Incumbent Tenure Crowds
Out Economic Voting
Martin Vinæs Larsen∗
September 2015
Prepared for 47th annual meeting of the Danish Political Science Association.
Abstract
We know that electoral support for political leaders are intimately related to
the economic circumstances they provide. However, we do not know whether the
importance of these economic circumstances change during the political leaders’
time in office. In this article I show that as an executive party’s time in office
increase, the electoral support for the executive party becomes more independent
of the economic situation. Additional analyses suggest that the reason for this is
that voters become more certain about the executive party’s quality as the party’s
time in office increases, and therefore the information conveyed by the present
economic situation is less relevant to voters when evaluating the incumbent.
∗
Department of Political Science, University of Copenhagen, [email protected]. The author would like
to thank Bernard Grofman, Robert Klemmensen, David Dreyer Lassen, Kasper Møller Hansen, Michael
Lewis-Beck, Richard Nadeau and Rune Stubager for comments on an earlier draft of this paper.
1
1
Introduction
The electoral support for political leaders is intimately related to the economic circumstances the leaders provide for their constituents (Healy and Malhotra, 2013; Lewis-Beck
and Stegmaier, 2013). A recession, a spike in unemployment or price instability, will, all
else equal, lead voters to unseat incumbent politicians and seat political opponents. However, all else is often not equal. As such, previous literature on the economic antecedents
of electoral behavior have identified extensive variation in the relationship between the
economic situation and electoral support for incumbents (Paldam, 1991; Van der Brug,
Van der Eijk and Franklin, 2007). But why is economic voting more prevalent in some
elections than it is in others? Focusing on how governments differ in their ability to
shape economic outcomes, previous literature has primarily tried to answer this question
by showing that institutions which constrain political power over the economy also constrain the economic vote. (Powell Jr and Whitten, 1993; Hellwig, 2001; Nadeau, Niemi
and Yoshinaka, 2002; Anderson, 2006; Hellwig and Samuels, 2007; Duch and Stevenson,
2008).
In spite of this comprehensive literature on the variation in economic voting, no article
have throughly and systematically examined how the duration of a governments time in
office condition the economic vote. Only two existing studies peripherally address the
relationship between time in office and economic voting (Nadeau, Niemi and Yoshinaka,
2002; Singer and Carlin, 2013), and neither of these directly test the relationship between
the effect of economic conditions on incumbent support and the incumbent’s tenure. As
a consequence, we do not presently know whether voters are more or less likely to punish
or reward incumbents for economic outcomes at different times of their tenure. An
important question, which might help us better predict at which elections the economy is
gonna matter, and provide insight into the mechanisms underpinning the economic vote.
Political leaders vary greatly in how long they are in office. Some are there for a
year. Some for a decade. Some are there for several decades. Conceivably, voters have
stronger or at least different ties to a government which has been around for ten years
than a government which has been around for two years. There is some reason to expect
2
that such changes in the ties between the voters and their government might change
the extent of economic voting. As such, a few studies have examined how the nature of
the relationship between voters and their government can condition the economic vote.
Kayser and Wlezien (2011), for instance, finds that voters who feel close to or identify with
governing politicians from a particular party are less likely to hold this party accountable
for economic outcomes. Tilley and Hobolt (2011) and Malhotra and Kuo (2008) find
similar patterns using experimental data. To the extent that incumbents’ tenure creates
a change in the incumbent-voter relationship, these findings suggest that tenure could
potentially condition the economic vote.
In this article, I argue that as a government’s tenure increases, the ties between the
voter and the government changes in a way, which makes the voter discount economic
conditions when deciding whether to reelect the incumbent. In effect, I argue that incumbents tenure crowds out the economic vote.
Based on previous literature on the theoretical mechanisms underlying the economic
vote, I provide three different explanations for why changing ties between the incumbent
and the voter might lead the latter to discount economic performance as the tenure of
the former increases. (1) A first-impression explanation which suggests that once voters
have gotten an impression of the executive based on the economic performance in the
executive’s first-term, they lock-in on this impression and will not let subsequent economic
performance change their initial beliefs about the executive. (2) A learning explanation
which suggests that the economy is less valuable when evaluating experienced executives,
because voters’ beliefs about the experienced executives have been hardened by years of
exposure to information about their quality. (3) A persuasion explanation which suggests
that experienced executives are able to persuade voters that the country is doing good
economically, irrespective of the actual economic circumstances. These explanations all
predict a negative relationship between tenure and the economic vote, and they are all
predicated on the notion that the basic relationship between voters and governments
change as time in office increases. However, they disagree about how the relationship
change; do voters convince themselves of the virtue or wickedness of the incumbent, do
3
they simply learn more about the incumbent’s quality, or are they more easily persuaded
by the rhetoric of an incumbent with more tenure.
Empirically, I find support for the proposition that economic voting decreases with
time in office using two different datasets. First, I use country-level data on election
results in 327 elections in 33 different countries. Estimating a time-series cross-sectional
model I find that the effect of economic growth on the executive (i.e. presidential or
prime-minister) party’s vote share depends negatively on the tenure of the executive
party. Second, I use the European Election Studies to attain a pooled cross-section of 60
representative national surveys from 10 Western European countries. Using these data
I find that the effect of voters’ perception of the national economy depends negatively
on the tenure of the executive party. Having provided evidence that economic voting
decreases with time in office, I then turn to parsing out which of the different theoretical
explanations that seem most plausible. Based on various empirical tests, I find suggestive
evidence for the learning explanation, but not the first-impression explanation or the
persuasion explanation.
The article concludes that incumbent tenure crowds out economic voting. A conclusion which contributes to the comparative economic voting literature in several ways.
Most importantly, this article demonstrates that it is worth while to go beyond notions
of political control over economic conditions, when trying to explain why the economy
matters more in some elections than it does in others. Specifically, the present study
highlights differences in the voter-incumbent relationship, a factor which have received
scant attention in the extant literature, but which can be used to further our understanding of when the economy matters. Relatedly, the results of this article suggest, that while
economic voters might very well be myopic when it comes to evaluating the economy
(Healy and Malhotra, 2009; Healy and Lenz, 2014), this does not mean that we can avoid
looking at how the past performance of political leaders shape the electoral context the
economic voters operate within (see Krause and Melusky 2014 for a similar point). Finally, the examination of the different theoretical mechanisms underlying the relationship
between tenure and economic voting also adds to the literature which tries to understand
4
the mechanisms underlying the economic vote (e.g. Kinder and Kiewiet 1979; MacKuen,
Erikson and Stimson 1992; Alt, Bueno de Mesquita and Rose 2011). Specifically, the
findings laid out in this article implies that economic voting is a tool voters use to assess
the quality of politicians, and once it has served this purpose, once voters have a good
idea about how qualified politicians are, economic performance becomes less important
for voters.
2
Economic voting and time in office
Research on how the economy shapes the electoral support for incumbents have generally been sensitive to the role timing of economic outcomes might play for both voters
and politicians. For instance, several studies have investigated the extent to which voters are myopic, caring exclusively about their present (and recent) economic situation
(Healy and Malhotra, 2009; Achen, 2012; Healy and Lenz, 2014; Hellwig and Marinova,
2014), and the degree to which incumbent politicians can use this myopia strategically
by creating political business cycles and timing elections (Nordhaus, 1975; Smith, 2003;
Kayser, 2005). Similarly, some studies have tried to discern whether voters focus on past
or (their expectations of) the future economy (Lanoue, 1994; Singer and Carlin, 2013;
Soroka, Stecula and Wlezien, 2015). Even so, extant research on economic voting have
rarely examined the potential role tenure of political leaders might have in moderating
the economic vote.
