– DRAFT - PLEASE DO NOT CITE OR QUOTE WITHOUT AUTHOR'S PERMISSION –
– COMMENTS ALWAYS WELCOME –
Presidential Activism in Central and Eastern Europe (CEE): A statistical
analysis of the use of presidential vetoes in the CEE EU member states,
1990-2010.
Philipp Köker
PhD candidate in Political Science
University College London – School of Slavonic and East European Studies
[email protected]
Abstract
The power to veto legislation is often referred to as the most important presidential power.
While the use of presidential vetoes by the U.S. president is well studied, scholarship dealing
with this issue in the European context is rare. The existing literature largely focuses on directly elected presidents and is mainly aimed at measuring presidential veto powers; yet, only
few researchers have analysed how presidents actually use them.
This paper aims to fill this gap by providing the first time-series cross-country analysis of the
use of presidential vetoes. I argue that the use of presidential vetoes is influenced not only by
factors pertaining to the political environment but also by the mode of election of the president. While until now the lack of appropriate data made the test of such hypotheses impossible, I use a new, original data set comprising data on presidential veto usage in nine Central
and East European democracies from 1990-2010. The results show that direct presidential
elections, forthcoming parliamentary elections and an increase in the amount of legislation
passed by parliament increases presidents’ use of their veto power.
keywords: presidential activism, veto, semi-presidentialism, parliamentary systems, Central Eastern
Europe
Paper prepared for presentation at the 63rd Political Studies Association
Annual International Conference, 25 - 27 March 2013, Cardiff
1
The power to veto legislation is traditionally seen as one of the most important presidential
prerogatives. By sending bills back to parliament for repeated debate and potential re-passage,
presidents can block or at least suspend unfavourable legislation and thus avert changes to the
status quo. Even the mere threat to veto already serves as a powerful bargaining tool (Spitzer
1988, xvi). Not the least due to its versatility but also its high symbolic importance, the presidential veto has been at the heart of U.S. presidential studies and is well studied both theoretically (Hammond/Miller 1987; Shapiro/Kumar/Jacobs 2000; Cameron/McCarty 2004) and
empirically (Rohde/Simon 1985; Shields/Huang 1995, 1997; Gilmour 2002). In the European
context, however, the empirical study of presidential veto use has only recently received
scholarly attention as part of more general studies of presidential activism (Tavits 2008). The
use of presidential vetoes in parliamentary systems of Europe is thus understudied and still
hampered by the fact that many authors choose to study exclusively countries with popularly
elected presidents. Given that using their veto power presidents can – independently from
their mode of election – not only delay but also potentially block important legislation (or
eventually push towards amendments to bills being made), the study of this specific presidential power also has very practical relevance.
The aim of this paper is twofold: First, it will bring together a coherent set of testable
hypotheses on the use of presidential vetoes in parliamentary systems with directly and indirectly elected presidents. Thereby, it combines insights from both European- and US-focussed
studies. In adaption of Tavits’ (2008) ‘political opportunity framework’ I argue that the use of
presidential vetoes is shaped by the political environment. However, I assume that these factors only come into play within the framework set by constitutional stipulations, i.e. first and
foremost the mode of presidential election. Hereby, I concur with traditional assumptions and
argue that directly election presidents should be more active than their indirectly elected counterparts. Second, this paper provides a test of the aforementioned hypotheses through – to my
knowledge – the first time-series cross-country analysis of the use of presidential vetoes in
parliamentary systems. While until now the lack of appropriate data made the test of such
hypotheses impossible, I use a new, original data set comprising data on presidential veto usage in nine Central and East European (CEE) democracies from 1990-2010. The results support my hypotheses: they show that direct presidential elections and forthcoming parliamentary elections as well as an increase in legislation passed by parliament increase veto use by
presidents.
1
An earlier version of this paper was presented at the 2nd Annual Conference of the European Political Science
Association (EPSA), 20-23 June 2012, Berlin.
Determinants of the use of presidential vetoes
As mentioned in the introduction, the empirical study of presidential veto activity has until
now been constrained to the US president. Some theoretical approaches for the Latin American (Tsebelis/Alemán 2005) and (East) European context (Tsebelis/Rizova 2007) exist, yet
they are still in want of empirical testing. In the context of parliamentary systems, scholars
have rather concentrated their efforts in this respect on presidential interference in government formation (Amorim Neto/Strøm 2006; Tavits 2008; Schleiter/Morgan-Jones 2009b,
2010). Furthermore, the data used to test hypotheses in these studies is far from being adequate (discussion see below). On the other hand, the accounts of presidential activism that
include legislative powers rely on only little more than anecdotal reference to their use (e.g.
Krupavicius 2008; Tavits 2008; Amorim Neto/Costa Lobo 2009) and thus do not allow for
making broader generalisations. To assemble a set of testable hypotheses regarding use of
veto powers, it is therefore necessary to also to take into account the theoretically and empirically more advanced approaches from the American context.
In U.S.-presidential studies, explanations of presidential activism – and the use of the
presidential veto in particular – are commonly divided into presidency- and president-centred
approaches (Hager/Sullivan 1994). Proponents of presidency-centred accounts assert that
‘most presidents would behave similar in similar contexts’ (ibid. 1081) as the main factors
influencing presidential decision-making – the institutional and political environment – lie
outside incumbents’ influence. President-centred explanations on the other hand stress the
importance of presidents’ individual characteristics and abilities and only ascribe a minimal
role to institutional and environmental factors (ibid). As president-centred variables have
proven difficult in terms of their theoretical foundation (Tavits 2008, 135), measurement and
their ability to produce significant results (e.g. Lee 1975), I will limit my discussion to presidency-centred variables. Thus, I will structure my discussion by differentiating between hypotheses associated with the constitutional framework and the political environment the president faces.
