in 2009, one notices that two new political groups were formed2 after
the dissolution of the Union for a Europe of the Nations (UEN) group, which was formed
in 1994, but was not reassembled after 2009 (McElroy and Benoit 2012: 153-154). At the
same time, the composition of the European Parliament in terms of national groups has
also changed. Seventy national parties gained seats in the European Parliament, while 50
national groups that were represented in the 6th legislature failed to secure representation
in the current term (McElroy and Benoit 2012: 153-154).
To a certain extent this makes the picture even more complex. There is a mutually dependent relationship between national parties and European Party Groups. This implies
that agreement has to be reached between the national and transnational organizations
in order to insure a recursive and long-lasting collaboration. There is strong evidence
2
Europe of Freedom and Democracy group, and European Conservatives and Reformists group
12
CHAPTER 2. THEORETICAL FRAMEWORK
that the main transnational groups have become strong and cohesive, if not collusive,
over time (Hix, Noury and Roland 2007, Raunio 1997, McElroy and Benoit 2012). Given
these opposing forces at work, it is necessary to establish a high level of cohesion in the
plenary. Parties need to act in a cohesive manner in order to maximize their influence
in the legislative decision making process (Cox and McCubbins 2007). It can be argued
that to a certain extent national parties have become less important in the European
Parliament.
At the same time, the internal structure of the parties has developed to a great extent
in the latter years. One such important development is the assignation of members to
committees. Jacobs, Corbett and Shackleton (1990: 348) propose that the position of the
EP is both formed and negotiated at the committee level. Yordanova (2009) argues that
the committees face a large degree of both internal and external pressure. This provides
a possible explanation for the observed level of cohesion. Shifting negotiations to the
committee level provides an opportunity to reach a grand coalition before the plenary.
2.2.1
Negotiations at the committee level
There are several reasons why it is desirable to shift negotiations at the committee level.
First, the committees have their own resource structure and the members are specialized
in their respective area. This is highly advantageous for the parties as they will gain
more inside knowledge. A side effect of this is that the position the EP will be strengthened during the negotiation process with the Council and Commission, as the EP will
have an informational advantage (Raunio 1997, Mamadouh and Raunio 2003). Furthermore, Neuhold (2001) proposes that committees also serve as a means of a developing
a majority in the EP on an issue-by-issue basis. Yordanova (2009: 254) adds that the
“committee system also provides an arena for developing strong cohesive position among
party groups”.
In a context where it seems that a large part of the focus European Party Groups
lies upon maintaining and enforcing discipline (Carrubba, Gabel, Murrah, Clough, Montgomery and Schambach 2006, Carrubba 1997, Carrubba, Gabel and Hug 2008, Hug 2003)
it remains slightly puzzling why strong parties would shift a part of their decision making
capabilities to individual legislators. Arguably, at least in part the answer to this question is a matter of scarcity of resources. It is doubtful that the transnational parties have
enough resources to follow the entire decision making process in the legislative. Transnational parties are formed based on policy preferences. Therefore, it is safe to assume that
the members of the party will have similar preferences to the central leadership. This is
seen by looking at the high rates of voting cohesion in the current European Parliament.
The cohesion rates are displayed in the Figure 2.1. We see that the only group with a
cohesion rate below 50% is EFD, while all the other groups have a cohesion of over 80%.
2.2. UNDERSTANDING THE EUROPEAN PARLIAMENT TODAY
13
Figure 2.1: Aggregate voting cohesion per European Party Group. Source: votewatch.eu
In order for such cohesion rates to be attained and maintained, a intermediary step
between the reviewing the proposal from the Commission and plenary voting is needed.
The Committee stage provides this intermediary step. I argue that parties prefer to
invest resources in a beneficial committee assignment on their behalf, in order to insure
a cohesive position of the party in the plenary. This argument is in line with Raunio
(1997) and Neuhold (2001). European Party Groups assign members to the committees
in a way in which they can maximize their gains. For the purpose of this thesis the
way in which members are assigned to committee is not important. Several authors such
as Yordanova (2009), Bowler and Farrell (1995), McElroy (2006; 2008) and Whitaker
(2005; 2001) have explored the factors that lead to committee assignments. However,
the aspect of relevance is that transnational party groups incur a certain costs when
appointing members to committees and that EPGs have an incentive to assign members
to committees.
Cox and McCubbins (1993) provided evidence for the US Congress that parties are
able to enforce discipline through the institutions of legislative in itself. Hix, Noury and
Roland (2007) argue that there are similar mechanisms that influence the behaviour of
parties in the European Parliament. Theoretical models of party discipline are relatively
common in the literature (Carrubba et al. 2006, Carrubba 1997, Carrubba, Gabel and
Hug 2008, Hug 2003).3 If parties did not care about committee assignment they will suffer
considerably higher costs and increased uncertainty at the plenary level. Therefore, there
are few reasons to believe that higher costs and a high level of uncertainty in coalition
formation is desired by the rational EPG leadership. These costs could be seen as political
transaction costs.
The transaction costs at the plenary level are composed of the two same elements as
discussed by Furubotn and Richter (1997: 47-48). Their content is different since we are
referring to political parties and not national states, yet the rationale remains unaltered.
First, there are the costs of setting up and maintaining the political organization of
the system. These costs encompass the formal negotiations, the ability of the party
to impose its own view and the costs of withstanding external pressure. The external
3
Although some contributions contents the party’s disciplining power, for example Mühlböck (2012)
14
CHAPTER 2. THEORETICAL FRAMEWORK
pressure can either come from national states, other European level institutions or from
interest groups. Secondly, there are the costs of running a polity (Furubotn and Richter
1997: 47-48). Here, we aim to capture both the costs linked to internal organization (such
as committee or rapporteurship appointment), the costs of maintaining and imposing a
cohesive polity line and the costs associated with monitoring and enforcing the agreed
policy line.
European Party groups have to balance out these two types of costs. It is widely
accepted that transnational parties in the European Union have few resources. Therefore,
in order to maximize their influence in the legislative arena parties have to carefully choose
their battles, in other words, carefully assess their costs versus their benefits. The costs of
internal organization and monitoring and enforcing behaviour come at two levels. First at
the macro-level of inter-institutional bargaining and secondly when parties address their
internal organization.
One highly efficient way of reducing the costs, while at the same time increasing
the efficiency of legislative negotiations is to shift the inter-institutional organizational
costs, to the intra-party level. If transnational parties were not to shift the core of the
negotiations to the committee level they would suffer the same costs twice. Therefore, it
is rational for parties to shift negotiations at the committee level as it maximizes their
utility while at the same time it minimizes the costs. Firstly, when they appoint members
to the committees and yet again when they would incur the same costs at the plenary
level. Therefore, by shifting the real bulk of debate and negotiations at the committee
level parties can more effectively allocate their resources.
Albeit strong theoretical reasons to expect committees to be the optimal solution to
the negotiation problem in the context of MEPs with split loyalty, committees are not
always the best way to represent the view of the entire European Parliament. In the
words of the former vice-president of Party of European Socialists as quoted by (McElroy
2006: 13):
Committees are definitely and regrettably not representative of the Parliament
in plenary, they are not microcosms; this results in legislative distortion. The
environmentally minded from all groups are on the Environment committee,
giving it a distinctly green outlook; likewise there are too many farmers on
Agriculture. The result of this specialization and lack of representativeness
is that policy is not reflective of the majority view of the Parliament and we
frequently have to spend hours in Parliament voting to correct the committee
report and proposed legislation. (Personal interview)
consequently, even if committees are for many motives optimal for insuring a cohesive
position in the plenary the also present some drawbacks, as illustrated above.
In conclusion, it seems that successful negotiations in the committee stage lie at the
heart of insuring a cohesive position in the plenary. Consequently , it can be proposed that
2.2. UNDERSTANDING THE EUROPEAN PARLIAMENT TODAY
15
in cases where these negotiations fail, or are burdened with disagreements, that European
Party groups, unable to rely on their alliance partner, will use the party pressure in order
to insure that at least their MEPs will act in a cohesive manner and not defect from
the party line. This is further explored in Section 2.3. Nonetheless, even if negotiations
at the committee level are likely to impact the level of agreement within the European
Parliament, there are also other factors that affect the legislator’s behaviour in the plenary.
This is further discussed in the next section.
2.2.2
Voting in the European Parliament
Carey (2007) proposes that legislative unity is mainly driven by the following three factors:
similar preferences, discipline and agenda control. At the same time, it can be argued
that within party cohesion is mainly driven by the division-of-labour (Carey 2007) and
the costs of collecting information (Hix and Noury 2009).
The literature aiming at explaining the existing cleavages and party cohesion in the
European Parliament has been on the ideological dimension (Hix, Noury and Roland 2007,
Hix and Noury 2009, McElroy and Benoit 2012; 2007). Many of these studies aimed at
investigating the dimensionality of the policy space. Some studies have used the available
roll call data and employed spatial models and thereafter regressed the dimensions of
conflict (Hix and Noury 2009, Hix, Noury and Roland 2005, Hix and Høyland 2013).
Despite its inherent advantages, this approach has received some criticism (e.g. Mühlböck
and Yordanova 2012). Others have looked at ideological party positioning by employing
expert surveys (McElroy and Benoit 2012; 2007).
These studies show that there is overall one dominating dimension of voting in the
European Parliament, namely the traditional left and right, while at the same time some
variation in voting explained by another pro- / anti-integration dimension (Hix, Noury
and Roland 2007). One of the weaknesses of spatial models of legislative voting (Poole
2005) is given by the lack of control variables in the analysis - the algorithms aim to
maximize the correct number of classifications given the specifications of the model.4
Secondly, the content of the dimensions needs to be inferred post-hoc (Hix, Noury and
Roland 2006: 495). Even, with these minor caveats the results have proven to be very
robust (Hix and Høyland 2013). This finding is also supported by studies employing a
different method - namely expert coding. The series of paper by McElroy and Benoit
(2007; 2012) also reach a similar conclusion. McElroy and Benoit (2012) show that
European Party Groups occupy the entire left-right political spectrum. Furthermore,
they add that national parties are relatively cohesive in terms of their placement as well.
This leads one to expect that most of the variations in legislative unity are explained by
ideological factors.
Arguably, a high level of cohesion has been a scope in itself. For example, Hix and
4
This is further discussed by (Poole 2005)
16
CHAPTER 2. THEORETICAL FRAMEWORK
Noury (2009) show that the high level of cohesion that is observed in the European
Parliament implies, at least to a certain extent, that the EP has started to resemble
national parliaments. Furthermore, the extensive literature on the European Parliament
has also given us increased understanding the balance of power between institutions,
especially when it comes to the need for cohesion as a way for the traditionally relatively
weak and new European Parliament to impose itself.
When it comes to explaining the Parliament’s interaction with the Council and Commission one of the main focuses of the research has been on procedural differences
(Crombez 1996; 1997; 2000, Tsebelis 1994, Tsebelis and Garrett 2000). This approach
has offered important insights on how the European Parliament functions. It has explained the relative power of EPGs and has aided research aimed at understanding the
dimensionality of this institution. Nonetheless, in the post-Lisbon setting, co-decision II
became the ordinary legislative procedure. This implies that the power of the European
Parliament has increased considerably. Another implication for the theoretical advances
in the field is that procedural aspects have become so sparse, that analyzing them in the
present setting is unlikely to provide us with new insights.
Before proceeding to a discussion of other potential factors affecting voting I will
draw upon some of the main finding from the literature focusing on ideology. Given the
increased powers of the EP, and despite the high level of agreement, defections are rare,
yet they still occur. Hix, Noury and Roland (2006) use nominate scaling methods as
a proxy for ideology, in order to estimate the number of dimensions that exist in the
European Parliament. They find that the main dimension of politics in the European
Parliament is the traditional left and right dimension (Hix, Noury and Roland 2006).
The evidence of the existence of a second dimension is relatively weak and has declined
in importance over time (Hix, Noury and Roland 2006: 507). Hix and Noury (2009)
conducted a similar study after the 2004 round of expansion, when ten new member
states were added and reached similar conclusions.
The findings from Hix and Noury (2009) largely support those from Hix, Noury and
Roland (2006). How can the EP be conceptualized? The European Parliament is much
like any normal legislature and is to be conceptualized in a low dimensional policy space.
The levels of party cohesion remained relatively stable when compared to those from the
Fifth European Parliament and ideological distance remained the most important predictor of voting behaviour, even if nationality played a role in the case of the controversial
Services Directive (Hix and Noury 2009). This provides strong evidence that the most
important dimension of ideology is the left and right ideological battle, as highlighted by
(Hix and Høyland 2013).
The study conducted by Hix and Noury (2009) shows that the voting behaviour of
MEPs was not affected by the size-effect created by the pure increase in the number of
legislators, nor by a composition effect, created by the socio-economic differences between
2.3. COHESION FOR FREE?
17
the old and new members. Overall, new MEPs from the 10 new members states behaved
slightly different from the remainder, however these differences are more visible in the
case of highly salient directives Hix and Noury (2009).
There is one puzzle that still needs to be addressed: are there only ideological differences that drive defections? Is the observed level of cohesion an artifact of strong party
leadership that mastered rapporteur appointment and that have shifted negotiations to
the committee level in order to insure cohesiveness in the plenary? The European Parliament is a heterogeneous organization with a hierarchical structure. Therefore, the nature
of the cases there is bound to spark a higher polarization in some cases compared to others. Especially in situations, where it is against the legislator’s own interests to present a
unified front that will accept a proposal that will be either against her ideological position,
or the desires of her constituency.
This thesis sets out to provide one possible explanation for cases where there is increased polarization at the EP level and an alternative mechanisms for why individual
defections occur in some cases. Aside procedural differences other structural characteristics of the bills discussed have been largely ignored by the literature in the case of the
European Parliament. New research (Bailer, Mattila and Schneider 2010) on the Council
highlights the importance of economic factors with regards to the level of polarization or
cohesion. More specifically, I am to investigate whether bills with budgetary implications
alter the level of cohesion within the European Parliament given the existing preconditions. In the next section I will use theory developed for the Council and derive several
testable predictions.
2.3
Cohesion for free?
Overall, research on the legislative bodies of the European Union has ignored the structural and re-distributive implications of legislation. If Lasswell (1950) famous dictum
“politics is who gets what, when, where and how” is still of interest, it can be argued
that there is a re-distributive interests affect the conflict dimension in the European Parliament. To the best of my knowledge there are no studies investigating how budgetary
implications affect legislative cohesion in the European Parliament. This stands in stark
contrast to how such approaches have been recently employed on the Council (Bailer,
Mattila and Schneider 2010, Aksoy 2010, Kandogan 2000, Carrubba 1997). Given the
increase in importance and power of the European Parliament, extrapolation of this approach to EP may give us new insights in the dimensions of conflict that govern this
complex institution.
The aim of this thesis is to investigate how legislative unity is affected by the structural
characteristics of a bill controlled for other exogenous factors such as party of rapporteur
and his/her seniority. From a substantial point of view the most relevant case for such
18
CHAPTER 2. THEORETICAL FRAMEWORK
an analysis is the 7th European Parliament. The current EP is constituted after the
ratification of the Lisbon treaty, which increased the power of the Parliament, secondly
the co-decision became the ordinary legislative procedure and the EP gained for the first
time some budgetary power (Hix and Høyland 2011). These factors are likely to better
highlight the theoretical mechanisms proposed below.
Arguably, a natural starting point for the theoretical framework of this thesis are
the contributions of Hillman (1989), Peltzman (1984), Grossman and Helpman (1996)
and Stigler (1971). Peltzman (1984), for example, shows empirically that in many cases
economic interests are masked by ideological variables. The article is solely focused on
congressional voting, nevertheless, given the similarities between the congressional voting
and the European Parliament it might be argued that his main finding can be translated to
the setting of the European Parliament. The US Congress literature seems to agree that
economic interests are important determinants of the voting behaviour (Magee, Brock
and Young 1989).
A more general theoretical framework is proposed by Grossman and Helpman (1996)
who investigate how special interest groups affect the platforms of the respective candidates. They show that parties are willing, at least to a certain extent to give into the
demands of the special interest groups (Grossman and Helpman 1996). They further
show that parties aim at maximizing the sum of aggregate benefits between informed
votes and those of the special interest groups. Even if, the theoretical approach is relatively far away from looking at the individual reasons for defections, it gives a hint that
economic gains actually matter in the legislator’s, and to some extent party’s preferences.
In the context of the European Union, Bailer, Mattila and Schneider (2010) argue that
interest groups are also influential in the distribution of the European Union budget.
This highlights that defections based on economic concerns are not necessarily linked to
national interests. There are many possible explanations for why legislators are more
willing to act in a less cohesive manner on bills that have budgetary implications than
otherwise. Nonetheless, given the limited scope of this thesis the main aim is to establish
whether such interests are actually a polarizing factor within the European Parliament,
especially in relationship with national states.
The above gives some evidence that at least from a theoretical standpoint the economic
implications of legislative proposals are important for the legislator’s incentive to support
or not support a bill. We are forced to depart slightly from the framework used in the
case of the Congress, or settings where governments or national states are represented.
The European Parliament encompasses individual members from every member state.
Therefore, theoretically we can differentiate between two motivations for caring about
budgetary implications: national gains to their own constituency and gains to supporters
of the European Party group.
In the case of the Council it has been shown that the redistributive dimension has
2.3. COHESION FOR FREE?
19
gained more importance (Zimmer, Schneider and Dobbins 2005). Another interesting
mechanism that came forth in the case of the Council was that losers were not likely to
remain quiet (Zimmer, Schneider and Dobbins 2005). While the link between Council
members and their constituencies is clear, this link is, however, more blurred in the case of
the European Parliament. Drawing on the literature of democratic deficit (Føllesdal and
Hix 2006, Crombez 2003, Majone 1998), one notices that the weak connection between
the electorate and the European Parliament is frequently mentioned. Why should such
mechanism apply to the EP when the link between electorate and MEPs is frail and at
the same time legislators are organized in transnational party groups?
One answer to this question comes from the literature on the US Congress. Fenno
(1978) aims at answering how legislators are influenced in their decisions by their home
constituencies. He proposes that legislators have three main goals, re-election, maximization of power while in service, and good public policy Fenno (1978). These goals are
very similar to how MEPs have been conceptualized so far. Despite the frail electoral
connection, there are strong reasons to believe that legislators, also within the European
Parliament, care about their national constituencies, as they are the ones that insure
re-election. Fenno’s (1987) argument that legislators will employ a “home style”, while
they cultivate trust in their respective constituencies. Given the similarities between
the European Parliament and national parliament, more general legislative theory is also
supportive of this argument. For example Weingast, Shepsle and Johnsen (1981) argue
that districts are highly important in the allocation of political pie, especially in regards
to economic benefits and taxation. Therefore, as long as re-election within the European
Parliament, or a the continuation of a political career is a goal for legislators they will
have strong incentives to account for the interests of their “home constituencies”.
Granted that legislators are interested in constituency interests, it is only natural
that they will aim at insuring that benefits will be distributed towards the constituency,
while at the same time will try to avoid that detrimental economic measures will affect
these. These “sources of bias” in legislation are further discussed by Weingast, Shepsle
and Johnsen (1981) and to a certain extent by Shepsle (1979). Yet, in certain cases
legislators might be willing to accept short-term costs for their constituency in favour or
long term benefits, or for other reasons, such as maximizing individual utility be willing
to internalize these costs. While in other cases, given that the EP is trans-national these
costs might be so spread that the losses will be difficult to identify. Furthermore, there
is a high degree of cohesion within the European Parliament and there are relatively
few defections (Hix and Høyland 2011; 2013, Nordkvelle 2012). The juxtaposition of
these factors with a normative desire of good policy implies that whether a bill has not
a budgetary implication is in general irrelevant for its passage.
Another reason for why this might be the case is simply the multitude of factors affecting the legislator’s voting decision at the moment of the vote. Budgetary implications
20
CHAPTER 2. THEORETICAL FRAMEWORK
are not the only reason legislators vote the way they do. They are merely a factor that
is likely to affect their decision one way or the other. It can be argued that the failure or
rejection of a bill is dependent on the aggregate interest legislators have for such a bill,
not on whether it alters the legislative status quo. If group X desires bill Y, and she is
able to get support from groups A, B, C, those groups will per definition support that bill
regardless of the costs. If groups A, B, and C would have not desired that bill they would
have not agreed to support it in the first place. Therefore the first testable implications
is:
Hypothesis 1. Budgetary implications do not affect the probability of a bill being passed
or failed.
One example of a case with budgetary implication which was desired by a considerable
number of legislators and parties is “Mechanism for monitoring and reporting greenhouse
gas emissions and other information relevant to climate change”5 . Most of the costs of this
legislation were initially covered by the existing framework within the European Union
(European Commission 2011b), hence new budget lines were not requested. In this case
93% of the legislators voted for the proposal and the winning coalitions was formed by all
parties with the exception of EFD. Another similar legislation is “Accounting rules and
action plans on greenhouse gas emissions and removals resulting from activities related to
land use”6 . Consequently, when legislators endorse the costs of a legislation, and the type
of case does not have a strong ideological dimension, the effect of budgetary implications
will be minimal.
Even if legislation is not likely to impact the failure or passage of a bill, it is likely
to affect legislative unity. It can be argued that in cases of legislation which either will
affect specific national countries, or regions, or that will affect a specific sector legislative
unity is likely to be reduced. One example of the latter is the “Common fisheries policy”7 ,
where only 502 legislators voted for while 137 voted against. This polarization is likely
to be visible also in the case of resolutions, that in time will potentially affect either the
budget of the European Union, or of individual member states, such as: “Decision on the
opening of, and mandate for, inter-institutional negotiations on common organisation of
the markets in agricultural products (Single CMO Regulation)”. It can be distinguished
between two mechanisms that affect the legislator’s decision, on the one side national
party pressure, while on the other and perhaps most importantly the desire to represent
the national constituency’s needs. These mere examples seem supportive of the arguments
presented above, nonetheless, empirical tests of this are employed in Section 4.4.
Even if budgetary implications are unlikely to determine the chances of passage or
failure of bill, they are more likely to increase legislative polarization, therefore have a
5
case number: A7-0191/2012
case number: A7-0317/2012
7
case number: A7-0008/2013
6
2.3. COHESION FOR FREE?
21
negative impact on legislative unity. If it is assumed that losers are willing to use their
voice, in other words vote against a proposal that has detrimental budgetary implications
for their constituency, party or that supports something they are ideologically against, it
is better to look at overall legislative unity.
Hix and Høyland (2013) show that an overwhelming majority the winning coalition
has been formed by five or more parties. This framework does not imply that legislation
with budgetary implications will be passed with fewer parties, even if that might be the
case, but it implies that individual defections and polarization are more likely. Hence,
under the assumption that the discontent will show their position, it is expected that
overall legislative unity will be lower.
As discussed in the previous sections the European Parliament has large incentives to
present an unified front in order maximize its influence in the legislative process. Some
studies have shown that defections from the party line are rare, yet it is interesting to
look at which factors might motivate defections when studying aggregate legislative unity.
Factors such as pressure from European Party Groups, or national parties might exercise
varying pressure upon legislators. Given the diverse factors that affect the legislator’s own
voting decision factors such as membership in one or another European Party Group, or
even the pressure of a national group become less important. Local constituencies are
smaller than both national parties which usually represent the country and European
Party Groups. Of course, group pressure might have a different impact on different legislators, nevertheless, from a theoretical perspective this is not that relevant. As defections
will most likely take place when the interests of the constituency of the legislators are at
risk, as in these cases the legislator will be most interested in those that actually vote for
him (Fenno 1978).
Theoretically, it can be distinguished between two types of legislation, those bills
that are desired by a large majority of the legislators and do not spark conflict and
those cases where there is an a high level of internal polarization. The polarization can
either be on the necessity of the measures encompassed by the bill, the way in which
it alters the status-quo or the fact that it alters the existing consensus. In practice
quantifying this distinction is virtually impossible, as it is hard to trace the paternity
of a bill. Additionally, the level of polarization is dynamic. For example, a bill which
starts as relatively uncontroversial, for example ACTA, becomes controversial while it is
drafted and debated. At the other side of the spectrum, one can find bills which start
out as controversial, yet when the bill is parsed through the complex decision making
mechanisms within the European Union the bill ends up being relatively uncontroversial
by the time of the final vote.
Clarke and Primo (2007: 734-744) argue that theoretical models are to be conceptualized as “maps” and that even if all aspects cannot be tested they are still to retain these
are they aid one in understanding the phenomenon. A dynamic measure of polarization
22
CHAPTER 2. THEORETICAL FRAMEWORK
within the European Parliament is very hard to attain and construct. Therefore, I choose
to employ a proxy for this. In this framework the most most important level of polarization is the one a bill has right before its final vote. Consequently, the proxy employed
in this thesis to capture this latent concept is defection by the party of one of the rapporteurs. As discussed in Section 2.2.1, committee level negotiations are paramount for
legislative unity. Thus, in cases where there is a split within the party of the rapporteur,
or the party defects, the other parties are likely to use their available party pressure in
order to constrain the behaviour of legislators in the plenary.
Granted that the bulk of negotiations is shifted at the committee level, it can be
argued that in cases where bills have budgetary implications, it will be harder to get all
legislators to agree, and hence we will observe lower legislative unity. Additionally, when
budgetary implications are the only polarizing factors, and committee level negotiations
were successful, legislative unity will be lower in the plenary. In those cases the alliances
are already formed at the committee level and European Party Group have few incentives
to increase pressure on individual legislators. Therefore, the legislator that feels that the
proposal is detrimental for his home constituency will defect without incurring the risk
of losing the trust of his respective EPG, hence to a large extent he will continue to
maximize all of his three goals.
Hypothesis 2. Bills with budgetary implications are more likely to decease the overall
parliamentary cohesion in the absence of other polarizing factors.
Despite the high level of agreement within the European Parliament there might be
cases where negotiations at the committee level may stall, or be only partly successful.
For example, it has been shown that in general high-profile, salient cases tend to be
dominated by more mechanisms than normal bills (Hix, Noury and Roland 2005; 2007,
Hix and Høyland 2013). Alternatively, it can be argued that, regardless of the profile
of the case, negotiations at the committee level which are burdened with a large level of
disagreement. Such cases can be identified by mapping splits of the defection of the party
of one of the rapporteurs.
In such cases, party groups, will notice such defections. Party groups are strategic
actors that aim at getting their most desired policy through. In situations where they
know defection from another group is likely they are likely to increase pressure on legislators in order to further their goal. Under the relatively not problematic assumption
that party groups will increase internal pressure on legislators when there is an unstable
coalition let us investigate how this would potentially affect the behaviour of legislators.
As long as legislators have three main goals that they aim to maximize, namely reelection (e), power while in office (p), and good policy (g), they are bound to respond to
party pressure. Nonetheless, the response will be different for each legislator, depending
on her utility function. In this application the most interesting situation is when there is a
case which has a budgetary implication, and there is a split in the party of the rapporteur.
2.3. COHESION FOR FREE?
23
As discussed above, if budgetary implications are detrimental for the home constituency
of the legislator she is most likely to defect in order to protect her home constituency, as
in there are few consequences. In other words, it is conceptualized that in the absence
of additional polarizing factors, e and p will remain constant, while e will decrease if the
legislator does not respect the desires of the home constituency.
Nonetheless, in the situation at hand, with increased party pressure, both e and p
are likely to be affected. Thus, legislators are faced with a dilemma, where they have
to choose between reduced e and reduced p. Yet, there is a weak connection between
the electorate and legislators (Hix and Høyland 2011, Scarrow 1997) , while, as discussed
above party groups have very strong incentives to monitor the behaviour of legislators.
Given this, I propose that legislators are more likely to follow the line of the European
Party Group, and not defect from the party line. This is likely to lead to increased
legislative unity when compared to cases where budgetary implications are present on
their own.
If European Party Groups will increase legislative pressure the legislators will experience higher costs if they choose to defect from the party line. One potential reason is that
European Party Groups might sanction frequent defectors, for example by granting them
less speaking time, or less favorable committee assignment. If such a sanction would be
in place the legislator’s second goal, of power maximization, would be potentially threatened. Nonetheless, it remains unclear, which goal legislators value higher. Perhaps, the
answer to this question is highly dependent on the case analyzed and on the preferences
of the legislator at that point in time. For example, in cases where the national party is
preparing nominations for the next term, it might be the case that the legislator would
prefer to employ his/her home style and defect from the party line, despite sanctions, in
order to maximize the chances of re-election. This leads to a final testable implication:
Hypothesis 3. For cases with budgetary implications, and splits or defections on behalf
of the party of the rapporteur, legislative cohesion is likely to be higher.
One such example is “Fishing opportunities and financial contribution provided for
in the EC-Denmark/Greenland Fisheries Partnership Agreement”8 . In this case there
were clear national interests at stake, for example those of Denmark. Nonetheless, the
European Party Group Greens/EFA opposed this legislation, 574 legislators voted for
this legislative proposal. As discussed by the intergovernmentalist approach, national
interests can be important for legislators. (Moravcsik 1993: 482) proposes that interstate
negations are dependent on national state preference formation. European Party Groups
as strategic actors, will surely take this in their calculations and therefore increase pressure
in cases where such mechanisms are present. As anticipated above, in cases which have
both budgetary implications, and when the committee stage negotiations encountered
troubled waters European Party Groups are likely to put more pressure on legislators.
8
case number: A7-0358/2012
24
CHAPTER 2. THEORETICAL FRAMEWORK
In the cases where there is a split between the party of the rapporteur, the impact
of budgetary implication may be of the opposite direction. This follows from the fact
that splits in the party group of the rapporteur are more easily detectable and may thus
induce EPGs to employ the available “whips” to discipline their MEPs. Furthermore, the
costs of defections for the legislators will in these cases be significantly higher as defection
may hamper the other goals that legislators have. Hence, the collocation of polarizing
factors is most likely to increase legislative unity. This is tested in Section 4.5.
2.4
Summary
By drawing upon several literatures this thesis proposes a new framework for assessing
cohesion in the European Parliament. In this chapter I have first reviewed the literature
assessing legislative behaviour in the European Parliament. So far, the literature (Hix
2002, Hix, Noury and Roland 2007, Hix and Noury 2009, McElroy and Benoit 2012)
has focused on the ideological determinants of voting in the European Parliament or
on procedural differences (Crombez 1997; 2000). Nonetheless, after the ratification of
the Lisbon treaty the powers of the European Parliament have increased considerably,
especially as now the Parliament has budgetary power (Hix and Høyland 2011). By
extrapolating recent findings in the Council (Bailer, Mattila and Schneider 2010) and
employing classical theoretical arguments developed originally for the US Congress (Fenno
1978, Cox and McCubbins 2007) I propose that the budgetary implications affect the
behaviour of legislators in final votes.
As legislators are mainly responsive to the demands of the European Party Groups
(Mühlböck 2012), their behaviour is not only determined by the way in which they value
the concerns of their constituencies – versus their own gains while in term – but also
by party pressure as well. In order to capture this, I have argued that the initial level
of agreement before final votes, which can be seen by splits in the party of the rapporteur is likely to affect their payoffs. I have further proposed that legislators aiming at
avoiding sanctions will respect the desires of their party group in situations where they
add increased pressure and are more likely to defect in cases where party groups do not
mobilize. Consequently, the negative effect of budgetary implications is mitigated by increased party pressure. The next chapter presents the data and research design employed
in order to test theoretical propositions.
Chapter 3
Data and Research Design
In the first part of this chapter, I will describe the data collection process and present
and overview of the variables included in the dataset. Thereafter, I present several issues
linked to coding before proceeding to discussion of coding one of the substantially important variables. The discussion changes course to the operationalizations of the dependent
variable and the choice of model and is completed with a test of the linearity assumption.
3.1
Data collection
The focus on this thesis lies on the 7th European Parliament and it aims to investigate
the effect of budgetary implications of legislation on legislative unity. As the literature
has only marginally focused on how budgetary implications affect legislative cohesiveness
in the European Parliament, there is no available data on the budgetary implications
of bills. In order to be able to test the theoretical propositions quantitatively I have
collected data on the bills passed in the 7th European Parliament from the first sitting on
01.07.2009 to 14.03.2013. The adoption, and ratification of the Lisbon treaty increased
the powers of the EP. Firstly, the Parliament is now on equal footing with the Council,
as co-decision became the ordinary legislative procedure. Secondly, the Parliament has
gained increased budgetary power. Therefore, the European Parliament has now become a
stronger decision maker within the EU. Given the scope of this thesis it is the most relevant
to study the hypothesized relationship in the context of the 7th European Parliament.
As this is a novel dataset and the collection of this data is one small contribution made
to literature, it may be useful to go through the process of data collection and inherent
coding issues and decisions made while making the dataset in this section. The main
advantage of this dataset is that it consists of the entire population of cases available, has
very limited number of missing observations and spans over a relatively significant part
25
26
CHAPTER 3. DATA AND RESEARCH DESIGN
of the current term.1 The dataset records several characteristics of interests in the frame
of this thesis, which are presented in Table 3.1 on page 28, the coding of the variables in
presented in section 3.2. The units of analysis are final votes.
King, Keohane and Verba (1994: 23-25) set out five main guidelines for data collection
in qualitative research. Given their generality they are highly relevant in this setting as
well. The data collection process has been carried out in accordance with the guidelines
of King, Keohane and Verba (1994: 23-25).
I have collected the information on a set of variables deemed to be important for the
entire population of cases from the 7th European Parliament. This implies that issues
of sample selection and sample bias become redundant as long as the aim is to make
inferences about the current sitting of the European Parliament. At the same time, this
satisfies the second criterion set out by King, Keohane and Verba (1994: 24).
Ideally, similar data should have been collected for the previous sittings of the European Parliament. In that case one could have made more general inferences about
how budgetary implications affect legislative unity in situations where the parliament has
limited decision making power over the budget, compared to when it is one of the actors
that decides the budget. Given the relatively limited scope of this thesis I argue that the
population of cases voted upon in the European Parliament during the current sitting
(last date coded: 14.03.2013) will suffice for the purpose at hand. It has to be emphasized
that this is a first to test how budgetary implications affect legislative unity, which may
be extended by further research. The main scope of the analysis is to investigate whether
budgetary implications have an impact on legislative unity.
The next two guidelines are linked to validity and reliability. Validity refers to measuring the actual concept that we aimed to measure, namely that we capture the latent
concept (King, Keohane and Verba 1994: 25). Simply because of the nature of the variables, for example the result of a vote is either a pass or a fail, or if 200 voted for, there are
only 200 votes for. Of course one can question the indices used as dependent variables,
however, that is not a problem of data collection, but one of operationalization of latent
concept. Hence, such issues are discussed in section 3.3. As budgetary implications are
variable of substantial interest in this thesis, a further discussion of issues of validity and
reliability it taken in regard to the way budgetary implication was operationalized and
coded in practice.
Reliability encompasses that one is to employ the same procedure for every case coded
(King, Keohane and Verba 1994: 25). In order to insure the reliability of the dataset, I
have randomly chosen 25 cases and re-coded them. The results were the same. Even if
the method used to code each case was consistent, there is an inherent danger of typos,
such as writing 29 instead of 20. This danger is larger for the numeric variables. The 30
1
The main source of missing is logical missing. The overwhelming majority of the 253 missing observations on the number of terms, presented in table 4.1 on page 43, are logical missing as those cases did
not have rapporteurs
3.2. DATA GATHERING PROCESS AND CODING
27
most influential observations presented in section 4.7.1 are not coding errors.
I have performed several tests to check for coding errors. The errors were identified
and corrected before the analysis of the data was started. The first test was to check if
the numbers of legislators that voted, did not vote and were absent corresponded to the
total number of MEPs. The number of MEPs fluctuated slightly during the set period, a
± 10 boundary was used on legislators. Secondly, I have aggregated abstentions per day
and double checked the cases where such errors were present. Even if these tests are not
perfect, they should at least to a certain extent increase the reliability of the data.
One of the main goals was to maintain the dataset replicable. This is also the last
criterion set out by King, Keohane and Verba (1994: 26). The data is collected from
various publicly available sources. Each source is both an open source, and to the extent
of my knowledge is regularly updated, facilitating both the replication of the dataset,
as well as tests of coding. The case number, and the number of terms served by a
rapporteur are taken from http://www.europarl.europa.eu/portal/en, while the names
of the rapporteurs and their respective parties are retried from http://parltrack.euwik
i.org/. Budgetary implications are coded from both of the above, while depending on the
availability of the documents from the official website of the Commission and European
Parliament. The remaining variables are coded based on the information available on
http://www.votewatch.eu/.
3.2
Data gathering process and coding
The dataset represents an effort to code all the resolutions and bills voted upon in the
current term of the European Parliament which lies at the heart of the analysis. In developing the framework I have explicitly focused on two aspects, the budgetary implication
and the way in which legislators voted. While there are several sources of acquiring the
bills voted upon in the current term2 , the dataset collects information with other indicators that in some cases are available from other sources, such as the number of votes, rate
of absenteeism, or the rapporteur for each case in order to facilitate empirical investigation. The ID-indicators of the data set are the case number (i.e. A7-0010/2009) and the
date of the vote are coded, alongside with the full name of the bill, these will also enable
merging with other datasets. A full overview of the variables is presented in Table 3.1 on
page 28.