This is surprising for at least two reasons. First, a large set of studies within the
economic voting literature has identified substantial variation in the economic vote across
time and space (see for instance Paldam 1991; Nadeau, Niemi and Yoshinaka 2002; Duch
and Stevenson 2008), variation which could potentially be explained by differences in the
tenure of the executive parties up for election. However, in charting the sources of this
variation, extant literature has primarily focused on the relationship between politicians
and the economy, suggesting that the clarity of responsibility of the former for the latter
can explain the lion’s share of variation in the economic vote (Anderson, 2007; Duch and
5
Stevenson, 2008). Further, to the extent that the literature on variation in the economic
vote has gone beyond examining the relationship between politicians and the economy, it
has primarily focused on how voters’ predispositions might make them less or more prone
to punish and reward the government for economic outcomes. For instance, showing that
voters who identify with the governing party is less likely to punish it for poor economic
performance (Kayser and Wlezien, 2011) or examining whether politically sophisticated
voters are more or less likely to react to specific types of economic outcomes (Gomez and
Wilson, 2001, 2008).
Second, we know that time in office influences electoral support. After incumbents
are elected they often enjoy a so-called “honeymoon” period, where they enjoy very high
levels of popularity. Further, studies of long-term trends in incumbents popularity have
found that there is a depreciation-effect, a cost of ruling, which means that for each year
an incumbent is in office, the incumbent will, on average, lose public support (Mueller,
1970; Nannestad and Paldam, 1994; Paldam and Skott, 1995). If there are such temporal
trends in the incumbents popularity, it seems likely that there are also temporal trends
in the sources which underlie incumbents’ (un)popularity. That is, the relative weight of
different factors which shape electoral support might very well change during incumbents’
time in office.
In fact, only two studies have examined time in office in conjunction with the economic
vote. Nadeau, Niemi and Yoshinaka (2002) include time in office in a larger index of institional factors (e.g. ideological cohesion of government, number of parties in parliament)
and then looks at whether this index correlates with the economic vote. However, they
do not examine time in office in itself. Singer and Carlin (2013) do examine the effect of
only time in office, but do not look at how this affects the overall economic vote. Instead,
they are interested in whether certain components of economic perceptions play a role at
a certain point in time. Further, neither of these studies examine the effect of “objective”
economic conditions, such as economic growth and unemployment. As such, in spite of
these two interesting studies, we know little about the overall relationship between tenure
and the effect of economic conditions on support for governing politicians.
6
2.1
Explaining the effects of tenure
Why would we expect the economy to matter less for an executives electoral support as
time in office increases? Based on extant literature on economic voting three potential
explanations stand out. Below I lay out each in turn, and then turn to examining empirically whether the economy matters less for the electoral support of more experienced
incumbents.
A first-impression explanation: Drawing on theories of belief formation developed
in social psychology (Lord, Ross and Lepper, 1979; Jonas et al., 2001) studies of political
cognition have shown that once voters have developed a set of beliefs about a political
entity, they are likely to ignore evidence which puts this belief into question (Lodge
and Taber, 2013), what is conventionally called confirmation bias. That voters display
such confirmation bias when attributing blame for economic outcomes has already been
established. For example, studies have shown, that people who identify with a party
in office are more likely to say that the economy is doing well; irrespective of how the
economy is actually doing. (Gerber and Huber, 2010; Tilley and Hobolt, 2011). One can
imagine a similar mechanism explaining why the economy matters less as incumbent’s
time in office increases. If one assumes that when an incumbent is first elected, voters
have very few preconceptions about said incumbent, then voters’ beliefs are likely to be
malleable in this period. Specifically, one of the things these initial beliefs might be
shaped by is the state of the economy. Once a first-impression is formed, however, voters’
beliefs will no longer be responsive to economic performance due to confirmation bias.
Instead, they will ignore any evidence which seems to disconfirm this first-impression;
creating a negative relationship between tenure and the economic vote.
A learning explanation: The theoretical literature on economic voting suggests
that one reason voters are more likely to reelect incumbents in economically good times
is that voters believe that a good economic situation is a signal of a high quality incumbent (Alesina, 1995; Alt, Bueno de Mesquita and Rose, 2011; Duch and Stevenson,
2008). If voters believe the economy conveys such a signal about incumbent quality, this
might explain why voters begin to discount economic performance as the tenure of an
7
executive increase. How so? Imagine we are dealing with a new incumbent which has to
be re-elected in a year in which the economic situation is very good. Since we are dealing
with a new incumbent, voters know very little about the incumbent’s competence; very
little, that is, apart from what they can infer from the economic situation. Accordingly,
they infer that the incumbent is competent. In the next election-cycle the economy is still
doing good, but in the third election cycle the economic situation changes for the worse.
However, when voters have to decide whether to reelect the incumbent for the third time,
they have already been able to partly asses the incumbent’s competence from the incumbent’s previous economic accomplishments; along with other information about the
incumbent’s performance which have come forward during the incumbent’s three terms in
office (e.g. scandals). Therefore the election year economy plays a smaller role in voters
assessment of the competence of the now third-term incumbent than it did in the assessment of the same incumbent after the first term. Accordingly, the learning explanation
suggests that since voters get more and more information about the incumbent as time
in office increases, they are able to form stronger beliefs about the incumbent’s quality,
and these these stronger beliefs are naturally less malleable. Less likely to be swayed by
any specific signal of the incumbent’s quality. The beliefs about incumbent competence
are less likely, in other words, to be swayed by the signal of competence conveyed by
economic situation.1
A Persuasion-explanation: Recent literature on economic voting has focused on
the fact that the economy is mediated; that voters perception of the economy is filtered
through a number of actors. Actors which can potentially determine how voters perceive
and act on the economy. For instance, Kayser and Peress (2012) suggest that the way the
media reports on the economy, leads voters to benchmark the economic performance of
their country against that of other similar countries. Other studies have shown that the
incumbent’s themselves might be able to shape how the economic situation is perceived
(e.g. Huber, Hill and Lenz, 2012). Following this, one can imagine that as an incumbent
becomes more experienced and well-known it can more easily shape voters perception
1
The learning explanation is a rational choice explanation, and as such it should be formalized. I do
this in Appendix A. The key result of this model is similar to what is presented above.
8
of this economy. In this view, incumbents should become more skillful manipulators of
voters economic perceptions as time in office increase through, for instance, extended
control of government agencies which report on the economy or a more sophisticated
handling of the media. This would lead to a negative relationship between time in office
and the economic vote as time in office increased, since experienced incumbents should
be able to dislodge voters perception of the economy from the actual economic situation.
Convincing voters that the economy is doing well even if the economy is doing poorly.
3
Country-level empirics
Is there in fact an empirical relationship between tenure and the extent of economic
voting? I begin answering this question by examining a country-level dataset of elections.