The constitutional framework: Why direct elections matter
The most prominent constitutional factor used in the explanation of presidential activism is
the mode of election. Since Duverger (1978; 1982), most scholars have argued that directly
elected presidents should be more active than their indirectly elected counterparts. It is commonly asserted that indirectly elected presidents lack a ‘legitimacy comparable to that of dep-
uties’ (Duverger 1978, 29; translation by the author) whereas popularly elected presidents
have their own source of legitimacy and are therefore under less constraints when using their
powers. This assumption is shared by the majority of later works (Bunce 1997; Elgie 1999a;
Elster 1997; Metcalf 2002; Protsyk 2005; Shugart 1993; Siaroff 2003). However, according to
Tavits (2008), this argumentation is flawed as the electoral mandate of both directly and indirectly elected presidents ‘is still tied to the specific constitutionally prescribed tasks of the
president’ (Tavits 2008, 33) and there should thus be no difference in the activism of directly
and indirectly elected presidents. While this is generally true, in the early years of democratisation (or after the radical change of the constitution) constitutional stipulations tend to be
comparatively vague and give political actors room for manoeuvre (see de Raadt 2009 with
reference to CEE). Through active early and continued constitutional practice, presidents
might thus be able to carve out a more prominent role for themselves, and thereby promote
their veto power from something originally foreseen as a safeguard against unconstitutional
and controversial bills to a pivotal part of the legislative process.
Scholars have also applied principal-agent models to highlight the increased independence of directly elected presidents compared to their indirectly elected counterparts
(Samuels/Shugart 2006; Tavits 2008). Here, it is argued that indirectly elected presidents are
agents of parliament and depend on the assembly for re-election. Therefore, they should be
less active and confrontational in office. Directly elected presidents lack these constraints and
not only can they be more active, they also need to in order to remain popular with their electorate to ensure re-election (Elster 1997, 227; Samuels/Shugart 2006, 8ff; Tavits 2008, 33f).
Again, this argument is challenged by Tavits (2008). She asserts that ‘punishment’ by the
assembly is unlikely due to fixed presidential terms and high hurdles for presidential impeachment as well as the fact that ‘the assembly that put an indirectly elected president into
power is not the same assembly that decides on his or her reappointment’ (Tavits 2008, 34f).
Indirectly elected presidents should therefore be as independent as popularly elected presidents and principal-agent models cannot be applied (ibid.). However, as super-majorities are
needed to elect a president in the assembly and a successful candidate thus needs support from
several parties (including at least some that are not part of the government), it is likely that the
coalition of parties that elected a president in the first place still disposes of enough votes for a
re-election (or at least enough seats to block the election of another candidate). Presidents are
thus still dependent on parliament even if its composition changes. Furthermore, Elster (1997)
argues that parliament (and also governments) should prefer ‘weak’ (i.e. less influential or
ambitious) candidates as they expect less activism/interference in their affairs. This in turn
means that differences between directly elected presidents – who are chosen by voters for
their promises of policy change – and indirectly elected presidents could exist because of different selection criteria in the respective presidential electorates. On the basis of these theoretical considerations one should therefore conclude that directly elected presidents are more
politically active than their indirectly elected counterparts. Popularly elected presidents do not
need to fear punishment by parliament for their interference in the legislative process and will
thus be able to veto more often. Moreover, these presidents need to show to their electorate
(i.e. the public) that they are actively pursuing voters’ interests and will thus be compelled to
veto more frequently to satisfy these expectation. Indirectly elected presidents on the other
hand are punished for veto activity as it interferes with the business of their voters – the
members of the legislature. One should therefore assume that they veto less frequently.
H1: Directly elected presidents veto more frequently than indirectly elected presidents.
Unfortunately, the effect of direct presidential elections on presidential activism has
until now not been satisfactorily tested empirically. Amorim Neto and Strøm (2006) introduce
the share of non-partisan ministers in government as a proxy to measure presidential interference in government formation. However, this measure does not gauge actual presidential activism but rather the success of assumed (yet not directly observed) presidential interference
in government formation. It is thus hardly a reliable or valid predictor of presidential activism
as – when using the share of non-partisan ministers as the dependent variable – one can only
determine under which conditions presidents are more likely to get what they want rather than
explain when presidents actually use their powers. Other studies have also used Amorim Neto
and Strøm’s measure (Schleiter/Morgan-Jones 2009b; Tavits 2008) but results are overall
inconclusive2.
The political environment: Creating opportunities for activism
While the mode of presidential election is obviously not a factor considered in U.S.-American
presidential studies, factors pertaining to the political environment are the focus of explanations of both European- and American-centred explanations of presidential activism. These
factors change much more frequently than the constitutional order, vary on a quarterly or even
monthly basis and cannot be immediately influenced by the president. As Margit Tavits
2
Tavits argues that the amount of presidential powers is a more valid predictor of presidential activism. However, this argumentation – especially with regard to the success of presidential interference in government formation is at risk of being tautological. I do therefore not include this argument in my discussion and analysis.
(2008) brings the most important of these variables together in her ‘political opportunity
framework’, said framework will constitute the basis for the following discussion.
The general logic of Tavits’ (2008) framework – the only coherent theoretical account
of presidential activism in parliamentary systems to date – builds on the study of intraexecutive conflict and consensus in parliamentary systems with presidents. Tavits argues that
activism by the president is the result of the relative ‘strength of other political institutions and
the constellation of political forces in government and parliament’ (Tavits 2008, 35). These
determine the level of consensus between the president and other institutions and thereby the
political opportunities in which presidents choose to make use of their powers. The lower the
consensus between president and government or parliament, respectively, ‘the greater the incentive and opportunity for presidents to assert their influence’ (ibid. 36). Based on this general set of mechanisms, Tavits identifies several variables which I will discuss below in conjunction with similar approaches from the literature.