The actual coding was done in a case-by-case manner. If the same case has been
voted upon twice, for example a first reading and then the document returned for a
second reading both cases are coded, on different dates. As long as two votes are taken
upon the same bill they will share the same case number. In cases where a split vote was
2
For example votewatch.eu .
CHAPTER 3. DATA AND RESEARCH DESIGN
28
Variable:
source: parltrack.euwiki.org, europarl.europa.eu
source: europarl.europa.eu
All names are separated by commas for ease of disaggregation
source: votewatch.eu
source: europarl.europa.eu
If the rapporteur changed party,
the party coded is the one at the time the vote was taken,
source: parltrack.euwiki.org
source: europarl.europa.eu
Description and source:
Source: europarl.europa.eu
The full name of the document, information about type of
legislation and vote, procedure
source: votewatch.eu
Passage or failure of a bill on the respective vote, source: votewatch.eu
Information about the policy area.
Can be used as a proxy for the lead committee. source: votewatch.eu
source: votewatch.eu
source: votewatch.eu
source - votewatch.eu
Records the number of MEPs that
were present, but did not vote. source: votewatch.eu
The variable was coded per case, however as a check
I have aggregated the data by session date and checked for coding errors
Source: votewatch.eu
source: votewatch.eu
Bills have budgetary implications only when they alter the existing status quo.
Source: parltrack.euwiki.org, europarl.europa.eu
Table 3.1: Overview of dataset
Type of variable (possible values):
Character - example: A7-0001/2009
Character - example: Agreement EC/
Mongolia on certain aspects of air services Legislative resolution : Single vote - consultation
Categorical, 0 for rejected proposals, 1 for approved ones - example: 1
Character - example: Transport & tourism
Case number
Name of document
Result of vote
Policy Area
voted yes - example: 456
voted no - example: 21
voted no - example: 2
did not vote - example 218
Continuous,
Continuous,
Continuous,
Continuous,
that
that
that
that
Yes votes
No votes
Abstain votes
Didn’t vote
Continuous, number of MEPs that
were absent from the session - example 17
MEPs
MEPs
MEPs
MEPs
Absent
Categorical, 0 for resolutions, 1 for legislative proposals - example: 1
Categorical, 0 for no budgetary implication,
1 for budgetary consequences - example: 0
of
of
of
of
Type of vote
Budgetary implication
Character, name of other rapporteurs,
one column per each - example: COSTA Paolo
Character, EPG of the each other
rapporteur at the time of the voting - example: ALDE
Continuous - example: 3
number
number
number
number
Majority formed by
Character, names of parties in winning coalition example: GUE/NGL, Greens, S&D, ALDE, EPP, ECR, EFD
Character, name of lead rapporteur - example: SIMPSON Brian
Character, EPG of the lead rapporteur at the time of the voting - example: S&D
rapporteur at the time of the voting - example: S&D
Lead Rapporteur
Party of lead rapporteur
Number of terms served
by lead rapporteur
Other rapporteurs
Party of other rapporteurs
3.2. DATA GATHERING PROCESS AND CODING
29
requested on the final vote - the case will appear twice, the full name allows the user to
differentiate between the two, as it specifies which part of the document was voted upon.
Most of the other information coded is dependent on the case, therefore the variables
are coded per case, however there is one exception, namely the budgetary implication.
If a vote taken upon an entire bill which originally had a budgetary implications and a
split is proposed only for an amendment, which did not have a budgetary implication,
the first vote on the amendment will have no budgetary implication, the second vote on
the bill will have a budgetary implication and the final vote on the entire bill will also
have a budgetary implication. It has to be emphasized that this was a rare case, the total
number of rows affected by this coding decision is very small (≤ 3).
In order for the reader to get a sense of how the data was collected and coded I present
an overview of the variables, possible doing categories and a short description, in Table
3.1 on page 28. All the examples used are from the same case, hence show how a row
was coded. The example was chosen arbitrary. In this paragraph I will employ a different
example, in order to highlight the coding of the variables. Firstly, I coded variables that
aid one in identifying the case, such as the case number, for example A7-0329/2012 and
the full name of the document. One example is, “Tariff-rate quotas applying to exports
of wood from Russia to the EU - Draft legislative resolution : single vote - ordinary
legislative procedure, first reading”. Based on this, I have coded whether the case was
a legislation (coded 1), or a resolution (coded 0). Thereafter, I have coded the policy
area, which in this case is International trade. Next, the voting behaviour of legislators
was coded, how many voted yes, no, and abstain, as well as information about those who
were present but did not vote.
The budgetary implication was coded as 0 for cases that were encompassed by the
existing framework, and 1 for cases which alter the existing status-quo. This is further
discussed in the next section, Section 3.2.1. Finally, with the aid of information from vote
watch.eu, I have coded the European Party which formed the winning coalition. Lastly,
the names, and European Party Groups of the rapporteurs were coded. As mentioned, I
have coded the EPG to which the rapporteur belonged at the time of the vote.
As mentioned the variables are coded per case, therefore each new row in the dataset is
either a new case, a recurring case which was voted again, for example a second reading, or
in very few cases a split vote. It has to be emphasized that the overwhelming majority of
the cases were new cases. This allows one to explore the both the structural characteristics
of the case and the aggregated behaviour of legislators and to a certain extent of European
Party Groups at the moment of the voting.
Most of the concepts covered are self-explanatory and require a relatively low level
of abstraction. A standard code-book was not required for the data gathering process.
Variables such as the number of yes, no, abstentions, or the number of absent legislators
only required a transcription of the numbers to the data sheet. Yet, in order to maximize
30
CHAPTER 3. DATA AND RESEARCH DESIGN
coherence in the data gathering process I developed strict coding rules for the variables
where a subjective decision was required. In the next section I will present a discussion
of the decision made while coding budgetary implications.
3.2.1
When do bills have budgetary implications?
A budgetary implication can be simply defined as: a bill has a budgetary implication when
its enforcement has an impact, either positive or negative impact on the budget. Nevertheless, under this definition any bill act would have a budgetary implication, as its mere
application, and enforcement entails at least certain administrative costs. Furthermore,
the European Union budget is designed as a framework, where the distribution of certain
resources is allocated from the start. This entails that all European Party Groups and
legislators are aware that certain sums will be spent on certain areas. It is outside the
scope of the thesis to discuss what mechanisms affect the distributive decisions around
budgetary allocation.
Do bills with financial implications which are restricted to the existing framework
actually have budgetary implications? This question is debatable. It can be argued that
these bills do not have in practice any budgetary implications, as they do not alter the
status quo. The conflicts between legislators and European Party groups are already
resolved at the time of the vote on the actual proposal. A bill which has financial implications which are already covered by the existing framework will not change the status
quo on the area. The main focus lies on legislative unity, and the interest lies on how
budgetary implications alter the voting behaviour of legislators. If a proposal is covered
by the framework, there are no theoretical grounds that it will alter the behaviour for
legislators as the status quo will remain unchanged.
From a theoretical perspective the interests lies on polarization within the legislative,
as discussed in Chapter 2. In order to capture the theoretical concepts the definition of a
budgetary implication used in this thesis is: A bill which either adds a positive contribution
or has a negative impact on the budget and which is not covered by an existing framework
or goes beyond it, in other words a bill which does change the existing budgetary status
quo. In other words, bills which alter the existing framework. Under the assumption
that every legislative act will have an impact on the budget, no matter how small - this
definition allows one to distinguish between two types of legislation. Those that will
trigger economic polarization within the legislative on the one side, and those bills that
only have regulatory outcomes, where economic concerns are largely redundant in final
votes.
How to differentiate between the two possible situations? In many respects this has
been the hardest variable to code. Nonetheless, the European Union’s commitment to
openness and the availability of internal documents have enabled the coding of this variable. In order to code the budgetary implication, I have looked first at the published
3.2. DATA GATHERING PROCESS AND CODING
31
legislative report, thereafter at the committee draft report and in some cases at the documents attached to the procedure from the Official Journal. The published legislative
report proved to be a very good proxy for coding the budgetary implication.
During the coding process, I noticed that bills which had budgetary implications in the
Commission’s proposal retained them. Conversely, bills that started without budgetary
implication, seem to have remained without budgetary implication to the final vote.
Legislation with budgetary implications is required to have an addition set of documents
attached, even if such documents are not always available, the reports mention the their
presence.
In order to give the reader a better understanding of how the variables were coded and
where the information was retrieve from I will choose two arbitrary examples, one without and one with budgetary implication. For example case A7-0223/2011 - Derivatives,
central counter-parties and trade repositories does not have a budgetary implication. By
looking at the published legislative proposal - COM(2010)0484 - it is noted in the section
Budgetary implication that “The proposal has no implication for the Union budget.” (European Commission 2010). In the committee draft report there is no mention of budget, or
financial implications (PE456.945). Finally, when we look at the documents attached to
the procedure published in the Official Journal we see that there is no mention of financial
or budgetary implications of the legislation (OJ C126 2011: 9). Therefore, A7-0223/2011
was coded as not having a budgetary implication.
On the other hand let us shift focus to a bill with a budgetary implications. For example, case A7-0218/2012 - European statistical program 2013-2017 which has a budgetary
implication. The published legislative proposal clearly states that the legislation has a
budgetary implication –
Total amount to be borne by the budget of the EU is 299.4 million EUR
(current prices) for the duration of the programmer from 2013 to 2017, of
which 57.3 million EUR is covered by the programming period 2007 to 2013
and 242.1 million EUR by the programming period 2014 to 2017. (European
Commission 2011c)
A similar statement is also made in the Committee Draft report (ECON, Committee
on Economic and Monetary Affairs 2012), thus budgetary implications that exceed the
allocated framework are in most cases easy to recognize. At the same time, it has to
be mentioned that a careful analysis of the above mentioned documents was required in
order to identify budgetary implications in certain cases.
In some cases the difference between bills with budgetary implications and those
without, or already covered by the budget is blurred. Generally, legislation which was fully
covered by the existing framework was simply stated not to have a budgetary implication,
or in other cases the Commission’s report only stated that the bill has a minor impact
32
CHAPTER 3. DATA AND RESEARCH DESIGN
on the EU budget, without further details. As mentioned, in those cases, after a careful
analysis of the documents from the Commission and Committees was required, and by
following the definitions presented in this section a decision was made in each case.
Arguably, in the terms discussed by King, Keohane and Verba (1994) and Adcock
and Collier (2001) the concepts of reliability and validity of the measurements are largely
satisfied. Given the low degree of abstaction, the standard formatting of the documents
and their availability there were few hinders when going from a theoretical concept to
empirical observations.
3.3
Operationalization of legislative unity
In order to measure legislative unity, I choose to employ two indices used for measuring
party unity (Rice 1924, Hix, Noury and Roland 2005). The classical approach was developed by Rice (1924). The index is given by the absolute difference between Yes and No
votes divided by the sum of Yes and No votes. This shown in equation 3.1.
RICE =
|Y es − N o|
Y es + N o
(3.1)
The RICE index ranges from 0 to 1, where 0 is reached when an equal number of
legislators vote Yes and No, while 1 is reached when all vote either Yes or No. A vast
majority of the cases in the European Parliament are voted upon under the simple majority rule (Hix and Høyland 2011), where legislators have three voting options, namely:
Yes, No, Abstain. Unfortunately, the RICE index in its original form is ill-equipped to
deal with abstentions.
Nevertheless, abstentions in the EP have no impact on the outcome of the vote under
simple majority. Thus, ignoring abstentions is not per se substantially problematic. At
the same time, ignoring abstentions totally, implies the strengthening of the assumptions
that legislators are hard liners and will choose direct defection from the party line. At
the same time, a conventional approach has been to treat abstentions as no votes. For
example several empirical studies of ideology using nominate scores operate with this
assumption (Hix, Noury and Roland 2005; 2006; 2007, Hix and Noury 2009). One way
of understanding this in the context of the EP is to assume that the same mechanisms
that leads to no votes lead to abstentions. Hence, the index becomes:
RICEabs =
|Y es − (N o + Abstentions)|
Y es + (N o + Abstentions)
(3.2)
In order to better highlight this, I present a hypothetical example. If 100 legislators
vote “Yes” and 10 “No”, while another 20 abstain from voting: RICE reports a relativly
high cohesion score of 0.81, while RICEabs reports a much lower score of only 0.53.
One implication of this difference is that RICE is likely to over-estimate legislative unity,
3.3. OPERATIONALIZATION OF LEGISLATIVE UNITY
33
while RICEabs is likely to under-estimate it. Therefore, using both indices as dependent
variables might prove a be useful and simple solution to avoiding and controlling for this
bias problem.
Adding an extra assumption on behaviour is not the optimal way of dealing with
abstentions. Another better approach would be to differentiate the group of legislators
that did not vote from those that votes “Yes” and those that voted “No”. A solution to
this problem is presented by Hix, Noury and Roland (2005: 215) initially to deal with
party group cohesion. However, legislative cohesion is just aggregated party cohesion,
therefore there should be no constraints in using the Agreement Index as a proxy for
overall legislative cohesiveness.
The Agreement Index is calculated via the following formula (Hix, Noury and Roland
2005: 215):
AI =
max{Y, N, A} − [ 21 (Y + N + A) − max{Y, N, A}]
Y +N +A
(3.3)
The Agreement Index (AI) treats abstentions, labelled with A as a separate category.
Unlike, the RICE index the interest here lies on the largest group of votes against the
others. RICE and the Agreement Index are similar, thus in many respects the Agreement
Index can be conceptualized as a more generic version of RICE which is capable of dealing
with more than two categories. Treating abstentions as a different category affects the
final outcome. Using the same example from above we can illustrate how the results
change. The Agreement Index in this case is of 0.538. We see that AI score is very close
to the score on RICEabs when we account for abstentions, however it still remains much
lower than the original RICE score.
Even if, the Agreement Index can be seen as an improved version of RICE it still
fails to deal with one issue. It does not account for those that were present, but did not
vote. Arguably there are two ways of including them, either as a separate category, or
like in the case of RICE combine them with one of the existing categories. Those that
did not vote have absolutely no impact on the outcome of the vote, hence it is hard to
conceptualize them as a separate category. It can be further argued that the lack of a
vote from legislator cannot affect legislative unity.
A more natural way of treating those that did not vote might be to include them in the
abstention category. The reason for combining these categories is that from a theoretical
standpoint they have the same impact on legislative unity. They do not impact the
outcome of the vote, but are two categories that should be accounted for, at least in
empirical tests. In order to test how this impacts legislative cohesion I chose to include
them, this slightly alters the results and formula, becoming:
AD = Abstain + Did not vote
(3.4)
34
CHAPTER 3. DATA AND RESEARCH DESIGN
AIn =
max{Y, N, AD} − [ 21 (Y + N + AD) − max{Y, N, AD}]
Y + N + AD
(3.5)
Including them as a separate category will mathematically change the result only
when those that did not vote are the largest group. In the sample at hand this is seldom
the case. In order to have a basis for comparison I will slightly modify the example above,
by assuming that there were 20 more members that were present at that session but did
not vote. When those that did not vote are included we see that the AIn index decreases
to 0.33, from 0.58. This implies that dealing with those that did not vote can lead to
a problem of under-estimating legislative cohesion. Nevertheless, since the theoretical
reasons for excluding those that did not vote are weak I choose to run the empirical tests
on AIn as well.
3.4
What about ideology?
Most of the existing literature in the field has argued that ideology is the main predictor
of the voting behaviour of MEPs in the European Parliament, as discussed in Chapter
2. It has been shown by employing nominate models that the left-right dimension is the
main dimension of voting within the European Parliament (Hix and Noury 2009, Hix and
Høyland 2013) and that European Party Groups occupy the entire range of the left-right
spectrum (McElroy and Benoit 2012).
This thesis does not aim to investigate the importance of ideology for legislative unity
in the European Parliament. It focuses on how the budgetary implications of bills affect
legislative unity. Nonetheless, ideology, or party positions are important factors that one
should control for. There are several ways of doing this, nevertheless the structure of the
data employed in the empirical analysis and the length of the time series impeded me for
using better controls for ideology, such as nominate scores. An alternative operationalization would have been a measure of ideology similar to that developed by McElroy and
Benoit (2012). Nevertheless, data for such a measure on the current European Parliament
is not available. If the measure is not adjusted in a case by case manner it might lead
to misleading results. In this context I will employ an alternative solution to control for
party positions, namely party dummies for each of the parties in the winning coalition of
each case.
Let us take a closer look at the party system in the European Parliament. In 2003,
Hix, Kreppel and Noury (2003) concluded that the party system in the EP “has become
more consolidate and more competitive as the powers of the EP have increased”. Nevertheless, a lot has changed from 2003, the European Union had two rounds of enlargement
and in 2009 the Lisbon treaty was ratified increasing again the powers of the European
Parliament. In broad terms the party system has remained relatively unchanged, however
after the 2009 election new parties were formed, while others ceased to exist (McElroy
3.4. WHAT ABOUT IDEOLOGY?
35
and Benoit 2012), however there were not radical changes in the overall ideological position of parties, only in the names of the European Party Groups. Secondly, parties have
increased in size, but the composition of the largest parties remained relatively stable.
Nonetheless, new national parties joined the European Parliament (further discussed in
McElroy and Benoit (2012)).
In the seventh European Parliament, the European People’s Party is the largest group
with 270 members, followed by S&D with 190 members and ALDE with 85.3 Even if these
three parties to a certain extent have divergent ideologies, their policy interests overlap
in many areas. If these three groups reach agreement on a certain vote the legislation will
pass with a majority in favour of at least ∼ 72%. A center left-coalition without ALDE
would mean a support of only ∼ 36%, while a center-right with ALDE would mean a
support of ∼ 47%. Therefore, if S&D and EPP have divergent preferences on an issue,
they will both battle for the support of ALDE. Nevertheless, it has been shown that in ∼
70% of the cases a coalition of either center-left, with EPP and ALDE or one of center-left
and EPP against ALDE is formed (Hix and Høyland 2013: 9). The initial reaction is to
conclude without further ado that the party system has become collusive and that there
is very little competition between the largest parties within the European Parliament.
One explanation for this type of behaviour is the ideological proximity of the two
largest parties (Hix, Raunio and Scully 1999, Hix 2001, Hix, Kreppel and Noury 2003,
Kreppel and Tsebelis 1999). The literature points out that EPP and S&D tended to
be clustered closer together on the pro/anti-integration dimension rather than on the
conventional left and right one. However, the integration issues have become less salient in
the past years. When looking at the voting patterns in the seventh European Parliament,
we notice that there are three areas where the left and the right tend to compete more
when compared to the remainder. There are two areas where a center-right coalition tends
to be more successful and there is competition between the left and right on Economic and
Monetary questions and in Employment and Social matters (Hix and Høyland 2013: 9).
The left seems more successful in Gender Equality and Environment and Public health
(Hix and Høyland 2013: 9). Both of these trends are a continuation of the pattern from
the sixth EP. As the European Parliament got increased power in the budgetary sector
we see that there is increased competition compared to the sixth EP between EPP and
S&D on that area.
There is little doubt that ideology is a very complex concept. As discussed in the
beginning of this section two of the most reliable measures of ideology cannot be used in
the application at hand due to data availability issues. Using party dummies is therefore
the next best alternative.
Even if, this operationalization comes with several caveats, it is the best alternative to
control for party positions, given the nature of the data. It captures how party alliances
3
There a some small variations as MEPs sometimes change affiliation during the term
36
CHAPTER 3. DATA AND RESEARCH DESIGN
vary from case to case. Given that the position of parties is only secondary to this analysis
the models which control for ideology are presented in the Chapter 5 of the thesis.
3.5
Choice of model
The dependent variable in this case is the overall unity of legislators in the European
Parliament. This is a latent concept that is operationalized in several distinct ways, as
discussed above, in Section 3.3. In accordance with the definition from Greene (2003:
896) the dependent variable is censored. Censoring implies that the researcher is able to
observe the independent variables for the entire sample, but she is not able to observe
the full spectrum of the dependent variable (Long 1997: 187). In other words, censored
observations occur when “some observations on the dependent variable, corresponding
to the known values of the independent variable(s) are not observable” (Kennedy 2011:
262).
Kennedy (2011: 264) illustrates that omitting or not accounting for limit observations
creates bias. In order to avoid such problems, Kennedy (2011: 264) proposes to include
the limit observations via maximum likelihood. This means that in the case of censored
variables the parameter estimates from simple OLS regressions are likely to be biased
(Henningsen 2010: 1). If we were to fail to account for censoring we could have estimated
an OLS, the problem would have been that censored observations would pull down the
line, and hence cause an underestimated intercept while at the same time it would overestimate the slope (Long 1997: 189). James Tobin proposed a way to deal with this in
his 1958 study of household expenditures (Tobin 1958).
Why not an OLS? At a first glance the model seems to require the basic requirements
for an OLS model. The dependent variable must be continuous. If this is not the case.
An OLS-model will return inconsistent estimates for the data (Wooldridge 2002: 524). If
we are to restrict the sample to the observations of yi > 0, in other words looking only
at the data on the uncensored observations, we are going to create omitted variable bias.
Tobit or a probit estimation? These two models employ the same structural model,
however there are some differences between the two. In a Tobit model it is assumed that
we know the value of the latent dependent variable y* when y* > 0, however in a probit
model it as assumed that the researcher knows only if y* > 0 (Long 1997, Greene 2003).
Greene (2003: 776) further points out that the results from the probit model can be
1
(βtobit ). It should be emphasized that this holds only if the tobit model
derived by σtobit
is correct. Therefore, it can be argued that using a tobit model is a more efficient way of
estimating the regression at hand.
There are two assumptions within the model that need to be checked - first that the
disturbance i is not heteroskedastic. If this condition is not satisfied the estimates are
likely to be inconsistent. Nevertheless, if the error are heteroskedastic this can be cor-
3.6. HOW DOES THE TOBIT MODEL WORK?
37
rected by modelling this directly. However, in the application at hand heteroskedasticity
is not a problematic. The tests for heteroskedasticity are presented in Appendix C. As it
can be seen from Figure C.1, there are relatively few reasons to doubt that the assumption
of heteroskedasticity is violated.
The second assumption cannot be checked so easily. It implies that the same data
generation process that determines the censoring is the one that determines the outcome
variable (Long 1997). Kennedy (2011: 264) argues that this assumption in some cases
is problematic. In this example we must be willing to assume that there are the same
mechanisms that determine the overall agreement level in the European Parliament as
those that determine absolute agreement.
The Tobit model, or censored regression model uses the information provided by
censoring and provides consistent estimates of the parameters (Long 1997: 189). I have
used the AER and VGAM packages in R in order to estimate the Tobit models presented in
the next chapters. The results of the models are identical when using the two libraries.
For robustness checks I have estimated fixed and random effects linear models. These
models are presented and discussed in Chapter 4 and 5, respectively.
3.6
How does the Tobit model work?
Using the typology from Wooldridge (2002: 517-520) the applications of censored regression models can be divided into two categories. First, when the dependent variable y* is
observed for the entire sample but is censored below or above a value and secondly when
we are interested in the features of the distribution of y (Wooldridge 2002: 517-520). The
latter approach is closer to the application in this thesis. Nevertheless, such uses of tobit
models have been criticized, for example by Sigelman and Zeng (1999).
Before proceeding to estimating a tobit model, let us look at the structural characteristics of the model. The equation of the tobit model is presented in Equation 3.6
(Wooldridge 2002: 517-520). Where y* is the latent variable that is observed for values
that are larger than τ and otherwise censored. The error term has a normal distribution
given by i ∼ N (0, σ 2 ).
yi∗ = Xi β + i
(3.6)
In order to reach the log-likelihood function, the standard likelihood function for censored normal distributions is used, where τ is the censoring point (Wooldridge 2002).
The log-likelihood for a standard tobit model is presented in Equation 3.7 (Wooldridge
2002). The log-likelihood function of the tobit model is composed of two parts. The first
part corresponds to a standard regression which does not account for censored observations (Wooldridge 2002). The latter part deals with the relative probabilities that an
observation is censored.
38
CHAPTER 3. DATA AND RESEARCH DESIGN
ln(L) =
N X
di
− ln(σ) + ln(φ)
i=1
y i − Xi β
σ
Xi β
1 − di ln 1 − Φ
σ
(3.7)
Given the structure of the model we get three types of coefficients, nevertheless there
seems to be little agreement on what is best to report (Greene 2003: 764, Wooldridge
2002: 520). First, we have the expected value of the latent variable (y*), secondly of y|y
> = 0 and lastly the expected value of y. According to Greene (2003: 764) if the data
is always censored there is little point in reporting the coefficients for the latent variable.
In the case of the analysis at hand the data is not always censored, in fact there are few
right-censored observations, namely between 1 and 5 depending on the specification of
the dependent variable. Because there are relatively few censored observations and the
effect on the latent variable is what is of interest, the mean effect of y* is what will be
reported. This is in accordance with Greene (2003: 764).
Wooldridge (2002: 572) recommends to report the coefficients and their respective
standard error. This is reported in the tables in Chapter 4. When it comes to the
interpretation of the model, it should be interpreted as if there was no censoring in the
data (Wooldridge 2002: 572). The reason for this is that the “population model is a
linear conditional mean” Wooldridge (2002: 572). The conditional mean assumption
implies that information about the expected value of the disturbances is not contained
by x (Greene 2003: 14).
Given the formal structure of the tobit model one can also extract three different
types of marginal effects. Just as in the case of expected values we retrieve the same
three types of marginal effects. First we have the marginal effects on y*, thereafter those
for values of y for uncensored observations and those for both censored and uncensored
observations Wooldridge (2002). Yet, are these effects of interest? Given the application
at hand marginal interests are not of substantial interest. There are few data points that
are censored, and there are also limited theoretical reasons for which these would be of
interest.
3.7
Linearity?
There are several very good reasons why a tobit model, or other linear specifications
are optimal given the data and variables at hand. Nevertheless such estimations make
a very stark assumption, namely that the relation between the dependent variable and
the independent ones is linear. Given the similarities between the dependent variables
I choose to present the tests only for the Agreement Index (AI). There are only minor
changes in the results when we test for the other specifications of the dependent variable.
3.7. LINEARITY?
39
Standardized Residuals Plot
●
●
●
●
●
●
2
●
●
●●
●
●●
●
●
●●
●
●
●
●
●
●●
0
●
●●
●
●
●
●
●
●●
●
●
●
●●
●●●
●
●●●
●
●● ●
●
●
●
●●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
−2
Standardized Residuals
●
●
●
●
●
●
●
●
●
● ●● ●
●
●
●●
●
●
●
● ●
●● ●
●●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
● ●
●
●
●
●●
●●
●
●●
●
●
●●
●● ●
●
●
●●●
●●● ●
●●
●●
● ●●
●
●●
●
●
●●
●
●
●
●●
●
●●
●
●
●●
●●
●●
●
●● ●
●
●●●
●
●●
●●
●
●
●●
●
●
●
●
●
●
● ●● ●
●●
●
● ●
●
●●
●●
●
●
●●●●
●
●
●
●
●●●
●● ● ●
●●
●
●●
●
●
●
●
●
●
●●
●●●
●●
● ●●●●
●
●●
●●
●
●
●
●
●
●●●
●●
●●
●● ●●
●
●
●
●
●
●
●
●●●
●
●
●
●
●●●
●●
●●
●●
●
●
●●
●
●
●
●●
●
●
●● ●
● ●
●●●●
●
●●●
●
●●●●●
●●●●
● ●
●
●●
●●
●●
●
●●●
●
●
●
●●●
●
●●
●
●●
●
●
●
●●
−4
●
●●
●
●
●
●●
●
●●
●
●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●●
●
●
●
●●
●
●●
●
●
●●
●●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●●
●
●
●
●●
●
●●
●●
●
●
●●
●
●
●
●●
●
●●
●●
●●
●●
●
●●
●●
●
●
●●●
●
●
●●
●
●
●
●
●
●
●
−6
−5
−4
−3
−2
−1
0
Standardized Predicted Values
Figure 3.1: Standardized residuals vs. predicted values for the AI tobit model
First, in order to check for linearity a model needs to be estimated. The model tested
in this section is the one presented in Section 4.5 and specifically in Table 4.5. There are
several ways of assessing linearity in a model, I will start by first looking at the residuals
versus the predicted values. This is presented in Figure 3.1.
The linearity assumption is supported in this case as long as amount of observations
below and above the 0 line are relatively equally distributed. In this case it can be argued
that they are relatively equally distributed, but the relationship is not perfect. Values
that are located close to the 0 line are relatively well predicted. In this case the main
concentration of points is around 0 for high values on the dependent variable. It has
to be noted that there are some points which are under-predicted and some that are
over-predicted. The further away an observation from 0 in a negative direction, it is
more over predicted. Observations that have positive standardized residuals are more
over predicted. Influential observations will be discussed in Chapter 5 of this thesis.
Generally, the model is more likely to under-predict legislative unity.
Another assumption of linearity is the homogeneity of variance. This can also be
assessed from Figure 3.1, by evaluating whether the vertical scatter of the points is similar,
or ideally the same across all values of the standardized predicted values. In this case
there seem to be only some problems of homogeneity of variance. Figure 3.1 shows that
the model is better at predicting higher levels of legislative unity. Yet, they are not so
grave that a linear approach should be dropped.
Furthermore, we need to assess whether the residuals are normally distributed, a histogram is presented in Figure 3.2a. The interpretation of this figure is straightforward, it
the better the histogram matches the normal distribution, the more normally distributed
the residuals. Again, in this case perfection is not reached, nevertheless, it can be argued
40
CHAPTER 3. DATA AND RESEARCH DESIGN
PP Plot
0.6
0.4
0.0
0.0
0.2
0.1
0.2
Density
0.3
Expected Probability
0.8
0.4
1.0
0.5
Histogram of standardizedResiduals
−4
−2
0
2
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.0
0.2
0.4
standardizedResiduals
(a) Distribution of residuals
0.6
0.8
1.0
Observed Probability
(b) PP Plot
Figure 3.2: Test of normal distribution for residuals
that the distribution at least resembles a normal distribution.
Another way of evaluating this is by looking at the PP Plot, which is also presented
in Figure 3.2b. Overall, it can be argued that the data is relatively well-behaved, yet the
relation is not perfectly linear. As highlighted by Greene (2003: 11), linearity doesn’t
necessarily denote the relationship between the variables, but the way in which parameters
and their disturbances enter the equation. In most cases, this assumption is not restrictive
(Greene 2003: 14). On a final note, the assumptions regarding the data generation process
for the regressors, namely “x may be fixed or random, but it is generated by a mechanism
that is unrelated to ” (Greene 2003: 17) is relatively unproblematic in this application.
Albeit some minor concerns, the assumptions for linear regression models hold relatively well, even if the relationship between the dependent variables and the independent
ones is not perfectly linear.
3.8
Summary
In order to test the theoretical implications I have collected data on all final votes taken in
the European Parliament in the period 14.09.2009 - 14.03.2013. The data was collected
from publically available sources and represents a first attempt to code the budgetary
implications of bills voted upon in the European Parliament in the above mentioned
period. The datasets encompasses information about the type of bill, and its budgetary
implication, how legislators voted and the rapporteurs and their respective parties for
each case. A full overview of the variables is presented in Table 3.1.
A bill is considered to have a budgetary implication when it enactment changes the
existing budgetary status quo, as discussed in Section 3.2.1. Legislative unity is opreta-
3.8. SUMMARY
41
tionalized by employing the standard RICE index Rice (1924) and the Agreement index
as developed by Hix, Noury and Roland (2005). In order to address the caveats of these
approaches I have constructed two modified version. The robustness of the results over
these various operationalizations should serve as an indication that the relationship is
not only dependent on the operationalization of legislative cohesion. Party positions are
used a proxy for ideology. Sections 3.5 and 3.6 address the choice of model. As the
dependent variable is right-censored the optimal model is a tobit. The assumption of linearity is tested in Section 3.7, despite some minor caveats, the assumption is not overall
problematic.
Chapter 4
Results
In this chapter, I shall first present the descriptive statistics for the variables included in
the model. Thereafter, a short discussion of simple bivariate correlations of substantial
interest is undertaken. The results from a standard OLS model are presented in order to
evaluate the results in the presence of controls. Given the structural form and the fact
that it does not account for censoring the OLS models have, the substantial interpretation
will be succinct. Hypothesis 1 is tested in Section 4.4. The main body of this chapter
encompasses the results of the Tobit regression models, their substantial implications,
in other words the empirical tests of Hypothesis 2 and 3. A closer discussion of the
interaction between budgetary implications and defection on behalf of the party of the
rapporteur is presented in Sections 4.6 and 4.6.1. Lastly, a discussion of the observations
with most leverage on the results is presented in Section 4.7.
4.1
Descriptive statistics
The descriptive statistics for the variables to be included in the main statistical models
are presented in Table 4.1. The dependent variables which capture legislative unity, as
discussed in Section 3.3, are continuous. The Agreement Index ranges from 0 to 1. Where
0 denotes the no legislative unity, while 1 denotes perfect legislative unity. The value 0
is attained when an equal number of legislators vote yes, no and abstain, while 1 when
all the legislators vote either yes, no or abstain. The RICE index reaches its minimum
value when half of the legislators vote yes and the other half vote no, and its maximum
when all the legislators vote either yes or no.
The independent variables in this models are all categorical. Given the nature of the
concepts these variables aim to capture, it is only natural to operationalize them as such.
For example, a bill either has, or does not have budgetary implications. Yet, there is one
exception. The number of terms served by the lead rapporteur, which is a continuous
variable.
42
4.2. BIVARIATE CORRELATIONS
43
Table 4.1: Descriptive statistics
AI
AIn
RICE
RICEabs
Budgetary implication
Legislation
No. Terms of rapporteur
Defection
EPP dummy
Mean
0.76
0.67
0.68
0.61
0.18
0.38
1.82
0.08
0.95
SD
0.19
0.18
0.42
0.41
0.38
0.48
0.83
0.27
0.23
Min
0.12
0.03
-0.97
-0.97
0.00
0.00
1.00
0.00
0.00
Max
1.00
0.95
1.00
1.00
1.00
1.00
5.00
1.00
1.00
Missing
0.00
0.00
0.00
0.00
2.00
0.00
253.00
0.00
0.00
We see that none of the dependent variables reach their minimum value, but for instance AI, and RICE reach their maximum value, highlighting the issues of right-censoring
discussed in Section 3.5. At the same time, all the specifications of the dependent variables reach values close the maximum points. This indicates that a tobit model is an
appropriate way to deal with this type of a dependent variable, as it is right-censored. In
the next section, I will look at bivariate correlations and thereafter, despite the censoring issues I will estimate OLS models and evaluate how the relationship changes in the
presence of controls.
4.2
Bivariate Correlations
Bivariate correlations highlight the patterns present in the data, therefore I will start by
assessing the simple correlations between the co-variates and the dependent variables.
There are two reasons for presenting such an analysis, firstly for the intrinsic interests
of the patterns present in the data and secondly as a check for the regression models
estimated. It is problematic to base a large amount of substantial interpretation on
bivariate correlations, other than a check for the presence of correlations and expected
the regression results.
Table 4.2: Bivariate correlations for the variables of substantial interest
AI
AIn
RICE
RICEabs
Budget
Legislative
No. Terms
Defection
EPP
AI
1.000
0.878
0.657
0.712
0.003
0.204
0.062
-0.422
0.409
AIn
0.878
1.000
0.696
0.734
-0.004
0.218
0.086
-0.365
0.379
RICE
0.657
0.696
1.000
0.982
0.007
0.120
0.057
-0.300
0.270
RICEabs
0.712
0.734
0.982
1.000
-0.005
0.130
0.044
-0.309
0.277
Budget
0.003
-0.004
0.007
-0.005
1.000
0.189
-0.034
-0.013
0.030
Legislative
0.204
0.218
0.120
0.130
0.189
1.000
0.115
-0.078
0.077
No. Terms
0.062
0.086
0.057
0.044
-0.034
0.115
1.000
-0.006
0.018
Defection
-0.422
-0.365
-0.300
-0.309
-0.013
-0.078
-0.006
1.000
-0.275
EPP
0.409
0.379
0.270
0.277
0.030
0.077
0.018
-0.275
1.000
The bivariate correlations are presented in Table 4.2. They give one a measure of
the relationship between two variables and ranges from -1 to 1 (Hellevik 2006: 236-242).
44
CHAPTER 4. RESULTS
The closer the correlation to ±1, the stronger the relation. It can be seen that all the
independent variables have a degree of correlation with the dependent ones. It seems that
budgetary implications are weakly correlated with the Agreement Index and its modified version which includes those that did not vote. Furthermore, the direction of the
relationship changes. There is a small negative correlation between AIn, and RICEabs
and budgetary implications. Between AI, and RICE and budgetary implications the correlation is positive. As discussed in Chapter 2, the impact of budgetary implication is
dependent on the existent level of polarization within the European Parliament. However,
this does not imply that the relationship is non-existent or not significant. The weakness
of bivariate correlation is that they do not control for the presence of other factors that
might in turn affect the relationship. In order to determine the true nature of the relationship, or whether it is a mere artifact of statistical modelling, further tests are needed.