This type of data has been used to study variation in the economic vote in numerous other
studies (Whitten and Palmer, 1999; Powell Jr and Whitten, 1993; Hellwig and Samuels,
2007; Kayser and Peress, 2012). The methodology underlying these studies and this study
is that one can measure the level of economic voting by looking at the correlation between
economic indicators and the support for leading politicians, and, in turn, use variation
in this correlation to infer whether specific factors, such as tenure, change the degree to
which voters hold politicians accountable for economic outcomes. For instance, Hellwig
and Samuels (2007) use such an approach to show that the effect of economic growth on
incumbent politicians vote share is moderated by economic openness.
The chief advantage of this approach is that it sidesteps problems of endogeneity
related to using voters perception of the economy by using objective economic indicators
(Kramer 1983, Van der Brug, Van der Eijk and Franklin 2007, 26). The chief disadvantage
is that the economic indicators used are aggregates, which make them quite noisy (Duch
and Stevenson, 2008, 26). To overcome this this problem I try to use a large sample of
elections and carefully specify the statistical model.
9
3.1
Data and model
I use a dataset of 327 elections across 33 countries (see appendix B for a list of countries
and elections). The key dependent variable is percentage point change in electoral support
for the executive party at legislative elections (∆yit ).2 The executive party is the party
which had primary control with the executive branch at the time of the election (i.e.
the party of the prime-minister or the president). Using the executive party rather
than the parties in government is common in the literature (see for instance Duch and
Stevenson 2008). Further, several studies have showed that the executive party is much
more prone to electoral judgment among voters than other governing parties (Fisher and
Hobolt, 2010). To get data on a wide cross section of elections I use and extend datasets
developed by Kayser and Peress (2012) and Hellwig and Samuels (2007). These include
information on the vote share the executive party received at each election as well as the
previous election.
The key independent variables are economic growth (gr) and tenure in office (ten).
Economic growth is a proxy for economic conditions in the country and is measured as
GDP per capita growth in the election year (pct.). This variable is used because it is
available for a large cross section of elections and because it has been widely used in the
previous literature. Data on economic growth was taken from the World Bank’s database.
Tenure in office is measured as the number of years since the election in which the current
executive party got into power. Data on tenure is taken from the database on political
institutions (Beck et al., 2001), and has been extended by the author to create better
coverage for the electoral variables. See appendix C for descriptive statistics.
Turning to modeling, I set changes in electoral support as a linear function of tenure,
economic growth and an interaction between the two. I also include time fixed-effects (π)
in order to control for any time-trends in tenure, growth and electoral support. Finally, I
include the level of electoral support for the executive party at the last election to control
for the size of the executive party (yit−1 ). As such, the model I estimate can be described
2
As such, even in presidential systems where the president is directly elected by the voters, it is not
support for the president which is used but rather votes for the president’s party in the legislature. This
makes the different system’s more comparable. The election results used are from the lower house if the
legislature is bicameral.
10
as:
∆yit = αyit−1 + β1 grit + β2 tenit + γgrit × tenit + πt + it
(1)
Where the coefficient of interest is γ, which signifies the change in the effect of economic growth as tenure increases. A coefficient which I expect to be negative.
3.2
Results
Table 1 presents the key estimates from model (1) in column one. It is estimated using
an OLS regression, and the standard errors are estimated with clusters on the countrylevel. The base-line growth and tenure effects should be interpreted as the effect of the
variable when the other variable is held at zero. The base-line effect of economic growth
is estimated to be 0.42, and can thus be understood as the (theoretical) effect of economic
growth on change in electoral support if you run for reelection without any tenure.
The variable of interest is the interaction between economic growth and tenure. The
interaction is statistically significant and negative, suggesting that the positive effect of
economic growth at the beginning of an executive party’s tenure diminishes over time.
Specifically, the estimate suggest that each year the effect of economic growth on electoral
support drops 0.04 from the starting point of 0.42. As such, this model suggest that after
about ten years in office the effect of economic growth is essentially zero.
How sensitive is this finding to the current model specification? To investigate this,
the model is extended in two ways. The first extension is inclusion of country fixed-effects
to control for any time invariant country-specific factors. This is done in column two of
table 1. This leaves the results substantively unchanged.
The second extension is the inclusion of leader-fixed effects. That is, a dummy for
each of the 128 different incumbents in the dataset.3 Including the leader-fixed effects
means that any factors which are constant within the same incumbent is omitted from the
estimation of the interaction. As such, the model estimates the interaction by comparing
3
The leader-fixed effects counts an executive party which returns to power after being defeated as a
new incumbent. For instance, the United Kingdom has five different incumbents in the data set across
11 elections, in spite of the fact that only two different parties were in power in at these elections.
11
Table 1: OLS regression of change in electoral support for the executive party
(1)
(2)
(3)
∗
∗
Lagged executive party voteshare -0.19 -0.28 -0.33∗
(0.05) (0.07) (0.12)
Economic growth
0.42+ 0.44+ 0.54∗
(0.22) (0.22) (0.21)
Tenure
-0.02 -0.17 -1.39∗
(0.08) (0.13) (0.61)
Economic Growth * Tenure
-0.04∗ -0.04∗ -0.05∗
(0.02) (0.02) (0.02)
Time fixed-effects
X
X
Country fixed-effects
X
X
Leader fixed-effects
X
2
R
0.25
0.42
0.69
RMSE
6.95
6.49
5.32
Observations
327
327
327
1
.5
0
−.5
Effect of growth on electoral support
Country-clustered standard errors in parentheses.
Model (3) also includes a linear control for time.
+
p < 0.10, ∗ p < 0.05
0
5
10
15
Years in Office
Figure 1: Effects of economic growth on support for executive party across levels of
tenure with 90 pct. confidence intervals. Bar plot shows the density of the years in office
variable.
12
the degree to which the same executive party is punished (or rewarded) for the economic
situation across elections, rather than by comparing how harshly different executive parties with different levels of tenure were punished. The leader fixed-effects are included in
the model estimated in the third column of table 1. Because the leader fixed effects takes
up many degrees of freedom, the year fixed-effects are omitted. Instead a linear control
for time is included. In this specification the interaction-effect is virtually unchanged and
statistically significant. Figure 1 plots the interaction using this specification.
Based on the consistencies in the significance, size and direction of the interaction
effect, there seems to be strong evidence that economic growth becomes a less important
determinant of the executive party’s vote share as tenure increases. The economy seems
to matters less for more experienced incumbents.
Before moving on, one alternative explanation for the findings deserves to be discussed;
namely, that the negative correlation between tenure and economic voting is due to
strategic election-timing. As such, some might argue, that what we see above has nothing
to do with how voters evaluate political leaders with different levels of tenure, but is simply
a reflection of the fact that certain types of leaders call early elections. In appendix D
I examine this alternative explanation by trying to control away election-timing in two
different ways; (1) including a control for number of times an incumbent has called
an election and (2) restricting the sample of elections to countries with fixed terms.
The results presented in the appendix show that in the most demanding specification,
including leader fixed-effects, the interaction remains negative and statistically significant.
4
Individual-level empirics
Having established a relationship between economic voting and the tenure of the executiveparty at the country-level, I now explore the same relationship at the individual-level.
Specifically, I try to delineate whether the relationship between voters perceptions of the
economy and support for the executive party follow the same pattern as the one identified above. That is, to what extent voters rely less on their perceptions of the national
13
economy when deciding whether to vote for a more experienced incumbent.