The constellation of partisan forces in presidency, government and parliament
The partisan identities of president and prime minister as well as the partisan composition
determine the level of consensus within the executive. When president and government are in
cohabitation, i.e. there is no overlap between their party affiliation or ideological position (in
case the president is an independent), preferences of both executive branches are unlikely to
coincide. Presidents are likely to disagree with legislation initiated by the government and
have an incentive to become active. During unified government (i.e. when presidents face no
partisan opposition in the government), however, these incentives are smaller as presidential
preferences are already being implemented. As the presidential veto is a power to avert
changes to the status quo that are unfavourable to the president, presidents can be expected to
veto more during cohabitation than when relations are unified (or at least neutral 3). Besides
Tavits (2008), Schleiter and Morgan-Jones (2009b; 2010) and Amorim Neto and Strøm
(2006) also argue that cohabitation (defined as the presidential party not being part of the
government) increases presidential activism. Furthermore, the constellation of partisan forces
is used to explain activism of the American president. American scholars have found that partisan opposition in House or Senate increases the likelihood of presidential vetoes (Gilmour
2002; Lee 1975; Rhode/Simon 1985; Shields/Huang 1995; 1997). Given the fact that authors
across regions and approaches agree on the effect of cohabitational/unified intra-executive
3
See the methodology section for a better overview of coding practices and this paper’s approach to model intraexecutive relations.
relations on presidential activism, it is reasonable to conclude that this effect should also be
visible for the use of presidential vetoes.
H2: Presidents will veto most during cohabitation, less when intra- executive relations are neutral
and least when the executive is unified.
Nevertheless, the participation of the president’s party (or their declared preferred party) is
not the only measure of the constellation of partisan forces in the institutional triangle of presidency, government and parliament. Here, an additional factor from American presidential
studies can also be accommodated under the umbrella of Tavits’ framework and applied to the
European context. Scholars of presidential vetoes in the US have argued that a larger seat
share of the presidential party in either chamber of Congress increases the likelihood of legislation being passed that is favourable to the president. According to a logic similar to the one
presented above, activism decreases as with an increasing presidential seat share as presidents’ preferences are already being implemented without their interference (Rhode/Simon
1985; Shields/Huang 1995; 1997). In an extension of the argument by Tavits and others presented above, one can therefore sensibly assume that the higher the seat share of the president’s party the less vetoes will be issued by the president and vice versa.
H3: Presidents veto less when the seat share of their (preferred) party is larger; they veto
more when the seat share is small or when they do not have parliamentary representation.
The weakness of parliament and government
According to Tavits (2008), incentives for presidential activism also arise from the weakness
of parliament and government. A parliament is considered weak when it exhibits a high degree of fragmentation (as measured by the effective number of parties after Laakso and Taagepera, 1978) as this increases transaction costs and makes coordination among parliamentary
actors more difficult. Schleiter and Morgan-Jones (2009b; 2010) concur with this argument –
yet only in relation to presidential interference in government formation. Based on this logic,
it follows with regard to presidential vetoes that higher fragmentation makes it less likely for
parliament to override vetoes of the president, rendering the president’s power not only more
effective but also making it a more frequent phenomenon. While this effect seems logical
enough, it is also possible that increasing parliamentary fragmentation has exactly adverse
consequences. Fragmentation also increases coordination costs on the side of the presidency
and makes it more difficult to foresee whether parliament will override a veto or not. Furthermore, in a highly fragmented legislature a higher degree of consensus among parliamentary parties is required to pass bills (especially one takes into account that majority governments are not the norm many countries).4 Vetoes of bills that are passed based on such a
broad compromise are thus very likely to be overridden. Consequently, parliamentary fragmentation may also decrease the likelihood of presidential vetoes. Nevertheless, for the paper
at hand, I will assume the former given its support in the literature and the fact that the membership of party alliances in fragmented assemblies (both in terms of participating parliamentary parties and deputies) are prone to be relatively volatile and cannot always ensure the repassage of a bill after a veto.
H4: Presidents veto more when parliamentary fragmentation is high.
A measure of fragmentation could also be applied when dealing with the weakness of governments. Nevertheless, another avenue proposed by Tavits (2008) and others proves more
promising. Authors generally agree in the context of presidential activism that minority governments are ‘weak’. With regard to presidential vetoes, this means that governments without
a majority in the legislature will be less likely to withstand presidential vetoes, i.e. muster
majorities to pass the bill again and override them. Again with regards to presidential interference in government formation, Amorim Neto and Strøm as well as Schleiter and MorganJones (2009b; 2010) concur with Tavits (2008) that minority status of the government increases activism in parliamentary systems as it weakens this part of executive branch. Yet, for
the study at hand the minority/majority status might not be the most informative indication.
Given the characteristics of parliamentary reality, such as fluctuations in the government’s
seat share, imperfect attendance of deputies at votes in the legislature, and individual deputies
ignoring the government’s line, even majority governments (especially those with only a slim
majority) can at times have problems to override presidential vetoes. It therefore seems more
adequate to think of the weakness of governments in terms of the size of their seat share (i.e.
the seat share of the parties which are officially part of the government) and express its relationship with presidential vetoes as follows.
H5: Presidential veto more when the government’s seat share is small.