That there is a certain level of correlation between the independent variables means that
they all have to be included in the model in order to avoid bias in the estimates. As it
can be seen from Table 4.2 multicollinearity is not a problem in this case.
At the same time the bivariate correlations help us asses the relationship between the
various specifications of the dependent variables. There is a 87.8% correlation between
the Agreement Index as proposed by Hix, Noury and Roland (2005) and the modified
version, AIn. The high correlation implies that the two indices measure almost the same
thing. That they are not perfectly correlated implies that the concepts captured are
slightly different. This explanation is analogous for RICE and RICEabs. The bivariate
correlation between the RICE index and the modified version of RICE that accounts for
abstentions is even higher. A correlation of 0.982, implies that they almost capture the
same concept.
The relationship between the Agreement Index and the RICE index is presented in
Figure 4.1. It illustrates that the relationship is linear. Most of the data points are
clustered around the value 1, indicating a high level of agreement within the European
Parliament. The correlation between the AI and RICE is of only 0.657. This is expected
as the two measure slightly different theoretical concepts and deal with abstentions in
different ways. The Agreement Index deals with abstentions as a new category, equally
important to yes and no votes, whereas RICE completely ignores abstentions. Solely
based on the bivariate correlation one can expect to a certain extent that the effect of
budgetary implications is likely to differ across the different specifications of the dependent
variable. However, when the nature of the data is accounted for, the relationship stabilizes
as discussed in the following sections.
Let us take a closer look at scatter plot between the AI unity measure and the RICE
index, presented in Figure 4.1. As the relationship between the two indices is linear,
observations that attain high values on one, will most likely attain high on the other.
We notice that there is a lot of clustering around the 1 region, while there are fewer
1.0
4.2. BIVARIATE CORRELATIONS
45
● ● ●
●●
●●●
●●●
●
●
●
●
●
●
●
●
●●
●
● ●
●
●
●
●●
●●
●
●●●
●
●
●
●
●
●
●
●
●
● ●●●●
●
●
●●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●● ●●
●
●
●
●
●
● ●
●
●
●●
●
●
●
●
●
●
●
●
●
●●●
●
●
● ●
●
●●●
●
●
●
●
●
●
●●
●
●
●●
● ●● ●●● ● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●●
● ● ●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
● ●
●
●
●
●
●
●
●
●
●
● ●●
●
●
●●
●
●
●
●
● ● ●●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ● ●●
●
●
●
●
●
●
●
●●● ●● ●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●● ●●
●●
●●
●
●
●
●
●
●
●
●
●●
●
● ● ●
●
●
●
●
●●
●
●
●●
●
●
●●
●
●●●
●●
●
●
●
●
●
●
●
●
●●
● ●●
●
●
●
●
●
●●● ●
●
●
●●
●
●
●●
●
●
●
●
●
●
●●●●
●
●
●
●
●● ● ●
●●
●●
●
● ●
●
●●
●●
●
●●● ●
●●
●●●
●
●●
●●●●●
●●● ●●
● ●●
●● ●
●
●
●●
●
●
●
●
●● ●●● ●
●●
●
●
●●●
●
●●
●
●
● ●●● ●●
●●●●
●●
●●
●
● ● ●
●
●
●●
●
●
●●
●
●
● ● ●● ●● ●
●
●
●
●
●
●
●● ●
●
●
●
●●
● ●●● ●●●
●●
●●●
●●
●
●●
●●
● ●
●
●
●
●
●●
●● ●● ●● ●
●●
●●
●●
●
●
●
●
●
●●
● ●● ● ● ●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●●● ●
●
●
●●●●
●●
●
●
●
●
●
●●
●
●
●●
●●
●●
●●
●●● ●
●●
●
●●●●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●● ● ● ● ●●
● ●
●●●
●●
●●
●
● ●●
●
●●
●
● ● ●●
●
●●
●
●●
●●
●
●●
●●
●●
●
●
●●
●●
●
●
●
●●
●●
●
●●
●
●
●●
● ● ●●
●
●●●
●●
●●●
●●●
●●
●
●●●●
●
●
● ●●●●
●
●●●
● ●●
●
●
●
● ● ● ●●●
●
●
● ●●●
●●●●● ●● ●●
●●● ●
● ●
●
●●●
●●
●
●●●●
●●
●●
● ●● ●
●
●● ●
●●●
●● ● ●
●
●
●
●
●
●●
●●● ● ●
●
● ●
●
●●
●
●
● ● ●●●●●●
●
●
●
● ●●● ● ●
●
●
●
●
●
●
● ●●
● ●
●
●● ●●
●
●●
●
●●●●
●
●
●
● ●●●
●●●
● ●
●●
●
●
●
●
●
●
●
● ● ●
●
●
●
● ●
●
●
●
●
●● ●
●●
● ● ●
●
●●
●
●
●
●
●●
●
●
● ● ●
●
●
●●
● ●● ●
●●
●
●● ● ● ●
●
●
●
●
●● ●
●
●
● ●
●
●
●●
● ●●
●● ●
● ●●
●
●
● ●● ● ●
●● ●
●
●
●
●●●
● ●●
●
●
●
●
●●
● ●●
● ● ●●
● ●●
●
● ●● ●
● ● ●● ●
● ●●
●
0.6
0.4
●
0.2
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.0
abs(data$RICE)
0.8
●
0.2
0.4
0.6
0.8
1.0
data$AI
●
●
●●
●
●● ●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●●
●
●
●
●
●
●●
●
●●●
●
●
●
●
●●
●●
●
●●
●
●●●
●
●●
●
Figure 4.1: Scatter plot between AI and RICE index
observations that attain minimum values. This supports the theoretical consensus that
the European Parliament is a united legislative body. It seems that also in the current
European Parliament, legislators tend to agree and vote in a similar fashion.
Furthermore, there are several observations that tend to receive lower scores on AI
than compared to RICE. One explanation for this is that the Agreement index deals with
abstentions, while the RICE index ignores them. Thus, cases with a large number of
abstentions will receive a different score on AI on RICE. One example of this is in the
case of “Exclusion of certain countries from trade preferences”, which is a draft legislative
resolution voted upon by co-decision. In this case, namely A7-0207/2012, 322 legislators
votes for, 78 against, while 218 abstained. The AI score is of 0.28 which is considerably
lower than the RICE score of 0.42.
Arguably, AI and RICE are likely to over-report legislative cohesion while their respective modified versions under-report it. In order to determine whether the results are
dependent on the specification of the dependent variable I will estimate all the regressions
with these various specifications of legislative unity. Obtaining similar results across these
various model specifications can be seen as convincing evidence that the results are not
dependent on the measure of cohesion used. At the same time, using diverse specifications of the dependent variable is a good way of dealing with the issue of under or over
reporting legislative cohesion by setting different constraints on certain categories.
There is one problematic aspect that remains, namely dealing with those that were
46
CHAPTER 4. RESULTS
present but did not vote. There are no theoretical reasons to assume anything about this
group. Not voting can either be a random event. For example, legislator X had a meeting
to attend at the time of the vote. At the same, not voting can be a strategic decision.
One example of this could be that legislator Y is against a proposal, but does not want
to defect, and hence decides not to vote. As discussed in Section 3.3 I choose to include
those that were present, but did not vote in AIn.
There are few, if any accurate ways of empirically testing whether the reason for
not voting is strategic or simple random. Making stalk assumptions about what not
voting means would only increase the distance between the theoretical and the empirical
models. One such assumption is made, namely I have assumed that those that did not
vote are equally important as abstentions. This is incorporated by the AIn. As there
are no theoretical reasons to exclude those that did not vote, I have incorporated this
this constraint on the AI. AIn can serve as a test of how much the results are driven
by those that did not vote. Despite this change AIn remains relatively highly correlated
with RICE (0.87), with RICEabs (0.73), and AI (0.87).
RICE by budgetary implication
0.5
0.0
RICE
0.6
−1.0
0.2
−0.5
0.4
Agreement Index
0.8
1.0
1.0
AI by budgetary implication
0
1
Budgetary Implication
0
1
Budgetary Implication
Figure 4.2: Party unity by budgetary implication. The green lines illustrate individual
observations, while the purple area shows the distribution.
An alternative way of displaying bivariate correlation between two variables of substantial interest is by looking at bean plots. Bean plots are a version of box plots, which
unlike box plots reveal the true form of the density, therefore making them more informative. The polygon represents the shape of the density, while the short horizontal lines
represent each data point (Kampstra et al. 2008). The data points that are duplicates
are represented by longer thin horizontal lines (Kampstra et al. 2008). The long thicker
horizontal lines represent the mean of each distribution (Kampstra et al. 2008).
For the Agreement Index the distribution of no budgetary implication is slightly dif-
4.3. OLS REGRESSION RESULTS
47
ferent than that of budgetary implications, therefore it can be argued that there is a
relation between the two. When it comes to other dependent variables we see a similar
pattern, the means of the two distributions vary. Without controlling for other factors it
seems that budgetary implications increase legislative cohesion in both the case of AI and
RICE, while they decrease it for AIn and RICEabs. One possible reason is that bivariate
correlations fail to account for the initial level of polarization. Although offering some
evidence for the proposition that budgetary implications affect legislative unity, any firm
conclusions require moving beyond these bivariate plots. Furthermore, the relationship
is expected to change in the presence of controls. In the next section, I will proceed to
present the results from the OLS regression model and discuss briefly how the relationship
of interests changes in the presence of controls.
4.3
OLS Regression Results
In order to facilitate interpretation I first estimate standard OLS-regression models disregarding the censoring issues of the dependent variable. In terms of magnitude and
implicitly significance the results are likely to be biased, as discussed in section 3.5. It
is still interesting to examine the results even if we fail to account for the nature of the
data. As can be seen from Table 4.3, the OLS results provide some preliminary support
for the hypothesized relationship.
Firstly, when no interaction is assumed between the defection by the party of the
rapporteur and budgetary implication we see that budgetary implications have an overall
negative effect on legislative cohesion, yet the coefficient is not significant. In other words,
in the absence of controls for the initial level of polarization budgetary implications does
not affect legislative unity. Despite this caveat, the interesting theoretical relationship is
for cases where there is a higher level of polarization. As discussed in Section 2.3, one
way of studying this more in depth is to set up an interaction between defection by the
party of the rapporteur and budgetary implications. This is presented in Table 4.3 in
models (2) - (5).
Once this interaction is in place, the coefficient for budgetary implications remains
negative, but gains significance. This implies that in the case of OLS, the effect of
budgetary implication is not strong enough to have an impact on its own on legislative
cohesion, indicating perhaps that there is not enough variation in the data. Nonetheless,
when the interaction is removed for tobit model specifications budgetary implications
retain their significance, as presented in Appendix B, in Table B.1. In this case the
models have been estimated on all types of votes, we see that legislative bills increase
the overall legislative unity in the plenary. Therefore, it can be argued that legislators
value legislative proposals higher than resolutions, as they have a tendency to be more
cohesive on such votes. At the same time, parties are more likely to exert party pressure
48
CHAPTER 4. RESULTS
Table 4.3: OLS Models
Dependent variable:
AI
AIn
absRICE
(1)
(2)
(3)
(4)
(5)
Budget
−0.018
(0.011)
−0.034∗∗∗
(0.012)
−0.035∗∗∗
(0.012)
−0.027∗
(0.014)
−0.036∗∗
(0.015)
Result
0.150∗∗∗
(0.022)
0.139∗∗∗
(0.022)
0.211∗∗∗
(0.022)
0.150∗∗∗
(0.027)
0.052∗
(0.028)
Legislative
0.050∗∗∗
(0.009)
0.048∗∗∗
(0.009)
0.054∗∗∗
(0.009)
0.056∗∗∗
(0.012)
0.067∗∗∗
(0.012)
No. Terms
0.007
(0.006)
0.007
(0.005)
0.011∗∗
(0.005)
0.013∗∗
(0.007)
0.006
(0.007)
Defection
−0.165∗∗∗
(0.016)
−0.202∗∗∗
(0.018)
−0.157∗∗∗
(0.018)
−0.270∗∗∗
(0.022)
−0.274∗∗∗
(0.023)
0.268∗∗∗
(0.023)
0.265∗∗∗
(0.023)
0.240∗∗∗
(0.023)
0.347∗∗∗
(0.029)
0.338∗∗∗
(0.030)
0.177∗∗∗
(0.038)
0.139∗∗∗
(0.038)
0.261∗∗∗
(0.046)
0.229∗∗∗
(0.049)
0.361∗∗∗
(0.032)
0.381∗∗∗
(0.032)
0.221∗∗∗
(0.032)
0.284∗∗∗
(0.039)
0.326∗∗∗
(0.041)
944
0.343
0.338
0.140(df = 937)
944
0.358
0.353
0.138(df = 936)
944
0.343
0.338
0.139(df = 936)
944
0.377
0.373
0.171(df = 936)
944
0.330
0.325
0.180(df = 936)
EPP dummy
Budget:Defection
Constant
Observations
R2
Adjusted R2
Residual Std. Error
Note:
abs(RICEabs)
∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
on the legislators on these types of votes. Legislative proposals can alter the existing
status-quo, which compared to resolutions, which in many cases can be conceptualized
as policy statements. The number of terms served by the lead rapporteur has a positive
impact on legislative unity only in models (3) and (4).
It is worth noticing that despite the different specifications of the dependent variable
the results do not change much in terms of significance and direction. This implies that
regardless of the way in which we deal with abstentions, or even when we include the
legislators that did not vote, the results remain robust. There is one exception – the
coefficient which controls for the number of terms the lead rapporteur has served loses
significance in the AI model. The number of terms served by the lead rapporteur is has
a small positive impact on overall legislative cohesion in models (3) - (5). Overall, the
coefficients for seniority, even if positive, remain weak in the OLS models. The evidence
about the impact of seniority in the literature have been split. Yet, given the size of these
coefficients and the possible bias in OLS estimation I will explore their impact further in
the Section 4.5.
In general, defections by the party of the rapporteur are rare, with a total of 97
defections in this period. Nevertheless, when we use the Agreement Index and its modified
version as a dependent variable, the coefficient is significant and has a negative impact
4.4. COHESION, NOT REJECTION
49
on the overall cohesion, when there are no budgetary implications. The relation does not
change much for RICE and its modified version. In the cases where there is no defection
by the party of the rapporteur budgetary implications will have a negative impact on the
overall cohesion. For example in the case of AI, legislative unity will decrease with 3.4%
and for RICE with 2.8%.
The interaction term between budgetary implication and the defection by the party
of rapporteur is positive. This implies that defection by the party of rapporteur does
not decrease cohesiveness for cases with budgetary implication. On the other hand, the
effect of budgetary implications has a negative impact on cohesion for cases where there
is no such defection. As presented in Section 2.3, there are several substantial ways
of understanding this relation, yet given the problems of model estimation a check of
robustness might be needed before proceeding to substantial conclusions on the empirical
tests.
When it comes to the control variables, the result of the vote and the presence of
the largest party in the winning coalition we see that both have a positive impact on
legislative cohesiveness. As discussed in Section 3.5, the results from the OLS models
are to be interpreted with caution as they from a statistical standpoint per definition are
biased. In the next section, I will present simple logit models with result as the dependent
variable in order to test the first hypothesis. Thereafter, I shall proceed to a discussion
of the tobit models, which account for the censoring of the dependent variable. Towards
the end of the chapter, after a discussion of the substantial results, and a discussion of
influential observation is undertaken in order to check whether the results obtained are
only driven by a small number of cases.
4.4
Cohesion, not rejection
The first hypothesized relationship is that budgetary implications do not affect the failure
or passage of a bill. In order to test this proposition I have estimated a simple logit with
the outcome of the vote as the dependent variable and the remainder of the co-variates
as independent. I have included, in successive models, the AI, AIn, RICE and RICEabs
as controls for cohesion. The results are presented in Table 4.4. Removing the controls
for cohesion does not alter the results substantially.
The outcome of the vote is affected by other factors, and not by budgetary implications. The results show that the defection by the party of the rapporteur reduces the
changes for a passed bill, while the presence the European People’s Party in the winning
coalition increases the chances of passage for a bill. However, EPP is the largest party,
and represents around ∼ 36 % of the legislative. Arguably, there are other factors that
might predict the outcome of the vote, yet they are outside the theoretical realm of this
thesis.
50
CHAPTER 4. RESULTS
Table 4.4: Logit models with outcome of the vote as the dependent variable
Dependent variable:
Outcome of the vote
(1)
(2)
(3)
(4)
Budget
0.111
(0.414)
0.128
(0.414)
−0.022
(0.420)
−0.316
(0.435)
AI
−0.412
(0.879)
AIn
0.602
(1.101)
−2.408∗∗∗
(0.564)
RICE
−4.602∗∗∗
(0.684)
RICEabs
Legislative
0.186
(0.339)
0.120
(0.341)
0.406
(0.347)
0.682∗
(0.365)
No. Terms
0.066
(0.189)
0.051
(0.189)
0.145
(0.192)
0.164
(0.198)
Defection
−1.313∗∗∗
(0.379)
−1.265∗∗∗
(0.378)
−1.356∗∗∗
(0.388)
−1.309∗∗∗
(0.408)
EPP Dummy
0.952∗
(0.503)
0.883∗
(0.501)
1.131∗∗
(0.510)
0.771
(0.526)
Constant
2.332∗∗∗
(0.663)
2.101∗∗∗
(0.641)
2.861∗∗∗
(0.637)
3.834∗∗∗
(0.695)
Observations
Log likelihood
Akaike Inf. Crit.
944
−167.575
349.149
944
−167.531
349.061
944
−158.284
330.567
944
−143.132
300.264
Note:
∗
p<0.1;
∗∗
p<0.05;
∗∗∗
p<0.01
Claiming that budgetary implications have an effect on the failure or passage of a bill
would be theoretically inconsistent. In order to better illustrate this, let us focus on a
hypothetical example. We have a set of legislators A, B, C and D, which all have equal
voting rights. Assume that A, B and C desire a costly legislation, while D does not. In
order for A, B and C to be rational to desire such a legislation they must be willing to
internalize its costs, despite opposition from D. If they would not have been willing to do
so, it would not have been rational for them to desire the set legislation in the first place.
The example above illustrates another aspect, that as long as there is a high enough
number of legislators that desire a costly bill, they are more likely to focus on insuring
that necessary majority is in place before reaching the final stage of the negotiations and
4.5. TOBIT REGRESSION RESULTS
51
tolerate strong disagreement from a minority. This section offers supportive evidence of
Hypothesis 1. In other words, as discussed in Section 2.3 budgetary implications do not
impact the passage or failure of a bill.
This section has illustrated one important substantial difference. There is no statistically significant relationship between budgetary implications and the failure or passage
of a bill. In other words, budgetary implications have no effect on the passage or failure
of a bill. This is supportive evidence of Hypothesis 1. Still, at the moment nothing can
be said with certainty yet about the effect of budgetary implications on legislative unity.
This is addressed in the next sections.
4.5
Tobit Regression Results
The failure to account for the censoring of the dependent variable leads to biased estimates, as discussed in Section 3.5. In order to correct for such bias, I estimate several
tobit models, which differ in how the dependent variable is treated. As mentioned, a tobit
model, unlike an OLS model, accounts for the right-censoring on the dependent variable.
It is estimated via maximum likelihood. The potential problem with the OLS models
estimated is that they are likely to underestimate the intercept while at the same time to
overestimate the slope (Long 1997: 189). Given the small magnitude of the coefficients
a wrongful estimation of the model might lead one to misleading conclusions about the
significance and importance of the covariates, and thus of their substantial implications.
Furthermore, using an OLS in these circumstances per definition creates omitted variable
bias as discussed in Section 3.5. Therefore, Tobit models are estimated as an empirical
test of the hypotheses presented in Section 2.3.
The models are estimated with four specifications of the dependent variable, legislative
agreement. In short, the models show that budgetary implications in the absence of other
polarizing factors, and defection on behalf of the party of the rapporteur in the absence of
budgetary implications decrease legislative unity. When there is both defection and the
bill has a budgetary implication legislative unity is increased. Furthermore, legislative
bills – and in Models (2) and (3) also seniority – have positive effects on legislative unity.
Using legislative agreement aggregated to the entire legislative has one down side, namely
that it aggregates individual level data to the level of the entire legislature. Many nuances
are lost in this aggregation process, making us unable to differentiate between defections
from the European Party group, or from national parties.
Nevertheless, given the that budgetary implications have not been directly discussed in
the literature on the European Parliament, I am more interested in defections in general,
not necessarily their source. Differentiation between individual behaviour requires very
detailed individual level data, which unfortunately is not available. Nevertheless, there
is a way to circumvent this issue. No assumptions are made upon the effect of party
52
CHAPTER 4. RESULTS
group pressure on MEPs. Recent evidence also shows that national parties are not likely
to defect from the European counterpart (Nordkvelle 2012: 34). In only 3.76% of the
cases the national party line defected from the European Level one (Nordkvelle 2012: 34).
Therefore, not being able to differentiate between MEPs that defected from the European
party line to support their national party, or vice-versa is not problematic.
The models are estimated on all the final votes, irrespective of type of vote taken
in the European Parliament in the current session. The reason for this is that there
are no theoretical reasons to expect that the effects will be stronger or absent on some
types, while they will be present on other types of votes. There has been a debate
in the field in how the type of procedure affects the behaviour of legislators and the
Parliament’s power relative to the other institutions within the European Union (Crombez
1996; 1997; 2000, Tsebelis 1994, Tsebelis and Garrett 2000). This approach has offered
important insights in how the European Parliament functions. It has explained the
relative power of EPGs and has aided research aimed at understanding the dimensionality
of this institution. However, in the post-Lisbon setting co-decision II became the ordinary
legislative procedure. This implies that variations in the procedural aspects have become
so sparse that in the context of the current European Parliament such an approach has
become moribund.
The tobit regression results are presented in Table 4.5. Firstly, I estimate a model
with the Agreement Index as the dependent variable, thereafter with AIn, its modified
version. In order to check whether the results are dependent on the measure of legislative
unity employed the tobit regression is also estimated with RICE and RICEabs. In order
to capture the theoretical argument that the effect of budgetary implications is dependent
on the initial level of polarization, I set up an interaction between budgetary implications
and defection by the party of the rapporteur. The models further control for the result of
the vote, whether the bill was a legislation, or a resolution, the number of terms served by
the lead rapporteur, and whether EPP was in the winning coalition. Below, I will proceed
to the interpretation of the results. Overall, given the significance and magnitude of the
scaling parameter, σ, tobit model specifications are a significant improvement from OLS,
as they account for the censoring on the dependent variable.
The Agreement Index takes values between 0 and 1, where 0 represents the lowest
level of agreement possible. In other words, when for example 100 vote yes, 100 vote
no and 100 abstain AI will be 0. Let us focus on the results when we disregard those
that were present but did not vote. In this case the lowest score on AI is reached in the
case of “ Motions for resolutions - G-20 Summit in Pittsburgh”, voted upon on 08.10.2009
where 158 legislators voted yes, 318 no, while 164 abstained. The agreement score for
case, B7-0086/2009 is 0.125. It might be surprising that the highest level of disagreement
is reached on a resolution. One way of understanding this is the that the conclusions
of the G-20 summit in Pittsburgh led to substantial implications for the regulation of
4.5. TOBIT REGRESSION RESULTS
53
Table 4.5: Tobit models with different specifications of the dependent variable
Dependent variable:
AI
AIn
(1)
abs(RICE)
abs(RICEabs)
(3)
(4)
(2)
∗∗∗
Budget
−0.034
(0.012)
−0.035
(0.012)
−0.028
(0.014)
−0.036∗∗
(0.015)
Result
0.139∗∗∗
(0.022)
0.211∗∗∗
(0.022)
0.150∗∗∗
(0.027)
0.052∗
(0.028)
Legislative
0.048∗∗∗
(0.009)
0.054∗∗∗
(0.009)
0.057∗∗∗
(0.012)
0.067∗∗∗
(0.012)
No. Terms
0.007
(0.005)
0.011∗∗
(0.005)
0.014∗∗
(0.007)
0.006
(0.007)
Defection
−0.202∗∗∗
(0.018)
−0.157∗∗∗
(0.018)
−0.271∗∗∗
(0.022)
−0.274∗∗∗
(0.023)
EPP dummy
0.265∗∗∗
(0.023)
0.240∗∗∗
(0.023)
0.347∗∗∗
(0.029)
0.338∗∗∗
(0.030)
Budget:Defection
0.177∗∗∗
(0.038)
0.139∗∗∗
(0.038)
0.261∗∗∗
(0.046)
0.229∗∗∗
(0.049)
Constant
0.381∗∗∗
(0.032)
0.221∗∗∗
(0.032)
0.284∗∗∗
(0.039)
0.326∗∗∗
(0.041)
σ
−1.981∗∗∗
(0.023)
−1.975∗∗∗
(0.023)
−1.768∗∗∗
(0.023)
−1.717∗∗∗
(0.023)
Observations
Log likelihood
Wald Test (df = 7)
944
529.119
525.511∗∗∗
944
525.504
493.260∗∗∗
944
325.851
570.476∗∗∗
944
280.819
464.899∗∗∗
Note:
∗∗∗
∗
∗
p<0.1;
∗∗
p<0.05;
∗∗∗
p<0.01
financial markets in the European Union, while at the same time the summit increased
the external pressure on the EU as a whole to increase regulation on the financial markets
and hinder further losses because of the financial crisis. Additionally, given the timing
there was a lot of pressure on legislators from their national constituencies and perhaps
governments. The low level of agreement is easy to understand in this case.
The results from the tobit models show that legislators are more inclined to reach a
grand coalition in absolute terms 1 when they vote over legislation compared to resolutions. As a large part of the literature has argued, it seems that legislators place more
importance on legislative proposals than on resolutions. The costs are lower for legislators
to disagree on resolutions and set up working bodies on the task or area, than to fail to
agree on legislation. If legislation is failed in the first reading, an absolute majority is
1
and not in terms of how many parties vote
54
CHAPTER 4. RESULTS
required to pass it in the second reading (Hix and Høyland 2011), thus creating several
procedural effects. These effects produce an extra incentive for legislators to present a
cohesive behaviour in the first reading in the plenary.
One possible explanation is that legislative proposals bear greater importance compared to resolutions, which can be considered as mere policy statements. Legislative
proposals can in certain cases radically change the existing legislation in a country or at
the European level. Furthermore, given either the difficulty, at least in terms of procedural aspects, or the length of time necessary, European Party Groups, and maybe even
national ones may increase the constraints placed on legislators. Such arguments are supported by the tobit models presented in this section. The effect of legislative proposals is
small, yet significant at the 1% level and robust over the different operationalizations of
legislative unity.
The argument that legislators that have served more terms gain leeway and are able
to create a more cohesive behaviour in the European Parliament is only partly supported.
The evidence in the literature has been mixed so far, as they are from this model specification. The coefficients for seniority are small, and they are only significant at the 5%
level in the AIn and RICE. The number of terms can also be conceptualized as a proxy
for legislative leverage of certain MEPs. It might be that the position and reputation of
these legislators aid them when leading discussions within the European Parliament.
Scarrow (1997) shows that the European Parliament seems to attract members with
long-standing careers in the national politics. It can be argued that this implies that senior
legislators are able to forge a higher level of cohesion within their national delegations as
well, given their prestige and reputation. McElroy (2006: 14) finds that both experience
and expertise are important factors that determine the committee allocation of MEPs.
Her findings are contrasting to those of Bowler and Farrell (1995). The finding can be
extrapolated to legislative unity. Yet, caution is advised when arguing that rapporteurs
which served more than one term have a positive effect on legislative unity, as this finding
is supported by only two of the four models presented. This finding is further explored
in the Robustness chapter, where it is shown that its significance is dependent on model
specification.
The presence of EPP in the winning coalition also has a moderate positive effect on
legislative unity. The result is robust across the different specifications of the dependent
variable. Unlike seniority, the magnitude of the effect is relatively larger, varying between
0.240 in the AIn model and 0.347 in the RICE specification. The finding is significant at
the 1% level. In the 7th European Parliament the European People’s Party is the largest
group with 270 members. If all the members of the European People’s Party would vote
for a proposal, that would represented ∼ 36.7% of the votes. Conversely, if the European
Party does not support a bill, it is much harder to the remaining parties to force a grand
coalition, as parties on the Left side of the political spectrum, and Center would have
4.5. TOBIT REGRESSION RESULTS
55
to collaborate with those on the far Right. This positive effect is not surprising, yet at
the same time it highlights the importance of including a control for the largest party
within the European Parliament. Alongside, the interaction of substantial interest this is
perhaps one of the most robust finding. The effect of party positions is further explored
in Section 5.2.
Largely, the results remain robust regardless of the specification of the dependent
variable in terms of direction and statistical significance, with the exception of seniority.
This finding provides us with some indication that the results are not so dependent on
the measure of cohesiveness used. More specifically, regardless of the way we deal with
abstentions, and even when we include those that did not vote the results remained
relatively unchanged in terms of direction, relative magnitude, and significance.
The theoretical section of this thesis has mainly focused on the effect of budgetary
implications on legislative unity. There are only 210 bills with budgetary implications
that exceed the existing framework. In other words, in only ∼ 17% of the cases legislators voted upon bills with such implications. As we have seen in the previous section,
budgetary implications do not have an impact on failure or passage. Given the theoretical framework employed this finding is expected. However, as discussed the impact
of budgetary implications is dependent on the initial level of polarization. In order to
capture this an interaction term is set up between budgetary implications and defection
by the party of the rapporteur. This will enable one to see how the effect of budgetary
implications varies in the absence of other polarizing factors, compared to cases where
there is a known level of disagreement. Even if legislative unity is observed at the same
time as defection, defection is decided before the actual moment of the vote. Further
endogenity issues are discussed in section 5.1.
Using defection as a proxy for legislative polarization gives one a tool which enables
her to study this effect more in-depth. Both budgetary implications and defections seem
to be relatively rare events. When there is no budgetary implication, defections of behalf
of the party of the rapporteur have a negative effect on legislative unity. This effect is
robust regardless of the specification of legislative unity. This implies that the effect is
presented regardless of the way we deal with abstentions, or even when we include those
legislators who did not vote. This is supportive of Hypothesis 3.
When there is no defection on behalf of the party of the rapporteur, budgetary implications have a small negative effect on the level of cohesion within the European Parliament.
The effect is significant at the 1% level for AI and AIn, while at 5% for the modified version of RICE which accounts for abstentions, and only at the 10% level for RICE. The way
in which we deal with abstentions does not only affect the significance or RICE, but also
the magnitude of the coefficient. In the versions of the dependent variable that deal with
abstentions, the magnitude of the effect of budgetary implications ranges between -0.034
and -0.036. Nonetheless, when abstentions are ignored, namely in the RICE model, the
56
CHAPTER 4. RESULTS
effect is smaller, of only -0.028. This implies that legislators seem to prefer to abstain, on
in some cases not to vote, rather than vote against the European Party Group’s wishes.
In other words, European Party Groups seem to have a certain leverage over the
behaviour of legislators. This leeway is not so strong that it deters individual members
from expressing their own wishes, but they do it in such a way that it doesn’t defy the
party line, for example by abstaining on the respective vote. This is supportive evidence
of the theoretical arguments presented in Section 2.3. The presence of this tendency does
not imply that legislators are willing to alter their behaviour totally, or that they never
defect, but merely that in certain cases they prefer to be soft-liners and abstain, instead
of voting the opposite of their party. Hence, they aim at minimizing the potential costs
of defection, while at the same time aiming at maximizing their power in the legislature.
These findings are supportive of the hypothesized relationship in the theoretical chapter. The coefficient of the interaction term is supportive of Hypothesis 3. The interaction
term is robust and significant at the 1% level in all the four models presented in this section. This offers relatively convincing evidence that the coefficient is not only a matter
of statistical randomness.
As negotiations are shifted to the committee level, and rapporteurs have a rather
high leverage in the decision making process, the other legislators present at the case
negotiations will know when the rapporteur, and implicitly her party, is likely to defect.
Given that parties have accepted and constantly invest in appointing skillful rapporteurs
it is likely that they will follow the position of the rapporteur, or in some cases there might
simply be a split in this party. Knowing this, provides the other legislators, or even party
groups with a very strong incentive to find other alliance partners and to try to induce
a cohesive behaviour within the party at the time of the vote. Therefore, the observed
overall cohesion will be higher in cases where legislators already have an incentive to
disagree, namely cases with budgetary implications, and knowledge of a defection on
behalf of the party of the rapporteur. This interaction is further explored in the two next
sections.
In conclusion, the tobit models provide convincing evidence in favour of the hypothesized relationship. Firstly, the behaviour of individual members, is shaped, at least to
a certain degree by budgetary implications. Secondly, these findings suggest that there
may be an economic dimension of voting within the European Parliament. This highlights
that the conclusions reached by (Fenno 1978) for the US Congress are highly relevant for
the European Parliament and again show the similarities between the US Congress and
the European Parliament.
This is further explored in section 3.4. So far it has been shown that, almost regardless
of the way in which one deals with abstentions, and those that did not vote budgetary
implications have a certain impact on legislative unity. The theoretical predictions presented in Hypothesis 2 are supported, at least to a certain degree. Furthermore, by only
4.6. BUDGETARY IMPLICATIONS AND DEFECTIONS
57
looking the coefficients we see that budgetary implications seem to be more important
also for bills where there is relatively little initial polarization. In order to further discuss
whether the empirical results actually support Hypothesis 3, I will focus on the interaction
term between budgetary implication and outcome of the vote in the next section.
4.6
Budgetary implications and defections
0.2
Figure 4.3 presents the interaction between bills where the party of the rapporteur was
in the winning coalition versus those were it was not, with regards to their budgetary
implications for the AI model. If the dependent variable is changed the changes in
the plot are only minimal, given the small differences between the coefficients. The
points represent the estimate, while the red lines represent the confidence interval of the
estimate. The confidence interval for bills with defection and budgetary implications and
defection is relatively large, given that there are few bills that fit in the category. Even
so, the confidence intervals of the two terms do not overlap, supporting that the effect is
statistically significant.
0.0
−0.4
−0.2
●
−0.6
Effect of budgetary implications on AI
●
No Defection
Defection
Figure 4.3: Interaction plot based on Model (1) showing the effect of budgetary implications for cases with and without defection on behalf of the party of the rapporteur
58
CHAPTER 4. RESULTS
The importance of inherent characteristics of a bill has been largely neglected in
the literature. As presented in Chapter 2, most of the work has focused on the type of
procedure, ideology or in some cases the salience of the case as explanations for the voting
behaviour in the plenary. The economic traits of a bill have been largely ignored for the
European Parliament, and only recently studied for the Council. These results show that
the budgetary implications have a significant impact on to cohesiveness in the European
Parliament, even if their effect is small. This can be seen as a suggestion that there is an
economic determinant of voting in the current European Parliament.
Figure 4.3 allows one to see how the effect of budgetary implications differs in two
cases and to better link the empirical evidence to Hypothesis 2 and Hypothesis 3. In
normal cases, where there is no evidence of disagreement in the negotiation phase before
the vote, budgetary implications have a small negative effect on overall cohesion. Many
of the bills coded in this period had a positive implication the communitary budget,
therefore if the time series is extended to cover other sessions of the European Parliament
we would expect to see a stronger effect, especially if it is differentiated between bills with
a negative implication for the budget and those with a positive one.
Another way of looking at the mechanisms that lead to this effect is by focusing on
successful negotiations at the committee stage and party pressure. When there is no
defection on behalf of the party of the rapporteur, it is an indication that negotiations
at the committee stage were relatively successful and that the bill is likely to have an
acceptable level of support in the plenary. This proposition is in accordance with previous
findings for example those of Neuhold (2001) and Yordanova (2009). As argued in Section
2.3, European Party Group leaders will in those situations have fewer incentives to enforce
the party line in the plenary, therefore individual legislators, that still disagree with the
agreed proposal are more likely to defect, either by voting against the party line or
abstaining. In such cases it can argued that parties are less focused on legislative unity,
and hence less likely to apply sanctions to legislators who defect. One possible explanation
for this is that the leadership is to a large extent convinced of the passage of the bill, as
they also rely on their alliance partners as well.
The opposite happens in cases where we have both budgetary implications and defection on behalf of the party of the rapporteur. Given the nature of the negotiation process
before the plenary, European Party leadership will know that at least one party is likely
to defect from the agreed consensus, either to support or to vote against the bill. Therefore, being convinced, at least to a certain extent, they will increase party pressure on
legislators in those cases in order to insure the desired outcome in the final vote. Hence,
we observe a higher level of legislative unity. If legislators value at least equally the goals
of protecting the interests of their home constituency and those of insuring power in the
legislative, they are less likely to defect in cases where there is increased party pressure,
hence in cases where their behaviour might be sanctioned. In the next section I will
4.6. BUDGETARY IMPLICATIONS AND DEFECTIONS
59
provide several case examples to better illustrate the mechanisms at work.