To do this I closely follow a recent study by Nadeau, Lewis-Beck and Bélanger (2013),
who investigated the relationship between national economic perceptions and the vote for
executive parties in 10 Western European countries across the past twenty years. This
gives me a well-established model of the economic vote, and allows me to simply extend
this model to include an interaction between tenure and the economic vote.
It is important to highlight what I hope to gain from this replication of the findings laid
out above. First, since results in the comparative economic voting literature are known
to be quite unstable (Paldam, 1991), it makes sense to replicate the findings made above
on a new dataset. As such, this will make it more likely that what I found above is in fact
a generalizable result. Second, using the European Election Study, which runs outside
the national election cycles, makes it possible to sidestep any additional concerns that
one might have concerning the extent to which the results are driven by election-timing.
Finally, using a data-set which focuses on economic perceptions rather than objective
economic conditions help rule out certain theoretical explanations for the findings (cf.
section 5).
4.1
Data and model
I use the European Election Studies (ESS). The EES is a survey of all EU countries which
have been conducted every fifth year since 1979. They are fielded in the year of European
Parliamentary elections, and their timing is therefore somewhat independent of national
elections. I use the six Europe-wide studies which have been conducted since 1989 (i.e.
’89, ’94, ’99, ’04, ’09 and ’14), as these are the only surveys which include questions about
national economic perceptions as well as vote intention in national elections. I focus on ten
countries which have participated in all five survey-rounds: Denmark, France, Germany,
Greece, Ireland, Italy, the Netherlands, Portugal, Spain and the United Kingdom.4 This
4
This is all countries which have been members of the EU since 89 except Luxembourg and Belgium.
Luxembourg was excluded because its national sample size was so much smaller than that of the other
countries. Belgium was excluded because the vote intention question was not asked in 2004, and because
the bilingual nature of the country makes it a hard fit for a pooled cross-section (see Nadeau, Lewis-Beck
and Bélanger 2013)
14
gives me 60 cross-sectional national surveys, which can be pooled to test whether the role
of economic perceptions change across the tenure of the executive party.
Turning to indicators, the key dependent variable is whether voters would vote for
the executive party if a national legislative election was held tomorrow (reelect).
The key independent variables are tenure and national economic perceptions. Tenure
is measured as the number of years the executive party had been in power at the time of
the survey (ten). National economic perceptions (N EP ) are measured using a question,
which asked the respondent whether the economic situation in their own country had
gotten better or worse in the past 12 months.
I employ the same control variables as Nadeau, Lewis-Beck and Bélanger (2013),5
that is, respondents ideology (ideology), self-perceived class (class) church attendance
(religion), a dummy indicating whether the respondent voted for the executive party at
the last election (reelectlag , see appendix C for exact question wording of all questions).
All variables were rescaled to go from zero to one, and recoded so that higher values were
likely to mean a higher probability of voting for the executive party.6
I model the probability that voters vote for the executive party as a logistic function of
national economic perceptions, tenure, an interaction between the two and the individual
level controls. As such, the model I estimate can be described as:
P r(reelect) = logit(α0 + α1 N EPijt + α2 tenit + γtenit × N EPijt + Xijt β + ijt )
(2)
Where i indicates country, t year and j the respondent. X is row vector of the control
variables ideology, class, religion and reelectlag and β is a column vector of coefficients
attached to these controls. The coefficient of interest is once again γ, which signifies the
change in the effect of national economic perceptions as tenure increases, and which I
expect to be negative.
5
I exclude a control measuring time since last election, since this is very closely related to tenure.
As such, religion, class, and ideology were coded differently across different surveys to take differences
in the ideological position of the executive into account.
6
15
4.2
Results
I estimate the parameters of model (2) using a multi-level logistic regression, clustering
the standard errors at the country-level, and estimating random effects at the survey-level.
Column one in table 2 presents the key estimates from the model. Ideology, class,
religion and lagged executive party vote all have the expected signs, and, apart from
religion, are statistically significant. The base-line economy and tenure effects should
(once again) be interpreted as the effect of the variable when the other variable is held at
zero. The base-line effect of national economic perceptions is estimated to be 1.86, and
can thus be understood as the (theoretical) effect of going from one end of the national
economic perceptions scale to the other on the logit probability of voting for an executive
party without any tenure.
The key variable of interest is the interaction between economic growth and tenure,
which signifies how this effect changes as tenure increases. The interaction is statistically
significant and negative, suggesting that the positive effect of the respondents’ perception
of the national economy at the beginning of an executive party’s tenure diminishes over
time.
We also investigate the sensitivity of these individual-level findings to the model specification. In column 2 we include leader fixed-effects (cf. the country-level data). This
more demanding model does not change the conclusions, as the interaction remains negative and statistically significant. In column 3 we introduce leader by time fixed-effects,
in effect introducing a dummy for each of the sixty surveys (survey fixed effects). The
interaction remains negative and statistically significant.
In order to investigate the consequences this negative interaction has for the effect of
economic perceptions on the actual probability of voting for the executive party I derive
average marginal effects across different levels of tenure based on the model with survey
fixed-effects and plot them in figure 2. This figure reveals that the average marginal
effect of national economic perceptions is substantially reduced as tenure increases. For
an executive party with one year of tenure the effect of a voter going from one end of
the economic perception scale to the other is a change in probability of voting for the
16
Table 2: Logistic regression of voting for the executive party
National Economic Perceptions
Tenure
NEP * Tenure
Lagged executive party vote
Ideology
Religion
Class
Survey random effects
Leader fixed effects
Survey fixed effects
Log Likelihood
Observations
(1)
(2)
(3)
1.86∗
(0.24)
0.05+
(0.03)
-0.05∗
(0.02)
4.38∗
(0.16)
2.32∗
(0.24)
0.07
(0.09)
0.32∗
(0.15)
X
1.89∗
(0.22)
-0.07∗
(0.03)
-0.06∗
(0.02)
4.39∗
(0.16)
2.34∗
(0.25)
0.08
(0.09)
0.30+
(0.16)
X
X
1.88∗
(0.22)
.
.
-0.06∗
(0.02)
4.37∗
(0.16)
2.34∗
(0.25)
0.08
(0.09)
0.30+
(0.16)
-10,231
39,556
-10,393
39,556
X
X
-10,129
39,556
Country-clustered standard errors in parentheses.
Tenure omitted in model (3) because of the survey fixed-effects.
+
p < 0.10, ∗ p < 0.05
17
executive party of about 15 percentage points. For an executive party with about 15
years of tenure the same change leads to a seven percentage point change. A comparison
of the average marginal effect at one years of tenure and fifteen years of tenure reveals
.15
.1
.05
0
Effect of NEP
.2
that this decline is statistically significant (p < 0.05).7
0
5
10
15
Years in Office
Figure 2: Average marginal effects of national economic perceptions on probability of
voting for the executive party across levels of tenure with 90 pct. confidence intervals.
Bar plot shows the density of the years in office variable.
Taken together, these findings point in the same direction as those found in the
country-level data. As such, the economy loses at least part of the influence it has on the
electoral support for the executive party, as the tenure of the executive party increases.
5
Assessing alternative explanations
Above we identified a negative relationship between time in office and economic voting. As
such, voters seemed to disregard the executives economic performance as time in office
increased. However, what theoretical mechanism underlies this negative relationship?