4
It could be argued that political polarisation of the legislature also plays a role here as lower polarisation facilitates cooperation among parliamentary parties whereas higher polarisation makes compromise more difficult.
The electoral cycle
The electoral cycle is another factor mentioned by both European and American scholars (yet
not by Tavits (2008)). Due to the fixed terms of the legislature and the executive in the United
States, scholars of the American president frequently treat the electoral cycle as a constitutional variable. However, most European countries allow for calling early elections. This
means that any variables relating to the electoral cycle exhibit a certain degree of variation
and therefore belong to the group of variables shaping the political environment (and not in
one group with the mode of election which is usually stable over several legislative terms).
Schleiter and Morgan-Jones (2009a) argue with regard to presidential activism in government
dismissal that in these cases the more recently elected branch of the executive enjoys an advantage due to ‘fresher’ legitimacy which gives them more leverage in negotiations with the
other branch. Generalising this logic would then mean that presidents should be more active
in the beginning of their term and the more time has passed since the last parliamentary elections. Interestingly, several scholars of the U.S.-American president come to the same conclusion – although by the ways of a different argument. Cameron (2009), Gilmour (2002),
Shields and Huang (1995; 1997) and Woolley (1991) show that presidents can be expected to
veto more bills in the beginning of their term than in the end. According to Cameron (2009),
this is because presidents will attempt ‘to build a more extreme reputation and thereby extract
concessions in subsequent, related legislation’ (Cameron 2009, 376). In years or Congressional elections, presidents should also veto more highlight key differences between the president’s party and the other parties. While the dichotomy of the American party system and the
fixed election schedule makes the analysis and modelling of such veto strategies easier, the
different nature of parliamentary systems can still be accounted for. Here, we can assume that
not election years but the closeness to the expected date of parliamentary and presidential
elections will influence presidents’ veto behaviour. Therefrom follow the last two hypotheses.
H6: Presidents veto more in the beginning of their term than towards its end.
H7: Presidents veto more when parliamentary elections come closer.
Methodology
The statistical analysis of presidential activism in the European context has until now been
made impossible by the fact that adequate data was unavailable. Furthermore, despite the fact
that event count models have become the norm to analyse the use of vetoes in the U.S.American context, it is not clear which exact model with its underlying assumptions is actually most appropriate for the statistical analysis of this phenomenon. In the following, I discuss
both issues and thereby introduce both my original data set and a negative binomial regression
model for the analysis of presidential vetoes in parliamentary systems.
A new data set on the use of presidential vetoes in CEE
Analyses of presidential interference in government formation often use the share of nonpartisan ministers as a proxy in their models (Amorim Neto/Strøm 2006; Tavits 2008; Schleiter/Morgan-Jones 2009a; 2010) arguing that presidents will under certain conditions prefer a
non-partisan to the prime minister’s nominee. However, the share of non-partisan cabinet
members is then only a measure of the success rate of presidential interference rather than
interference itself and is thus a very indirect measure of presidential activism. Consequently,
data that captures the actual use of presidential powers is needed.
For the study at hand, I have therefore created a comprehensive, new and original data
set on the use of presidential vetoes in nine CEE EU member state ranging from the respective first democratic presidential election in the beginning of the 1990s until December 20105.
With regard to parliamentary systems with presidents, the new EU member states in CEE6
present a kind of natural experiment. All democracies in the region emerged during the same,
comparatively short period of time and share a similar past. Furthermore, all countries had to
master the same challenges. Internally, political actors had to choose a new political system
and adapt to the new institutional and political context. As most states made it their aim to
become members of NATO and EU, they also experienced the same outside pressures and
demands from external actors. The regional closeness also holds factors such as culture and
history relatively constant. On the other hand, there is much variation on the variables of interest to this study, especially patterns of parliamentary and governmental politics vary greatly
from country to country. Furthermore, only few countries had a significant pre-communist
democratic history so that one can observe the influence of different factors without them
5
Due to problems with data availability, the data set does unfortunately not contain data for Bulgaria until
01/2002 as well as data for Romania until 10/2004.
6
Bulgaria, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia, Slovenia.
being blurred by historical effects that would skew the results. Last and for the purpose of this
study most important, all but the president of Slovenia have the power send back bills to parliament for re-consideration and the respective procedures are sufficiently similar7.
The data is structured by monthly observations for every country. For every month, I
have determined how often presidents have used their veto power (which then serves as the
dependent variable in the following analysis) and exactly specified all independent variables,
including even the slightest changes in parliamentary composition and seat shares. To my
knowledge, this presents the first ever cross-country longitudinal data set on the use of presidential powers. To guarantee a high degree of accuracy, data has exclusively been collected
from primary sources, such as databases of the respective parliaments and presidential offices,
parliamentary publications, national law gazettes and reports of the parliamentary research
offices. Secondary sources such as scholarly publications and newspaper articles have only
been used to validate the data. In case of doubt, presidential offices and parliaments were contacted to confirm the correctness of the data.
[Table 1 about here]
An event-count model of presidential vetoes in parliamentary systems
For a long time, political scientists have used ordinary least square (OLS) regression analysis
to analyse the use of presidential powers (Rhode/Simon 1985; Hoff 1991; Woolley 1991).
However, as Shields and Huang (1995; 1997) show this model is far from well-suited because
it predicts values below zero and tends to overestimate coefficients if used on count data. In
the case at hand, the dependent variable – the number of times presidential powers have been
used during a given period of time – only takes integer, non-negative values. Therefore, an
event count regression model should be used (Hilbe 2011).