As I have noticed from coding the dataset, budgetary implications are not always
detrimental to constituencies, sometimes they add increased revenue to the EU budget.
In most cases, however, these budgetary implications make losers and beneficiaries easier
to identify. Some of the bills that are covered in the dataset had a positive budgetary
implication for the communitary budget, yet their impact was different from member
state to member state depending on the existing regulation. Another reason for why this
relationship may be observed and why the impact of budgetary implications is negative,
yet weak on legislative unity. One potential explanation for this is that a large majority
of the bills actually increase the revenue of the European Union. The evidence so far
suggests that alongside the Left-Right ideological dimension that has been proved to
drive votes in the plenary (Hix 2004, Hix, Noury and Roland 2007, Hix and Noury 2009)
there are other mechanisms at work.
The apparently high level of cohesion in the European Parliament might be slightly
more superficial than previously assumed. Looking at cases with budgetary implications
might reveal new ways of moving beyond the surface and understanding the strategic
choices legislators make in the European Parliament. Nonetheless, the plot presented in
Figure 4.3 only presents the interaction term. Empirically it might be more interesting
to look at the interaction term while setting various levels on the other covariates. A plot
of simulated probabilities is presented and discussed in section 4.6.1.
4.6.1
Simulated Probabilities: Substantial difference?
It might be interesting to investigate whether the effects are strong enough to produce
substantially different results. This is done in the simulation plots. The simulated effects
for budgetary implications for bills where defection is present, and respectively for those
without, are presented in Figure 4.4 and in Figure 4.5. Unlike the interaction plot, the
simulated effects account for the effect of the other covariates present in the model. It
produces random samples from the specified multivariate normal distribution. The distribution is given by the beta coefficients and the co-variance matrix of the model. 1000
random draws are taken in this case. The simulation shows that there is a substantial
difference in cohesion also when other variables are taken into account and kept at median. At the same time, it allows one to grasp a more detailed picture of the interaction
between budget and defection, which allows one to make more detailed inferences about
the hypothesized relationship. The predicted cohesiveness does not change substantially
when we only look at legislation and keep the rest of the coefficients at their mean values,
of course with the exception of the interaction term.
For ease of interpretation, I choose to present the simulated probabilities as box
plots. The box is bordered by the 25th and 75th quartile, while the dotted lines rep-
60
CHAPTER 4. RESULTS
0.85
Effect for bills with defection
0.75
0.70
0.65
●
●
●
0.55
0.60
Predicted cohesiveness
0.80
●
●
●
●
●
●
●
No
Yes
Budgetary implication
Figure 4.4: Simulated effect of budgetary implication for bills with defection when the
other co-variates are kept at their median values, and EPP at 0
resent the “whiskers” and are bordered by the lowest and the highest values within the
1.5 * range(25th , 75th quartile). The small circles in the outskirts of the graph show the
extreme values, or outliers. Outliers in this care not necessarily problematic, as they can
be seen as an artifact of the algorithm used to randomly draw the distributions and hence
of little substantial interest. The overall interpretation of the figure is the standard one
of a box and whisker diagram.
In Figure 4.4 the effects of budgetary implications for bills where the party of the
legislator defected is presented. When there is a defection, we see that the impact on
predicted cohesiveness is positive. This is much smaller for bills without budgetary implications, compared to those which have such implications. This supports the theoretical
arguments presented in the theoretical section of this thesis and shows that the relationship holds even in the presence of other co-variates. The mere defection of the party of
the rapporteur, increases legislative unity, as the other parties get together to to insure
that their most desired outcome will actually happen in the plenary, while at the same
time they increase pressure on legislators. Therefore, European Party Groups are more
likely to use their available “sticks and carrots” in order to insure that their legislators
4.6. BUDGETARY IMPLICATIONS AND DEFECTIONS
61
will act in a cohesive manner in the plenary.
There is little doubt that cases where both budgetary implications are presented
and there is a defection on behalf of the party of the rapporteur are important and even
salient cases. In more general terms these cases usually have clearly identifiable losers and
beneficiaries. Several examples of this are “Security of gas supply”, “EU guarantee to the
EIB against losses under loans and guarantees for projects outside the EU”, “Temporary
suspension of autonomous Common Customs Tariff duties on imports of certain industrial
products into the Canary Islands”, or “Common system for taxing financial transactions”,
are only some of the cases that have both budgetary implications and where the party of
the rapporteur defected.
Simulated effects for bills without defection
0.86
●
●
●
0.84
0.82
●
●
●
0.78
0.80
Predicted cohesiveness
●
●
No
Yes
Budgetary implication
Figure 4.5: Simulated effect of budgetary implication for bills without defection, when
the other co-variates are kept at their median values and EPP at 1
This effect is even stronger when we have budgetary implications. There are some
reasons to believe that budgetary implications are likely to increase the existing level of
polarization within the European Parliament. Furthermore, the leadership of the parties
might be even more motivated to put pressure on individual legislators as they are aware
that national parties might also put pressure on their legislators.
62
CHAPTER 4. RESULTS
Figure 4.4 provides relatively good evidence of the theoretical mechanisms discussed
in the theoretical chapter. First, it shows that budgetary implications have an impact on
legislative cohesion, and this effect is stronger and positive in the presence of other factors
that on their own reduce the expected level of unity. Even if budgetary implications on
their own have a weak, and non-significant impact on legislative unity, this section has
shown that this effect is dependent on the initial level of polarization.
As noted previously, defections are relatively rare. Let us look at the effect of budgetary implications for normal cases. Given that the party of the rapporteur is in the
winning coalition and an overwhelming majority of the cases, it is likely that European
Party Groups have fewer incentives to apply party pressure on their legislators in these
cases. Therefore, Figure 4.5 allows us to see a more detailed picture of the effect of
budgetary implications when there are no defections.
Based solely on Figure 4.5 one can argue that budgetary implications decrease overall
legislative unity. There is a higher level of agreement in the European Parliament in the
absence of budgetary implications. Even if, legislative unity is positive in cases where
there are no defections on behalf of the party of the rapporteur, it still remains slightly
lower compared to cases where there are no defections and no budgetary implications.
Even if, there were not so many cases which had both budgetary implications and
where the party of the rapporteur defected, one notices that these cases are relatively
high-profile. Several examples where the confluence of these polarizing factors was exhibited are: “Trans-European energy infrastructure”, “Fishing opportunities and financial
contribution provided for in the EC-Denmark/Greenland Fisheries Partnership Agreement”, “Financing instrument for development cooperation - banana accompanying measures”, “EU guarantee to the EIB against losses under loans and guarantees for projects
outside the EU ”, or “Security of gas supply”.
Another highly interesting case is “Common system for taxing financial transactions”,
which has a positive impact on the EU budget - “Preliminary estimates indicate that,
depending on market reactions, the revenues of the tax could be 57 EUR billion on a
yearly basis in the whole EU.” (European Commission 2011a). The winning coalition was
formed by GUE/NGL, Greens, S&D and EPP, and in total encompassed 487 legislators
that voted for the proposal, while 152 voted against and 46 abstained. Given that the
nature of the case and the way it impacts the EU budget one would have perhaps expected
a higher level of cohesion. Alongside ideological differences, it can be argued that this
case highlights that legislators are also concerned with the interests of their constituencies.
This implies that legislators which are aware that higher taxes at the national level will
have a negative impact on chances for their re-election are likely to have voted against,
or abstained in this case. As in that all of the parties of the three rapporteurs were in the
winning coalition and knew that they had the vote secured, it is unlikely that the added
increased pressure on their legislators, hence lowering the costs of defection.
4.7. MODEL DIAGNOSTICS
63
In conclusion, legislative unity is affected by budgetary implications. In cases where
there are other polarizing factors and pressure on the legislators is increased, Figure 4.4
shows that the confluence of such factors with budgetary implications increases legislative
unity. On the other hand, in cases where there are no other polarizing factors and only the
budgetary status-quo is altered, Figure 4.5 shows that budgetary implications will have a
negative impact on legislative unity. This is in accordance to the theoretical propositions
presented in Section 2.3.
4.7
Model diagnostics
The first question that comes to mind is which model is better? Nevertheless, as substantially important effects remain significant over various specifications of the dependent
variable this question is not so relevant. An aspect that might be more relevant for model
fit is to look at the residual plots and investigate the cases which have leverage over the
results.
4.7.1
What drives the estimates?
Even if the substantial effects are stable across various specifications of the dependent
variable it is interesting to look at which cases drive the estimates for each model. The
main focus in this section will lie on the interaction term and its components. Given the
robustness of the results over different specifications of the dependent variable, I choose to
focus on the tobit model with the unaltered Agreement Index as the dependent variable.
The residual plots for this are presented in figure 4.6. Each point is labelled with the row
number of ease of identification. For the other models, the residual plots are presented
in Appendix A.
In order to test whether the results are only driven by a small number of cases I
remove the 5 of most influential observations for each term of the interaction on both
the negative and positive side. In total, I re-estimate the Tobit models without the 30
most influential observations. The results are presented in Table 4.6. As it can be seen
from the table the results remain largely unchanged in terms of direction and significance.
Consequently, it can be argued that the results are not only driven by a small number of
cases, even though these cases have the highest leverage over the results. In other words,
the 30 most influential observations on the interaction term are removed.
One of the cases which has the highest negative leverage on the coefficient for budget
is “Surveillance of budgetary positions and surveillance and coordination of economic
policies”2 , voted on 23.06.2011. In this case the vote was split, with 333 legislators
voting for, 303 voting against and only 26 abstentions. The winning coalition was formed
2
case number: A7-0178/2011
−0.002
Defection
−0.001
Budget
601
812
1125
1025 1180 206
519
522 652
1024
164218
655
1093
499
1199
11011198
575
854
550
405
599
1015
512
900973
757
416
365
98
1031
1072
1121
887
1106
1122
1107
1053
1030
1200
707
106
1042
16215
1105
955
1176
99
855
704
238 329391
739
228
1033
105
890
750
771
635
773836
909
18
441 532
1141
148
19
202
11054
843
392
217
749
84
149
1178
656
85899
1056
1055
762
615
384
981
581
9944
25 320
653
24
20
447
1061
399
557
491
473
171
985
578
760
580
475
527
478
358
572
178
610
176
688
904
602
1114
906
188
135
259
417
738
553
469
486
46
86
980
121
5
386
36
347
369
654
2
573
1021
79
462
569
1100
130
92
212
666
239
179
609
172
343
886
186
518
107
64
122
146
608
47
520
182
567
258
521
1190
127
586
1019
383
485
781
370
482
568
881
984
8
721
110
68
175
918
326
675
78
444
96
465
705
774
321
11
731
740
579
123
463
345
449
582
1096
177
333
844
919
442
273
229
108
61
44
798
198
434
905
971
934
921
813
814
464
373
197
613
33
1152
466
83
587
10
1064
694
307
134
869
450
785
998
23
548
506
662
634
1142
780
566
311
951
853
946
508
503
353
914
349
928
874
316
858
390
804
88
129
845
1133
663
636
948
646
289
662
241
400
366
443
660
661
857
528
297
312
437
299
1058
65
298
109
324
493
920
181
884
930
1078
325
1026
81
658
933
1035
124
507
817
282
697
514
818
304
496
271
481
1132
664
4
398
426
453
67
612
1147
734
203
166
150
410
457
903
155
268
167
784
902
436
950
1131
1130
1007
545
676
362
55
544
725
718
926
337
715
385
1013
611
116
143
338
537
607
368
290
351
563
1011
240
264
309
230
118
935
214
75
999
157
376
317
885
494
722
846
1012
265
619
989
990
1010
480
103
958
993
1005
452
883
28
431
319
204
559
74
782
413
1001
285
523
672
89
1059
776
800
1000
994
17
995
1003
979
1008
716
201
7
516
702
852
992
972
966
1006
161
1014
1092
807
827
1004
199
22
57
27
779
91
95
37
87
964
1079
517
761
882
1002
421
498
674
30
764
196
789
991
169
891
1009
830
838
209
515
38
533
786
374
538
1161
140
671
826
788
829
281
158
472
831
361
922
1120
163
371
162
429
435
76
996
173
534
670
828
48
94
961
536
53
1159
306
1163
200
723
790
1034
535
744
174
796
318
1166
651
879
315
293
360
97
396
300
274
314
93
585
1162
539
236
470
633
1164
823
117
346
792
793
978
270
310
982
649
308
945
211
1050
227
82
856
70
801
895
875
894
865
1165
440
305
40
618
700
808
26
868
833
968
867
642
643
159
588
119
956
160
589
339
791
701
954
136
1145
703
1139
907
835
1113
34
959
332
960
787
690
717
913
562
927
699
232
344
42
977
69
125
350
334
336
692
479
21
698
863
862
1023
860
650
225
983
967
783
614
335
677
423
1094
861
456
526
180
39
1187
291
408
641
1148
283
73
1197
851
29
451
459
778
460
775
590
102
678
139
1103
292
969
495
648
850
348
839
965
889
584
591
667
1073
687
1158
1196
949
987
748
439
488
929
915
720
901
1115
1070
288
501
631
397
322
571
976
295
822
695
706
1022
428
570
269
49
962
1195
916
1160
165
168
63
1194
1134
1140
598
665
997
137
424
43
561
287
752
372
747
859
141
546
795
1029
126
943
947
492
637
638
112
910
222
583
454
226
821
866
138
552
133
187
104
1065
1138
647
864
504
877
471
759
210
296
988
897
430
438
825
689
101
60
763
114
841
223
147
719
940
192
1091
957
313
221
474
932
224
418
794
1137
623
736
876
72
71
401
625
1174
152
1020
696
644
645
205
145
840
577
832
286
630
657
208
448
144
242
629
132
142
207
433
626
628
963
1169
627
549
249
170
389
737
266
267
1016
131
234
621
525
624
797
1135
404
758
815
524
190
691
1057
1066
427
547
500
673
1167
411
898
679
1136
216
367
1172
543
1168
668
1173
564
942
622
1175
1017
1181
640
819
189
58
195
502
754
1186
31352
235
834
156
153
708
476
824
735
1170
41
639
191
505
251
1171
917
574
284
820
878
184
1071
59
880
659
193
403
892
194
542
402
576
1129
551
1001067
1177
387
931
592
154
497
388445
709
742
1128
1124
1126
746
1127
477
669
255332
446
84890
458
363
375
419
1051
45
727
257
111
256
1063
594
1123
467
237
185
490
0.002
CHAPTER 4. RESULTS
0.001
64
593
595
596
1 113
2 3 4 5 6 7 8 9
0.00
705
1 113
2 3 4 5 6 7 8 9
Observation
1056
1055
1054
705
518
891968
987
196
32
239305
308
186
10181120 172
596
333 4451
677
44 520593
595
331
781 902
597
607
694
723
140
371
594
185
1127
490
817
892
856
256
676
1126
633
111
1124
1128
257
497
727
45
388
90
1079
300
458
848
1063
472
332
306
255
352
950
477
1051
988
363
746
82
446
184
669
1115
709
742
706
375
1123
969
592
931
387
100
237
754
1129
467
189
1067
668
53
673
880
691
131
266
267
433
963
1071
132
411
427
832
1020
142
794
313
932
1091
957
445
719
834
940
1186
296
71
759
1181
72
564
104
866
147
454
583
910
126
947
795
287
546
561
659
665
752
1160
112
428
570
962
269
295
49
695
322
571
639
397
501
901
187
720
133
479
552
1158
1196
640
775
584
591
889
965
850
292
838
657
102
460
678
778
590
1197
29
551
644
645
851
283
976
180
39
1177
194
402
542
576
1094
193
403
456
647
861
335
783
421
967
284
637
638
820
983
191
251
505
574
1103
735
1022
153
156
476
708
824
235
698
860
862
863
336
408
58
819
1175
125
57
216
367
543
547
69
170
232
524
525
59
699
73
927
3
448
878
913
917
1113
1145
1170
205
34
577
667
690
787
907
959
960
31
401
648
703
736
101
192
195
223
502
589
701
791
841
1017
1168
1173
119
154
160
210
430
504
588
622
689
825
942
956
1016
1057
1066
1135
1136
1167
159
190
234
500
621
624
641
679
758
797
867
898
977
1169
43
549
626
627
628
629
808
833
868
943
1134
1171
144
145
152
286
40
618
625
630
63
700
840
136
41
623
650
801
822
865
875
1023
1050
1065
221
224
70
897
114
226
821
1172
137
270
310
344
348
598
793
839
997
222
236
346
459
470
539
585
792
916
1070
1187
293
314
315
360
396
404
424
631
747
815
915
93
97
1073
1174
207
208
249
318
389
439
642
643
696
748
879
1137
1159
200
225
418
438
535
60
763
790
876
877
964
978
1029
1138
138
372
536
692
859
864
94
961
1140
1194
1195
162
168
173
423
429
435
534
562
649
670
76
828
996
141
163
288
334
339
361
488
48
687
831
929
949
1148
139
158
26
281
291
651
671
788
826
829
835
954
274
38
440
533
538
614
786
982
1009
1139
242
350
515
789
796
823
830
945
991
1002
1162
1164
1165
1166
174
30
471
474
674
761
1161
1163
209
374
37
517
764
779
882
87
922
95
1004
1006
1014
161
376
42
717
807
827
852
883
979
1008
201
21
385
431
436
516
563
702
725
885
926
992
1000
1003
1059
116
419
545
672
716
75
800
89
933
994
995
1001
1058
117
155
204
319
366
413
559
782
1005
103
1133
452
508
91
951
958
993
1010
1012
181
211
265
453
548
619
744
785
989
990
998
27
373
437
493
844
905
199
22
317
390
579
928
157
230
309
601
984
999
1011
1130
290
351
485
537
1013
1131
609
611
715
337
62
813
814
946
108
23
362
46
980
1007
134
450
738
869
325
740
83
110
1132
167
268
784
985
258
521
646
1142
166
410
449
903
203
731
935
36
612
67
774
398
426
486
86
884
920
553
664
188
304
481
818
105
514
64
697
178
466
507
587
658
904
464
478
739
349
475
329
1078
271
298
469
47
25
299
312
457
463
65
1096
297
506
124
212
443
528
660
661
857
289
400
465
662
416
636
663
129
845
1092
804
84
858
88
24
316
578
353
503
99
473
311
853
921
444
566
780
6
634
919
1035
343
1141
238
462
569
666
98
307
654
1064
1100
10
206
918
33
1152
613
197
881
934
61
798
971
229
273
1190
572
582
886
123
442
599
179
321
1114
177
96
106
326
675
175
573
68
854
11
653
8
568
482
164
370
656
383
586
874
135
447
567
608
146
512
1125
580
107
127
930
122
130
92
899
79
198
557
966
347
5
491
1061
28
386
20
1033
498
9
981
121
615
85
906
417
532
722
176
688
345
1178
1030
749
894
149
522
358
118
519
392
527
944
760
1200
843
19
202
148
1031
18
217
441
773
171
391
750
384
890
771
228
836
169
215
704
575
1176
855
955
1105
776
2
1198
707
1021
655
499
1053
1101
11042
1107
1199
369
218
1093
1106
1122
909
1072
285
973
721
757
17
1121
1024
900
365
264
78
610
240
652
1015
405
550
1025
1180
399
1019
581
812
887
718
734
895
74
972
480
523
227
846
494
150
762
109
526602
1147143
635
241 320
165
1034
914
7
16214
368 492
338
1026
81
496
324
544
948
737
55
259
495
282
−0.02
Budget * Defection
0.02
Observation
−0.006
−0.003
597
331
1018
259
948
544
7
324
143214 338
368 492
737 81
165
241
1147
399 496
526 610
109 150
1034
16
369 494
846914972
1021
419480 581635 74
2
1019
185
762 895
734
345 467 580
227
602 718
240
264320
237
887
874
198
375
776
1123
78
909
721
171
894
1 107 118
527
358
206 28329
169
519
522
384
760
164
5
79
854
1035
176
688
386
599
416
146
608
906
567
498
417
512
722
121
568
8836
347
1042
68
122
130
92
586
127
383
930
370
106
739
105
177
238
457
482
197
123
613
1152
582
1141
175
326
10
675
96
321
11
84
971
934
935
273
442
229
61
566
25
798
985
24
353
33
1092
1064
307
473
6
88
634
475
478
578
178
311
904
400
443
857
738
188
316
780
46
980
65
553
129
486
853
86
469
503
36
528
858
804
609
845
636
663
891
289
662
196
660
212
661
297
485
312
258
521
64
299
298
426
984
507
1078
47
697
67
612
110
405
481
579
658
1015
203
740
550
124
398
410
514
731
774
365
818
304
844
757
449
271
905
664
973
108
900
1106
373
1121
1072
1107
813
814
1053
490
903
1122
855
704
707
83
1176
167
784
785
1105
166
998
955
1007
134
1120
215
228
548
890
869
362
268
450
750
23
951
508
773
148
202
771
337
391
62
944
1013
18
611
843
149
928
946
1133
1142
19
390
723
392
217
366
749
140
557
1011
1178
441
532
615
1058
437
85
9
999
981
1051
1061
181
493
20
491
715
1063
899
933
199
230
363
537
135
290
351
446
669
75
309
317
1012
157
573
265
453
619
989
990
1010
179
103
1190
886
1005
155
993
452
607
881
436
918
472
559
91
545
725
1001
319
919
921
926
958
204
385
445
782
672
1000
116
563
413
506
659
994
995
1003
376
639
800
1008
349
89
1059
201
640
644
645
716
884
885
920
647
1132
637
638
657
883
992
431
516
702
1006
648
667
1022
1130
1131
161
641
650
979
1014
344
642
643
807
1023
649
651
852
1004
827
779
48
882
325
764
87
95
209
37
517
646
978
1096
1100
374
587
761
1161
136
27
464
466
674
22
30
463
465
977
1002
666
789
922
343
408
462
569
830
1103
515
654
796
966
991
1163
533
786
1114
572
1009
1166
1200
444
538
976
447
671
826
1162
653
788
823
829
1164
38
982
133
552
656
744
1198
187
831
945
1101
158
1093
117
1199
163
285
26
281
440
1024
1165
429
1033
670
173
534
162
211
828
1025
1180
1181
1186
17
339
361
94
961
1030
536
564
954
834
562
835
964
1031
1071
200
499
790
1139
435
535
575
76
996
1067
692
717
1129
174
523
880
334
42
350
360
97
100
396
318
93
1159
225
236
470
539
655
21
315
218
293
346
792
812
793
270
592
931
314
387
423
1187
274
614
652
879
459
310
70
291
1148
348
585
839
865
40
618
1050
139
709
1073
742
833
868
875
746
748
439
867
915
687
949
159
700
1070
160
488
631
822
929
956
57
288
791
701
801
63
916
1134
495
43
703
588
1195
959
119
787
960
137
598
690
943
997
168
424
589
1194
747
808
1140
232
699
255
332
141
372
859
222
226
504
1029
210
477
821
101
430
689
825
907
1065
1113
125
34
69
841
223
138
192
897
1138
114
736
458
698
863
864
877
401
862
438
205
221
224
60
763
848
860
623
471
577
625
152
448
90
1145
45
1137
145
286
418
630
840
170
876
1174
336
144
525
3
629
696
1169
474
524
626
627
628
549
913
547
927
1016
216
234
367
621
624
797
1135
208
257
543
758
190
1057
1066
207
1175
249
335
500
1167
389
898
1136
679
727
819
242
55
1168
58
1173
404
456
815
861
111
622
942
1017
1172
256
180
235
156
39
153
195
983
502
708
476
594
824
31
967
735
783
191
29
505
251
574
775
1170
284
820
590
102
41
917
878
1171
193
403
59
584
591
194
542
851
402
576
678
1094
720
292
850
551
889
965
283
571
1197
1177
269
49
778
460
421
901
838
397
752
126
962
1158
1196
910
112
583
454
154
104
597
665
759
287
501
546
940
947
147
322
1091
957
295
695
428
570
72
282
71
1160
593
595
832
296
132
963
331
142
561
719
266
267
131
795
53
313
932
794
691
673
427
596
411
1020
1018
866
668
189
433
754
73
969
99
479
706
182
98
82
1115
184
988
434
950
352
306 388
1125
601
633
1126
1124
892
1128
1127
300
1079
817
856
694
676
371
1054
1055
1056
902
497
781
333 4 44 520
677
186
172
239308
32 451
305
987
518
968
1026
182
434
1 113
2 3 4 5 6 7 8 9
Observation
Figure 4.6: Most influential observations for interaction term
only of ALDE and EPP. The lead rapporteur, Corien Wortmann-Kool has served two
terms and the bill had an implication for the overall EU budget. In this case the party
of the rapporteur did not defect, yet budgetary implications alone caused polarization.
Given the theoretical framework employed, it is not problematic that this case has a high
leverage over the estimates.
Granted the nature of the case there is little doubt that national interests another
polarizing issue in this case, especially as the regulation aims “to monitor and coordinate
Member States’ budgetary policies, by way of a preventive measure to ensure budgetary
discipline within the European Union” (Europa.eu 2012). Another dimension of conflict
in this case was the amount of power that was to be delegated to the EU as institution
(euractiv.com 2013). This case highlights that even if defection by the party of the
rapporteur does not capture all the dimensions of conflict within the European Union,
such as national or pro - anti- integration defection remains a stable and relatively reliable
proxy for the initial level of polarization in the EP before final votes.
The example brings one close to a major weakness of this thesis, its inability to accurately capture national polarization on cases. This implies that the intergovernmentalism
4.7. MODEL DIAGNOSTICS
65
Table 4.6: Tobit models without influential observations
Dependent variable:
AI
AIn
(1)
abs(RICE)
abs(RICEabs)
(3)
(4)
(2)
Budget
−0.029
(0.011)
−0.033
(0.011)
−0.023
(0.013)
−0.036∗∗
(0.014)
Result
0.163∗∗∗
(0.021)
0.232∗∗∗
(0.021)
0.155∗∗∗
(0.024)
0.030
(0.029)
Legislative
0.061∗∗∗
(0.009)
0.065∗∗∗
(0.009)
0.075∗∗∗
(0.010)
0.079∗∗∗
(0.011)
No. Terms
0.004
(0.005)
0.009∗
(0.005)
0.011∗
(0.006)
0.001
(0.007)
Defection
−0.180∗∗∗
(0.017)
−0.149∗∗∗
(0.018)
−0.261∗∗∗
(0.020)
−0.280∗∗∗
(0.021)
EPP dummy
0.289∗∗∗
(0.022)
0.266∗∗∗
(0.023)
0.408∗∗∗
(0.026)
0.350∗∗∗
(0.028)
budget:defection
0.158∗∗∗
(0.038)
0.156∗∗∗
(0.040)
0.240∗∗∗
(0.044)
0.227∗∗∗
(0.045)
Constant
0.332∗∗∗
(0.030)
0.174∗∗∗
(0.030)
0.221∗∗∗
(0.035)
0.344∗∗∗
(0.039)
σ
−2.072∗∗∗
(0.023)
−2.050∗∗∗
(0.023)
−1.917∗∗∗
(0.023)
−1.813∗∗∗
(0.023)
Observations
Log likelihood
Wald Test (df = 7)
919
598.596
643.357∗∗∗
917
579.201
600.534∗∗∗
918
453.003
823.404∗∗∗
920
361.760
570.399∗∗∗
Note:
∗∗∗
∗∗∗
∗
∗
p<0.1;
∗∗
p<0.05;
∗∗∗
p<0.01
critique on the ideology-approach that national interests drive the behaviour of legislators cannot be dismissed or incorporated more than already presented in the theoretical
section. On the positive side the thesis shows that there are other dimensions of polarization, or even conflict within the European Parliament except the legislators own ideology.
Hence, it shows the importance of economic polarization.
The cases that have a positive leverage over the estimates for budgetary implications
are mainly resolutions that aim at changing the existing budgetary framework in the
future, for example: “Draft amending budget No 3/2012: surplus from the 2011 financial
year”; “Draft amending budget No 3/2011: 2010 budget surplus”, or “Draft amending
budget No 8/2010: Section III - Commission - European Solidarity Fund: floods in Ireland
- completion of ESF - Objective 1 (2000 to 2006)”. Given the nature of these two cases,
with the exception of the latter they were accepted by a grand-majority and generally
66
CHAPTER 4. RESULTS
there was little polarization. One potential reason for this is simply that these cases are
pure formalities, even potentially alter the budget they encompass mainly technicalities.
The latter case is a good example of a highly uncontroversial issue, given the norms and
values upheld by the European Union. Yet again there are few reasons for which the
leverage of these cases is problematic.
If one is to contrast this the theoretical framework it becomes rather expected that
such cases have leverage. Furthermore, the cases which have positive leverage over the
estimates for budgetary implications also have a positive implication for the overall EU
budget. Legislators have few incentives not to support such cases, or to polarize in these
circumstances.
Table 4.6 presents the tobit models without the 30 most influential observations on the
interaction term. The results remain relatively robust. The greatest loss in significance
is in the RICE model for the coefficient for budgetary implications. This implies that
the RICE model is more dependent on those high-leverage cases compared to the other
models. One of the most influential cases for the RICE model is the “Implementation
of excessive deficit procedure”3 , one of the few bills decided upon by consultation. As
expected given the high leverage of this proposal upon member states the vote has been
split with 339 legislators voting for, 304 against and 26 abstentions. The small number of
abstentions can be seen as results of high party pressure. Again, in this case the majority
was formed by two parties, EPP and ALDE. Given the theoretical framework employed
in this thesis, it is expected that such bills will have a negative leverage in the models.
It should not be problematic that these observations drive the estimates, as they are
in line with the theoretical arguments. The persistence of the statistical relations between
the variables, and the results remain supportive of the theoretical framework. Nevertheless, these cases are only a handful, therefore they may not be representative. Even if the
tobit models are robust, it can be argued that a better robustness check would be either
to include other co-variates that might affect the studied relationship. Alternatively, I
will also check whether the results are robust enough to tolerate different model specifications, namely multi-level fixed and random effects. Such tests are presented in Chapter
5.
Even when removing some of the observations with most leverage the results of these
Tobit models provide empirical support for the hypothesized relation. In other words,
budgetary implications have a negative impact on legislative cohesion, in the absence of
other polarizing factors. Yet in cases where the party of the rapporteur also defects the
effect of budgetary implications is positive on legislative unity.
3
case number: A7-0179/2011
4.8. SUMMARY
4.8
67
Summary
Simple bivariate correlations show that there is a weak relationship between budgetary
implications and legislative unity. The OLS model allows one to better grasp the patterns
in the data, nonetheless the estimates are potentially bias, consequently tobit models are
estimated. Hypothesis 1 is supported by the models presented in Section 4.4. As predicted
in the theoretical section, budgetary implications do not affect the failure or passage of
a bill. Alike the OLS models, the tobit specifications show that budgetary implications
affect legislative cohesion. Hypothesis 2 and 3 are supported. The relationship between
budgetary implications and the voting behaviour of legislators in final votes is further
explored in Sections 4.6 and 4.6.1.
Figure 4.3 shows that budgetary implications decrease legislative unity in the absence
of splits in the party of the rapporteur, or in other words in the absence of other polarizing
factors. This relationship remains robust even when the effect of the other co-variates
is taken into account as show in Figure 4.5. This is supportive evidence of Hypothesis
2. Hypothesis 3 is also supported, the coefficient for the interaction between budgetary
implications and defections on behalf of the party of the rapporteur is significant and
positive in all the models estimated in this Chapter. The relationship is illustrated in
Figure 4.4.
The major caveat of the results is the weakness of the coefficients, warranting some
concern that the models might be driven by a few cases only. However, in Section 4.7
I have removed 30 of the observations that had most leverage on the result. Even if
the coefficients became weaker they retained their direction and to a large extent their
significance level. In the next section, I will explore whether the results are a mere artifact
of model specification, by employing hierarchical models and adding more controls.
Chapter 5
Robustness
In the previous section I have presented the results from the empirical testing of the
theoretical arguments proposed. I have argued that the results are robust over different
specifications of the dependent variable. The results were robust, yet in order to insure
that the findings are not driven by the type of model or by the set of covariates employed,
I will test several other model specifications and include some new control variables. I
will include controls for ideology, based on the findings from Hix and Noury (2009), and
for time as recommended by Cox and McCubbins (2007). Before proceeding to the above
mentioned, a discussion of the caveats of the empirical specification is undertook.
5.1
Shortcomings of the empirical approach
A tobit model seems to be the optimal model given the specification of the dependent
variables. As discussed in Section 3.5 the tobit model is able to eliminate a great deal
of bias linked to model estimation. Nevertheless, it does not steer clear of all potential
sources of bias. In this section, I am aim to highlight several other potential sources of
bias, and thereafter propose several remedies. However, Clarke (2005) shows that this is
not so problematic as previously assumed.
Omitted variable bias is perhaps the biggest concern given the specification employed
in the Results chapter. Therefore, there are considerable reasons to believe that the
estimates presented suffer from a great deal of bias. For example, Johnston and DiNardo
(1972: 110) argue that the inclusion of irrelevant variables in the model is much less
problematic, when compared to excluding variables that might be correlated with the
co-variates of interest. The same dictum is presented by King, Keohane and Verba (1994:
173).
Arguably, there are two such variables that have been ignored so far, firstly party
positions, or ideology, and secondly policy areas. In the following sections I present
models with controls for party positions and thereafter proceed to an estimation of multilevel models, both fixed and random effects in order to account for both omitted variable
68
5.1. SHORTCOMINGS OF THE EMPIRICAL APPROACH
69
bias as an excluded co-variate and in order to account for the nature of the data.
Is this enough? Ideally, I would include more control variables. Nonetheless, in this
case, even if it is desired, it is practically impossible. Mainly, because of the nature of
the dataset, the time-series is recent and there are few, if any available datasets that
are compatible. The models employed in this thesis are all linear, therefore limiting the
sources of omitted variable bias to other variables that are correlated with the variables
included in the regression (Clarke 2005: 348).
Clarke (2005) shows that unless the researcher know the true specification of the
model “[t]he addition may increase or decrease the bias, and we cannot know for sure
which is the case in any particular situation” (Clarke 2005: 342). There is relatively
limited research on budgetary implications of legislation within the European Union, and
even less within the European Parliament. It can be concluded that omitted variable
bias remains a problem for this thesis. As presented in the next two sections the main
findings remain robust even when dealing with the two most likely sources of bias.
A second potential problem of the general model specification is endogenity. Ideally, it
would be desired that all the independent variables occur and are observed, in time, and
prior to the dependent one. This aspect is mainly problematic for the outcome of the vote
which is observed at the same time as the level of legislative unity. When this variable
is removed from the model specification there are only minor changes in the coefficients.
Alternatively, even if the defection of the party of the rapporteur is known a priori, it is
observed also at the same time as legislative unity.
Confluence of events makes causal inferences more problematic. From a chronological
perspective, defection on behalf of the party of the rapporteur occurs before the observed
legislative unity in final votes. According to Cox (1992: 293) this temporal ordering serves
as a way of showing, or at least arguing for the existence of a causal effect. Causality is
only secondary here. Even if it can be argued that from a temporal perspective defection
is observed prior to cohesion, it might be better to attempt to remedy this problem.
Arguably, a more suitable way to deal with this endogeniety problem is by employing a system estimation by instrumental variables (Wooldridge 2002: 183-205). Simply
put, in order to employ such a model specification one is to find an instrument that predicts the endogenous independent variable, but does not predict the dependent variable
(Greene 2003, Sovey and Green 2011). The purpose is to isolate exogenous variation in
defections (Sovey and Green 2011). According to Greene (2003) one way of doing this
is by employing a two-stage least squares regression. In the first step, the instrument,
alongside the other co-variates of interests is used to estimate the predicted values for the
endogenous independent variable, in this case defection (Greene 2003). Secondly, these
predicted values are used to regress the dependent variable (Greene 2003).
Nonetheless, finding a good and strong enough instrument is not an easy task. Unfortunately, there is little theoretical guidance in choosing an instrument in this case.
70
CHAPTER 5. ROBUSTNESS
Arguably, one potential instrument is the country of the rapporteur. It can be argued
that rapporteurs which come from influential member states have higher leverage within
the European Parliament and enjoy a stronger backing within the Council, and hence
are more likely to lead successful committee level negotiations. Yet, the correlation has
proven to be too low for this instrument to be valid. I have also employed the GDP
level within the national country of the lead rapporteur, Yet again the conclusions was
the same, the instrument was too weak. As these instruments were too weak, I have
abandoned this strategy. Consequently, even if endogenity is not critical in this example,
it still remains slightly problematic.
5.2
What else is there? Controlling for ideology and
time.
Given the theoretical focus on the structural characteristics of bills, especially on their
economic implications, the models so far neglected ideology. This could be seen as a
potential weakness, especially as one of the arguments from the literature is that the
effect of budgetary implications is often hidden by ideology (Peltzman 1984). If budgetary
implications are actually significant, we should expect this relationship to hold even when
we control for ideology, or in other words for party positions. Ideally, I would have
employed a similar method to Hix, Noury and Roland (2006) to capture ideology, by
using the nominate scores. Nevertheless, corresponding data to this time series was not
available at the time of the writing, as discussed in Section 3.4. The proxy for ideology
employed in this analysis is party positions on each case, whether a party was or was not
in the winning coalition on each vote.