Below I try to discern which of the three already proposed explanations (persuasion,
7
This result is robust to a two-step estimation procedure, cf. appendix E.
18
first-impression and learning, cf. section 2) are likely to be responsible for the negative
relationship between economic voting and tenure. To do this I reexamine the findings
already laid out and offer some new empirical evidence. However, before engaging with
the different explanation, a caveat is in order. Rival theoretical mechanisms are difficult
to disentangle (Bullock, Green and Ha, 2010), and as such the presented evidence for
and against the different mechanisms proposed here will not be as strong as the evidence
presented for the main empirical finding; that economic voting decreases with time in
office. However, even if no definitive evidence can be laid out, it seems valuable to try
and leverage the empirical material at hand to get a better handle on why economic
voting decreases with time in office.
The easiest explanation to discount is perhaps the persuasion explanation. The persuasion explanation suggested that the decrease in economic voting as time in office
increased was a result of a similar increase in the ability of incumbents to divorce voters perception of the economy from its objective nature. That experienced incumbents
were able to convince voters in a bad economy that the economy was doing well. While
the persuasion explanation could account for the pattern identified in the country-level
data, where economic growth became more independent of electoral support as tenure
increased, it has a harder time explaining the individual-level data. As such, the individual level data revealed that not only did objective economic indicators matter less as
time in office increased, so did perceptions of the economy. The persuasion explanation
cannot account for a decrease in the extent to which voters punish incumbents based on
perceptions of a deteriorating economy. As such, the persuasion explanation offered no
account of why voters, who are not persuaded by the incumbent to perceive the economy
as doing well, should neglect to hold the incumbent electorally accountable. However, in
the individual level data such a decline in the effect of economic perceptions is just what
we see. This casts serious doubts upon the merit of the persuasion explanation.
This leaves the first-impression and the learning explanations. These are harder to
separate as they both predict a decreasing effect of both economic conditions and perceptions. In the first-impression explanation the economy is ignored because of confirmation
19
bias. In the learning explanation the economy is discounted because voters already know
the lessons about the incumbent which the economy can impart. Even so, after examining the data in more detail, the learning explanation seems more plausible. This is based
on two observations.
(1) The first-impression explanation was premised on the notion that in the beginning
of the incumbent’s term voters beliefs about the incumbent were not strong enough for
their information processing to be marked by confirmation bias, but as time in office
increased beliefs hardened and confirmation bias kicked in. An implication of this seems
to be that all types of confirmation bias related to the incumbent increase during an
incumbents time in office. Specifically, the same mechanisms underlying the confirmation
bias in the evaluation of the incumbent should create confirmation bias when it comes to
evaluating the incumbent’s results (Tilley and Hobolt, 2011). One such result is providing
a good economic situation (Gerber and Huber, 2010). As such, those who vote for the
incumbent in the first election should not subsequently be strongly motivated to perceive
the economy differently than those who did not. Conversely, later in the incumbents term,
those who have voted for the incumbent have formed strong beliefs about the incumbent,
and accordingly those who voted for the incumbent should be more motivated to view
the economy in a positive light, whereas those who did not vote for the incumbent should
be more motivated to view the economy in a negative light. Accordingly, if the first
impression explanation is correct, this would imply that as the incumbents time in office
progresses, the perception of the economic situation among those who voted for the
incumbent at the previous election to divert more strongly from the perception of those
who did not vote for the incumbent.
In order to examine whether any such polarization occur, I examine the individuallevel data. Specifically, I estimate a linear regression with national economic perceptions
as the dependent variable. As the independent variables I include lagged executive party
vote share, tenure and an interaction between the two. The coefficient attached to this
interaction tells us whether the gap in economic perceptions between those who voted
for the incumbent at the previous national election and those who did not increased with
20
time in office. Something we would expect if the first-impression explanation is accurate.
I also include the remaining individual level controls used in table 2. I estimate three
different versions of this model, which use either random effects, leader fixed-effects or
time by leader fixed effects. The key estimates from these three models are displayed in
table 3.
Table 3: Does motivated reasoning increase with tenure? Linear regression of NEP
Lagged executive party vote
Tenure
Vote * Tenure
Ideology
Religion
Class
Survey random effects
Leader fixed effects
Survey fixed effects
Observations
(1)
(2)
(3)
0.06∗
(0.01)
-0.00
(0.01)
-0.00
(0.00)
0.11∗
(0.02)
-0.01
(0.01)
0.03+
(0.02)
X
0.07∗
(0.01)
-0.02
(0.01)
-0.00
(0.00)
0.11∗
(0.02)
0.02
(0.04)
0.01
(0.02)
X
X
0.06∗
(0.01)
-0.09∗
(0.00)
-0.00
(0.00)
0.11∗
(0.02)
-0.01
(0.01)
0.03+
(0.02)
43,771
X
X
43,771 43,771
Country-clustered standard errors in parentheses
+
p < 0.10, ∗ p < 0.05
As can be seen from these models there is a gap in economic perceptions between those
who voted for the incumbent at the last election, and those who did not. However, this
gap does not increase with tenure. The interaction is substantively small, negative and
statistically insignificant. As such, unlike what the first-impression explanation implies,
there is no increasing polarization in economic perceptions as time in office increases.
(2) The learning explanation suggested that the reason voters became less responsive
to economic performance as incumbents’ tenure increased was that voters became more
sure about incumbents quality as tenure increased. This, as time invariably meant more
information about incumbent quality, and as a result the marginal value of information
about the incumbent’s quality relayed by the (present) economic situation decreased. As
21
such, the learning explanation suggests that the important difference between new and
experienced incumbents resides in the fact that voters are more certain of experienced
incumbents quality.
To examine whether voters became more certain in their beliefs about incumbents’
quality as time in office increased, we once again turn to the individual-level data. Specifically, we examine a question posed in all sixty surveys; a question which asked how
probable it was that the respondent would ever vote for the executive party. Answers
were given on a 1 to 10 scale going from “not at all probable” to “very probable”. Following the learning explanation we would expect that as the tenure of the executive party
increase, voters become either more sure that they will vote for or more sure that they
will not vote for the executive party. Accordingly, it is interesting to examine whether
voters with more experienced incumbents move towards the end-points of this scale. In
order to investigate this, I constructed two variables; one variable which measures the
percent of voters that picked the two end-ponts of the scale (1,10), and another variable
which simply measured the distance from the midpoint of the scale (5.5). Both variables
were recoded to go between 0 and 100.
To examine the effect of tenure on certainty I estimate three regressions for each of
the two dependent variables. The models estimated were multi-level logit models for
the binary dependent variable and multi-level linear regression for the certainty index. I
estimate one model with just the certainty variable as the dependent variable and tenure
as the independent variable, one model which includes country fixed-effects as well as one
model which includes both country and time fixed effects.8 Estimates from these models
are reported in table 4.
The models all show a positive relationship between tenure and the certainty variables. The tenure-coefficients are not all statistically significant. However, in the most
demanding specifications, which include both time and country fixed effects, the coefficients are statistically different from zero for both dependent variables. As such, there
seems to be evidence for voters becoming more certain about their vote choice as tenure
8
Since I am interested in the main effect of tenure, rather than the interactive effect, and since I only
have sixty independent observations on tenure it is not feasible to include the leader fixed effects.