Event count models are based on a probability distribution, the simplest being the
Poisson distribution. However, as a Poisson regression can only be used when the variance of
the sample does not exceed its mean. As this is rarely the case in the study of presidential vetoes8, Shields and Huang (1997) therefore propose to use a negative binomial distribution
which can be derived as a generalisation from the Poisson distribution (ibid. 443). The negative binomial distribution uses an additional parameter ѵ to account for overdispersion in the
data, i.e. heterogeneity caused by the variance of the data exceeding its mean (Hilbe 2011, 3)9.
7
See Table 3 in the appendix of this paper.
This is also true for the data set used in the study at hand.
9
The Poisson is thus a special case of the negative binomial where ν = 0.
8
Due to the fact that more sophisticated negative binomial regression models are a rather recent addition to the repertoire of statistical models for social scientists, they do not allow for
as much specifications as for instance OLS-models, yet these are not critical for the study at
hand. Thus, a negative binomial regression is most appropriate for the data at hand and can be
expected to yield robust results.
Vetoes in any political systems are a rather rare event (even in Lithuania – the country
exhibiting the most frequent use in my data set – there are only 0.8 vetoes per month) which
means that there are a lot of excess zeroes in the data. These in turn can potentially skew the
results (Hilbe 2011, 346). Studies of vetoes by the U.S.-American president have therefore
often resorted to using years or congressional terms as observational units to avoid this problem. However, legislative and executive terms in parliamentary and semi-presidential systems
are not fixed and key variables such as parliamentary fragmentation can even vary from
month to month. Excess zeroes have thus to be dealt with in another way. Problems with data
collection and coding can be excluded. For the sake of statistically modelling the veto use one
has to assume that zero counts and non-zero counts stem from two different distributions, i.e.
zeros and event counts are created by two independent statistical processes. Depending on the
relationship between these processes, such data can be modelled either by negative binomial
hurdle models or zero-inflated negative binomial models (Cameron/Trivedi 1998, 123; Hilbe
2011, 346ff). Hurdle models assume that one process is based on the other, meaning that the
first process determines whether an event generally occurs or not and the second process then
decides on the number of observed events. This means that some observations – those that are
assigned a zero in the first process – are never at risk of an event whereas the others always
are10. Zero-inflated models on the other hand assume that both processes are independent and
their outcomes are merely mixed in the data. Furthermore, while ‘hurdle models […] estimate
zero counts using different distributions, zero-inflated models incorporate zero counts into
both the binary and count processes’ (Hilbe 2011, 370) by mixing them in accordance with
their proportion in the data (Freese 2001).
In the case at hand, an argument could be made for either one of the models to be used
in the analysis. However, the zero-inflated model is more adequate for the following two reasons. First, the assumptions underlying the zero-inflated model resemble the actual process
more closely than the hurdle model. It may well be that there is an observation where no veto
occurs even though the president would be willing to do so, e.g. when parliament simply does
10
Cameron and Trivedi (1998, 127) exemplify this using the example of patients’ visits to the doctor modelled in
Pohlmeier and Ulrich (1995). In the first process patients first decide whether they go to the doctor or not. In the
second process the doctor specifies the intensity of the treatment for those patients who decided to come.
not pass bills which could potentially be vetoed (after all, presidents will not veto a bill blindly just for the sake of exercising their power). Second, most observations show counts of only
one or two vetoes (61.7% and 22.9% of all non-zero counts, respectively). The results of a
hurdle model would therefore hardly informative as the second step of its calculations is eventually only concerned with the observations in which the remaining 38.3% of vetoes occurred
(n=157).
Statistical model and empirical findings
My hypotheses are tested with a zero-inflated negative binomial regression model as
described above. While a full variable description can be found in the appendix of this paper
(Table 4), I shall note here that the closeness of presidential and parliamentary elections and
the effective number of parties are used in form of their natural logarithm. In addition to the
variables described in my hypotheses, I include two further control variables in the model.
First, the number of bills passed by the legislature during any given month of observation
(also in logged form). The passage of legislation is the natural prerequisite for a presidential
veto; in studies of the U.S.-American president the amount of legislation during a given period of observation has proven to be an important control for the frequency of veto use11 and
also helps to control for country-specific differences in patterns of legislation. Second, I include a dummy for the second presidential term12 to control for the potential existence of a
‘lame duck’-effect. Last, two-stage models such as the one used in this paper often specify
different variables for both processes. However, in the study at hand one can sensibly assume
that both processes are determined by the same variables. Even if the exclusion of some variables from the first stage might be theoretically justifiable at first glance (e.g. the mode of
presidential election or the presidential term as they should have to direct bearing on the passage of objectionable bills) it is more reasonable to assume that the government and legislators take these into account.
As described above, the zero-inflated model consists of two stages; while the first one
(bottom half of Table 2) determines which variables are associated with the inflation in zerocounts using a logit function, the second stage (upper half of table) runs a negative binomial
regression based on all non-zero observations and a number of zero observations (proportional to their occurrence in the data) which according to their constellation of independent varia-
11
Studies of presidential veto use by the U.S.-American president have also used this variable arguing that a
higher amount of legislation increases the likelihood of presidential vetoes. This is because a higher number of
bills should theoretically also be associated with the passage of more bills that are objectionable to the president.
12
In the countries of CEE president can only serve two terms – either consecutively or in general.
bles should have experienced a non-zero count in the first part of the model. When interpreting the coefficient signs, one thus needs to account that in the upper half a positive sign signals a positive association of the variable and an increase in vetoes. In the bottom half, on the
other hand, a positive sign shows a positive association of the respective variable with zeroes
in the data. As the interpretation of the zero-prediction is more difficult to interpret and the
results of the negative binomial half of the model more relevant for the test of the hypotheses,
I turn first to the latter.