The usage of this proxy for ideology allows us to capture which parties supported
each bill, and how this impacts legislative unity. Using European party group dummies
requires us to make a secondary assumption, namely that MEPs join the party that
best represents their ideological standpoint. Given the recent evidence (Hix and Høyland
2013, McElroy and Benoit 2010) there seems to be a certain level of empirical support for
this assumption. Therefore, there are few reasons that make us consider this assumption
problematic. Most of the empirical and theoretical evidence suggests that the main
dimension of voting in the European Union is the Left-Right one. This proxy allow one
to isolate whether there are parties that are neto-defectors.
There are numerous empirical accounts showing the European Parliament operates in
a low dimensional policy space, where the most important dimension is the left - right one.
The evidence shows that the importance of pro and anti-integration issues has decreased
throughout time (Hix and Høyland 2013). Therefore, it can be argued that a proxy for
ideology which captures this, is a good substitute for nominate scores. More specifically
5.2. WHAT ELSE IS THERE? CONTROLLING FOR IDEOLOGY AND TIME.
71
Table 5.1: Tobit models with controls for ideology and number of sessions
Dependent variable:
AI
AIn
abs(RICE)
(1)
(2)
(3)
(4)
Budget
−0.010∗
(0.006)
−0.015∗
(0.008)
−0.005
(0.008)
−0.004
(0.008)
Result
0.059∗∗∗
(0.011)
0.139∗∗∗
(0.016)
0.043∗∗∗
(0.016)
−0.056∗∗∗
(0.014)
Legislative
0.005
(0.005)
0.022∗∗∗
(0.007)
0.018∗∗
(0.007)
0.012∗
(0.006)
No. Terms
−0.0003
(0.003)
0.005
(0.004)
0.003
(0.004)
−0.003
(0.004)
Defection
−0.023∗∗
(0.010)
−0.007
(0.013)
−0.054∗∗∗
(0.013)
−0.037∗∗∗
(0.012)
−0.00003∗∗∗
(0.00001)
−0.00002∗∗
(0.00001)
−0.00003∗∗∗
(0.00001)
−0.00004∗∗∗
(0.00001)
EPP dummy
0.337∗∗∗
(0.012)
0.298∗∗∗
(0.017)
0.424∗∗∗
(0.017)
0.433∗∗∗
(0.015)
S&D dummy
0.322∗∗∗
(0.012)
0.285∗∗∗
(0.016)
0.432∗∗∗
(0.016)
0.423∗∗∗
(0.014)
ALDE dummy
0.141∗∗∗
(0.012)
0.120∗∗∗
(0.017)
0.197∗∗∗
(0.017)
0.198∗∗∗
(0.015)
Greens dummy
0.134∗∗∗
(0.007)
0.112∗∗∗
(0.010)
0.154∗∗∗
(0.010)
0.176∗∗∗
(0.008)
GUE/NGL dummy
0.056∗∗∗
(0.005)
0.037∗∗∗
(0.007)
0.035∗∗∗
(0.007)
0.074∗∗∗
(0.007)
ECR dummy
0.102∗∗∗
(0.005)
0.079∗∗∗
(0.007)
0.097∗∗∗
(0.007)
0.129∗∗∗
(0.006)
EFD dummy
0.058∗∗∗
(0.005)
0.054∗∗∗
(0.007)
0.075∗∗∗
(0.007)
0.079∗∗∗
(0.006)
budget:defection
0.078∗∗∗
(0.019)
0.055∗∗
(0.027)
0.142∗∗∗
(0.027)
0.098∗∗∗
(0.024)
Constant
−0.259∗∗∗
(0.023)
−0.325∗∗∗
(0.032)
−0.504∗∗∗
(0.032)
−0.528∗∗∗
(0.028)
σ
−2.671∗∗∗
(0.023)
−2.334∗∗∗
(0.023)
−2.333∗∗∗
(0.023)
−2.448∗∗∗
(0.023)
944
1, 179.445
4, 888.157∗∗∗
944
863.677
1, 998.342∗∗∗
944
857.460
3, 729.560∗∗∗
944
969.669
5, 123.591∗∗∗
Time
Observations
Log likelihood
Wald Test (df = 14)
Note:
abs(RICEabs)
∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
in the case at hand the interest lies mainly on the impact of controlling for party positions
with regard to issues.
A problem with this proxy is that it fails to capture the individual ideological position.
Therefore, within the realm of this thesis it cannot be claimed that legislators value
economic concerns more than their ideological position, or vice-versa. Even if nominate
scores would have been calculated, such inferences would remain problematic.
72
CHAPTER 5. ROBUSTNESS
There is another important aspect that has been largely ignored both in the theoretical
discussion and in the empirical approach, namely the time perspective. Some studies of
the European Parliament, and mainly research on the US Congress (Cox and McCubbins
2007) highlight that legislators tend to act in a more heterogeneous way in the beginning
of the parliamentary mandate and that their behaviour gets more and more cohesive as
sessions go by. In order to capture this aspect, I have designed a variable that simply
counts every session that went by. If there are 50 cases discussed in the same session
they will have the same number, while the next session when legislators meet will be
incremented by 1. This variable is continuous. Taking the natural logarithm of this
variable does not substantially affect the results presented in Table 5.1.
Tobit models specifications with the above mentioned control variables are included
Table 5.1. Perhaps the most contrasting result in this table is the very small, yet significant negative coefficient for time. In the case of the 7th European Parliament it seems
that cohesion decreases slightly or, marginally as time goes by. The effect is so weak in
this case. Overall, the introduction of controls for party potions and time decreases the
magnitude and in some cases also the significance of the results. For example, seniority
is not significant in any of the models. Legislative bills retain their positive effect on
legislative unity only in models (2) - (4). Controlling for ideology and the number of
sessions that passed seems to reduce the impact of legislative bills. Yet, for the models
AIn and RICE it can be argued that legislative bills have a small, but positive effect on
legislative unity in the European Parliament.
The coefficients for budgetary implications become smaller and lose significance in
the RICE and RICEabs models. The effect remains negative for bills where the party of
the rapporteur does not defect. The effects of defection by the party of the rapporteur
remain unchanged and are significant in all the models with the exception of the AIn. The
interaction term, is significant and positive in all the models. Although, the magnitude
of the effects of the interaction get weaker the substantive interaction presented in the
previous Chapter is largely unaltered. In short budgetary implications reduce legislative
cohesion. Yet, when the party of the rapporteur defect on a case with a budgetary
implication legislative unity is likely to increase.
The coefficients for the parties are positive and significant in all the models presented
in Table 5.1. The largest effects are for the two largest parties within the European Parliament in the current sitting: EPP and S&D. The coefficients for all the party dummies
are significant at the 1% level.
Given the small changes in several of the co-variates might also argue that the models
presented in Section 4.5, might have been affected by omitted variable bias. The substantial interpretation of the relationship of interest, the interaction between budgetary
implications and the proxy for polarization has not changed to a large extent. Highlighting the robustness of the hypothesized relationship presented in Section 2.3. Adding
5.3. ACCOUNTING FOR POLICY AREAS
73
controls for party positions shows another substantial difference, specifically the importance of abstentions. Yet, the models became more sensitive to the way in which we deal
with abstentions. One can infer that legislators are soft-liners and prefer to abstain, or
not to vote, rather than directly defect from the party line. This is supportive evidence
of the hypotheses presented in the theoretical Chapter of this thesis. Overall the relationship of substantial interest remains robust even in the presence of more controls, for
the Agreement Index and its modified version AIn.
There are many other controls that could have been included in this section, nevertheless, given that the time series employed in the analysis is relatively recent there are
few corresponding datasets publicly available. These circumstances make the inclusion of
new control variables impossible. Nevertheless, in order to investigate whether the relationship remains robust I will estimate multi-level models control for policy areas. This
is presented in the next section. By doing this I am also able to show how and whether
policy areas affect the relationship of substantial interest. Hix and Høyland (2013) show
the cohesion varies across policy areas.
5.3
Accounting for policy areas
Differences between policy areas have so far only received limited attention. This may
be problematic. Previous research found that there are different level of cohesion along
various policy areas (Hix and Høyland 2013). Furthermore, it has been shown that there
are also differences in the coalitions that are formed across policy areas (Hix and Høyland
2013). Not accounting for differences between policy areas may be seen as a source of
omitted variable bias, as it means excluding unobserved differences between policy areas
from the analysis. Furthermore, ignoring the hierarchical character is a problem in itself.
Ignoring the hierarchical character of the data may introduce bias in the standard
errors, and thus increase the chances for Type I errors (Steenbergen and Jones 2002:
219). At the same time accounting for the hierarchical character of the data will allow
one to test whether the results are robust enough to support both a different estimation
and a new variable.
There are different ways of accounting for the multi-level structure of the data. One
approach is to assume fixed effects, the other is random effects (Gelman and Hill 2007:
Chapter 14). Both of these approaches are pursued in the following. I first estimate a set
of fixed effects models. I then move on to estimate a set of models where I include both
a random intercept and a random effect of budget. The reported results show that the
relationship of interests remains significant.
The data at hand is both highly clustered and at the same time it has a clear hierarchical structure, as each case can be seen as nested within a policy area and the data
at hand has both subsets, fitting nicely in the definition developed by Steenbergen and
74
CHAPTER 5. ROBUSTNESS
Table 5.2: Fixed Effects Multi-level linear models
Dependent variable:
AI
AIn
abs(RICE)
(1)
(2)
(3)
(4)
Budget
−0.036∗∗∗
(0.012)
−0.043∗∗∗
(0.012)
−0.028∗
(0.015)
−0.036∗∗
(0.016)
Result
0.140∗∗∗
(0.022)
0.209∗∗∗
(0.022)
0.155∗∗∗
(0.027)
0.054∗
(0.028)
Legislative
0.076∗∗∗
(0.012)
0.093∗∗∗
(0.013)
0.085∗∗∗
(0.015)
0.101∗∗∗
(0.016)
No. Terms
0.012∗∗
(0.006)
0.018∗∗∗
(0.006)
0.019∗∗∗
(0.007)
0.012∗
(0.007)
−0.180∗∗∗
(0.018)
−0.140∗∗∗
(0.018)
−0.248∗∗∗
(0.022)
−0.246∗∗∗
(0.023)
0.259∗∗∗
(0.023)
0.230∗∗∗
(0.024)
0.341∗∗∗
(0.029)
0.331∗∗∗
(0.030)
Budget
0.034
(0.026)
0.030
(0.026)
0.044
(0.032)
0.053
(0.033)
Budgetary control
−0.002
(0.027)
0.012
(0.027)
0.021
(0.034)
0.003
(0.035)
Civil liberties, justice & home affairs
−0.037
(0.028)
−0.071∗∗
(0.028)
−0.001
(0.034)
−0.037
(0.036)
Constitutional & inter-institutional affairs
−0.032
(0.040)
−0.041
(0.041)
−0.046
(0.050)
−0.036
(0.052)
Culture & education
0.013
(0.045)
0.009
(0.045)
0.107∗
(0.056)
0.074
(0.058)
Development
−0.025
(0.035)
−0.040
(0.035)
0.005
(0.043)
−0.021
(0.045)
Economic & monetary affairs
−0.098∗∗∗
(0.026)
−0.082∗∗∗
(0.026)
−0.083∗∗∗
(0.032)
−0.120∗∗∗
(0.033)
Employment & social affairs
−0.060∗
(0.034)
−0.072∗∗
(0.035)
−0.074∗
(0.043)
−0.069
(0.044)
Environment & public health
−0.030
(0.028)
−0.024
(0.029)
−0.024
(0.035)
−0.024
(0.037)
Fisheries
0.036
(0.031)
0.011
(0.031)
0.046
(0.038)
0.060
(0.040)
Foreign & security policy
−0.015
(0.030)
−0.003
(0.030)
0.032
(0.037)
−0.001
(0.038)
−0.068∗
(0.038)
−0.056
(0.039)
−0.050
(0.048)
−0.083∗
(0.050)
Industry, research & energy
−0.043
(0.030)
−0.045
(0.031)
−0.048
(0.038)
−0.052
(0.039)
Internal market & consumer protection
0.022
(0.035)
0.028
(0.035)
0.034
(0.043)
0.041
(0.045)
International trade
−0.029
(0.027)
−0.036
(0.027)
0.0002
(0.034)
−0.027
(0.035)
Legal affairs
−0.002
(0.029)
−0.050∗
(0.029)
0.012
(0.036)
0.011
(0.037)
Petitions
0.025
(0.082)
0.029
(0.083)
0.026
(0.102)
0.042
(0.106)
Regional development
0.005
(0.033)
0.006
(0.034)
0.029
(0.041)
0.031
(0.043)
Transport & tourism
−0.026
(0.032)
−0.041
(0.033)
0.005
(0.040)
−0.001
(0.042)
budget:defection
0.136∗∗∗
(0.038)
0.106∗∗∗
(0.038)
0.224∗∗∗
(0.047)
0.176∗∗∗
(0.049)
Constant
0.378∗∗∗
(0.037)
0.224∗∗∗
(0.037)
0.259∗∗∗
(0.046)
0.310∗∗∗
(0.048)
944
0.400
0.383
0.135
23.518∗∗∗
944
0.379
0.362
0.137
21.562∗∗∗
944
0.410
0.394
0.168
24.539∗∗∗
944
0.378
0.360
0.175
21.445∗∗∗
Defection
EPP dummy
Gender equality
Observations
R2
Adjusted R2
Residual Std. Error (df = 917)
F statistic (df = 26; 917)
Note:
∗
abs(RICEabs)
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
5.3. ACCOUNTING FOR POLICY AREAS
75
Jones (2002: 219). Alongside accounting for the nature of the data, multi-level models
present several other advantages. They allow one to specify predictors at different level
and to investigate whether the effect of lower level predictors is conditioned or modified
by the higher level ones Steenbergen and Jones (2002: 219). Multi-level modelling allows
one to make more specific inferences about the nuances of the effects investigated.
The reference category for fixed effects models, presented in Table 5.2, is the Agriculture committee. It can be argued that the bills discussed in the Agriculture committee
might concern structural funds to a larger degree when compared to bills from other
committees. The Common Agricultural Policy has been one of the most salient topics in
the European Parliament since its entry into force. Aksoy (2010: 174) points out that
Agriculture is one of the largest expenses in the European Union budget. Given that the
importance of the agricultural sector varies across members states it can be argued that
this policy area is rather polarized. Furthermore, this is one of the areas where, according
to the theoretical framework, the legislator’s interest on their constituencies is likely to
be strong.
Therefore, it can be argued that agriculture is a substantially interesting reference
category. From a theoretical perspective it is suggested that budgetary implications will
differ across policy areas. This motivates letting the effect of budgetary implications vary
in these model. The results from the random intercept and random effect models are
presented in Table 5.3. The fixed effects models are presented in Table 5.2.
The fixed effects models only capture variation within each policy area when estimating the models, hence effectively controlling for all observed and unobserved differences
across policy areas. The results are presented in Table 5.2. The coefficient for budgetary
implications retains its significance in all the models with fixed effects. When it is assumed that policy areas have a fixed effect, budgetary implications for bills where there
was no defection on behalf of the party of the rapporteur will have a negative effect on
overall legislative unity, regardless of the way in which one deals with abstentions and
those that did not vote. The magnitude of the coefficient for budget varies between -0.028
in the RICE model to -0.043 in the AIn model. When the MEPs that did not vote are
accounted for, the effect of budgetary implications is slightly stronger, enforcing again
the claim that legislators prefer a form of mild defection, either not to vote, or to abstain,
rather than directly defect. Unfortunately, the distinction between defecting by voting
the opposing from the party line, for example voting No, when the party line is Yes,
versus abstaining or not voting remains blurred. Hix (2002) considers defections those
cases where legislators vote differently from the party line.
It seems that European Party Groups mobilize and induced, more pressure on legislators, and hence a higher level of cohesion among their members when they know the
initial level of disagreement is high. This is shown by the interaction term between defection and budgetary implications. Setting aside the interaction term one notices that the
76
CHAPTER 5. ROBUSTNESS
coefficient for seniority regains its significance. Thus, it seems that members that served
more than one term are better equipped at forging legislative unity within the European
Parliament. Although, it has to be mentioned that the effect are relatively small, yet
the coefficient is significant in all the model specifications. At the same time, legislative
proposals have a relatively higher positive impact on legislative unity. Even when accounting for the hierarchical nature of the data by estimating a fixed effects linear model
the results are analogous to those from the initial tobit models, presented in section 4.5.
The fixed effects model includes a dummy variable for each policy areas, which produces less efficient estimates. Only looking at variation within policy areas means avoiding
all possible sources of omitted variable bias from variables at the policy area level. This
does however come at the cost that all variation between policy areas is discarded (Gelman and Hill 2007: Chapter 12). It may also be interesting to investigate whether the
reported results hold when the effect of budgetary implication itself is allowed to differ
across these policy areas and controls for party positions are introduced.
Random effects models can be understood as an approximation of the “weighted average of the mean of the observations in the [group] and the mean over all [groups]” (Gelman
and Hill 2007: 253). In contrast to fixed effects models, which accept estimations on very
little data, random effects models tend to constrain the estimates from groups with very
little individual data towards the mean of all the data available (Gelman and Hill 2007).
The fixed effect model imposes several restrictions on the variations of coefficients. By
looking at table 5.2 we see that even when the assumption about the effects of policy
areas is relaxed the results remain largely unaltered in terms of significance across the
various specifications of the dependent variable.
I have included party dummies in the models with random intercept, random effects
of budget across policy areas. As discussed in section 5.2 controlling for party positions
decreases the importance of the effect of budgetary implication, especially for bills where
there is not defection. These results hold even when we account for policy areas as it can
be seen from Table 5.3, the substantial effects are not different from those presented in
Section 5.2. The effect of budgetary implications for cases where there are not defections
are only significant in the AI and AIn model, showing again that legislators try to avoid
defection and select a softer approach. At the same time it shows that accounting for
party positions affect the relationship between budgetary implications and legislative
unity. This is expected in light of Peltzman’s (1984) argument that economic factors are
often hidden by ideological variables.
Albeit the loss of significance for the coefficient of budget alone in the RICE and
RICEabs model, the interaction term remains significant in all the models as it can be
seen from Table 5.3. Even so, there is a decline in the estimated effect. The theoretical
argument that in polarized cases, where there is a high risk of both individual and party
defection, the remaining European Party Groups will push for a cohesive behaviour within
5.3. ACCOUNTING FOR POLICY AREAS
77
Table 5.3: Random intercept and effect of budget controlled for party positions
Dependent variable:
AI
AIn
(1)
abs(RICE)
abs(RICEabs)
(3)
(4)
Budget
∗
−0.022
(0.012)
∗
−0.019
(0.010)
−0.005
(0.010)
−0.005
(0.009)
Result
0.050∗∗∗
(0.011)
0.138∗∗∗
(0.016)
0.043∗∗∗
(0.016)
−0.058∗∗∗
(0.014)
Legislative
0.015∗∗
(0.006)
0.035∗∗∗
(0.008)
0.020∗∗
(0.009)
0.020∗∗∗
(0.008)
No. Terms
0.002
(0.003)
0.008∗
(0.004)
0.006
(0.004)
−0.0001
(0.004)
Defection
−0.022∗∗
(0.010)
−0.006
(0.013)
−0.054∗∗∗
(0.013)
−0.036∗∗∗
(0.012)
EPP dummy
0.333∗∗∗
(0.012)
0.292∗∗∗
(0.017)
0.423∗∗∗
(0.017)
0.427∗∗∗
(0.015)
S&D dummy
0.323∗∗∗
(0.012)
0.287∗∗∗
(0.016)
0.430∗∗∗
(0.016)
0.422∗∗∗
(0.015)
ALDE dummy
0.145∗∗∗
(0.012)
0.121∗∗∗
(0.017)
0.200∗∗∗
(0.017)
0.202∗∗∗
(0.016)
Greens dummy
0.133∗∗∗
(0.007)
0.109∗∗∗
(0.010)
0.154∗∗∗
(0.010)
0.174∗∗∗
(0.009)
GUE/NGL dummy
0.060∗∗∗
(0.006)
0.043∗∗∗
(0.008)
0.039∗∗∗
(0.008)
0.079∗∗∗
(0.007)
ECR dummy
0.106∗∗∗
(0.005)
0.084∗∗∗
(0.008)
0.102∗∗∗
(0.008)
0.136∗∗∗
(0.007)
EFD dummy
0.058∗∗∗
(0.005)
0.051∗∗∗
(0.007)
0.078∗∗∗
(0.007)
0.079∗∗∗
(0.006)
budget:defection
0.088∗∗∗
(0.020)
0.052∗
(0.027)
0.146∗∗∗
(0.027)
0.099∗∗∗
(0.024)
−0.289∗∗∗
(0.022)
−0.350∗∗∗
(0.031)
−0.542∗∗∗
(0.031)
−0.569∗∗∗
(0.028)
944
1, 121.495
−2, 206.991
−2, 119.688
944
815.201
−1, 594.403
−1, 507.101
944
811.101
−1, 586.202
−1, 498.900
944
913.988
−1, 791.976
−1, 704.674
Constant
Observations
Log likelihood
Akaike Inf. Crit.
Bayesian Inf. Crit.
Note:
(2)
∗
p<0.1;
∗∗
p<0.05;
∗∗∗
p<0.01
78
CHAPTER 5. ROBUSTNESS
their own parties in order to insure their most desired outcome in the plenary.
Contrariwise to the fixed effects model the number of terms sever by the lead rapporteur looses significance in models (1), (3) and (4). It barely remains significant in
the AIn model at the 10% level and has a very small magnitude of only 0.008. Consequently, there is little that can be said about the actual effect of seniority on legislative
unity as the effect seems to be dependent on both model and specification. On the other
hand, legislators seem to act in a more cohesive way when they vote upon legislative bills
compared to resolutions. This effect has proved robust across various model types and
specifications.
5.4
Summary
A caveat of the approach that remains unsolved is dealing with the potential endogenity.
Several instruments for defection were tested, yet, the correlation was to weak for such
an approach to be fruitful. Therefore, such a course was not pursued. Sections 5.2
and 5.3 aims to address the omnipresent threat of omitted variable bias. The results are
relatively robust, yet in the presence of party controls the effects of budgetary implications
on their own loose significance for the RICE and RICEabs models, yet the remainder of
the interaction remains significant. This suggests that the findings of this thesis are
relatively vulnerable to the way in which ideological positions are captured. Despite
this caveat, the robustness tests remain supportive of the hypothesized relationships.
Budgetary implications on their own have a negative effect on legislative unity, yet this
effect is mitigated when the part of the rapporteur also defects.
In order to test whether the results were driven by the type of model I have estimated
fixed and random effects models, which also account for the hierarchical structure of the
data. This chapter has shown that in the presences of controls for party positions the
relationship between budgetary implications and legislative unity is weaker when there
are no other polarizing factors. Yet, irrespective of the controls for party positions the
relationship remains significant in situations where there are both budgetary implications
and defections by the party of the rapporteur, even when policy areas are accounted for,
as presented in Table 5.3.
Chapter 6
Conclusion
This thesis has set out to investigate whether budgetary implications affect the behaviour
of legislators in final votes. The main focus in the literature has been on the ideological
determinants of voting (Hix, Noury and Roland 2006, Hix and Noury 2009), and in
some cases on the procedural differences between the bills (Kreppel 2002). After the
ratification of the Lisbon treaty, the European Parliament has become an equal legislator
to the Council in most policy areas and has gained budgetary power (Hix and Høyland
2013). Peltzman (1984) has shown that ideological variables may mask economic interests.
Economic determinants of voting in the European Parliament have been overlooked by
the literature so far. This thesis makes a contribution to filling this gap in the existing
literature by investigating how the budgetary implications of bills affect the behaviour of
legislators in final votes.
Drawing upon a recent study (Bailer, Mattila and Schneider 2010) on the Council
and using classical theory developed for the US Congress (Cox and McCubbins 2007,
Fenno 1978), I have proposed that, depending on the initial level of polarization in the
European Parliament, budgetary implications alter the behaviour of legislators. At the
same time, it has to be mentioned that budgetary implications do not affect the chances
for passage or failure of a bill. Budgetary implications are important for the level of
agreement, not for the outcome of the vote. It is conceptualized that legislators have
three main goals: re-election, power while in office and good policy (Faas 2003, Fenno
1978). In order for them to attain the latter two goals, they are to further the interests
of their home constituencies (Fenno 1978).
For the setting of the European Parliament I have, however – in line with much
of the existing literature (Hix 2002, Hix, Raunio and Scully 1999, Hix and Noury 2009,
Mühlböck 2012) – argued that legislators are responsive to the demands of their European
Party Groups, and hence less likely to defect in situations where there is increased party
pressure. This is especially so because the electoral connection between legislators and
their constituencies is frail (Carey 2007). Such situations are captured by looking at
situations where there is a breakdown in the negotiations and the party of the rapporteur
79
80
CHAPTER 6. CONCLUSION
either defects or is split. From this proposition I derive two testable hypotheses. First, I
hypothesize that in situations where the bill has implications for the status quo allocation
of resources, legislative cohesion is likely to be lower. Secondly, legislative unity is likely
to increase in situations where the party of the rapporteur defects, as legislators respond
to increased party pressure. Failure to respond to party pressure might lead to sanctions,
consequently making the attainment of the second goal harder.
To the best of my knowledge, such propositions have not been tested on the European
Parliament previously. In order to rectify this omission, I have collected data on all final
votes in the European Parliament in the period 14.09.2009 - 14.03.2013. The dataset
contains information about the type of bill, information about how the legislators voted,
as well as the rapporteurs on the case. Bills are coded to have budgetary implications
when they alter the existing status quo on the set area, as discussed in Section 3.2.1.
By employing the RICE index (Rice 1924) and the Agreement Index (Hix, Noury and
Roland 2005) in their original form and with some modifications, I have tested the impact
of budgetary implications. Despite the small level of censoring, OLS estimates are likely
to be inconsistent. Consequently, I have employed Tobit model specifications in order
to draw inferences about the hypothesized relationship. The initial level of polarization
before final votes has been captured by the interaction between budgetary implications
and defection on behalf of the party of the rapporteur. This interaction has been shown
to be robust across model specifications. Yet, as predicted by Peltzman (1984) it seems
that ideological variables mask in part the effect of budgetary implications. Even when
controlling for party positions, budgetary implications in the absence of other polarizing
factors retain their significance in the models with the Agreement Index. Accounting for
the hierarchical nature of the data does not alter the results substantially, as shown in
Section 5.3.
In conclusion, it can be argued that the interaction between budgetary implications
and legislative unity is significant and relatively robust. Generally, members of the European Parliament aim at pleasing their home constituencies, while at the same time
remain loyal to their European Party Group. In cases where the party of the rapporteur
defects and the bill has budgetary implications, it seems that parties use their pressuring
mechanisms and, as a consequence, higher cohesion in the legislative is likely.
Despite the relative robustness of these findings, caution is warranted. First, the effects
are weak in terms of magnitude. Secondly, even if defection is known chronologically
before the final vote, it is only observed at the time of the vote, thus raising problems
of endogenity. Several instruments have been tested in order to remedy the problem,
nonetheless they have proven too weak. Given the weak correlation such an approach has
not been pursued.
This thesis has been unable to differentiate between national interests and partisan
ones. For example, Grossman and Helpman (1996) propose that interest groups are also
81
able to alter the behaviour of legislators, hence legislators might also defect because of
partisan interests and not only because of national constituencies. Further research is
needed in order to map the exact mechanisms that lead to defections. Another aspect
that could be potentially interesting for future research is the dis-aggregation of budgetary
implications further in order to investigate how these effects differ when bills add increased
revenue, or in cases in which they are extractive from the national constituencies versus
in cases where they only impact the budget of the EU as an institution.
This thesis provides some evidence that altering the existing budgetary allocation
is bound to lead to a higher level of disagreement between legislators. As mentioned,
future research could dis-aggregate budgetary implications further and investigate how
legislative unity is influenced when national party positions are accounted for. This
is perhaps most interesting in situations when the implications are detrimental for the
respective member states.
These caveats notwithstanding, this thesis has provided robust evidence for how budgetary implications impact the level of cohesiveness in the European Parliament, yet the
effects are dual as they are dependent on the initial level of polarization. Nevertheless,
in order to better map the dynamics of the European Parliament, future research could
elaborate on both the factors that influence committee level negotiations, while at the
same time differentiate more clearly between the mechanisms that affect the legislator’s
behaviour in the plenary,or example, by differentiating between partisan interests and
national ones.
In conclusion, this thesis has shown that budgetary implications are important for
the level of cohesion within the European Parliament. For cases where the bill alters
the existing budgetary status-quo, while there are few other polarizing factors, legislative
unity is lower. The main exceptions are the cases where the party of the rapporteur
either defects or is split. In these cases party groups are able to forge a higher degree of
cohesion.
Chapter 7
Bibliography
Adcock, Robert and David Collier. 2001. “Measurement validity: A shared standard for
qualitative and quantitative research.” American political science review 95(03):529–
546.
Aksoy, Deniz. 2010. “Who gets what, when, and how revisited: Voting and proposal
powers in the allocation of the EU budget.” European union politics 11(2):171–194.
Aspinwall, Mark D and Gerald Schneider. 2000. “Same menu, seperate tables: The institutionalist turn in political science and the study of European integration.” European
Journal of Political Research 38(1):1–36.
Bailer, Stefanie. 2004. “Bargaining Success in the European Union The Impact of Exogenous and Endogenous Power Resources.” European Union Politics 5(1):99–123.
Bailer, Stefanie, Mikko Mattila and Gerald Schneider. 2010. Money Makes the EU Go
Round: The Objective Foundations of Conflict in the Council of Ministers. In APSA
2010 Annual Meeting Paper.
Bowler, Shaun and David M Farrell. 1995. “The Organizing of the European Parliament:
Committees, Specialization and Co-Ordination.” British Journal of Political Science
pp. 219–243.
Carey, John M. 2007. “Competing principals, political institutions, and party unity in
legislative voting.” American Journal of Political Science 51(1):92–107.
Carrubba, Clifford J. 1997. “Net financial transfers in the European Union: who gets
what and why?” Journal of Politics 59:469–496.
Carrubba, Clifford J, Matthew Gabel, Lacey Murrah, Ryan Clough, Elizabeth Montgomery and Rebecca Schambach. 2006. “Off the record: unrecorded legislative votes, selection bias and roll-call vote analysis.” British Journal of Political Science 36(04):691–
704.
82
83
Carrubba, Clifford, Matthew Gabel and Simon Hug. 2008. “Legislative voting behavior,
seen and unseen: A theory of roll-call vote selection.” Legislative Studies Quarterly
33(4):543–572.
Clarke, Kevin A. 2005. “The phantom menace: Omitted variable bias in econometric
research.” Conflict Management and Peace Science 22(4):341–352.
Clarke, Kevin A and David M Primo. 2007. “Modernizing political science: a model-based
approach.” Perspectives on Politics 5(4):741.
Cox, David R. 1992. “Causality: some statistical aspects.” Journal of the Royal Statistical
Society. Series A (Statistics in Society) pp. 291–301.
Cox, Gary W and Mathew D McCubbins. 1993. “Legislative leviathan: Party government
in the House.”.
Cox, Gary W and Mathew D McCubbins. 2007. Legislative leviathan: Party government
in the House. Cambridge: Cambridge University Press.
Crombez, Christophe. 1996. “Legislative procedures in the European Community.” British
Journal of Political Science 26:199–228.
Crombez, Christophe. 1997. “The co-decision procedure in the European Union.” Legislative Studies Quarterly pp. 97–119.
Crombez, Christophe. 2000. “Institutional reform and co-decision in the European Union.”
Constitutional Political Economy 11(1):41–57.
Crombez, Christophe. 2003. “The Democratic Deficit in the European Union Much Ado
about Nothing?” European Union Politics 4(1):101–120.
Diermeier, Daniel and Timothy J Feddersen. 1998. “Cohesion in legislatures and the vote
of confidence procedure.” American Political Science Review pp. 611–621.
ECON, Committee on Economic and Monetary Affairs. 2012. “DRAFT REPORT on
the proposal for a regulation of the European Parliament and of the Council on
the European statistical programme 2013-2017 (COM(2011)0928 – C7-0001/2012 –
2011/0459(COD)).”.
URL:
http://www.europarl.europa.eu/sides/getDoc.do?pubRef=//EP//NONSGML+COMPARL+PE-486.047+01+DOC+PDF+V0//ENlanguage=EN
euractiv.com. 2013. “Brussels wins new powers over national budgets.”.
URL: http://www.euractiv.com/future-eu/commission-gets-new-powers-natio-news517959
84
CHAPTER 7. BIBLIOGRAPHY
Europa.eu. 2012. “Surveillance of budgetary policies.”.
URL: http://europa.eu/legislations ummaries/economica ndm onetarya f f airs/stabilitya ndg rowthp act
European Commission. 2010. “Proposal for a REGULATION OF THE EUROPEAN
PARLIAMENT AND OF THE COUNCIL on OTC derivatives, central counterparties
and trade repositories.”.
URL: http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:2010:0484:FIN:EN:PDF
European Commission. 2011a. “Proposal for a COUNCIL DIRECTIVE on a common
system of financial transaction tax and amending Directive 2008/7/EC.”.
URL: http://www.ipex.eu/IPEXL-WEB/dossier/document.do?code=COMyear=2011number=0594a
European Commission. 2011b. “Proposal for a REGULATION OF THE EUROPEAN
PARLIAMENT AND OF THE COUNCIL on a mechanism for monitoring and reporting greenhouse gas emissions and for reporting other information at national and Union
level relevant to climate change.”.
URL: http://www.ipex.eu/IPEXL-WEB/dossier/document/COM20110789.do
European Commission. 2011c. “Proposal for a REGULATION OF THE EUROPEAN
PARLIAMENT AND OF THE COUNCIL on the European statistical programme
2013-2017.”.
URL: http://www.ipex.eu/IPEXL-WEB/dossier/ document/COM20110928.do
Faas, Thorsten. 2003. “To defect or not to defect? National, institutional and party group
pressures on MEPs and their consequences for party group cohesion in the European
Parliament.” European Journal of Political Research 42(6):841–866.
URL: http://dx.doi.org/10.1111/1475-6765.00106
Fenno, Richard F. 1978. Home Style: House Members in Their Districts (Longman
Classics Series). London: Harper Collins Publishers.
Føllesdal, Andreas and Simon Hix. 2006. “Why there is a democratic deficit in the EU:
a response to Majone and Moravcsik.” JCMS: Journal of Common Market Studies
44(3):533–562.
Furubotn, Eirik G and Rudolf Richter. 1997. “Institutions and economic theory: The
contribution of the new institutional economics.”.
Gelman, Andrew and Jennifer Hill. 2007. Data analysis using regression and multilevel/hierarchical models. Cambridge: Cambridge University Press.
Greene, William H. 2003. Econometric analysis. Prentice Hall, New Jersey: Upper Saddle
River.
85
Grossman, Gene M and Elhanan Helpman. 1996. “Electoral competition and special
interest politics.” The Review of Economic Studies 63(2):265–286.
Hellevik, Ottar. 2006. Forskningsmetode i sosiologi og statsvitenskap. Oslo: Universitetsforlaget.
Henningsen, Arne. 2010. “Estimating Censored Regression Models in R using the censReg
Package.” R package vignettes .
Hillman, Arye L. 1989. The political economy of protection. Vol. 32 Newark: Harwood
Academic Pub.
Hix, Simon. 2001. “DECISION RULES IN THE EUROPEAN UNION: A RATIONAL
CHOICE PERSPECTIVE.” Journal of Common Market Studies 39(2):367.
Hix, Simon. 2002. “Parliamentary behavior with two principals: preferences, parties, and
voting in the European Parliament.” American Journal of Political Science pp. 688–698.
Hix, Simon. 2004. “Electoral Institutions and Legislative Behavior: Explaining Voting
Defection in the European Parliament.” World Politics 56(2):pp. 194–223.
URL: http://www.jstor.org/stable/25054255
Hix, Simon, Abdul G Noury and Gérard Roland. 2007. Democratic politics in the European Parliament. Cambridge: Cambridge University Press.
Hix, Simon and Abdul Noury. 2009. “After enlargement: voting patterns in the sixth
European Parliament.” Legislative Studies Quarterly 34(2):159–174.
Hix, Simon, Abdul Noury and Gérard Roland. 2005. “Power to the parties: cohesion
and competition in the European Parliament, 1979–2001.” British Journal of Political
Science 35(2):209–234.
Hix, Simon, Abdul Noury and Gerard Roland. 2006. “Dimensions of politics in the
European Parliament.” American Journal of Political Science 50(2):494–520.
Hix, Simon, Amie Kreppel and Abdul Noury. 2003. “The party system in the European
Parliament: Collusive or competitive?” JCMS: Journal of Common Market Studies
41(2):309–331.
Hix, Simon and Bjørn Høyland. 2013. “Empowerment of the European Parliament.”