22
Table 4: Do voters become more certain of their executive party vote choice as time in
office increases?
Distance (midpoint)
(1)
(2)
(3)
Tenure
Constant
Country fixed effects
Year fixed effects
Log Likelihood
Observations
0.28
(0.18)
65.65∗
(2.05)
0.49∗
(0.22)
66.22∗
(1.02)
X
-295,815
59,753
-295,805
59,753
0.30∗
(0.10)
74.86∗
(0.95)
X
X
-295,780
59,753
Pr(min,max)
(4)
(5)
(6)
0.03
(0.02)
-0.52∗
(0.18)
0.04
(0.02)
-0.52∗
(0.12)
X
-37,645
59,753
-37,640
59,753
0.03∗
(0.01)
0.09
(0.08)
X
X
-37,585
59,753
Country-clustered standard errors in parentheses.
+
p < 0.10, ∗ p < 0.05
increases.
Taken together, the above exploration of the different theoretical explanations of why
economic voting decreases with time in office seem to vindicate the learning explanation.
As such, while we failed to find evidence for empirical implications of the persuasion and
first impression explanations, we did find evidence for an empirical implication of the
learning explanation.
6
Concluding remarks
The British parliamentary elections of 1997 and 2001 featured two very different incumbents. One was the Conservative party. In power for 18 years and headed by John
Major. Prime Minister for seven years and cabinet member for the last ten. The other
was the Labour party. In power for four years headed by Tony Blair. Prime minister
with a shiny new cabinet. As British voters searched for clues in 1997 and 2001 about
the quality of the incumbent, some probably examined the economic situation. When
these voters decided to what extent the economic situation should be leveraged in their
eventual electoral decision, did the fact that these incumbents were so different matter?
Did the fact that the incumbent up for election in 1997 had been in power for almost
two decades make them use the economy differently than in 2001 when the incumbent
23
had only been in power for four years? To a large extent questions like these can not be
answered by examining existing literature on economic voting, which has generally given
sparse attention to how differences in the relationship between voters and the incumbent
might moderate the economic vote. Instead, this literature has primarily focused on how
political and economic institutions affect voters propensity to hold incumbents electorally
responsible for the economy. Institutions which did not change between the 1997 and
2001 elections. In this article I have tried to amend this by examining whether a potentially important factor in the incumbent-voter relationship affects the extent of economic
voting; the time in office of the incumbent.
Accordingly, I have examined whether there is a negative relationship between economic voting and time in office. Specifically, I have shown that the electoral support for
executive parties become more independent of the economic situation as their time in
office increases. A finding which was documented using two markedly different datasets,
and which was robust across various demanding statistical specifications within each
dataset. I presented three different theoretical explanation for why this might be so; a
first-impression, a learning and a persuasion explanation. The explanation which was
most empirically persuasive was the learning explanation, which suggested that when an
incumbent with more tenure runs for office the voters have stronger prior beliefs about
this incumbent’s quality than when an incumbent with less tenure runs for office. This,
as voters have used past economic accomplishments along with other information to form
these strong beliefs about the experienced incumbents competence. As a result voters are
less likely to be persuaded about the incumbents quality based on the single data point
which the economic situation constitutes. As such, returning to the example above, the
present study suggest that when deciding whether to reelect the British incumbents,
voters relied more on the economic accomplishments of the relatively new Labour administration than they did the economic accomplishments of the relatively old Conservative
administration. A difference borne out of voters multitude of past experiences with the
Conservative incumbent, and their lack of similar experiences with the Labour incumbent.
Most directly, these findings add to the large literature on how economic outcomes
24
shape candidate selection. Specifically, this study adds to this literature by identifying a
new source of variation in this economic vote. Put in more general terms, this study shows
that systematic differences in the incumbent-voter relationship might interact with the
economic vote, and accordingly it highlights the importance of examining this relationship
if we want to get a better understanding of how and when the economy matters.
These findings also suggest that more research is needed on which factors become
more important as the explanatory power of the economy wanes. Do voters, for instance,
become more interested in the ideological position of the executive party and/or specific
policies as tenure increases, or do electorates simply become less malleable as governing
politicians become more experienced?.
Finally, on the face of it, the findings above seem to have some potentially worrying
implications for electoral accountability. As such, if politicians know that as time in office
increase they are less likely to be punished for adverse economic outcomes, this naturally
reduces their disincentive to provide such adverse outcomes. However, these potential
concerns seem to be somewhat alleviated by the fact that the apparent theoretical mechanism which drives this behavior is learning.9 As such, the reason voters rely more on
the election year economy in the beginning of an incumbents term, is simply that the
election year economy is one of the only available pieces of information about the incumbents quality. Accordingly, the negative relationship between time in office and economic
voting cannot be interpreted as the electorate loosening the reins on the incumbent. Instead, it seems that as time in office grows the potential sources of information on the
quality of incumbents grows along with it. This makes voters less attentive to the sign
of incumbent quality provided by the election year economy, but it does not make voters
are less attentive to the total sum of signs of incumbent quality.
9
These concerns are also alleviated by the fact that tenure does not seem to have adverse effects on
economic growth or voters perceptions of economic conditions, cf. appendix F.
25
References
Achen, Christopher H. 2012. When Is Myopic Retrospection Rational? In APSA 2012
Annual Meeting Paper.
Alesina, Alberto. 1995. Partisan politics, divided government, and the economy. Cambridge University Press.
Alt, James, Ethan Bueno de Mesquita and Shanna Rose. 2011. “Disentangling accountability and competence in elections: evidence from US term limits.” The Journal of
Politics 73(01):171–186.
Anderson, Cameron D. 2006. “Economic voting and multilevel governance: a comparative
individual-level analysis.” American Journal of Political Science 50(2):449–463.
Anderson, Christopher J. 2007. “The end of economic voting? Contingency dilemmas
and the limits of democratic accountability.” Annu. Rev. Polit. Sci. 10:271–296.
Beck, Thorsten, George Clarke, Alberto Groff, Philip Keefer and Patrick Walsh. 2001.
“New tools in comparative political economy: The Database of Political Institutions.”
the world bank economic review 15(1):165–176.
Bullock, John G, Donald P Green and Shang E Ha. 2010. “Yes, but what’s the mechanism?(don’t expect an easy answer).” Journal of personality and social psychology
98(4):550.
Duch, R.M. and R.T. Stevenson. 2008. The Economic Vote: How Political and Economic
Institutions Condition Election Results. Political Economy of Institutions and Decisions
Cambridge University Press.
Fisher, Stephen D and Sara B Hobolt. 2010. “Coalition government and electoral accountability.” Electoral studies 29(3):358–369.
Gerber, Alan S and Gregory A Huber. 2010. “Partisanship, political control, and economic
assessments.” American Journal of Political Science 54(1):153–173.
26
Gomez, Brad T and J Matthew Wilson. 2001. “Political sophistication and economic
voting in the American electorate: A theory of heterogeneous attribution.” American
Journal of Political Science pp. 899–914.
Gomez, Brad T and J Matthew Wilson. 2008. “Political sophistication and attributions of
blame in the wake of Hurricane Katrina.” Publius: The Journal of Federalism 38(4):633–
650.