[Table 2 about here]
The regression output shows that the coefficients for three out of my seven variables
as well as one of my controls are statistically significant. First and foremost, the results confirm my hypothesis on the effect of popular presidential election on the president. Directly
elected presidents veto more often than their indirectly elected counterparts. This also concurs
with the traditional argument in the literature that the former should be more politically active
due to their increased independence from the legislature and the government as well as – possibly – their increased legitimacy.
While the nature of intra-executive relations and the fragmentation of parliament do
not appear to have statistically significant effects on the number of presidential vetoes, the
variable relating to the seat share of the president does. The coefficient estimates for the presidential seat share is negative and exhibits a high level of significance. This corroborates the
assumptions that presidents who dispose of parliamentary representation should veto less frequently as they are able to influence the content of legislation (indirectly) at an earlier stage.
Presidents with little or no representation, however, lack this possibility and therefore have to
resort to their veto more often. Even though the coefficient estimate for the seat share of the
government does not reach statistical significance, one can at least establish that the sign is
negative and thus pointing in the expected direction.13
In contrast to these findings, only one of the variables relating to the electoral cycle
produces a statistically significant coefficient estimate (although only at 0.1-level). Nevertheless, my assumptions are supported by the result. The negative sign shows that the number of
vetoes increases the closer one moves towards parliamentary elections. Although the specific
mechanisms should still be subject to further testing an theorisation, the assumption that pres13
Another model using a dummy variable for minority government did bring better results (not displayed here);
in fact, the model exhibited a worse overall fit.
idents veto more often before parliamentary elections to highlight policy differences between
parties appears to be valid.
Last, the control variable for the amount of legislation also produces a statistically significant result. As could be expected based on previous studies of vetoes in the United States,
an increase in the amount of legislation passed by parliament and consequently to the president for signature also increases the use of vetoes. Nevertheless, the question if an increase in
legislative activity simply increases the likelihood of bills being objectionable to the president
on political/ideological grounds or if it rather (or at the same time) decreases the quality of
legislation making the president refer the bills back to parliament for procedural reasons still
needs to be answered.
As mentioned above, the logistic part of a zero-inflated model is more difficult to interpret in relation to the hypotheses presented here given that it simply predicts zeroes in the
data (thus ignoring the specific number of vetoes in the non-zero counts). This means that like
a usual logistic regression it only distinguishes between 0 and 1 with the only difference that
it predicts zeroes and not ones. In this section, three none of the seven variables or the controls show significant coefficient estimates. While this shows that theorisation of this process
is needed and possibly the exclusion of variables (or inclusion of new ones), the coefficient
signs can still give us some limited information about the patterns of veto use. Generally, we
should expect coefficient signs to switch between the two parts of the model. For the effect of
direct election (where such a change can be witnessed) this means that popular presidential
elections are not only associated with a higher number of vetoes per month (negative binomial
part) but also with a more observations/months in which the veto is used (logistic part). A
similar change can be seen in the sign of the coefficient estimates of the amount of legislation
passed. While in the negative binomial part of the model one could see that an increase in the
number of bills passed by parliament leads to more vetoes in that observation, the negative
sign in the logistic part suggests that an increase means that there are overall less
months/observations in which no vetoes have been recorded.
Nevertheless, due to the fact that none of the estimates reaches statistical significance
and because the zero-generating process of this model is not sufficiently theorised yet, any too
specific interpretation of these results should be avoided.
Conclusion & discussion
In this paper I have presented the first cross-country statistical analysis of the use of
presidential vetoes in parliamentary and semi-presidential systems. Combining insights from
U.S.-American and European-centred scholarship, I argued that presidents’ use of the veto
can be explained by factors pertaining to the constitutional setting and the political environment as well as those relating to the president as an individual, whereby the latter should be
seen a complementary and not a rival group of explanatory factors.
Using an original data on the use of presidential vetoes in nine CEE democracies from
1990 until 2010, I tested my hypotheses and found most of them confirmed by the results. The
first and probably most important finding is that directly elected presidents are more active in
the use of their veto. It lends credence to the argument political scientists have traditionally
made but never tested empirically on such a large scale and calls for a revision not only of the
arguments made by Margit Tavits (2008) but also the findings of this paper by the ways of
more detailed qualitative analysis.
Another important implication of this analysis is that studies of American and European presidents generally use the same set of explanatory variables (or that variables from one
context can easily be adapted for another). As scholars have not yet set these two bodies of
theory in relation to each other, this opens a common perspective for research on presidents
independently from the parameters of the political system they operate in.
Nevertheless, the paper has also shown that coding practices (especially for intraexecutive relations) need to be reviewed. To allow for generalisation of results, the theoretical
framework also needs to be reviewed and the more light needs to be shed on the motivations
underlying presidential behaviour in different political systems. This is also necessary to benefit from all possibilities of the statistical model. Contrary to hurdle models, zero-inflated
models actually allow for specifying different variables for the two stages. While the ‘all-in’
approach was certainly adequate for this study, one should find a theoretically sound and justified way for choosing the right variables for each step to increase model fit and the predictive power of the model. Only then can one also offer better interpretations of the results of
the logistic part of the zero-inflated regression model.
References
Amorim Neto, Octavio/Kaare Strøm. 2006. “Breaking the Parliamentary Chain of Delegation: Presidents and Non-partisan Cabinet Members in European Democracies”. British Journal of Political
Science 36 (4) 619-643.
Amorim Neto, Octavio/Marina Costa Lobo. 2009. “Portugal's semi-presidentialism (re)considered: An
assessment of the president's role in the policy process, 1976-2006”. European Journal of Political
Research 48 (2) 234-255.