Annual Review of Political Science (0).
Hix, Simon and Bjørn Høyland. 2011. The political system of the European Union. Houndmills: Palgrave Macmillan.
86
CHAPTER 7. BIBLIOGRAPHY
Hix, Simon and Bjørn Høyland. 2013. “Empowerment of the European Parliament.”
Annual Review of Political Science 16(1):171–89.
URL:
http://www.annualreviews.org/doi/abs/10.1146/annurev-polisci-032311110735
Hix, Simon, Tapio Raunio and Roger Scully. 1999. “An Institutional Theory of Behavior
in the European Parliament.”.
Hug, Simon. 2003. “Endogenous preferences and delegation in the European Union.”
Comparative Political Studies 36(1-2):41–74.
Hug, Simon and Thomas König. 2002. “In view of ratification: Governmental preferences
and domestic constraints at the Amsterdam Intergovernmental Conference.” International Organization 56(02):447–476.
Jacobs, Francis, Richard Corbett and Michael Shackleton. 1990. The European Parliament. London: Longman.
Johnston, John and John DiNardo. 1972. Econometric methods. Vol. 4 Wiley Online
Library.
Kampstra, Peter et al. 2008. “Beanplot: a boxplot alternative for visual comparison of
distributions.”.
Kandogan, Yener. 2000. “Political economy of eastern enlargement of the European Union:
Budgetary costs and reforms in voting rules.” European Journal of Political Economy
16(4):685–705.
Kennedy, Peter. 2011. A Guide to Econometrics. Malden, Mass.: Blackwell Publishing.
King, Gary, Robert O Keohane and Sidney Verba. 1994. Designing social inquiry: Scientific inference in qualitative research. Princeton: Princeton University Press.
Kreppel, Amie. 2002. The European Parliament and Supranational Party System: a study
in institutional development. Cambridge: Cambridge University Press.
Kreppel, Amie and George Tsebelis. 1999. “Coalition formation in the European Parliament.” Comparative Political Studies 32(8):933–966.
Lasswell, Harold Dwight. 1950. Politics: Who gets what, when, how. New York: P. Smith.
Lindberg, Björn, Anne Rasmussen and Andreas Warntjen. 2008. “Party politics as usual?
The role of political parties in EU legislative decision-making.” Journal of European
Public Policy 15(8):1107–1126.
87
Long, J Scott. 1997. Regression models for categorical and limited dependent variables.
Vol. 7 Thousand Oaks: SAGE Publications, Incorporated.
Magee, Stephen P, William A Brock and Leslie Young. 1989. Black hole tariffs and
endogenous policy theory: Political economy in general equilibrium. Cambridge: Cambridge University Press.
Majone, Giandomenico. 1998. “Europe’s ‘democratic deficit’: The question of standards.”
European law journal 4(1):5–28.
Mamadouh, Virginie and Tapio Raunio. 2003. “The Committee System: Powers, Appointments and Report Allocation*.” JCMS: Journal of Common Market Studies 41(2):333–
351.
Mattila, Mikko. 2009. “Roll call analysis of voting in the European Union Council of Ministers after the 2004 enlargement.” European Journal of Political Research 48(6):840–857.
McElroy, Gail. 2006. “Committee representation in the European Parliament.” European
Union Politics 7(1):5–29.
McElroy, Gail. 2008. “Committees and party cohesion in the European Parliament.”.
McElroy, Gail and Kenneth Benoit. 2007. “Party groups and policy positions in the
European Parliament.” Party Politics 13(1):5–28.
McElroy, Gail and Kenneth Benoit. 2012. “Policy positioning in the European Parliament.” European Union Politics 13(1):150–167.
Moravcsik, Andrew. 1993. “Preferences and Power in the European Community: A
Liberal Intergovernmentalist Approach.” JCMS: Journal of Common Market Studies
31(4):473–524.
URL: http://dx.doi.org/10.1111/j.1468-5965.1993.tb00477.x
Moravcsik, Andrew. 1998. The choice for Europe: social purpose and state power from
Messina to Maastricht. Ithaca: Cornell University Press.
Mühlböck, Monika. 2012. “Linking Council and European Parliament? Voting unity of
national parties in bicameral EU decision-making.” Journal of European Public Policy
(ahead-of-print):1–18.
Mühlböck, Monika and Nikoleta Yordanova. 2012. “Party group cohesion in the European
Parliament: Tracing the bias in roll-call votes.” Available at SSRN 2173785 .
Neuhold, Christine. 2001. “The’Legislative Backbone’Keeping the Institution Upright?
The Role of European Parliament Committees in the EU Policy-Making Process.” European integration online papers (EIoP) 5(10).
88
CHAPTER 7. BIBLIOGRAPHY
Nordkvelle, Jonas. 2012. “Legislative voting and party unity in the European Parliament.”.
OJ C126. 2011. “Official Journal of the European Union.”.
URL: http://eur-lex.europa.eu/JOHtml.do?uri=OJ:C:2011:216:SOM:EN:HTML
Peltzman, Sam. 1984. “Constituent Interest and Congressional Voting.” Journal of Law
and Economics 27(1):181–210.
URL: http://EconPapers.repec.org/RePEc:ucp:jlawec:v:27:y:1984:i:1:p:181-210
Persson, Torsten and Guido Enrico Tabellini. 2005. The economic effects of constitutions.
Cambridge, MA: The MIT Press.
Poole, Keith T. 2005. Spatial models of parliamentary voting. Cambridge: Cambridge
University Press.
Raunio, Tapio. 1997. The European Perspective: Transnational Party Groups in the
1989-1994 European Parliament. Farnham: Ashgate Aldershot.
Rice, Stuart Arthur. 1924. Farmers and workers in American politics.
Scarrow, Susan E. 1997. “Political career paths and the European Parliament.” Legislative
Studies Quarterly pp. 253–263.
Schneider, Gerald and Lars-Erik Cederman. 1994. “The change of tide in political cooperation: A limited information model of European integration.” International Organization
48(04):633–662.
Selck, Torsten J and Bernard Steunenberg. 2004. “Between Power and Luck The European
Parliament in the EU Legislative Process.” European Union Politics 5(1):25–46.
Shepsle, Kenneth A. 1979. “Institutional Arrangements and Equilibrium in Multidimensional Voting Models.” American Journal of Political Science 23(1):pp. 27–59.
URL: http://www.jstor.org/stable/2110770
Shugart, Matthew Soberg. 1998. “The inverse relationship between party strength and
executive strength: a theory of politicians’ constitutional choices.” British Journal of
Political Science pp. 1–29.
Sigelman, Lee and Langche Zeng. 1999. “Analyzing censored and sample-selected data
with Tobit and Heckit models.” Political Analysis 8(2):167–182.
Sovey, Allison J and Donald P Green. 2011. “Instrumental variables estimation in political
science: A readers’ guide.” American Journal of Political Science 55(1):188–200.
Steenbergen, Marco R and Bradford S Jones. 2002. “Modeling multilevel data structures.”
american Journal of political Science pp. 218–237.
89
Stigler, George J. 1971. “The theory of economic regulation.” The Bell journal of economics and management science pp. 3–21.
Tobin, James. 1958. “Estimation of relationships for limited dependent variables.” Econometrica: Journal of the Econometric Society pp. 24–36.
Tsebelis, George. 1994. “The power of the European Parliament as a conditional agenda
setter.” American Political Science Review pp. 128–142.
Tsebelis, George and Geoffrey Garrett. 2000. “Legislative politics in the European Union.”
European Union Politics 1(1):9–36.
Weingast, Barry R., Kenneth A. Shepsle and Christopher Johnsen. 1981. “The Political
Economy of Benefits and Costs: A Neoclassical Approach to Distributive Politics.”
Journal of Political Economy 89(4):pp. 642–664.
URL: http://www.jstor.org/stable/1833029
Whitaker, Richard. 2001. “Party control in a committee-based legislature? The case of
the European Parliament.” Journal of Legislative Studies 7(4):63–88.
Whitaker, Richard. 2005. “National Parties in the European Parliament An Influence in
the Committee System?” European Union Politics 6(1):5–28.
Wooldridge, Jeffrey M. 2002. Econometric Analysis of Cross Section and Panel Data.
Cambridge, MA: MIT press.
Yordanova, Nikoleta. 2009. “The Rationale behind Committee Assignment in the European Parliament Distributive, Informational and Partisan Perspectives.” European
Union Politics 10(2):253–280.
Zimmer, Christina, Gerald Schneider and Michael Dobbins. 2005. “The contested Council: Conflict dimensions of an intergovernmental EU institution.” Political Studies
53(2):403–422.
Appendix A
Influential observations
90
597
593
595656
596
331
1018
1
11 12 2
29 39 49 59 69 79 89 99
0.004
0.000
544
948
259
7
492
972
143
338
368
737 81
324
1034 1147
214
526
2
494
914
399
109
16
496
895
734
874
74
610 718
1021
894
227
241 345
369 480
1019
150
185
776
375
762
498
602
635
127
580
887
240 320
11035
198
169
78
909
352
419
118
79
171
934
413
910
1092
206
527
586
237
467
519
623
358
622
522
164
1124
568
1128
1126
818
99
1152
760
581
854
5
384
892
599
1123
98
1127
122
416
180
940
879
69
553
121
608
146
421
107
621
126
123
795
858
417
512
130
176
906
1042
329
1194
386
567
722
688
70
28
68
875
106
1064
739
105
347
798
793
238
92
46
1141
833
370
675
8
482
601
125
124
177
297
289
678
857
321
37
84
197
383
636
609
792
326
1125
178
582
96
335
475
780
486
904
166
400
1070
443
985
473
336
152
196
478
808
891
971
25
442
470
613
794
316
485
10
170
778
86
24
457
307
167
265
791
273
703
61
736
738
980
353
790
374
757
578
973
346
49
175
405
935
984
33
423
723
774
566
514
188
528
721
1072
334
365
503
845
474
65
88
228
656
290
229
299
3372
6
624
731
38
764
955
1015
11
392
6
18
391
615
1107
1177
19
202
532
634
660
704
1195
900
944
1106
215
491
311
469
771
789
836
137
550
785
843
855
20
376
481
890
951
288
508
740
110
662
881
361
573
813
814
919
9
135
258
918
981
312
920
1187
162
217
449
773
921
998
661
663
886
490
73
83
844
85
878
441
450
749
298
304
507
884
129
30
521
149
832
212
897
148
410
506
557
579
750
1133
153
707
108
349
548
390
1063
446
869
161
664
1061
285
933
1059
1105
653
373
585
899
163
173
325
471
48
1051
1190
281
658
905
1058
136
366
788
807
1029
47
651
654
669
6687
7
966
564
1130
1131
1132
363
133
140
424
64
445
647
650
804
1122
187
23
385
437
453
649
876
1096
138
158
181
408
644
648
1121
22
639
645
646
735
784
640
659
436
493
1100
398
426
758
853
1071
619
637
638
641
642
643
657
1022
1175
71
926
1023
1067
1142
1181
139
203
396
464
667
786
880
1101
351
462
463
592
465
466
674
852
928
387
447
452
614
823
903
1078
1093
1200
499
655
859
103
1198
134
165
1199
344
39
444
612
666
761
1053
827
877
1129
116
587
1024
317
1114
159
472
626
652
42
488
545
625
812
834
978
1025
1166
343
40
569
572
1180
1007
1013
1163
1174
309
43
830
831
946
979
1164
1178
865
91
339
697
835
230
62
826
100
1030
1033
1161
168
672
701
1031
160
523
829
977
994
174
435
744
454
999
1103
155
575
800
1165
291
362
725
72
1011
1162
117
209
218
552
742
828
976
271
559
563
1005
440
746
898
990
993
31
1010
1148
748
787
268
332
495
562
75
867
1008
17
539
801
958
1012
1139
1197
337
889
157
431
458
982
1000
1094
883
954
201
27
611
629
775
190
1001
1006
1196
319
55
995
1140
141
418
537
101
179
549
882
885
515
700
702
76
848
90
931
1003
1009
350
538
989
34
439
763
1014
26
709
839
945
992
211
747
782
87
1173
21
477
516
517
628
796
561
255
95
367
991
257
717
200
204
627
821
922
1002
1004
542
727
822
314
1172
45
671
996
1186
594
232
274
670
860
618
89
438
448
825
961
1137
293
630
1113
111
29
960
1073
1134
225
476
631
997
1138
318
459
598
779
840
93
112
841
949
1169
943
551
716
942
956
1065
1167
929
1120
1159
850
286
296
256
815
97
41
502
679
401
692
868
348
533
535
864
820
959
1057
1176
216
315
913
94
715
114
1168
145
536
708
322
500
504
524
915
205
534
797
863
1016
1170
360
144
1136
310
404
577
1171
207
226
210
242
295
916
1017
1066
1135
192
862
222
223
389
63
689
819
861
270
429
1145
430
696
824
221
525
907
208
547
119
234
224
249
102
1079
543
60
154
403
195
236
292
584
402
927
191
690
235
505
284
460
58
59
251
720
3
1050
156
282
574
698
983
287
576
597
456
967
193
962
194
588
783
699
147
901
917
851
589
932
593
595
5583
7607
331
501
104
142
596
1158
590
283
591
965
719
665
269
397
546
1091
571
673
752
963
313
759
957
866
1160
266
1020
433
570
695
411
1018
189
428
131
947
267
668
930
132
754
427
691
182
53
964
479
434
199 264
838
306
969
676
633
184
988
846
856902950
1115
706 817
300 371
1054
388
82
1055
694
1056
4
497
781
333
520
186
239 308
44
32
677
305
172
968
518
987
451
705
1026
1
11 12 2
29 39 49 59 69 79 89 99
Observation
0.00
1056
1055
1054
705
196
891 968
451 518 607 677
987
723
172
1018
44
32
308
186239 305
520 593
333388
596
595
331
597
781
694
82
656
706
4
817
1176
988
371
1115
256
285
184
185
633
300
497
594
1186
111
45
490
846
257
1063
255
969
950
306
856
966
727
709
931
179
90
848
477
264
676
446
902
458
838
964
332
140
199
375
746
53
653
976
552
691
325
1103
754
132
1051
427
668
930
267
131
189
654
669
100
651
977
266
742
759
957
947
752
963
411
1091
571
673
1020
1178
269
433
363
57
591
313
590
583
104
546
397
428
665
570
695
719
965
1160
650
866
142
445
834
1096
699
647
283
456
978
1123
1158
479
589
698
851
901
932
588
720
783
147
501
649
690
962
287
584
967
1050
1129
194
236
237
467
644
648
983
102
193
576
1053
156
251
270
292
429
574
861
58
862
191
235
284
3
505
639
646
99
119
360
402
403
460
534
645
863
907
536
715
927
94
310
533
535
543
868
917
959
98
430
525
547
63
819
824
97
1145
192
210
223
689
956
315
93
205
295
296
29
504
524
577
640
708
960
1079
216
322
348
59
716
820
850
961
195
232
401
618
659
670
692
779
860
89
913
1113
224
234
293
318
551
671
775
1159
200
221
459
75
841
943
1017
1066
1134
1135
1136
222
225
226
314
387
476
516
517
916
1016
1168
1170
144
145
204
448
500
797
825
87
915
1057
1100
112
502
515
537
538
542
679
782
822
996
1002
1004
1065
1125
114
1167
286
34
367
700
702
796
942
1073
1169
157
26
319
561
592
598
631
839
840
95
991
997
1003
1009
1014
101
268
539
630
76
787
867
882
885
958
989
992
1001
1006
1067
1196
154
208
249
60
627
821
880
889
995
1000
1012
1094
1171
1173
1181
201
207
274
337
389
404
431
439
562
611
628
696
828
945
982
1008
1010
1197
160
454
549
800
801
829
864
1005
1011
1138
190
362
41
559
629
701
815
826
883
91
929
949
954
990
993
1137
230
309
31
339
372
40
435
438
440
62
672
697
717
748
830
831
865
898
922
994
999
1007
1013
155
159
209
350
43
747
763
946
1139
1140
1148
1172
317
418
761
827
835
103
1078
1161
1162
1165
116
134
165
168
242
27
291
39
545
563
612
625
626
725
903
1163
1164
1166
21
271
351
396
452
488
601
674
786
823
852
859
877
928
979
1175
139
141
174
203
426
436
493
614
619
637
638
72
758
926
1029
1058
1071
117
138
181
366
373
385
398
419
437
735
784
853
876
905
1133
158
211
23
453
548
579
642
643
64
657
73
804
933
281
288
376
508
641
6687
7
844
951
998
108
1174
1195
390
42
585
658
744
785
788
807
897
1059
153
161
163
173
447
869
878
110
258
410
664
738
740
813
814
935
980
984
1142
129
137
450
507
71
83
1187
30
361
449
471
485
832
162
298
304
3
6
424
457
624
985
188
312
334
469
47
481
661
663
311
38
423
474
609
662
634
731
764
1022
212
46
6
65
789
88
105
566
660
739
86
25
1042
1194
152
229
299
329
353
464
478
528
578
667
1070
265
290
374
486
774
845
904
1177
475
503
736
10
33
408
462
514
778
170
175
178
24
346
613
167
416
444
49
808
187
443
790
84
11
273
316
400
473
61
703
791
794
463
307
465
971
238
336
721
470
1023
133
466
1131
1132
197
37
857
1130
166
582
1141
206
564
780
8
383
1121
335
442
621
96
655
678
177
326
553
1035
28
68
106
136
636
792
1190
321
370
567
22
289
297
599
1122
125
581
854
482
580
675
795
875
512
833
48
899
92
107
1101
146
386
793
347
608
798
1064
124
344
879
1105
164
421
5521
70
123
198
557
688
1061
622
499
623
176
906
130
858
1152
126
587
722
349
417
568
69
122
148
666
940
180
320
707
750
760
358
652
506
121
522
149
527
1093
1200
519
892
384
1092
1128
569
1127
572
1114
1126
884
343
818
472
586
345
1198
749
79
118
85
921
886
920
1124
773
171
135
369
1199
573
874
812
881
9
918
981
934
1021
919
890
240
217
855
1030
1033
441
843
20
910
413
836
1031
610
491
550
944
1106
202
704
1024
215
615
532
909
1107
78
900
169
352
771
1015
19
391
18
127
399
776
218
392
955
1
575
1019
365
887
228
2
480
602
498
1025
635
1072
405
762
1180
150
757
973
894
718
241
523
74
895
494
81
17
214
526
227
496
324
972
16
368
338
914
7734
492
10341091147 143
55
1026
737
544
948
1120
259
495
282
−0.02
Budget * Defection
Observation
−0.006
1180 218
575
1025
757
973
206
522 622
623
164
1024
1031
1072
405 519
1030
1033
228
365
955
854900944
392
812
599
18
19
1199
1015
1107
532
215
615
771
416
553
1198
704749
202
1106
550
621
343391
491
836
20
572
512
569
887
843
16
919
855
441
11042
918
981
890
1093
1114
1200
652
217
9
881
127
106
1194
573
666
135
886
909
773
85
920
921
238 329
587
1141
739
105
413
910
46
884
934
499
84
149
506
344
707
750
609
178
148
1056
349
79
171
1101
475
586
222
1055
486
818
904
399
473
1070
557
384
152
478
985
527
1054
1061
170
25
259
1105
655
358
180
24
760
940
610
86
69
121
635
172
485
580
568
1124
762
899
48
736
122
126
374
417
1126
858
602
1152
239
578
738
130
980
1021
466
5
176
906
1128
774
1127
518
352
1190
1023
123
688
136
423
369
70
186
182
465
1122
520
879
188
984
1064
1019
798
347
705
463
474
793
334
386
608
146
345
92
107
874
833
675
482
731
44
875
1177
125
764
78
36
795
564
624
1130
124
781
297
289
1131
1132
321
462
1121
567
370
636
133
792
326
333
68
469
434
678
96
1195
335
383
780
137
667
694
464
8
785
166
740
376
110
582
187
444
177
198
951
288
813
814
508
857
258
449
37
1187
197
320
442
892
470
408
83
212
914
450
336
878
998
581
1022
307
971
948
844
791
794
316
273
703
400
61
521
790
443
1035
897
167
808
49
1026
346
153
175
33
613
10
514
109
579
108
1133
778
390
503
869
845
817
548
265
290
229
299
528
734
11
353
471
324
47
933
566
660
241
607
6
496
789
81
1147
373
65
88
282
424
634
662
38
311
905
312
1058
481
366
661
663
1029
950
718
304
544
162
298
30
4
64
832
361
150
507
1120
129
641
143
453
721
118
23
338
368
664
385
410
437
1059
161
457
5
876
138
163
173
181
240
494
676
658
17
788
807
902
214
735
585
73
642
643
281
657
6
7
74
935
480
722
804
436
523
776
1071
493
1142
758
158
28
7
972
1092
784
398
91
426
71
853
165
472
1079
619
421
203
637
638
140
396
5
7
786
1175
966
351
452
926
199
674
903
103
139
169
196
891
447
612
1078
498
846
39
761
317
827
285
371
1007
1013
159
852
928
264
40
614
672
994
306
697
230
309
435
823
830
831
865
999
362
859
1011
856
826
559
1005
1034
174
701
990
993
838
134
800
1010
160
633
829
72
1008
1012
1197
454
801
337
53
877
1094
1000
271
300
42
201
611
1196
308
828
1001
1006
995
964
889
116
305
76
1003
1009
787
989
867
268
1014
958
968
992
539
75
1174
319
157
488
561
1181
626
95
625
723
991
702
545
700
1002
1004
227
34
880
1166
775
996
1067
782
895
537
1163
515
979
538
1164
204
744
87
314
946
274
516
517
43
592
1159
835
200
168
293
1113
1161
339
318
671
913
779
62
232
670
860
322
89
618
961
716
850
296
29
960
117
295
112
291
93
1165
155
495
725
1145
1162
315
387
479
956
563
209
930
97
310
868
535
533
32
440
959
526
898
927
94
1148
894
715
31
460
536
1100
907
677
119
534
863
360
748
1139
3
562
862
27
292
861
270
429
82
141
987
883
954
102
431
629
982
190
1140
451
418
1050
983
236
584
287
967
211
962
501
690
350
549
763
588
783
21
720
747
901
101
659
851
882
885
589
932
698
1158
439
283
945
456
628
1053
1173
839
26
640
717
922
147
699
796
1172
821
1160
866
627
367
492
1129
372
570
695
965
719
438
542
822
428
665
1137
737
397
546
104
142
1138
645
630
583
448
825
949
313
969
590
631
997
1073
591
978
598
639
840
646
1020
225
929
433
1134
687
1169
459
476
269
571
1091
815
673
942
841
1065
1167
41
834
943
752
947
963
864
551
286
759
957
266
502
679
411
1057
644
648
114
692
401
242
189
1168
145
348
404
820
131
915
500
216
649
797
207
267
1171
1170
1016
144
668
708
1115
1136
389
504
1125
524
205
226
696
916
1066
1135
1017
222
208
577
132
249
754
210
427
1178
192
60
221
223
689
234
63
154
224
1096
824
430
525
691
647
547
977
601
543
195
445
650
988
100
403
419
59
742
402
191
706
505
235
284
58
251
497
156
574
576
1103
193
363
917976
194
552
746 819
651
669
654
332
458
184
1051
325
99
848 90
1123
179
477
93198
709
653
727
237
467
255
388
257
446
111 1186
594
375 45
1063
256
1176 185
490
Defection
−0.001
−0.003
Budget
0.001
91
182
434
1
11 12 2
29 39 49 59 69 79 89 99
Observation
Figure A.1: Influential observations for the tobit model with AIn as the dependent variable. The dfbeta values are small, therefore observations with leverage are not a major
concern. I have removed the 30 of the most influential observations in this case and
re-estimated the model. Removing these observations did not impact the substantive
interpretation of the models.
1018
2 3 4 5 6 7 8 9
0.00
0.006
259
214
737
324
948
74 81
544
241 338
7
1147143
1026
895
165
368
399 492 610
16
1021
734
494
496
762 846914
419 520
264 345
369
109
526581
1034
602
1502 28320
972
375 480 580 718
198
118171
72178 874
887
909
930
527
206
11042
227
384
522
4 482
519
164
776
386
964
760
722
688
599
79
199
567
107
416
854
146
906
358
8838
92
512
568
608
5
613
127
240
329
175
739
105
130
106
370
84
176
383
934
675
798
498
1152
1141
400
634
68
443
971
985
566
442
6
33
177
1035
528
61
326
478
904
121
169
347
11
738
46
188
578
475
486
457
586
417
426
65
636
984
178
311
316
818
307
894
166
10
212
86
697
229
25
140
609
317
452
521
299
124
304
449
660
129
853
1059
24
966
258
469
905
928
75
197
663
298
662
661
702
807
238
271
514
658
672
761
230
508
664
47
612
857
196
268
319
312
903
437
891
1015
353
550
493
553
1053
281
88
548
161
83
900
1072
725
757
955
1133
405
740
155
309
365
410
845
265
273
318
671
890
723
133
855
116
362
385
537
707
773
973
392
453
1176
843
1106
1107
297
376
951
1105
148
201
391
158
321
396
771
85
228
669
670
96
20
314
749
844
552
95
1131
1132
704
858
215
490
532
9
202
36
446
491
1061
1121
149
503
880
899
715
886
1130
173
22
23
615
674
836
179
1122
441
585
64
1063
1178
564
750
918
122
163
413
209
881
1051
1120
445
780
162
431
981
557
944
18
19
831
89
919
363
48
827
830
884
563
110
174
87
920
921
217
982
100
1050
349
977
1142
829
978
869
657
865
867
659
21
315
429
639
644
826
93
641
648
828
833
135
274
516
645
325
637
638
640
667
808
94
1023
232
436
587
647
813
814
1022
1190
642
643
310
374
464
466
517
666
788
868
97
463
573
67
108
1100
450
465
538
650
651
931
343
462
515
646
649
835
1129
344
789
792
954
117
534
607
654
1096
200
653
136
717
786
30
444
447
533
536
569
852
926
160
572
744
656
1114
159
535
539
1064
588
1101
1200
289
37
839
1033
1199
1093
187
3
790
848
1166
408
423
506
1024
1198
1025
1058
1180
785
883
946
1161
440
485
499
979
1031
1113
1030
879
1164
507
793
935
1162
687
692
860
1148
1163
40
582
862
915
1165
1194
218
291
748
698
804
812
575
922
459
863
348
655
997
1070
823
137
366
943
1159
332
472
945
1186
652
1103
976
350
614
1139
288
958
69
703
822
211
373
70
1181
139
875
747
859
933
1195
701
980
1071
631
864
949
983
125
123
473
474
592
882
929
495
1134
579
916
1065
696
861
927
1140
418
1067
439
471
38
907
1137
1138
62
1073
168
334
424
999
210
791
438
141
192
60
629
774
361
834
226
618
689
956
1010
27
577
225
389
26
545
1135
1136
138
360
387
758
825
998
401
787
221
885
959
1057
223
524
625
700
1007
1145
1169
152
404
430
525
764
1029
390
597
623
628
630
234
598
114
222
286
372
42
157
216
339
624
626
627
821
960
1011
103
43
504
746
1078
1168
207
208
224
547
621
775
877
101
249
256
731
242
285
742
876
942
367
543
549
1008
1167
1187
181
448
145
255
622
841
205
699
76
1173
17
435
505
897
1066
1170
1172
204
235
562
195
31
523
720
763
961
1005
709
797
819
1012
170
190
203
779
815
589
989
283
488
574
591
63
1017
1175
851
1000
990
1016
1019
167
476
796
992
995
144
590
735
820
1013
251
840
102
156
993
502
1003
134
898
477
584
1001
1006
1014
500
1002
1004
878
917
576
619
994
290
351
236
336
292
1174
153
59
678
41
679
1009
119
191
193
458
58
90
45
782
194
257
337
708
784
34
736
111
284
716
996
403
824
1197
542
559
571
594
991
402
551
39
1177
913
1196
727
965
335
850
55
293
611
778
1115
470
801
1094
969
800
759
988
901
398
889
180
428
1158
460
695
456
154
570
783
967
29
501
546
104
346
481
962
270
296
397
690
49
832
947
71
295
940
910
1171
479
795
287
300
583
322
132
561
313
1091
719
269
932
131
72
1160
794
112
126
182
142
963
427
57
1020
147
665
454
1018
433
189
752
673
266
411
267
957
754
331
668
595
691
282
184
237
593
706
467
596
99
866
421
98
91
185
434 53 601
1123
635 73 82
1092
1125
388
352
1128
1124
1126
892
1127
306
950
817
633
1054
1055
1056
694
1079
371
856
676
497
902
333 44
172
677 781
186
239308
32 451518
705
305
968
987
1 113
2 3 4 5 6 7 8 9
Observation
1056
1055
1054
987
140
705
968
32 451518596 677
308
239305
186
172
781 891966
333 44
593
595 694
196 331
1018
723
902
856
676
371 458
633
1079
950
306
594
727
497
111
257
45
90
817
1127
1120
892
477
1126
1124
53
709
1128
352
255
746
256
669
1063
597
742
388
490
446
387
834
635
1067
375
82
445
592
1071
1051
91
1092
1181
1103
976
1186
332
363
706
691
668
184
754
421
957
266
267
189
57
673
752
866
454
657
411
963
126
433
131
147
269
1091
132
665
1020
427
142
583
112
72
910
794
940
49
690
832
932
270
313
719
346
104
29
73
287
456
408
659
71
180
296
397
947
988
1160
506
962
481
546
561
322
759
783
967
969
470
300
639
795
644
889
1115
295
848
335
398
901
187
39
800
501
1023
570
460
571
695
1158
293
641
648
428
850
965
1094
645
801
236
290
351
716
34
611
778
1196
402
551
584
784
913
1022
102
119
325
542
782
292
403
590
640
667
678
775
824
1197
284
336
559
591
736
991
194
637
638
708
996
1171
191
193
337
58
720
961
153
699
1009
1177
167
576
589
779
851
1001
1002
1004
1006
1014
1078
134
156
251
619
994
157
204
476
479
647
735
796
820
960
1003
1013
1175
170
574
63
993
1000
787
819
959
990
992
995
1005
1012
203
205
235
283
360
505
562
587
989
1187
435
448
618
841
956
101
367
543
59
679
700
731
76
1008
216
339
43
500
504
547
791
878
917
1011
1019
103
136
430
524
525
764
898
1016
1145
144
154
223
401
502
577
840
861
885
1007
1017
190
225
26
642
643
689
825
1010
1066
1170
1173
125
192
195
31
666
797
897
907
1167
145
181
210
361
549
622
62
701
1168
123
38
621
624
626
69
703
70
821
942
999
1057
1134
1169
1174
152
234
286
390
41
473
598
623
625
627
628
630
875
882
927
958
983
1135
1136
224
545
758
114
222
226
42
629
822
1073
221
334
464
863
1065
1159
439
466
698
804
916
943
1172
27
348
40
463
488
631
763
793
815
860
862
207
208
249
424
459
582
876
877
1029
1129
137
138
372
389
404
507
692
823
945
998
1070
1113
1137
1138
168
438
60
879
997
1140
418
579
696
748
790
864
915
929
946
949
1064
1139
1195
139
141
159
288
289
3
350
373
440
535
539
614
747
774
859
883
933
980
1163
1165
1194
160
291
366
37
423
588
839
922
1058
1148
1161
1162
1164
1166
200
30
465
485
533
536
687
785
786
935
979
242
534
792
852
926
108
1100
450
515
538
789
835
954
310
374
436
517
67
788
813
814
868
931
97
232
471
474
516
563
650
808
94
110
211
315
419
429
826
828
833
869
93
1050
651
844
865
867
982
209
376
573
649
829
951
1133
385
725
87
343
36
431
548
601
646
717
744
827
830
117
162
462
585
831
89
1190
23
508
64
780
116
21
444
155
163
274
413
740
905
95
1142
173
201
344
447
674
314
493
503
553
609
715
158
362
453
858
174
265
410
437
670
96
297
321
396
281
88
984
122
238
318
353
83
537
654
161
273
309
46
671
738
845
258
928
612
857
135
457
521
672
319
903
985
25
197
230
312
86
1035
268
24
469
449
702
75
178
664
761
452
47
105
514
653
658
739
129
807
237
486
661
662
298
329
475
663
894
10
1059
188
317
467
478
660
853
904
271
304
697
99
299
978
426
65
977
416
212
229
349
100
311
316
1096
569
98
124
166
578
838
920
921
307
572
586
1114
818
84
1141
557
566
636
656
884
443
68
177
400
1152
528
169
48
6
417
347
206
326
971
61
634
18
19
33
11
121
217
919
944
106
1125
854
383
442
934
240
370
5
1033
498
981
613
881
599
798
1123
608
564
675
568
176
750
1178
512
130
8
918
164
179
615
146
358
107
199
567
836
127
149
482
880
202
580
1130
175
886
899
1061
79
491
532
607
704
9
874
215
1131
1132
1122
185
4
1031
441
776
1030
552
1121
148
198
228
749
92
22
20
760
85
906
386
1106
391
1105
1107
1200
1101
519
843
1176
522
688
771
218
973
773
1199
855
392
499
890
365
707
722
964
1093
405
757
527
133
345
655
955
171
575
1072
384
900
1053
1024
369
1015
652
1198
480
550
1025
11042
2
812
1180
718
930
227
972
909
118
1021
887
610
399
581
78
721
150
28320
526
109
520
264
846
494
602
1034
1026
368 472
914
492
165
7734
76281
895
1147143
496
338
241
544
16
74
948
285
17214
324
523
49555
737
259
182
−0.02
Budget*Defection
0.02
Observation
−0.008
−0.004
331
595
593
596
1 113
−0.002 0.002
206
522
519
164
601
812
1180
652
10251125
1198
599
575
1024
854
1123
655
416 499
512
16218
1093
1042
550
1199
887
1015
900
1053
329
1101
98
1072
1200
955
106
1030
1031
909
739
105
757
84
133
762
707
405467 552
365
392
99
890
1141
855
773
1
973
185
320
1176
771
1105
843
391
384
1107
1106
85
237
602
20
985
749
171
228
581
148
22
399
578
610
478
904
1121
527
441
215
1131
1132
188
1033
259 386 46
580
1122
475
992981
738
1056
532
491
704
486
688
1055
1061
886
899
1130
202
880
1054
149
836
906
179
918
760
1021
1178
615
178
750
984
212
2
721
564
781
86
239
186
881
656
79
25
175
369
78
482
518
521
172
567
107
609
146
358
919
130
449
572
944
18
127
19
24
8
469
345
569
176
258
928
675
1096
217
568
905
48
608
798
5
1114
613
47
238
370
383
934
930
557
198
884
44
442
333
33
508
61
705
326
971
920
121
921
437
11
347
100
6634
653
1152
528
349
493
400
694
553
83
417
68
443
977
636
548
978
874
566
434
177
818
740
241
725
155
307
586
1133
166
116
453
182
311
316
385
229
426
65
324
124
299
914
304
951
376
697
948
660
734
853
317
654
1059
844
81
10
663
75
298
36
662
661
271
23
807
129
282
514
658
109
452
761
64
950
135
664
1147
702
268
817
496
312
209
197
230
1026
902
431
344
319
903
672
676
612
857
264
309
845
273
671
161
28
537
1142
118
214
55
353
318
281
285
88
297
964
462
1035
150
17
321
396
563
670
96
110
410
91
982
265
21
158
646
858
362
544
343
457
314
74
1190
503
715
173
674
338
201
869
143
163
413
95
718
523
368
722
122
649
780
494
447
585
651
831
846
89
162
827
174
444
830
573
87
472
829
1050
813
814
867
436
480
865
240
650
838
429
826
117
274
93
315
833
828
516
776
374
94
199
7
232
808
931
972
97
310
517
788
868
67
538
108
450
515
717
789
1100
169
792
498
534
935
835
891
196
200
954
1079
300
57
607
1092
744
786
30
533
536
421
160
1120
37
588
465
159
3
535
539
894
1064
289
790
4
1019
879
1113
926
507
852
582
1034
371
793
860
227
1159
40
862
804
698
1129
863
723
966
839
875
927
140
38
983
958
123
999
479
69
703
70
1166
423
306
361
82
1010
701
1115
125
907
785
1058
633
1007
463
1145
1161
861
979
485
103
883
1011
308
969
946
440
1008
466
791
700
76
435
1164
305
618
956
464
1005
1012
203
1162
687
988
968
989
283
360
1148
990
1000
787
995
1163
959
992
1013
1194
856
1165
993
291
1003
692
915
53
748
1001
1006
204
922
1014
157
960
1002
1004
619
994
1078
136
779
775
589
851
1009
167
699
337
997
1070
459
961
720
996
348
366
1197
991
559
591
336
666
292
823
678
137
211
590
119
102
526
945
782
913
350
895
1196
614
943
584
784
1139
34
611
288
642
643
778
495
716
290
351
236
373
801
1094
747
139
474
859
933
1195
980
822
864
949
965
428
850
293
1158
460
695
631
706
471
929
571
570
579
501
587
696
1140
39
520
800
418
677
901
882
916
473
398
295
335
1065
1137
1138
889
1134
439
795
168
424
141
774
438
470
647
1073
322
334
60
561
783
967
759
546
62
629
481
1160
451
32
962
27
389
397
210
296
226
947
998
138
180
192
545
456
287
689
221
758
1136
404
1135
1029
577
29
225
637
638
242
1022
26
104
372
625
825
1057
346
152
1169
313
187
270
719
222
628
401
42
492
71
390
623
987
114
877
630
690
932
165
49
598
832
885
207
208
249
794
223
234
667
848
640
184
286
524
940
876
627
224
525
764
430
910
624
821
325
626
1168
621
583
942
216
339
1020
43
549
504
547
181
665
1167
101
132
731
1172
622
145
1091
763
72
269
635
112
543
433
131
367
645
506
1187
448
126
142
1173
897
815
841
1170
1066
963
195
488
31
408
427
205
797
505
190
147
235
562
1023
1017
819
641
454
144
170
648
1016
866
63
411
574
189
752
840
673
1175
266
502
267
957
476
754
796
735
898
820
251
156
668
500
1174
917
691
878
134
41
73
59
644
576
679
639
153
191
193
58 659
194 284352
708
736
403 542
551
402
1177
737 824892 976
154
657
332388
1171
1186
1103
1128
1181
1124
1126
1071
497 592
363
1127
1067
834
387445
1051
597 709
742
256
746
255
477
446
490
1063
90
419
669
458
45
257 375
111
594 727
Defection
−0.002
0.000
APPENDIX A. INFLUENTIAL OBSERVATIONS
Budget
92
282
434
1 113
2 3 4 5 6 7 8 9
Observation
Figure A.2: Influential observations for the tobit model with RICE as the dependent
variable. The dfbeta values are small, therefore observations with leverage are not a
major concern. I have removed the 30 of the most influential observations in this case
and re-estimated the model. Removing these observations only slightly impacted substantive interpretation of the models. The model remains supportive of the hypothesized
relationship.