Healy, Andrew and Gabriel S Lenz. 2014. “Substituting the End for the Whole: Why Voters Respond Primarily to the Election-Year Economy.” American Journal of Political
Science 58(1):31–47.
Healy, Andrew and Neil Malhotra. 2009. “Myopic voters and natural disaster policy.”
American Political Science Review 103(03):387–406.
Healy, Andrew and Neil Malhotra. 2013. “Retrospective Voting Reconsidered.” Annual
Review of Political Science 16:285–306.
Hellwig, Timothy and Dani M Marinova. 2014. “More Misinformed than Myopic: Economic Retrospections and the Voter’s Time Horizon.” Political Behavior pp. 1–23.
Hellwig, Timothy and David Samuels. 2007. “Voting in Open Economies The Electoral
Consequences of Globalization.” Comparative Political Studies 40(3):283–306.
Hellwig, Timothy T. 2001. “Interdependence, government constraints, and economic
voting.” The journal of Politics 63(04):1141–1162.
Huber, Gregory A, Seth J Hill and Gabriel S Lenz. 2012. “Sources of bias in retrospective decision making: Experimental evidence on voters’ limitations in controlling
incumbents.” American Political Science Review 106(04):720–741.
Jonas, Eva, Stefan Schulz-Hardt, Dieter Frey and Norman Thelen. 2001. “Confirmation
bias in sequential information search after preliminary decisions: an expansion of dissonance theoretical research on selective exposure to information.” Journal of personality
and social psychology 80(4):557.
27
Kayser, Mark Andreas. 2005. “Who surfs, who manipulates? The determinants of opportunistic election timing and electorally motivated economic intervention.” American
Political Science Review 99(01):17–27.
Kayser, Mark Andreas and Christopher Wlezien. 2011. “Performance pressure: Patterns of partisanship and the economic vote.” European Journal of Political Research
50(3):365–394.
Kayser, Mark Andreas and Michael Peress. 2012. “Benchmarking across borders: Electoral accountability and the necessity of comparison.” American Political Science Review 106(03):661–684.
Kinder, Donald R and D Roderick Kiewiet. 1979. “Economic discontent and political
behavior: The role of personal grievances and collective economic judgments in congressional voting.” American Journal of Political Science pp. 495–527.
Kramer, Gerald H. 1983. “The ecological fallacy revisited: Aggregate-versus individuallevel findings on economics and elections, and sociotropic voting.” American political
science review 77(01):92–111.
Krause, George A and Benjamin F Melusky. 2014. “Retrospective Economic Voting and
the Intertemporal Dynamics of Electoral Accountability in the American States.” The
Journal of Politics 76(04):1102–1115.
Lanoue, David J. 1994. “Retrospective and prospective voting in presidential-year elections.” Political Research Quarterly 47(1):193–205.
Lewis-Beck, Michael S and Mary Stegmaier. 2013. “The VP-function revisited: a survey
of the literature on vote and popularity functions after over 40 years.” Public Choice
157(3-4):367–385.
Lodge, Milton and Charles S Taber. 2013. The rationalizing voter. Cambridge University
Press.
28
Lord, Charles G, Lee Ross and Mark R Lepper. 1979. “Biased assimilation and attitude
polarization: the effects of prior theories on subsequently considered evidence.” Journal
of personality and social psychology 37(11):2098.
MacKuen, Michael B, Robert S Erikson and James A Stimson. 1992. “Peasants or
bankers? The American electorate and the US economy.” American Political Science
Review 86(03):597–611.
Malhotra, Neil and Alexander G Kuo. 2008. “Attributing blame: The public’s response
to Hurricane Katrina.” The Journal of Politics 70(01):120–135.
Mueller, John E. 1970. “Presidential popularity from Truman to Johnson.” American
Political Science Review 64(01):18–34.
Nadeau, Richard, Michael S Lewis-Beck and Éric Bélanger. 2013. “Economics and elections revisited.” Comparative Political Studies 46(5):551–573.
Nadeau, Richard, Richard G Niemi and Antoine Yoshinaka. 2002. “A cross-national
analysis of economic voting: taking account of the political context across time and
nations.” Electoral Studies 21(3):403–423.
Nannestad, Peter and Martin Paldam. 1994. “The VP-function: A survey of the literature
on vote and popularity functions after 25 years.” Public Choice 79(3-4):213–245.
Nordhaus, William D. 1975. “The political business cycle.” The Review of Economic
Studies pp. 169–190.
Paldam, Martin. 1991. “How robust is the vote function? A study of seventeen nations
over four decades.” Economics and politics: the calculus of support pp. 9–31.
Paldam, Martin and Peter Skott. 1995. “A rational-voter explanation of the cost of
ruling.” Public Choice 83(1-2):159–172.
Powell Jr, G Bingham and Guy D Whitten. 1993. “A cross-national analysis of economic
voting: taking account of the political context.” American Journal of Political Science
pp. 391–414.
29
Singer, Matthew M and Ryan E Carlin. 2013. “Context counts: The election cycle,
development, and the nature of economic voting.” The Journal of Politics 75(03):730–
742.
Smith, Alastair. 2003. “Election timing in majoritarian parliaments.” British Journal of
Political Science 33(03):397–418.
Soroka, Stuart N, Dominik A Stecula and Christopher Wlezien. 2015. “It’s (Change in)
the (Future) Economy, Stupid: Economic Indicators, the Media, and Public Opinion.”
American Journal of Political Science 59(2):457–474.
Tilley, James and Sara B Hobolt. 2011. “Is the government to blame? An experimental
test of how partisanship shapes perceptions of performance and responsibility.” The
Journal of Politics 73(02):316–330.
Van der Brug, Wouter, Cees Van der Eijk and Mark N Franklin. 2007. The economy and
the vote: Economic conditions and elections in fifteen countries. Cambridge University
Press Cambridge.
Whitten, Guy D and Harvey D Palmer. 1999. “Cross-national analyses of economic
voting.” Electoral Studies 18(1):49–67.
30
Supplementary materials
Appendix A: A simple selection model which illustrates the learning explanation
In this model we examine a voter who has to decide whether to reelect an incumbent.
We start of by assuming that the voter is more likely to reelect the incumbent if the voter
believes the incumbent is more competent. Given this assumption, the voter’s goal is to
construct a set of rational beliefs about the incumbent’s competence given the available
evidence. Specifically, imagine an incumbent I which was elected at t = 0, and who is now
up for reelection at t = 1. We denote the incumbents competence as CI . Based on the
voter’s previous experience with incumbents (s)he starts of with a normally distributed
prior belief about CI , which we standardize to be centered on zero with a variance of 1.
Since voters are interested in the incumbents competence, CI , the voter tries to infer
how competent the politician is based on the economic situation, y, which is affected by
CI . However, the economy is also affected by a non-competence related shock , which
is independently and identically drawn in each period from a normal distribution with
mean zero and variance σ2 . Accordingly, the economic situation at t = 1 is:
y1 = CI + 1
(3)
In this equation, voters only observe y1 , however, voters know the distribution the
non-competence related shock is drawn from. That is, voters know how much of the
variance in the economic situation which is due to non-competence related shocks. As
such, voters are faced with a signal extraction problem, which the voter can solve by
updating his or her prior information about CI using y1 . This leaves voters with the
following posterior beliefs about CI :
CI |y ∼ N (
σ2
1
y
;
)
1
1 + σ2
σ2 + 1
(4)
From this we can see that the voter’s belief about the incumbents expected competence
is increasing in y1 . That is, a better economic situation leads the voter to infer that the
31
incumbent is likely to be more competent. Specifically, the effect of a one unit increase
in y1 on expected competence is
1
.