Bahro, Horst/Bernhard H. Bayerlein/Ernst Veser. 1998. “Duverger’s concept: Semi-presidential government revisited”. European Journal of Political Research 34 (2) 201-224.
Bunce, Valerie. 1997. “Presidents and the Transition in Eastern Europe”. in: Kurt von Mettenheim
(ed.) Presidential Institutions and Democratic Politics. Baltimore: Johns Hopkins University Press.
161–176.
Cameron, Charles M. 2009. “The Presidential Veto”. in: William Howell/George C. Edwards III.
(eds.) The Oxford Handbook of the American Presidency. Oxford: Oxford University Press. 362382.
Cameron, Charles M./Nolan McCarty. 2004. “Models of Vetoes and Veto Bargaining.” Annual Review of Political Science 7 400-435.
de Raadt, Jasper. 2009. “Contested Constitutions: Legitimacy of Constitution-making and Constitutional Conflict in Central Europe”. East European Politics and Societies 23 (3) 315-338.
Duverger, Maurice. 1978. Échec au Roi. Paris: Albin Michel.
Duverger, Maurice. 1980. “A New Political System Model: Semi-Presidential Government”. European Journal of Political Research 8 (2) 165-187.
Elgie, Robert. 1999a. “The Politics of Semi-Presidentialism”. in: Elgie, Robert (ed.) SemiPresidentialism in Europe. Oxford: Oxford University Press. 1-21.
Elgie, Robert/Sophia Moestrup (ed). 2008a. Semi-presidentialism in Central and Eastern Europe.
Manchester: Manchester University Press.
Elster, Jon 1997. “Afterword: The Making of Postcommunist Presidencies”. in: Ray Taras (ed.) Postcommunist Presidents. Cambridge: Cambridge University Press. 225-237.
Gilmour, John B. 2002. “Institutional and Individual Influences on the President's Veto”. The Journal
of Politics 64 (1) 198-218.
Hager, Gregory L./Terry Sullivan. 1994. “President-centered and Presidency-centered Expla- nations
of Presidential Public Activity.” American Journal of Political Science 38 (4) 1080-1103.
Hammond, Thomas H./Gary Miller. 1987. “The core of the constitution.” American Political Science
Review 81 (4) 1155-1174.
Hilbe, Joseph M. 2011. Negative Binomial Regression. 2nd edition. Cambridge: Cambridge University
Press.
Hoff, Samuel B. 1991. “Saying No: Presidential Support and Veto Use, 1889-1989.” American Politics Research 19 (3) 310-232.
Krouwel, Andre. 2003. “Measuring presidentialism of Central and East European countries.” Working
Papers Political Science No. 02/2003. Amsterdam: Vrije Universiteit Amsterdam.
Krupavičius, Algis. 2008. “Semi-presidentialism in Lithuania: origins, development and challenges”.
in: Robert Elgie/Sophia Moestrup (ed). Semi-presidentialism in Central and Eastern Europe. Manchester: Manchester University Press. 65-84.
Lee, Jong R. 1975. “Presidential Vetoes from Washington to Nixon.” The Journal of Politics 37 (2)
522-546.
Metcalf, Lee Kendall. 2000. “Measuring Presidential Power”. Comparative Political Studies 33 (5)
660-685.
Protsyk, Oleh. 2005a. “Politics of Intraexecutive Conflict in Semipresidential Regimes in Eastern Europe”. East European Politics and Societies 19 (2) 135-160.
Samuels, David A./Matthew Søberg Shugart 2006. “Presidents, Prime Ministers and Parties: A NeoMadisonian Theory of Party Organization and Behavior”. Paper presented at the Annual Meeting of
the American Political Science Association, Philadelphia.
Schleiter, Petra/Edward Morgan-Jones. 2009a. “Constitutional Power and Competing Risks: Monarchs, Presidents, Prime Ministers, and the Termination of East and West European Cabinets”.
American Political Science Review 103 (3) 496-512.
Schleiter, Petra/Edward Morgan-Jones. 2009b. “Party government in Europe? Parliamentary and semipresidential democracies compared”. European Journal of Political Research 48 (5) 665-693.
Schleiter, Petra/Edward Morgan-Jones. 2010. “Who's in Charge? Presidents, Assemblies, and the Political Control of Semipresidential Cabinets”. Comparative Political Studies 43 (11) 1415-1441.
Shapiro, Robert Y./Martha Joynt Kumar/Lawrence R. Jacobs (eds). 2000. Presidential Power. Forging the Presidency for the Twenty-First Century. New York: Columbia University Press.
Shields, Todd G./Chi Huang. 1995. “Presidential Vetoes: An Event Count Model.” Political Research
Quarterly 48 (3) 559-572.
Shields, Todd G./Chi Huang. 1997. “Executive Vetoes: Testing Presidency- Versus President- Centered Perspectives of Presidential Behavior.” American Politics Research 25 (4) 431-457.
Shugart, Matthew Søberg. 1993. “On presidents and parliaments”. East European Constitutional Review 2 (1) 30-32.
Shugart, Matthew Søberg/John M. Carey. 1992. Presidents and Assemblies. Cambridge: Cambridge
University Press.
Siaroff, Alan. 2003. “Comparative presidencies: The inadequacy of the presidential, semi-presidential
and parliamentary distinction”. European Journal of Political Research, 42(3), 287-312.
Tavits, Margit. 2008. Presidents with Prime Ministers: Do direct elections matter?. Oxford/New
York: Oxford University Press.
Tsebelis, George/Alemán, Eduardo. 2005. Presidential Conditional Agenda Setting in Latin America.
World Politics. 57(3) 396-420.