597
−0.004
331
1 113
593
595
596
2 3 4 5 6 7 8 9
Observation
0.015
16
762
602
0.005
320
−0.005
812
519
1025 1180
522 652
206
1024
164
218
1093
1199
655
499
1101
601
1125
854
1198
575
550
405
599
512
1015
1123
416
365
16
900973
757
887
1121
1031
1072
1106
106
1122
1107
1042
1200 215
1053
1030
329
707
1105
238 320
955
855
1176
739
704
105
762
228
98
890
909
1033
771
750
1141
391
773836
11054
148
18
202
441
84
217
944
843
19
467
392
99
149
602
749
1056
237
1055
384
1178
25
532
24
615
656
580
85
473
447
171
992
578
985
981
653
557
760
475
478
20
610
527
1061
581
358
491
178
176
399
688
904
899
906
188
1114
417
259321
469
738
572
486
121
86
721
46
980
369
2229
135
781
5 553
386
36
347
212
79
130
1021
239
172
1100
609
64
573
186
78
47
518
107
122
182
520
146
608
521
258
654
666
567
127
586
383
370
462
485
569
482
179
568
984
343
8
110
1190
175
326
675
68
96
731
886
774
11
705
930
345
740
449
881
579
123
198
582
177
442
333
273
108
844
61
798
434
44
971
934
813
814
905
83
33
444
373
197
918
613
134
1152
1064
465
307
1096
10
869
450
23
463
921
1142
62
785
998
780
6
634
548
946
853
919
311
566
874
503
951
508
928
858
914
390
353
804
316
241
464
845
948
663
129
636
88
289
662
466
1133
587
660
661
297
312
109
437
528
400
1035
299
443
857
366
324
298
493
1078
181
1058
65
506
81
694
658
124
282
507
514
817
818
933
304
697
271
457
496
349
453
664
481
734
1147
902
398
1026
426
150
67
920
612
166
884
676
203
155
268
903
410
646
950
167
784
545
718
436
544
55
1007
715
118
935
362
607
143
240
116
338
368
537
725
264
290
325
351
337
926
309
885
1013
214
611
385
75
157
230
1011
563
1132
999
494
317
431
846
376
480
722
28
958
883
1012
74
265
285
413
619
989
990
1010
103
523
89
776
993
1005
1059
17
319
452
91
204
782
27
559
7
800
1001
516
852
672
716
882
1000
1092
966
972
807
994
995
1003
199
764
1008
702
827
201
979
161
169
838
87
992
57
1006
517
374
1014
1079
421
761
779
964
674
1004
30
498
789
196
891
95
37
1131
830
515
209
533
786
1002
538
826
991
671
788
829
140
1009
894
472
38
831
163
371
1120
158
633
988
429
1161
281
1034
796
173
534
670
828
361
162
94
961
227
922
536
435
723
76
996
1130
200
790
535
174
53
306
117
1159
300
1163
360
97
318
396
982
744
93
1115
315
293
539
236
470
4
879
1166
274
314
346
792
823
793
270
969
585
1162
308
310
1164
945
70
82
26
1050
865
211
305
895
40
618
875
440
968
801
700
833
868
1165
867
339
856
808
159
956
160
588
791
119
701
717
954
22
562
589
479
703
835
959
42
960
787
690
907
1113
1145
34
48
978
1139
699
232
3
692
913
927
69
125
334
21
336
698
863
862
350
860
651
706
225
73
136
649
983
677
335
967
423
783
861
1187
456
1023
977
614
1094
180
39
526
642
643
459
184
451
291
29
283
32
1148
851
1197
775
590
102
778
348
678
460
839
495
139
344
292
850
584
591
1073
965
889
987
748
1158
1196
439
687
949
650
915
720
488
1070
901
929
631
822
288
501
397
571
322
269
49
295
641
695
916
962
570
428
63
1195
408
1134
165
1160
424
43
747
168
598
997
137
1194
752
1103
665
943
1022
1140
141
648
287
561
546
126
372
222
859
1019
667
795
112
492
947
1029
910
583
635
454
226
821
504
104
1065
210
866
138
430
825
689
101
114
1138
759
841
897
223
864
296
192
877
471
438
147
763
976
221
60
940
224
736
1091
719
957
401
313
623
932
72
794
205
474
418
71
1137
1174
625
152
876
1020
145
448
696
832
840
144
577
286
630
352
737
187
637
638
170
132
629
142
963
626
628
1169
208
627
549
433
525
242
207
266
524
267
131
249
552
1016
133
234
621
624
797
389
547
1135
647
758
190
691
1057
216
1066
427
367
404
673
543
411
500
815
1167
898
679
1136
892
1172
1175
1168
668
1173
819
942
622
1017
657
58
189
644
645
754
195
502
235
156
31
153
708
476
824
564
735
1170
191
41
505
251
574
1181
284
820
1171
917
640
878
834
59639
1186
388
193
403 497
194
402
576
880
1071
551
1177
1128
1129
1124
1126
154
1001067
387
445 542
931
592659
1127
709
742
746
477
255332
669
446
419
84890
458
363
45
375
1051
727
257
111
256
1063
594
1018
185
490
Defection
0.000
−0.002
Budget
93
1079
781
72178
676
419
581
902
856
930
259
737
496
914
948
544
497
324
81
338
368 480
214
492
1056
1055
241
165
1147143
1054
109
150
526
494
846
74
734
718
895
240
264
812
776
652
1025
1180
434
118
218
722
655
1024
550
575
499
1093
1199
754
1101
154
433
757
189
28
405
900
668
866
1198
1020
182
673
1015
1072
691
794
932
313
795
131
267
266
719
561
973
411
963
132
427
1031
365
1106
296
479
1160
832
1107
707
1053
142
428
570
1105
695
295
1091
1122
322
955
957
1030
1121
947
551
940
215
501
546
71
1177
287
72
665
759
1176
104
1200
147
1158
1196
402
454
576
583
59
73
836
194
228
542
855
910
962
704
126
391
403
193
878
917
1033
397
750
752
18
1170
773
890
112
901
284
460
574
820
191
19
251
505
778
1171
31
392
1197
269
49
195
283
441
502
735
41
476
571
708
749
771
824
843
148
153
815
889
965
1017
156
202
235
404
1173
1178
282
389
622
850
942
1168
207
249
292
1136
1167
208
500
58
679
898
944
1057
1066
1094
190
577
720
819
1135
1175
149
621
624
678
758
797
1016
234
696
85
1169
216
367
532
543
549
626
627
1137
217
547
628
629
851
876
1172
418
524
584
591
775
981
145
242
286
525
630
656
840
170
20
152
438
557
60
615
625
763
9
448
623
864
877
102
1138
1061
205
590
653
138
29
401
474
783
491
736
897
967
1029
983
1065
1092
1174
192
372
859
1140
144
180
223
226
39
471
821
841
899
101
1194
210
430
689
825
168
456
504
861
1195
927
137
335
598
913
997
221
224
3
447
572
1145
336
943
288
916
1134
114
43
135
488
63
929
949
1070
141
631
687
860
862
863
222
55
698
822
915
1113
34
439
748
907
1114
125
495
573
69
747
808
1073
139
424
654
1100
179
232
462
569
589
699
1148
119
291
588
666
886
690
787
801
959
960
343
348
703
839
1190
614
459
700
701
791
160
956
159
423
875
881
1050
1187
585
867
833
868
918
350
40
618
865
879
199
225
310
465
334
70
1096
1139
1159
314
463
293
315
692
919
921
270
318
346
792
793
835
236
470
539
76
954
996
1165
360
396
435
444
464
466
562
93
97
339
361
587
21
440
506
535
1162
1164
162
200
26
281
790
274
38
536
945
94
961
1009
1166
158
173
349
429
523
534
670
823
828
1163
163
42
920
982
991
1002
646
717
831
884
285
538
671
788
826
829
922
1161
211
37
515
533
786
796
95
1004
1132
174
325
789
830
1006
1014
209
30
674
761
779
966
161
374
517
87
992
1003
1008
201
702
744
764
807
827
882
979
1000
1131
117
516
672
716
852
994
995
1001
17
800
1059
1130
204
559
782
883
89
1005
319
376
413
431
452
75
993
1010
103
563
619
885
958
1012
116
265
385
926
989
990
725
27
48
545
978
317
436
607
157
651
723
838
1023
136
140
155
230
290
309
351
472
649
999
1011
537
977
642
643
715
1013
1120
22
453
611
337
344
362
933
1007
650
1022
181
268
493
641
1058
167
437
784
366
166
408
648
667
1133
903
1103
390
928
964
410
508
946
664
951
196
203
398
62
304
637
638
818
976
187
23
421
514
548
891
450
481
612
647
67
869
998
657
658
785
134
426
697
133
507
552
644
645
83
1078
1142
271
564
298
373
813
814
299
124
312
905
108
297
1181
640
660
661
91
289
449
662
844
636
663
834
845
528
731
639
740
804
858
880
579
1186
503
65
110
129
853
316
659
443
857
400
774
780
984
1071
258
311
521
485
64
387
88
6
634
1067
445
1129
353
1064
307
47
57
609
33
931
100
566
592
212
36
798
61
469
229
273
486
86
709
553
935
46
980
188
934
971
738
442
10
904
321
178
1152
613
96
197
326
582
675
175
478
123
578
475
11
473
482
742
985
25
746
457
177
24
370
383
586
130
68
92
669
84
127
255
8
446
477
848
568
347
122
90
363
567
458
1141
121
608
332
146
53
107
1035
906
1051
386
417
727
45
176
688
105
238
988
739
79
257
106
5
329
1063
256
1042
760
384
111
512
594
300
358
416
599
1018
527
894
82
854
1
171
633
909
164
490
817
874
198
522
519
206
950
887
371
331
580
169
345
969
597
306
1115
375
694
596
593
595
2239 333
1021
399
369
610677
44498
185
184
520
186
706
172
308
451
32
305
518
987
968
705
227
972
237
99
467
1019
635
98
4
352
388
1123
601
892
1124
1126
1128
1127
1026
10341125
7
1 113
2 3 4 5 6 7 8 9
Observation
0.008
434
0.004
164
0.000
Budget * Defection
206
331
16
182
259
171
596
595
593
519
522
887
597
854
416
909
599
705762
527
358
512
781
497
987
968
518
329
305
490
760
739
185
105
323844515 602
106
320
1018
610 73779
238
239308369
594
172
399
948
902
186
544
7
1034
520 608677
107
1021
1141146
176
111
386
688 812
154
324
906
333
2 256
282
417
44496567
676
568 652
257
1026
1063
121
81
143
8 84
338
368
68721
914
214
347
122
45492561
1125
433
1180
1025
866
985
727
127
856
754
25
24
586
130
92963
383
78
189
370
419
1127
1020
795
458
668
332
1128
345
794
932
177
313
1126
580
90
1051
673
719
1124
411
691
241
427
375
848
550
165
131
267
1160
218
1079
475
266
892
1024
296
1177
478
601
1147
473
570
132
255
363
695
142
655
295
578
551
178
322
832
405
1123
738
477
482
499
904
1093
501
197
1199
947
495
575
613
71
59
1101
123
1015
1152
109
46
72
980
900
757
930
446
188
546
582
957
1091
287
10
940
576
669
526
665
402
175
878
194
542
150
326
1198
675
147
917
553
96
1072
1196
494
1158
1171
365
403
193
581
198
973
759
486
11
86
874
321
1170
846
746
41
104
4428
1106
962
971
1121
388
934
352
36
469
454
1122
583
1107
910
31
609
1053
820
284
574
815
404
397
126
251
502
195
707
112
505
191
460
389
778
249
752
1031
371
442
1105
207
735
742
1017
901
273
955
476
824
208
622
942
229
1197
708
242
1173
566
1168
153
283
98
61
156
215
1136
679
798
1172
235
1167
898
500
212
696
1066
1176
1057
485
480
190
1019
758
1135
876
624
797
855
621
1016
1137
234
49
269
418
58
1030
704
836
889
353
965
819
474
549
1169
627
33
626
55
628
521
1175
228
258
571
629
850
1094
1200
891
635
984
577
292
64
60
763
543
367
286
630
438
216
709
1064
74
840
88
145
6
307
634
877
47
391
547
734
864
750
196
890
972
152
471
524
625
678
1138
525
720
771
1174
623
138
718
773
817
170
18
694
851
311
467
441
99
895
110
448
400
579
1029
443
857
144
19
1033
372
859
897
148
591
205
584
592
306
392
237
202
227
931
774
843
1140
401
100
316
1065
1194
780
740
217
736
65
168
749
944
821
226
129
192
775
844
224
221
1178
1195
223
853
731
149
783
102
1129
841
590
387
967
101
689
430
825
1067
210
29
503
114
905
950
983
504
240
137
997
598
532
449
141
85
264
528
108
300
858
288
373
445
804
927
498
39
180
615
845
913
916
981
929
488
943
222
3
1071
9
663
813
814
1035
636
456
20
289
662
861
949
687
1145
43
1134
335
557
660
1061
63
661
747
336
631
424
656
297
491
1070
312
83
880
785
299
915
1186
998
659
776
439
139
298
748
822
899
134
548
653
507
1078
426
834
1073
633
869
447
1148
697
450
82
860
291
639
1181
67
862
34
863
951
1113
23
612
698
808
1120
907
508
481
658
607
124
457
62
1142
135
125
614
69
118
589
640
203
564
514
801
398
818
839
946
119
348
304
572
928
232
479
1133
410
588
699
390
706
271
1114
133
552
664
690
787
959
960
366
459
423
573
723
645
703
184
644
187
700
1058
140
701
1187
437
657
73
791
28
350
903
179
647
160
875
956
976
585
1100
159
493
166
181
886
637
638
654
935
1050
867
167
1190
462
569
666
784
225
833
868
879
933
1139
334
268
53
1007
310
343
40
618
881
1103
722
865
21
362
408
1159
169
70
314
523
692
835
667
293
648
918
337
954
315
453
1013
1022
1165
318
611
270
562
641
996
1115
346
76
792
793
435
969
236
465
470
539
1011
650
715
921
93
1096
155
274
339
361
396
440
57
919
360
463
537
75
97
999
1162
1164
285
344
42
230
281
290
351
444
162
26
309
436
535
945
1166
157
200
642
643
717
790
977
1092
136
211
317
38
464
466
536
545
823
894
1023
1163
158
506
587
649
725
828
94
961
1009
116
173
472
534
651
670
926
1012
163
17
349
385
429
48
922
978
982
991
1002
1010
1161
265
563
619
831
885
920
91
958
988
989
990
1005
103
376
37
431
452
538
671
788
796
826
829
884
95
966
993
1001
1004
1006
1014
1059
1130
1132
199
204
209
30
319
325
413
515
533
559
646
674
779
782
786
789
800
830
883
89
964
1000
1003
1008
1131
161
174
201
374
516
517
672
702
716
761
764
807
827
838
852
882
87
979
992
994
995
744
22
117
421
27
1056
11055
1054
1042
1 113
2 3 4 5 6 7 8 9
Observation
Figure A.3: Influential observations for the tobit model with RICEabs as the dependent
variable. The dfbeta values are small, therefore observations with leverage are not a
major concern. I have removed the 30 of the most influential observations in this case
and re-estimated the model. Removing these observations had only a minor impact the
substantive interpretation of the models.
Appendix B
Tobit models without interaction
Table B.1: Tobit models with only budgetary implication. It presents the tobit models
without the interaction. Budgetary implications retain their negative effect on legislative
unity even in the absence of an interaction with defection of behalf of the party of the
rapporteur.
Dependent variable:
AI
AIn
abs(RICE)
(1)
(2)
(3)
abs(RICEabs)
(4)
Budget
−0.018
(0.011)
−0.022∗
(0.011)
−0.004
(0.014)
−0.015
(0.015)
Result
0.150∗∗∗
(0.022)
0.220∗∗∗
(0.022)
0.166∗∗∗
(0.027)
0.066∗∗
(0.028)
Legislative
0.051∗∗∗
(0.009)
0.056∗∗∗
(0.009)
0.060∗∗∗
(0.012)
0.070∗∗∗
(0.012)
No. Terms
0.007
(0.005)
0.012∗∗
(0.006)
0.015∗∗
(0.007)
0.007
(0.007)
Defection
−0.165∗∗∗
(0.016)
−0.128∗∗∗
(0.016)
−0.216∗∗∗
(0.020)
−0.226∗∗∗
(0.021)
EPP dummy
0.268∗∗∗
(0.023)
0.242∗∗∗
(0.023)
0.351∗∗∗
(0.029)
0.342∗∗∗
(0.030)
Constant
0.361∗∗∗
(0.032)
0.206∗∗∗
(0.032)
0.255∗∗∗
(0.039)
0.300∗∗∗
(0.041)
σ
−1.969∗∗∗
(0.023)
−1.968∗∗∗
(0.023)
−1.751∗∗∗
(0.023)
−1.706∗∗∗
(0.023)
Observations
Log likelihood
Wald Test (df = 6)
944
518.164
491.789∗∗∗
944
518.784
472.942∗∗∗
944
310.315
521.415∗∗∗
944
269.933
432.769∗∗∗
∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Note:
94
Appendix C
Heteroskedasticity
0.4
0.6
−3
2
2
0
−2
−4
3
●
●
●
●
●
●
● ●●
●
●
●
●
● ●
●
●
●
● ●
●●
●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
● ● ●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●● ●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●● ● ●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
● ●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●●
●
●
●●
●
●
●
● ●●
● ●
●
●
●
●
●
●
●
●
●
● ●● ●
● ●
●
●
●
●●
●
●● ●
●
●
● ●
●
●
●●
●
●
●● ●
0.2
0.4
0.6
0.8
Actual vs Pearson Residuals
Actual vs Fitted
−1
0
1
2
Theoretical Quantiles
3
−4
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.0
0.2
0.4
0.6
budget
0.8
1.0
●
●
●
●
●
●
●
●
●
●
●
●
0.4
−2
0
2
●
0.8
Normal Q−Q Plot
0.6
yhat
rp
2
0
−2
−4
1
●
●
●
●
●
Theoretical Quantiles
●
Sample Quantiles
0
●
●
yhat
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
−3
−1
●
rp
0.2
0.8
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
0.2
0.2
●
Fitted vs Pearson Residuals
yhat
−0.2
−0.6
rr
●
●
●
● ●
●
●
●
●● ● ●
● ●
●
●
●
● ●● ● ●
●
●
●
● ●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
● ● ●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ● ● ●● ● ●
●
●
●
●
●
●●
●
●
●
●● ● ●
●
● ●●
●
●
●
●
●
●
●
●
● ● ●●
●
●●
●
●
●
●
●●
●
●
● ● ●
●
●
●
●
●
●
●
●
●
●● ● ●●●
●●
●
●
●
● ●●
● ●●
●
●
●
●
●
●
●
● ●
●
●
●● ●
●
●
●
●
●
● ●
●● ●
●
●
●●
●
●
●
●
−0.2
0.2
●
−0.6
●
●
Normal Q−Q Plot
Sample Quantiles
Fitted vs Residuals
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.0
0.2
0.4
0.6
0.8
1.0
budget
Figure C.1: Test of heteroskedasticity for the AI tobit model. The figure shows that
heteroskedasticity is not a major concern for the analysis at hand.
95
Appendix D
R code
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
setwd ( ”d : /Workspace/R/ P r o j e c t s / CleanCode ” )
#s e t w d ( ”C : / U s e r s / A nd r e e a / D r o p b o x / T h e s i s ”)
l i b r a r y (AER)
library ( beanplot )
library ( car )
library ( f o r e i g n )
library ( l a t t i c e )
l i b r a r y ( lme4 )
library ( s t a r g a z e r )
library ( xtable )
library ( Z e l i g )
#### Read d a t a #####
#d a t a <− r e a d . c s v ( ”C : / U s e r s / A nd r e e a / D r o p b o x / J i n x / V o t e s / D a t a S e t . c s v ”) # remember t o
data <− read . csv ( ”D a t a S e t . c s v ” )
#v o t e s = r e a d . t a b l e ( ” v o t e s . t x t ”)
#c a s e s = r e a d . t a b l e ( ” c a s e s . t x t ”)
#B j o r n D a t a <− merge ( v o t e s , c a s e s ,
by . x = c ( ” c a se
i d ”) ,
a l l . x = TRUE)
#### F u n c t i o n s ####
# Extracting
the
parties
of
all
rapporteurs
for
each
case #
getRapporteursEPGs <− function ( epg1 , epg2 , epg3 , epg4 , epg5 , epg6 , epg7 , epg8 ) {
e p g s <− c ( ”GUE/NGL” , ” G r e e n s ” , ”S&D” , ”ALDE” , ”EPP” , ”ECR” , ”EFD” , ”NI ” )
rapEpgs <− c ( ” ” , ” ” , ” ” , ” ” , ” ” , ” ” , ” ” , ” ” , ” ” )
if
if
if
if
if
if
if
if
( is
( is
( is
( is
( is
( is
( is
( is
. na ( epg1 )
. na ( epg2 )
. na ( epg3 )
. na ( epg4 )
. na ( epg5 )
. na ( epg6 )
. na ( epg7 )
. na ( epg8 )
==
==
==
==
==
==
==
==
FALSE
FALSE
FALSE
FALSE
FALSE
FALSE
FALSE
FALSE
&&
&&
&&
&&
&&
&&
&&
&&
epg1
epg2
epg3
epg4
epg5
epg6
epg7
epg8
!=
!=
!=
!=
!=
!=
!=
!=
””)
””)
””)
””)
””)
””)
””)
””)
rapEpgs [ grep ( epg1
rapEpgs [ grep ( epg2
rapEpgs [ grep ( epg3
rapEpgs [ grep ( epg4
rapEpgs [ grep ( epg5
rapEpgs [ grep ( epg6
rapEpgs [ grep ( epg7
rapEpgs [ grep ( epg8
return ( rapEpgs )
}
g e t S t r i n g F r o m A r r a y <− function ( array ) {
i f ( i s . n u l l ( array ) ) {
return ( ” ” )
}
s t r i n g <− ” ”
f o r ( i i n 1 : length ( array ) ) {
i f ( array [ i ] != ” ” ) {
s t r i n g <− paste ( s t r i n g , array [ i ] ,
}
}
sep = ” , ”)
96
,
,
,
,
,
,
,
,
epgs
epgs
epgs
epgs
epgs
epgs
epgs
epgs
)]
)]
)]
)]
)]
)]
)]
)]
<−
<−
<−
<−
<−
<−
<−
<−
epg1
epg2
epg3
epg4
epg5
epg6
epg7
epg8
put
in
correct
path / use
s e t w d and r
97
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
return ( substring ( s t r i n g ,
2))
}
### E x t r a c t i n g w h e t h e r a p a r t y o f a r a p p o r t e u r h a s
g e t D e l t a B e t w e e n <− function ( m a j o r i t y , rapEpgs ) {
i f ( length ( m a j o r i t y ) == 0 ) {
return (NULL)
}
defected
from t h e
winning
c o a l i t i o n ####
d e l t a <− NULL
f o r ( i i n 1 : length ( rapEpgs ) ) {
i f ( rapEpgs [ i ] != ”NI ” ) {
f o u n d <− FALSE
f o r ( j i n 1 : length ( m a j o r i t y ) ) {
i f ( rapEpgs [ i ] == m a j o r i t y [ j ] ) {
f o u n d <− TRUE
break
}
}
i f ( f o u n d == FALSE) {
d e l t a <− c ( d e l t a , rapEpgs [ i ] )
}
}
}
return ( d e l t a )
}
#### C o n s t r u c t i n g and r e n a m i n g
v a r i a b l e s #####
data$ r e s u l t 1 <− i f e l s e ( data$ R e s u l t . o f . v o t e == ”+” , 1 ,
data$ a r e a 1 <− as . numeric ( data$ P o l i c y . a r e a )
p o l i c y a r e a N a m e s <− l e v e l s ( data$ P o l i c y . a r e a ) [ 2
:
i f e l s e ( data$ R e s u l t . o f . v o t e == ”−” , 0 , NA) )
21]
data$ b u d g e t <− i f e l s e ( data$ B u d g e t a r y . i m p l i c a t i o n . . 1 . . 0 . >= 1 , 1 , 0 )
data$ s i z e . w i n n i n g . c o a l <− NA
data$ M a j o r i t y . fo rmed . by <− as . character ( data$ M a j o r i t y . fo rmed . by )
c l a s s ( data$ M a j o r i t y . fo rmed . by )
data$ p a r t i e s . o f . rapp <− NA
data$EPG1
data$EPG2
data$EPG3
data$EPG4
data$EPG5
data$EPG6
data$EPG7
data$EPG8
<−
<−
<−
<−
<−
<−
<−
<−
as . character ( data$EPG1)
as . character ( data$EPG2)
as . character ( data$EPG3)
as . character ( data$EPG4)
as . character ( data$EPG5)
as . character ( data$EPG6)
as . character ( data$EPG7)
as . character ( data$EPG8)
f o r ( i i n 1 : nrow ( data ) ) {
m a j o r i t y . array <− u n l i s t ( s t r s p l i t ( data$ M a j o r i t y . fo rmed . by [ i ] ,
data$ s i z e . w i n n i n g . c o a l [ i ] <− length ( m a j o r i t y . array )
” , ”))
rapp . epg . array <− getRapporteursEPGs ( data$EPG1 [ i ] , data$EPG2 [ i ] , data$EPG3 [ i ] , data$EPG4 [ i ] , data$EPG5 [ i ] , data$EPG6 [ i ]
data$EPG7 [ i ] , data$EPG8 [ i ] )
data$ p a r t i e s . o f . rapp [ i ] <− g e t S t r i n g F r o m A r r a y ( rapp . epg . array )
r a p p E p g N o t I n M a j o r i t y <− g e t D e l t a B e t w e e n ( m a j o r i t y . array , rapp . epg . array )
data$ rapp . p a r t y . n o t . i n . m a j o r i t y [ i ] <− g e t S t r i n g F r o m A r r a y ( r a p p E p g N o t I n M a j o r i t y )
i f ( data$ rapp . p a r t y . n o t . i n . m a j o r i t y [ i ] != ” ” ) {
data$ d e f e c t i o n . by . p a r t y . o f . rapp [ i ] = 1
} else {
data$ d e f e c t i o n . by . p a r t y . o f . rapp [ i ] = 0
}
i f ( length ( grep ( ”EPP” , data$ M a j o r i t y . fo rmed . by [ i ] ) ) > 0 ) {
data$EPPdummy [ i ] = 1
} else {
data$EPPdummy [ i ] = 0
}
i f ( length ( grep ( ”S&D” , data$ M a j o r i t y . fo rmed . by [ i ] ) ) > 0 ) {
data$SDdummy [ i ] = 1
98
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
APPENDIX D. R CODE
} else {
data$SDdummy [ i ] = 0
}
i f ( length ( grep ( ”ALDE” , data$ M a j o r i t y . fo rmed . by [ i ] ) ) > 0 ) {
data$ALDEdummy [ i ] = 1
} else {
data$ALDEdummy [ i ] = 0
}
i f ( length ( grep ( ” G r e e n s ” , data$ M a j o r i t y . fo rmed . by [ i ] ) ) > 0 ) {
data$Greensdummy [ i ] = 1
} else {
data$Greensdummy [ i ] = 0
}
i f ( length ( grep ( ”GUE/NGL” , data$ M a j o r i t y . fo rmed . by [ i ] ) ) > 0 ) {
data$GUENGLdummy[ i ] = 1
} else {
data$GUENGLdummy[ i ] = 0
}
i f ( length ( grep ( ”ECR” , data$ M a j o r i t y . fo rmed . by [ i ] ) ) > 0 ) {
data$ECRdummy [ i ] = 1
} else {
data$ECRdummy [ i ] = 0
}
i f ( length ( grep ( ”EFD” , data$ M a j o r i t y . fo rmed . by [ i ] ) ) > 0 ) {
data$EFDdummy [ i ] = 1
} else {
data$EFDdummy [ i ] = 0
}
}
#### The A g r e e m e n t I n d e x and AIn a r e
calculated
b e l o w #####
c l a s s ( data$Yes )
data$AI <− (
ifelse (
data$Yes > data$No & data$Yes > data$Abs ,
data$Yes ,
ifelse (
data$No > data$Abs ,
data$No ,
data$Abs
)
) − (
0 . 5 * ( data$Yes + data$No + data$Abs − i f e l s e (
data$Yes > data$No & data$Yes > data$Abs ,
data$Yes ,
ifelse (
data$Yes > data$Abs ,
data$No ,
data$Abs
)
)
)
)
) / ( data$Yes + data$No + data$Abs )
data$AI <− i f e l s e ( data$AI == 0 , NA, data$AI )
#d a t a $AI2 <− i f e l s e ( d a t a $AI > . 7 ,
1 , 0)
data$ AbsDidntVote <− data$Abs+data$Didn . t . v o t e
data$AIn <− (
ifelse (
data$Yes > data$No & data$Yes > data$AbsDidntVote ,
data$Yes ,
ifelse (
data$No > data$AbsDidntVote ,
data$No ,
data$ AbsDidntVote
)
) − (
0 . 5 * ( data$Yes + data$No + data$ AbsDidntVote − i f e l s e (
data$Yes > data$No & data$Yes > data$AbsDidntVote ,
data$Yes ,
ifelse (
99
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
data$Yes > data$AbsDidntVote ,
data$No ,
data$ AbsDidntVote
)
)
)
)
) / ( data$Yes + data$No + data$ AbsDidntVote )
#### C a l c u l a t i n g
t h e RICE i n d e x #####
data$RICE <− ( data$Yes − data$No ) / ( data$Yes + data$No )
data$RICE2 <− i f e l s e ( data$RICE > 0 . 1 , 1 , 0 )
data$RICEabs <− NA
data$NoAndAbs <− data$No + data$Abs
data$RICEabs <− ( data$Yes − data$NoAndAbs ) / ( data$Yes + data$NoAndAbs )
###### C a l c u l a t i n g
t h e number o f
sessions
p a s s e d ###########
names ( data )
data$Date <− as . Date ( as . character ( data$Date ) , format = ”%d.%m.%Y” )
data$time <− as . numeric ( data$Date ) − as . numeric ( as . Date ( ”2009−09−14 ” ) )
data$ l o g t i m e <− l o g ( data$time )
#### D e s c r i p t i v e s #####
pdf ( ”scatterPlotDV . pdf ”)
par ( mfrow = c ( 2 , 1 ) )
s c a t t e r p l o t ( data$AIn , data$AI )
s c a t t e r p l o t ( data$RICE , data$RICEabs )
dev . o f f ( )
pdf ( ”scatterAIandRICE . pdf ”)
s c a t t e r p l o t ( data$AI , abs ( data$RICE ) )
dev . o f f ( )
s c a t t e r p l o t ( data$AI , abs ( data$RICEabs ) )
#B u d g e t a r y I m p l i c a t i o n T a b l e <− x t a b l e ( t a b l e ( d a t a $ b u d g e t ) )
boxplot ( data$RICEabs )
boxplot ( data$AI ˜ data$ B u d g e t a r y . i m p l i c a t i o n . . 1 . . 0 . )
h i s t ( data$AI )
#t a b l e ( d a t a $AI , d a t a $ B u d g e t a r y . i m p l i c a t i o n . . 1 . . 0 . )
p l o t ( data$AI , data$ budget , c e x = 0 . 5 , pch = 1 9 )
dev . o f f ( )
b u d g e t L e g i s l a t i o n <− w i t h ( data ,
xtable ( budgetLegislation ,
#x t a b l e ( summary ( d a t a [ ,
xtabs ( ˜ Budgetary . i m p l i c a t i o n . . 1 . . 0 . + Procedure . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . ) )
c a p t i o n = ”Budgetary
c ( ”AI ” ,
”AIn ” ,
”RICE ” ,
i m p l i c a t i o n by
legislative
votes ” ,
label = ”bugetLegislation ” ,
align =
'
l
”RICEabs ” ) ] ) )
numdata <− data [ , c ( ”AI ” , ”AIn ” , ”RICE ” , ”RICEabs ” , ” b u d g e t ” , ” P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . ” , ”No . . Terms ” ,
” d e f e c t i o n . by . p a r t y . o f . rapp ” , ”EPPdummy”
)
]
# ### D e s c r i p t i v e
t a b l e ###
# library ( xtable )
a l i g n = c ( ” l ” , rep ( ”c ” ,
d e s c . t a b l e <− cbind (
c o l M e a n s ( numdata [ , 1 :
apply ( numdata [ , 1 : 9 ]
apply ( numdata [ , 1 : 9 ]
apply ( numdata [ , 1 : 9 ]
colSums ( i s . na ( numdata [
)
3))
9 ] , na . rm = T) ,
, MARGIN = 2 , FUN = sd , na . rm = T) ,
, 2 , min , na . rm = T) ,
, 2 , max, na . rm = T) ,
, 1 : 9]))
colnames ( d e s c . t a b l e ) <− c ( ”Mean ” , ”SD” , ”Min ” , ”Max” , ” M i s s i n g ” )
rownames ( d e s c . t a b l e ) <− c ( ”AI ” , ”AIn ” , ”RICE ” , ”RICEabs ” , ” B u d g e t a r y
” D e f e c t i o n ” , ”EPP dummy” )
print (
xtable (
d e s c . table ,
i m p l i c a t i o n ” , ” L e g i s l a t i o n ” , ”No . Terms o f
rapporte
100
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
APPENDIX D. R CODE
digits = 2,
caption = ”Descriptive s t a t i s t i c s ” ,
l a b e l = ”tab : d e s c r i p t i v ” ,
a l i g n = c ( ” l ” , rep ( ”c ” , 5 ) )
),
size = ”footnotesize ” ,
f i l e = ” d e s c r i p t i v e s . tex ”
)
#### B i v a r i a t e
Correlations
between
variables
of
substantial
i n t e r e s t #####
data$absRICE <− abs ( data$RICE )
cor ( abs ( data$RICE ) , data$ budget , u s e = ” c o m p l e t e . o b s ” )
cor ( data$absRICE , data$ budget , u s e = ” c o m p l e t e . o b s ” )
b i v a r i a t e C o r <− cor ( data [ , c ( ”AI ” , ”AIn ” , ”RICE ” , ”RICEabs ” , ” b u d g e t ” , ” P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . ” , ”No . . T
” d e f e c t i o n . by . p a r t y . o f . rapp ” , ”EPPdummy” ) ] , u s e = ” c o m p l e t e . o b s ” )
colnames ( b i v a r i a t e C o r ) <− c ( ”AI ” ,
” d e f e c t i o n . by . p a r t y
rownames ( b i v a r i a t e C o r ) <− c ( ”AI ” ,
” d e f e c t i o n . by . p a r t y
”AIn ” , ”RICE ” , ”RICEabs ” , ” b u d g e t ” , ” P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . ” , ”No . . Te
. o f . rapp ” , ”EPPdummy” )
”AIn ” , ”RICE ” , ”RICEabs ” , ” b u d g e t ” , ” P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . ” , ”No . . Te
. o f . rapp ” , ”EPPdummy” )
print (
xtable (
bivariateCor ,
digits = 3,
caption = ”Bivariate c o r r e l a t i o n s ” ,
l a b e l = ”tab : b i v a r i a t e C o r ”
),
size = ”footnotesize ” ,
f i l e = ”bivariateCor . tex ”
)
p d f ( ”beanplotsAIandRICE . p d f ” )
par ( mfrow = c ( 1 , 2 ) )
beanplot (
data$AI ˜ data$ budget ,
c o l = c ( ”#CAB2D6” , ”#33A02C ” , ”#B2DF8A” ) ,
border = ”black ” ,
xlab = ”Budgetary I m p l i c a t i o n ” ,
y l a b = ”Agreement I n d e x ” ,
main = ”AI by b u d g e t a r y i m p l i c a t i o n ”
)
#g r e e n l i n e s show i n d i v i d u a l o b s e r v a t i o n s w h i l e
beanplot (
data$RICE ˜ data$ budget ,
c o l = c ( ”#CAB2D6” , ”#33A02C ” , ”#B2DF8A” ) ,
border = ”black ” ,
xlab = ”Budgetary I m p l i c a t i o n ” ,
y l a b = ”RICE ” ,
main = ”RICE by b u d g e t a r y i m p l i c a t i o n ”
)
dev . o f f ( )
the
purple
pdf ( ”beanplotsAIandRICElegislation . pdf ”)
par ( mfrow = c ( 1 , 2 ) )
beanplot (
data$AI ˜ data$ P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . ,
c o l = c ( ”#CAB2D6” , ”#33A02C ” , ”#B2DF8A” ) ,
border = ”black ” ,
xlab = ” L e g i s l a t i v e = 1 ” ,
y l a b = ”Agreement I n d e x ” ,
main = ”AI by l e g i s l a t i o n ”
)
#g r e e n l i n e s show i n d i v i d u a l o b s e r v a t i o n s w h i l e t h e p u r p l e
beanplot (
data$RICE ˜ data$ P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . ,
c o l = c ( ”#CAB2D6” , ”#33A02C ” , ”#B2DF8A” ) ,
border = ”black ” ,
xlab = ” L e g i s l a t i v e = 1 ” ,
y l a b = ”RICE ” ,
main = ”RICE by l e g i s l a t i o n ”
)
dev . o f f ( )
#### E s t i m a t i n g OLS m o d e l s #####
polygon
shows
the
distribution
polygon
shows
the
distribution
101
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
o l s A I . n o I n t e r a c t i o n <− lm ( AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r
+ EPPdummy, data = data )
summary( o l s A I . n o I n t e r a c t i o n )
o l s A I <− lm ( AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp + E
+ b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data )
summary( o l s A I )
o l s A I n <− lm ( AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp +
+ b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data )
summary( o l s A I n )
olsRICE <− lm ( absRICE ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . r
+ EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data )
summary( olsRICE )
olsRICEabs <− lm ( abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data )
summary( olsRICEabs )
stargazer ( olsAI . noInteraction ,
### E s t i m a t i n g
Tobit
without
olsAI ,
o l s A I n , olsRICE ,
olsRICEabs )
i n t e r a c t i o n ####
t o b i t . AI . n o I n t e r a c t i o n <− t o b i t ( AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy, data = data , l e f t = 0 , r i g h t = 1 )
summary( t o b i t . AI . n o I n t e r a c t i o n )
t o b i t . AIn . n o I n t e r a c t i o n <− t o b i t ( AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy, data = data , l e f t = 0 , r i g h t = 1 )
summary( t o b i t . AIn . n o I n t e r a c t i o n )
t o b i t . RICE . n o I n t e r a c t i o n <− t o b i t ( abs ( RICE ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy, data = data , l e f t = 0 , r i g h t = 1 )
summary( t o b i t . RICE . n o I n t e r a c t i o n )
t o b i t . RICEabs . n o I n t e r a c t i o n <− t o b i t ( abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Ter
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy, data = data , l e f t = 0 , r i g h t = 1 )
summary( t o b i t . RICEabs . n o I n t e r a c t i o n )
#### E s t i m a t i n g
tobit
m o d e l s ######
t o b i t . AI . d e f e c t <− t o b i t ( AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t
+ EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data , l e f t = 0 , r i g h t = 1 )
summary( t o b i t . AI . d e f e c t )
t o b i t . AIn . d e f e c t <− t o b i t ( AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data ,
summary( t o b i t . AIn . d e f e c t )
left = 0,
right
t o b i t . RICE . d e f e c t <− t o b i t ( abs ( RICE ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data , l e f t = 0 ,
summary( t o b i t . RICE . d e f e c t )
t o b i t . RICEabs . d e f e c t <− t o b i t ( abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp , data = data , l e f t = 0 ,
summary( t o b i t . RICEabs . d e f e c t )
s t a r g a z e r ( t o b i t . AI . d e f e c t ,
t o b i t . AIn . d e f e c t ,
t o b i t . RICE . d e f e c t ,
t o b i t . RICEabs . d e f e c t )
#s a v e . i m a g e ( ” MayImage . Rd a ta ”)
#### L i n e a r i t y
t e s t s #####
summary( t o b i t . AI . d e f e c t )
#g e t u n s t a n d a r d i z e d p r e d i c t e d and r e s i d u a l v a l u e s
u n s t a n d a r d i z e d P r e d i c t e d <− p r e d i c t ( t o b i t . AI . d e f e c t )
u n s t a n d a r d i z e d R e s i d u a l s <− r e s i d ( t o b i t . AI . d e f e c t )
#g e t s t a n d a r d i z e d v a l u e s
s t a n d a r d i z e d P r e d i c t e d <− ( u n s t a n d a r d i z e d P r e d i c t e d − mean( u n s t a n d a r d i z e d P r e d i c t e d ) ) / sd ( u n s t a n d a r d i z e d P r e d i c t e d )
s t a n d a r d i z e d R e s i d u a l s <− ( u n s t a n d a r d i z e d R e s i d u a l s − mean( u n s t a n d a r d i z e d R e s i d u a l s ) ) / sd ( u n s t a n d a r d i z e d R e s i d u a l s )
#c r e a t e s t a n d a r d i z e d r e s i d u a l s p l o t
pdf ( ”S t a n da r d i z e d v s P r e d ic t e d R e s id u a ls A I . pdf ”)
plot (
standardizedPredicted ,
standardizedResiduals ,
main = ” S t a n d a r d i z e d R e s i d u a l s P l o t ” ,
righ
rig
102
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
APPENDIX D. R CODE
xlab = ”Standardized
ylab = ”Standardized
Predicted Values ” ,
Residuals ”
)
abline ( 0 , 0 )
dev . o f f ( )
#c r e a t e r e s i d u a l s h i s t o g r a m
pdf ( ”ResidualsHistAI . pdf ”)
hist ( standardizedResiduals ,
curve (dnorm , add = TRUE)
dev . o f f ( )
f r e q = FALSE)
#g e t p r o b a b i l i t y d i s t r i b u t i o n f o r r e s i d u a l s
p r o b D i s t <− pnorm( s t a n d a r d i z e d R e s i d u a l s )
#c r e a t e PP p l o t
p d f ( ”PPPlotAI . p d f ” )
plot (
ppoints ( length ( s t a n d a r d i z e d R e s i d u a l s ) ) ,
sort ( probDist ) ,
main = ”PP P l o t ” ,
x l a b = ”Observed P r o b a b i l i t y ” ,
ylab = ”Expected P r o b a b i l i t y ”
)
abline ( 0 , 1 )
dev . o f f ( )
##### R e s i d u a l s and
influential
cases :
dfbeta
t e s t s ######
data$row <− rownames ( data )
attach ( data )
nona <− na . omit ( data . frame ( AI , AIn , RICE , RICEabs , budget ,
d e f e c t i o n . by . p a r t y . o f . rapp , EPPdummy, row ) )
detach ( data )
result1 ,
P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . , No . . Terms ,
summary( t o b i t . AI . d e f e c t )
d f b e t a <− as . data . frame ( r e s i d u a l s ( t o b i t . AI . d e f e c t ,
d f b e t a $ i d < rownames ( data )
type = ”dfbeta ” ) )
pdf ( ” R e s i d u a l s I n t e r a c t i o n D e f e c t i o n . pdf ”)
par ( mfrow = c ( 2 ,
2))
p l o t ( nona$row ,
text ( nona$row ,
dfbeta [ ,
dfbeta [ ,
2] ,
2] ,
t y p e = ”n ” , y l a b = ”Budget ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
dfbeta [ ,
dfbeta [ ,
6] ,
6] ,
t y p e = ”n ” , y l a b = ” D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
dfbeta [ ,
dfbeta [ ,
8] ,
8] ,
t y p e = ”n ” , y l a b = ”Budget * D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
dev . o f f ( )
data [ 1 0 5 6 ,
data [ 9 8 , ]
]
AI . n o I n f l u e n c e <− t o b i t (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp
+ b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = nona [ which (
nona$row != 1123 & nona$row != 1034 & nona$row != 1126 & nona$row != 184 & nona$row != 781
& nona$row != 98 & nona$row != 1126 & nona$row != 295 & nona$row != 1177 & nona$row != 576
), ],
left = 0,
right = 1
)
summary( AI . n o I n f l u e n c e )
AI . no . E f f e c t O n I n t e r a c t i o n <− t o b i t (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = nona [ which ( nona$row != 1054 & nona$row != 1056 & nona$row != 182 & nona$row != 4 3 4 ) ,
left = 0,
right = 1
)
summary( AI . no . E f f e c t O n I n t e r a c t i o n )
AI . no . E f f e c t O n I n t e r a c t i o n <− t o b i t (
],
103
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = nona [ which (
nona$row != 1054 & nona$row != 1056 & nona$row != 182 & nona$row != 434
& nona$row != 601 & nona$row != 812 & nona$row != 1125 & nona$row != 1180
& nona$row != 1025 & nona$row != 652 & nona$row != 1018 & nona$row != 596
& nona$row != 595 & nona$row != 331 & nona$row != 597 & nona$row != 256
& nona$row != 1026 & nona$row != 544 & nona$row != 948 & nona$row != 1026
& nona$row != 324 & nona$row != 705 & nona$row != 518 & nona$row != 968
& nona$row != 987 & nona$row != 305
), ],
left = 0,
right = 1
)
summary( AI . no . E f f e c t O n I n t e r a c t i o n )
#Df b e t a
f o r AIn
t o b i t . nona . AIn <− t o b i t (
AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp
+ EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
left = 0,
right = 1 ,
data = nona
)
summary( t o b i t . nona . AIn )
d f b e t a . AIn <− as . data . frame ( r e s i d u a l s ( t o b i t . nona . AIn ,
#d f b e t a . AIn$ i d 1 <− rownames ( d a t a )
type = ”dfbeta ” ) )
pdf ( ” R e s i d u a l s I n t e r a c t i o n A I n . pdf ”)
par ( mfrow = c ( 3 ,
2))
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . AIn [ ,
d f b e t a . AIn [ ,
2] ,
2] ,
t y p e = ”n ” , y l a b = ” Budget ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . AIn [ ,
d f b e t a . AIn [ ,
6] ,
6] ,
t y p e = ”n ” , y l a b = ” D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . AIn [ ,
d f b e t a . AIn [ ,
8] ,
8] ,
t y p e = ”n ” , y l a b = ” Budget * D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
dev . o f f ( )
#AI MODEL w i t h o u t
influential
cases
AIn . no . E f f e c t O n I n t e r a c t i o n <− t o b i t (
AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = nona [ which (
nona$row != 434 & nona$row != 182 & nona$row != 282 & nona$row != 1056
& nona$row != 1054 & nona$row != 1025 & nona$row != 1180 & nona$row != 575
& nona$row != 973 & nona$row != 757 & nona$row != 218 & nona$row != 1018
& nona$row != 331 & nona$row != 596 & nona$row != 656 & nona$row != 567
& nona$row != 1026 & nona$row != 259 & nona$row != 544 & nona$row != 7
& nona$row != 948 & nona$row != 972 & nona$row != 705 & nona$row != 451
& nona$row != 987 & nona$row != 518 & nona$row != 172
), ],
left = 0,
right = 1
)
summary( AIn . no . E f f e c t O n I n t e r a c t i o n )
#R e s i d u a l s RICE w i t h o u t
influential
cases
t o b i t . nona . RICE <− t o b i t (
abs ( RICE ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
left = 0,
right = 1 ,
data = nona
)
summary( t o b i t . nona . RICE )
d f b e t a . RICE <− as . data . frame ( r e s i d u a l s ( t o b i t . nona . RICE ,
#d f b e t a . AIn$ i d 1 <− rownames ( d a t a )
pdf ( ”ResidualsInteractionRICE . pdf ”)
par ( mfrow = c ( 2 ,
2))
type = ”dfbeta ” ) )
104
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
APPENDIX D. R CODE
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . RICE [ ,
d f b e t a . RICE [ ,
2] ,
2] ,
t y p e = ”n ” , y l a b = ” Budget ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . RICE [ ,
d f b e t a . RICE [ ,
6] ,
6] ,
t y p e = ”n ” , y l a b = ” D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . RICE [ ,
d f b e t a . RICE [ ,
8] ,
8] ,
t y p e = ”n ” , y l a b = ” Budget * D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
dev . o f f ( )
RICE . no . E f f e c t O n I n t e r a c t i o n <− t o b i t (
abs ( RICE ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = nona [ which (
nona$row !=434 & nona$row !=282 & nona$row !=128 & nona$row !=1056
& nona$row !=1054 & nona$row !=206 & nona$row !=522 & nona$row !=519
& nona$row !=601 & nona$row !=164 & nona$row !=812 & nona$row !=596
& nona$row !=593 & nona$row !=595 & nona$row !=331 & nona$row !=125
& nona$row !=259 & nona$row !=214 & nona$row !=737 & nona$row !=324
& nona$row !=544 & nona$row !=74 & nona$row !=987 & nona$row !=968
& nona$row !=705 & nona$row !=518 & nona$row !=305
), ],
l e f t = 0 , right = 1
)
summary( RICE . no . E f f e c t O n I n t e r a c t i o n )
#R e s i d u a l s RICEabs
t o b i t . nona . RICEabs <− t o b i t (
abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
left = 0,
right = 1 ,
data = nona
)
summary( t o b i t . nona . RICEabs )
d f b e t a . RICEabs <− as . data . frame ( r e s i d u a l s ( t o b i t . nona . RICEabs ,
#d f b e t a . AIn$ i d 1 <− rownames ( d a t a )
type = ”dfbeta ” ) )
pdf ( ”ResidualsInteractionRICEabs . pdf ”)
par ( mfrow = c ( 2 ,
2))
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . RICEabs [ ,
d f b e t a . RICEabs [ ,
2] ,
2] ,
t y p e = ”n ” , y l a b = ” Budget ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . RICEabs [ ,
d f b e t a . RICEabs [ ,
3] ,
3] ,
t y p e = ”n ” , y l a b = ” D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
p l o t ( nona$row ,
text ( nona$row ,
d f b e t a . RICEabs [ ,
d f b e t a . RICEabs [ ,
9] ,
9] ,
t y p e = ”n ” , y l a b = ” Budget * D e f e c t i o n ” , x l a b = ” O b s e r v a t i o n ” )
l a b e l = nona$row , c e x = 0 . 7 )
dev . o f f ( )
# RICEabs m o d e l
without
influential
cases
RICEabs . no . E f f e c t O n I n t e r a c t i o n <− t o b i t (
abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = nona [ which (
nona$row != 206 & nona$row != 596 & nona$row != 593 & nona$row != 887
& nona$row != 519 & nona$row != 812 & nona$row != 1025 & nona$row != 1180
& nona$row != 522 & nona$row != 652 & nona$row != 206 & nona$row != 596
& nona$row != 593 & nona$row != 595 & nona$row != 331 & nona$row != 397
& nona$row != 16 & nona$row != 752 & nona$row != 781 & nona$row != 602
& nona$row != 320 & nona$row != 902 & nona$row != 7 & nona$row != 1034
& nona$row != 1026 & nona$row != 1123 & nona$row != 892
), ],
left = 0,
right = 1
)
summary( RICEabs . no . E f f e c t O n I n t e r a c t i o n )
#### T a b l e
of
models
w i t h no
influential
observations
for
all
the
dependent
v a r i a b l e s ####
s t a r g a z e r ( AI . no . E f f e c t O n I n t e r a c t i o n , AIn . no . E f f e c t O n I n t e r a c t i o n , RICE . no . E f f e c t O n I n t e r a c t i o n , RICEabs . no . E f f e c t O n I n t e r a c
t i t l e = ”Tobit models without i n f l u e n t i a l o b s e r v a t i o n s ”)
105
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
#### I d e o l o g y AER m o d e l s
with
p a r t y dummy #####
t o b i t . AI . PartyDummies <− t o b i t (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + time + EPPdummy + SDdummy + ALDEdummy + Greensdummy + GUENGLdummy
+ ECRdummy + EFDdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
data = data ,
left = 0,
right = 1
)
summary( t o b i t . AI . PartyDummies )
t o b i t . AIn . PartyDummies <− t o b i t (
AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + time + EPPdummy + SDdummy + ALDEdummy + Greensdummy
+ GUENGLdummy + ECRdummy + EFDdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
data = data ,
left = 0,
right = 1
)
summary( t o b i t . AI . PartyDummies )
t o b i t . RICE . PartyDummies <− t o b i t (
abs ( RICE ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + time + EPPdummy + SDdummy + ALDEdummy + Greensdummy
+ GUENGLdummy + ECRdummy + EFDdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
data = data ,
left = 0,
right = 1
)
summary( t o b i t . RICE . PartyDummies )
t o b i t . RICEabs . PartyDummies <− t o b i t (
abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + time + EPPdummy + SDdummy + ALDEdummy + Greensdummy
+ GUENGLdummy + ECRdummy + EFDdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
data = data ,
left = 0,
right = 1
)
summary( t o b i t . RICEabs . PartyDummies )
s t a r g a z e r ( t o b i t . AI . PartyDummies ,
### ML random
#w i t h
party
intercept
with
t o b i t . AIn . PartyDummies ,
party
t o b i t . RICE . PartyDummies ,
t o b i t . RICEabs . PartyDummies )
p o s i t i o n s ####
positions
AI . MLbudget . p a r t y <− l m e r (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + SDdummy + ALDEdummy
+ Greensdummy + GUENGLdummy + ECRdummy + EFDdummy + ( 1 + b u d g e t | a r e a 1 ) ,
data = data
)
summary( AI . MLbudget . p a r t y )
AIn . MLbudget . p a r t y <− l m e r (
AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + SDdummy
+ ALDEdummy + Greensdummy + GUENGLdummy + ECRdummy + EFDdummy + ( 1 + b u d g e t | a r e a 1 ) ,
data = data
)
summary( AIn . MLbudget . p a r t y )
RICE . MLbudget . p a r t y <− l m e r (
abs ( RICE ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + SDdummy + ALDEdummy
+ Greensdummy + GUENGLdummy + ECRdummy + EFDdummy + ( 1 + b u d g e t | a r e a 1 ) ,
data = data
)
summary( RICE . MLbudget . p a r t y )
RICEabs . MLbudget . p a r t y <− l m e r (
abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + SDdummy
+ ALDEdummy + Greensdummy + GUENGLdummy + ECRdummy + EFDdummy + ( 1 + b u d g e t | a r e a 1 ) ,
data = data
)
summary( RICEabs . MLbudget . p a r t y )
s t a r g a z e r ( AI . MLbudget . p a r t y , AIn . MLbudget . p a r t y , RICE . MLbudget . p a r t y , RICEabs . MLbudget . p a r t y ,
106
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
APPENDIX D. R CODE
t i t l e = ”Random i n t e r e p t and
effect
o f budget
controlled
for
party
p o s i t i o n s ”)
#### ML F i x e d E f f e c t s ####
AI . MLbudgetfe <− lm (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp
+ b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + as . f a c t o r ( a r e a 1 ) ,
data = data
)
c l a s s ( data$ a r e a 1 )
summary( AI . MLbudgetfe )
AIn . MLbudgetfe <− lm (
AIn ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp
+ b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + as . f a c t o r ( a r e a 1 ) ,
data = data
)
c l a s s ( data$ a r e a 1 )
summary( AIn . MLbudgetfe )
RiceMLbudgetfe <− lm (
abs ( RICE ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + as . f a c t o r ( a r e a 1 ) ,
data = data
)
c l a s s ( data$ a r e a 1 )
summary( RiceMLbudgetfe )
R i c e a b s M L b u d g e t f e <− lm (
abs ( RICEabs ) ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + as . f a c t o r ( a r e a 1 ) ,
data = data
)
c l a s s ( data$ a r e a 1 )
summary( R i c e a b s M L b u d g e t f e )
s t a r g a z e r ( AI . MLbudgetfe , AIn . MLbudgetfe ,
#### L o g i t
Result
RiceMLbudgetfe ,
RiceabsMLbudgetfe )
m o d e l s ####
r e s u l t M o d e l . AI <− z e l i g (
r e s u l t 1 ˜ b u d g e t + AI + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp
+ EPPdummy,
model = ” l o g i t ” ,
data = data
)
summary( r e s u l t M o d e l . AI )
r e s u l t M o d e l . AIn <− z e l i g (
r e s u l t 1 ˜ b u d g e t + AIn + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp
+ EPPdummy,
model = ” l o g i t ” ,
data = data
)
summary( r e s u l t M o d e l . AIn )
r e s u l t M o d e l . RICE <− z e l i g (
r e s u l t 1 ˜ b u d g e t + abs ( RICE ) + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
model = ” l o g i t ” ,
data = data
)
summary( r e s u l t M o d e l . RICE )
r e s u l t M o d e l . RICEabs <− z e l i g (
r e s u l t 1 ˜ b u d g e t + abs ( RICEabs ) + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
model = ” l o g i t ” ,
data = data
)
summary( r e s u l t M o d e l . RICEabs )
s t a r g a z e r ( r e s u l t M o d e l . AI ,
###### I n t e r a c t i o n
r e s u l t M o d e l . AIn ,
P l o t AI m o d e l s #####
r e s u l t M o d e l . RICE ,
r e s u l t M o d e l . RICEabs )
107
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
r e q u i r e (AER)
t o b i t . AI . I n t e r P l o t <− t o b i t (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms+ d e f e c t i o n . by . p a r t y . o f . rapp
+ EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
data = data
)
t o b i t . AI . I n t e r P l o t 1 <− t o b i t (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ I ( d e f e c t i o n . by . p a r t y . o f . rapp − 1 ) + EPPdummy + b u d g e t * I ( d e f e c t i o n . by . p a r t y . o f . rapp − 1 ) ,
data = data
)
c o e f s <− cbind (summary( t o b i t . AI . I n t e r P l o t ) $ c o e f f i c i e n t s [ 2 ,
se <− cbind (summary( t o b i t . AI . I n t e r P l o t ) $ c o e f f i c i e n t s [ 2 ,
c o n f . i n t <− matrix (NA, nrow = 2 , ncol = 2 )
for ( i in 1 : 2) {
c o n f . i n t [ i , ] <− c ( c o e f s [ i ] − 1 . 9 6 * se [ i ] ,
}
1 ] , summary( t o b i t . AI . I n t e r P l o t 1 ) $ c o e f f i c i e n t s [ 2 ,
2 ] , summary( t o b i t . AI . I n t e r P l o t 1 ) $ c o e f f i c i e n t s [ 2 ,
coefs [ i ] + 1.96
*
1])
2])
se [ i ] )
pdf ( ” I n t e r a c t i o n P l o t D e f e c t B u d g e t . pdf ”)
par ( mfrow = c ( 1 ,
1))
p l o t ( c ( − 0 . 5 , 1 . 5 ) , c ( − 0 . 7 , 0 . 3 ) , t y p e = ”n ” , y l a b = ” E f f e c t o f
a x i s ( 1 , a t = 0 : 1 , l a b e l s = c ( ”No D e f e c t i o n ” , ” D e f e c t i o n ” ) )
for ( i in 1 : 2) {
segments ( y0 = c o n f . i n t [ i ,
}
1 ] , y1 = c o n f . i n t [ i ,
budgetary
i m p l i c a t i o n s on AI ” , x l a b = ” ” , x a x t = ”n ” )
2 ] , x0 = i − 1 , c o l = ” v i o l e t r e d 2 ” )
points ( c ( 0 : 1 ) , c o e f s , t y p e = ”p ” , pch = 1 6 )
segments ( y0 = 0 , x0 = −1, x1 = 2 , c o l = ” g r e y 5 0 ” )
dev . o f f ( )
##### S i m u l a t i o n ###########
r e q u i r e (MASS)
r e q u i r e (AER)
t o b i t . AI . S i m P l o t <− t o b i t (
AI ˜ b u d g e t + r e s u l t 1 + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms + d e f e c t i o n . by . p a r t y . o f . rapp
+ EPPdummy + b u d g e t * d e f e c t i o n . by . p a r t y . o f . rapp ,
data = data
)
beta <− summary( t o b i t . AI . S i m P l o t ) $ c o e f f i c i e n t s [ 1
vcov <− t o b i t . AI . S i m P l o t $var [ 1 : 8 , 1 : 8 ]
n s i m s <− 1000
set . seed (2709)
:
8,
1]
simb <− mvrnorm ( n = nsims , mu = beta , Sigma = vcov )
###i n t e r c e p t = 1 , b u d g e t = 0 , d e f e c t i o n = 1 , i n t e r a c t i o n = 0
x . 0 <− cbind ( 1 , 0 , median ( data$ r e s u l t 1 , na . rm = T) , median ( data$ P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . , na . rm = T) ,
median ( data$No . . Terms , na . rm = T) , 1 , min( data$EPPdummy, na . rm = T) , 0 )
###i n t e r c e p t = 1 , b u d g e t = 1 , d e f e c t i o n = 1 , i n t e r a c t i o n = 1
x . 1 <− cbind ( 1 , 1 ,
median ( data$ r e s u l t 1 , na . rm = T) ,
median ( data$ P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . , na . rm = T) ,
median ( data$No . . Terms , na . rm = T) ,
1,
min( data$EPPdummy, na . rm = T) ,
1)
x . beta . 0 <− x . 0 %*% t ( simb )
x . beta . 1 <− x . 1 %*% t ( simb )
exp . y . 0 <− x . beta . 0
exp . y . 1 <− x . beta . 1
q u a n t i l e . v a l u e s 0 <− q u a n t i l e ( exp . y . 0 ,
q u a n t i l e . v a l u e s 1 <− q u a n t i l e ( exp . y . 1 ,
sim <− data . frame ( c ( exp . y . 0 , exp . y . 1 ) )
sim $ b u d g e t <− NA
sim $ b u d g e t [ 1 : 1 0 0 0 ] <− ”No ”
probs = c ( 0 . 0 2 5 ,
probs = c ( 0 . 0 2 5 ,
0.5 ,
0.5 ,
0.975))
0.975))
108
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
APPENDIX D. R CODE
sim $ b u d g e t [ 1 0 0 1
:
2 0 0 0 ] <− ”Yes ”
pdf ( ”BoxPlotSimulationDefection . pdf ”)
boxplot (
sim $c . exp . y . 0 . . exp . y . 1 . ˜ sim $ budget ,
frame = T,
xlab = ”Budgetary i m p l i c a t i o n ” ,
ylab = ”Predicted cohesiveness ” ,
main = ” E f f e c t f o r b i l l s w i t h d e f e c t i o n ”
)
dev . o f f ( )
pdf ( ” S i m u l a t i o n P l o t D e n s i t y f o r F a i l e d . pdf ”)
p l o t ( c ( 0 . 4 , 0 . 9 ) , c ( 0 , 2 0 ) , t y p e = ”n ” , y l a b = ” D e n s i t y ” , x l a b = ” B u d g e t a r y i m p l i c a t i o n ” )
l i n e s ( density ( sim $c . exp . y . 0 . . exp . y . 1 . [ 1 : 1 0 0 0 ] ) , c o l = ”tomato ” )
l i n e s ( density ( sim $c . exp . y . 0 . . exp . y . 1 . [ 1 0 0 1 : 2 0 0 0 ] ) , c o l = ” s e a g r e e n ” )
segments ( x0 = q u a n t i l e ( sim $c . exp . y . 0 . . exp . y . 1 . [ 1 : 1 0 0 0 ] , p r o b s = 0 . 0 2 5 ) , y0 = 0 , y1 = 2 . 6 , c o l = ”tomato ” )
segments ( x0 = q u a n t i l e ( sim $c . exp . y . 0 . . exp . y . 1 . [ 1 : 1 0 0 0 ] , p r o b s = 0 . 9 7 5 ) , y0 = 0 , y1 = 2 . 6 7 , c o l = ”tomato ” )
segments ( x0 = q u a n t i l e ( sim $c . exp . y . 0 . . exp . y . 1 . [ 1 0 0 1 : 2 0 0 0 ] , p r o b s = 0 . 0 2 5 ) , y0 = 0 , y1 = 1 . 9 , c o l = ” s e a g r e e n ” )
segments ( x0 = q u a n t i l e ( sim $c . exp . y . 0 . . exp . y . 1 . [ 1 0 0 1 : 2 0 0 0 ] , p r o b s = 0 . 9 7 5 ) , y0 = 0 , y1 = 1 . 9 , c o l = ” s e a g r e e n ” )
legend ( 0 . 6 3 , 1 8 , c ( ”No D e f e c t i o n ” , ” D e f e c t i o n ” ) , f i l l = c ( ” s e a g r e e n ” , ”tomato ” ) , c e x = 0 . 8 , pt . bg = 1 )
dev . o f f ( )
p d f ( ”S i mu l a t i o n B u d g e t R I C E B e a n P l o t . p d f ” )
beanplot (
sim $c . exp . y . 0 . . exp . y . 1 . ˜ sim $ budget ,
frame = T,
xlab = ”Budgetary i m p l i c a t i o n ” ,
ylab = ”Predicted cohesiveness ” ,
c o l = c ( ”#CAB2D6” , ”#33A02C ” , ”#B2DF8A” ) ,
border = ”black ” ,
main = ” S i m u l a t e d e f f e c t s o f b u d g e t a r y i m p l i c a t i o n
)
dev . o f f ( )
#### S i m u l a t i o n
for
bills
f o r RICE ”
w i t h no d e f e c t i o n ####
#i n t e r c e p t = 1 , b u d g e t = 0 , d e f e c t i o n and i n t e r a c t i o n = 0
a . 0 <− cbind ( 1 , 0 , median ( data$ r e s u l t 1 , na . rm = T) ,
max( data$ P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . , na . rm = T) ,
median ( data$No . . Terms , na . rm = T) ,
0,
max( data$EPPdummy, na . rm = T) ,
0)
#i n t e r c e p t =1 , b u d g e t =1 , d e f e c t i o n and i n t e r a c t i o n =0
a . 1 <− cbind ( 1 , 1 , median ( data$ r e s u l t 1 , na . rm = T) ,
max( data$ P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . , na . rm = T) ,
median ( data$No . . Terms , na . rm = T) ,
0,
max( data$EPPdummy, na . rm = T) ,
0)
a . beta . 0
a . beta . 1
exp . ya . 0
exp . ya . 1
<−
<−
<−
<−
a . 0 %*% t ( simb )
a . 1 %*% t ( simb )
a . beta . 0
a . beta . 1
q u a n t i l e . v a l u e s a 0 <− q u a n t i l e ( exp . ya . 0 ,
q u a n t i l e . v a l u e s a 1 <− q u a n t i l e ( exp . ya . 1 ,
probs = c ( 0 . 0 2 5 ,
probs = c ( 0 . 0 2 5 ,
0.5 ,
0.5 ,
0.975))
0.975))
sima <− data . frame ( c ( exp . ya . 0 , exp . ya . 1 ) )
sima $ b u d g e t <− NA
sima $ b u d g e t [ 1 : 1 0 0 0 ] <− ”No ”
sima $ b u d g e t [ 1 0 0 1 : 2 0 0 0 ] <− ”Yes ”
names ( sima )
pdf ( ” S i m u l a t i o n P l o t D e n s i t y f o r F a i l e d . pdf ”)
p l o t ( c ( 0 . 1 , 0 . 4 ) , c ( 0 , 2 5 ) , t y p e = ”n ” , y l a b = ” D e n s i t y ” , x l a b = ” P r e d i c t e d c o h e s i v e n e s s ” )
l i n e s ( density ( sima $c . exp . ya . 0 . . exp . ya . 1 . [ 1 : 1 0 0 0 ] ) , c o l = ” h o t p i n k ” )
l i n e s ( density ( sima $c . exp . ya . 0 . . exp . ya . 1 . [ 1 0 0 1 : 2 0 0 0 ] ) , c o l = ” l i m e g r e e n ” )
segments ( x0 = q u a n t i l e ( sima $c . exp . ya . 0 . . exp . ya . 1 . [ 1 : 1 0 0 0 ] , p r o b s = 0 . 0 2 5 ) , y0 = 0 , y1 = 2 , c o l = ” h o t p i n k ” )
segments ( x0 = q u a n t i l e ( sima $c . exp . ya . 0 . . exp . ya . 1 . [ 1 : 1 0 0 0 ] , p r o b s = 0 . 9 7 5 ) , y0 = 0 , y1 = 2 , c o l = ” h o t p i n k ” )
segments ( x0 = q u a n t i l e ( sima $c . exp . ya . 0 . . exp . ya . 1 . [ 1 0 0 1 : 2 0 0 0 ] , p r o b s = 0 . 0 2 5 ) , y0 = 0 , y1 = 2 , c o l = ” l i m e g r e e n ” )
segments ( x0 = q u a n t i l e ( sima $c . exp . ya . 0 . . exp . ya . 1 . [ 1 0 0 1 : 2 0 0 0 ] , p r o b s = 0 . 9 7 5 ) , y0 = 0 , y1 = 2 . 6 , c o l = ” l i m e g r e e n ” )
legend ( 0 . 0 9 , 1 2 , c ( ”No b u d g e t a r y i m p l i c a t i o n ” , ” B u d g e t a r y i m p l i c a t i o n ” ) , f i l l = c ( ” l i m e g r e e n ” , ” h o t p i n k ” ) , c e x = 0 . 8 , pt
dev . o f f ( )
p d f ( ”SimulationBudgetRICEBeanPlotRESULT . p d f ” )
beanplot (
109
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
sima $c . exp . ya . 0 . . exp . ya . 1 . ˜ sima $ budget ,
frame = T,
xlab = ”Budgetary i m p l i c a t i o n ” ,
ylab = ”Predicted cohesiveness ” ,
c o l = c ( ”#CAB2D6” , ”#33A02C ” , ”#B2DF8A” ) ,
border = ”black ” ,
main = ” S i m u l a t e d e f f e c t s o f b u d g e t a r y i m p l i c a t i o n
f o r RICE f o r
votes
without
defections ”
)
dev . o f f ( )
pdf ( ”BoxPlotSimulationNODefection . pdf ”)
boxplot (
sima $c . exp . ya . 0 . . exp . ya . 1 . ˜ sima $ budget ,
frame = T,
xlab = ”Budgetary i m p l i c a t i o n ” ,
ylab = ”Predicted cohesiveness ” ,
main = ” S i m u l a t e d e f f e c t s f o r b i l l s w i t h o u t
)
dev . o f f ( )
defection ”
######## MISC #################
data [ which ( data$ b u d g e t == 1 & data$ d e f e c t i o n . by . p a r t y . o f . rapp == 1 ) ,
c l a s s ( data$ s i z e . w i n n i n g . c o a l )
]
poisson . WinningCoal <− glm (
s i z e . w i n n i n g . c o a l ˜ b u d g e t + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = data ,
family = poisson
)
summary( poisson . WinningCoal )
poisson . WinningCoal . i n t e r a c t i o n <− glm (
s i z e . w i n n i n g . c o a l ˜ b u d g e t + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + d e f e c t i o n . by . p a r t y . o f . rapp * budget ,
data = data ,
family = poisson
)
summary( poisson . WinningCoal . i n t e r a c t i o n )
poisson . AIn . WinningCoal <− glm (
s i z e . w i n n i n g . c o a l ˜ b u d g e t + AIn + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + d e f e c t i o n . by . p a r t y . o f . rapp * budget ,
data = data ,
family = poisson
)
summary( poisson . AIn . WinningCoal )
poisson . RICE . WinningCoal <− glm (
s i z e . w i n n i n g . c o a l ˜ b u d g e t + abs ( RICE ) + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy,
data = data ,
family = poisson
)
summary( poisson . RICE . WinningCoal )
poisson . RICEabs . WinningCoal <− glm (
s i z e . w i n n i n g . c o a l ˜ b u d g e t + abs ( RICEabs ) + P r o c e d u r e . . 1 . l e g i s l a t i v e . . 0 . r e s o l u t i o n . + No . . Terms
+ d e f e c t i o n . by . p a r t y . o f . rapp + EPPdummy + d e f e c t i o n . by . p a r t y . o f . rapp * budget ,
data = data ,
family = poisson
)
summary( poisson . RICEabs . WinningCoal )
s t a r g a z e r ( poisson . WinningCoal , poisson . WinningCoal . i n t e r a c t i o n )
save . image ( ”2Mai . Rdata ” )