Tsebelis, George/Rizova, Tatiana P. 2007. Presidential Conditional Agenda Setting in the Former
Communist Countries. Comparative Political Studies. 40(10) 1155-1182.
Woolley, John T. 1991. “Institutions, the Election Cycle, and the Presidential Veto.” American Journal of Political Science 35 (2) 279-304.
Appendix
Table 1: Geographical and temporal coverage of the data set
Country
Bulgaria
Lithuania
Poland
Romania
Slovakia
Period covered
Number of months/
observations
directly elected presidents
107
01/2002 – 12/2010
10/1992 – 12/2010
214
01/1991 – 12/2010
240
10/2004 – 12/2010
75
06/1999 – 12/2010
139
indirectly elected presidents
Czech Republic
01/1993 – 12/2010
215
Estonia
10/1992 – 12/2010
219
Hungary
08/1990 – 12/2010
245
Latvia
07/1993 – 12/2010
209
Slovakia
02/1993 – 05/1999
139
N=1738
Notes: a) Bills returned to first chamber only
Total number of vetoes
24
175
76
25a)
166
75
59
39
35
36
709
Table 2: An event count model of presidential vetoes in CEE
Variable
Coeff
SE
Veto prediction
Direct election
0.72***
0.25
Intra-executive relations
0.14
0.13
Presidential seat share
-2.20***
0.67
Fragmentation of parliament (log)
-0.08
0.25
Governmental seat share
-0.85
0.91
Time until of presidential elections (log)
0.04
0.10
Time until parliamentary elections (log)
-0.25*
0.13
No of bills passed (log)
Second term
0.24***
0.07
-0.14
0.14
0.32
0.67
Direct election
-4.54
5.72
Intra-executive relations
-0.61
1.73
-13.51
39.71
Fragmentation of parliament (log)
-0.50
4.69
Governmental seat share
14.27
23.07
Time until of presidential elections (log)
2.71
4.51
Time until parliamentary elections (log)
-2.31
3.85
No of bills passed (log)
-0.45
0.69
2.72
4.85
-11.25
16.10
constant
Inflate/Excess zero prediction
Presidential seat share
Second term
Constant
ln alpha
0.58***
0.18
alpha
1.79
0.34
N = 1738 ( non-zero observations = 406; zero observations = 1328)
Log pseudo likelihood = -1334.933
Notes: Zero-inflated negative binomial regression model; standard errors clustered on president-cabinet
configurations; *** p<0.005, *p<0.1
Table 3: Veto power of CEE presidents
Country
Override
majority
Relative/absolute
override majority
Further
information
Constitutional
stipulation
Bulgaria
>½
absolute
Czech Republic
>½
absolute
Estonia
>½
relative
§ 107
Hungary
>½
relative
Art 26.2
Latvia
>½
relative
Art 71
Lithuania
>½
absolute
Art 71
Poland
> ²/3 (1989)
> ³/5 (1997)
relative
Romania
>½
absolute
Art. 77.2
Slovakia
>½
relative (1992)
absolute (1999)
Art 87.3 (1992)
Art 87.2 (1999)
Slovenia
n/a
n/a
n/a
Art 101
Const. Acts cannot
be vetoed
Budget cannot be
vetoed
Art 50; 62
Art 27.5 (1989)
Art 18.3 (1992)
Art 122.5 (1997)
Notes: Constitutions and parliamentary rules of the countries studied stipulate a quorum of at least 1/2 of the
members of the assembly for any decision to be adopted.
Source: Krouwel (2001); Tsebelis/Rizova (2007); Elgie/Moestrup (2008).
Table 4: Variable description
Variable
Direct elections
Time until presidential elections
Time until parliamentary elections
Fragmentation of parliament
Governmental seat share
Presidential seat share
Intra-executive relations
No of bills passed (log)
Second term dummy
Description
Dummy variable
0=indirectly elected president
1=directly/popularly elected president
Natural logarithm of the number of months until the date of the next foreseeable presidential election. When earlier elections were scheduled, the
number of months is corrected from the month onwards in which the new
date was announced.
Natural logarithm of the number of months until the date of the next foreseeable parliamentary elections (lower house only). When earlier elections
were scheduled, the number of months is corrected from the month onwards in which the new date was announced.
Natural logarithm of the effective number of parliamentary parties as
measured by Laakso and Taagepera (1978). To reflect the high amount of
independent deputies, every independent is treated as a one-person parliamentary party in the calculation of the value. Furthermore, not the
number of available seats but the number of currently occupied seats in
parliament is used as the reference parameter to reflect changes in the
legislature’s membership more accurately.
Seat share of the government in the lower house of the legislature based
on the number of currently occupied seats.
Seat share of the president’s own party (also: party president was member
of at time of election if there were no other reasons but the presidency to
leave the party) in the lower house of the legislature based on the number
of currently occupied seats. Independent candidates without clear party
affiliation are coded as having no seats in the assembly.
1=cohabitation (no overlap between partisan affiliation of president and
government; in case of independent candidates this category is coded
when policy differences with the government were already clear at the
time of/before the election of president or inauguration of government)
0=neutral relations (no or small overlap between partisan affiliation of
president and government but no major policy differences known at time
of/before election of president or inauguration of government; this category is the default choice for independent candidates)
-1=unified relations (overlap of partisan affiliation of president and government; independent candidates who have shown significant policy overlap known at time of/before election of president or inauguration of government)
Natural logarithm of the bills passed by parliament in a given month and
presented to the president for signature. Excludes items which cannot be
vetoed (budget in Poland; constitutional acts in the Czech Republic).
Dummy variable
0=1st presidential term
1=2nd presidential term