1+σ2
As such, the effect of beliefs about competence
is moderated by σ2 . That is, as the variation in non-competence related shocks to the
economy increases, it is more likely that any variation in the economic situation is due
to non-competence related shocks, and accordingly the economy becomes a less efficient
estimator of the incumbent’s competence.
Now imagine the incumbent is reelected in t = 1. In period t = 2 the voter now has
to decide once again whether to vote for the incumbent. However, now the voter’s prior
beliefs about the incumbent incorporates the information obtained about CI at t = 1.
Voters update this prior based on the economic situation in t = 2, y2 , which is equal to:
y2 = CI + 2
(5)
This leaves the voter with the following posterior beliefs about the incumbents competence.
CI |y2 , y1 ∼ N (
1
σ2
1
y
+
y
;
1
2
2 )
(σ2 + 2)
(σ2 + 2)
σ2 + 2σ
(6)
σ +1
Just like in period 1 a better economic situation in period 2, y2 , is used to infer that
incumbents competence is higher, and just like in period 1 σ2 attenuates the degree to
which voters can use the economic situation to make inferences about CI .
However, there is one key difference from period 1; the effect of the economic situation
on the voter’s expectations about the incumbents competence have decreased. In period
one the effect of a one unit increase in y1 was
unit increase in y2 is
1
.
2+σ2
1
,
1+σ2
in period two the effect of a one
Accordingly, variation in the economic situation has less
bearing on how competent voters expect the incumbent to be when the incumbent is
up for re-election the second time. This as information about incumbent competence
from the last election is used to construct the voter’s prior beliefs about competence in
the next election. This is the key result from the model, which underlines the assertion
made above: as voters get more time to get acquainted with an incumbent’s competence,
32
the present economic situation plays a smaller role in shaping voters beliefs about the
incumbent and, in turn, whether they decide to reelect the incumbent.
Appendix B: Description of the samples
Table 5: Elections included in the country-level analysis
Country
Argentina
Australia
Austria
Belgium
Canada
Colombia
Costa Rica
Denmark
Dominican Republ
Ecuador
El Salvador
Finland
France
Germany
Greece
Iceland
India
Ireland
Israel
Italy
Luxembourg
Netherlands
New Zealand
Norway
Papua New Guinea
Portugal
Spain
Sweden
Switzerland
Trinidad and Tobago
Turkey
United Kingdom
United States
First election
1985
1961
1971
1961
1962
1982
1982
1964
1990
1984
1985
1962
1968
1972
1981
1963
1980
1973
1969
1972
1979
1963
1978
1969
1987
1980
1979
1976
1983
1991
1987
1964
1978
33
Last election
2001
2007
2008
2007
2008
2002
2002
2007
2002
1998
2000
2007
2007
2009
2009
2007
1998
2007
2006
2008
2009
2006
2008
2009
2002
2009
2008
2006
1999
2000
2002
2010
2002
# of elections
9
19
12
15
16
7
6
18
4
8
6
13
10
11
9
13
6
10
11
10
7
14
11
11
4
10
9
10
5
3
5
12
13
Table 6: Respondents included in the individual-level analysis
Country
1989
Denmark
576
425
France
Germany
601
401
Greece
Ireland
501
488
Italy
Netherlands 391
344
Portugal
Spain
379
458
UK
Total
4,564
1994 1999
1,113 682
671
271
890
709
1,049 267
929
270
608 1,398
649
694
773
214
718
451
743
540
8,143 5,496
2004 2009
863
801
607
409
308
576
328
597
725
564
795
414
597
713
450
492
696
596
761
537
6,13 5,699
2014 Total
1,004 5,039
901
3,284
1,369 4,453
892
3,534
840
3,829
764
4,467
1,025 4,069
746
3,019
956
3,796
1,027 4,066
9,524 39,556
Appendix C: Variable descriptions
The question wording for the different questions used in this individual-level analysis are
as follows.
Executive party vote: “If there were a general election tomorrow, which party would
you vote for?” Executive parties are coded 1, others are coded 0.
Executive party vote (last election): “Which party did you vote for at the General
Election of [Year]?” Same coding as for the vote variable.
Ideology: “In political matters people talk about ‘the left’ and ‘the right.’ What is
your position? Please indicate your views using any number on a 10-point scale. On
this scale, where 1 means ‘left’ and 10 means ‘right’ which number best describes your
position?”
Class: “If you were asked to choose one of these five names for your social class, which
would you say you belong to—the working class, the lower middle class, the middle class,
the upper middle class, or the upper class?”
Religion: “How often do you attend religious services: several times a week, once a
week, a few times a year, once a year or less, or never?”
National economic perceptions (NEP): In 1989, 1994, 2004, 2009 and 2014: “What
do you think about the economy? Compared to 12 months ago, do you think that the
general economic situation in this country is: a lot better, a little better, stayed the same,
a little worse, or a lot worse?” In 1999: “How about the state of the [country’s] economy?
34
Very satisfied, somewhat satisfied, somewhat dissatisfied, very dissatisfied?” (1999)
Probability of voting for executive party: “We have a number of political parties in
[country] each of which would like to get your vote. How probable is it that you will
ever vote for the following parties? Please answer on a scale where ’1’ means ”not at
all probable” and ’10’ means ”very probable”.” The scale was changed to 0-10 in 2009
and 2014. I only use answers which refer to the executive party. From this question two
variables were created. A variable which measure the distance from the midpoint, and
one which measure whether voters picked the endpoints of the scale (Pr(min,max)).
Descriptive statistics for these individual level variables are presented in table 7. Descriptive statistics for the country-level data are presented in table 8.
Table 7: Descriptive statistics, individual-level data
Executive party vote
Executive party vote (last election)
National economic perceptions
Time in office (years)
Class
Religion
Ideology
Survey year
Pr(min,max)
Distance from midpoint
Mean
0.32
0.36
0.43
5.18
0.46
0.52
0.51
2003
0.39
0.67
SD Min Max
n
0.47 0.00 1.00 39556
0.48 0.00 1.00 39556
0.28 0.00 1.00 39556
3.84
1
17 39556
0.30 -0.33 1.33 39556
0.32 0.00 1.17 39556
0.26 0.00 1.00 39556
8.71 1989 2014 39556
0.49 0.00 1.00 26340
0.35 0.00 1.00 26340
Table 8: Descriptive statistics, country-level data
Change in support for executive party (pct.)
Support for party last election (pct.)
Growth in GDP per capita (pct.)
Time in office (years)
Number of terms
Election year
Mean SD
Min
Max
-3.97 7.32 -42.80 20.70
37.02 12.15 0.00 67.30
3.07
3.07 -7.59 13.85
6.02
4.62
1
30
2.43
1.75
1
12
1989 11.95 1961 2010
n
327
327
327
327
327
327
Appendix D: Strategic election timing in the country level data
To probe the plausibility of the strategic election timing xplanation I introduce two
different type of controls for election timing.
35
The first type of control I employ to disentangle election-timing and tenure is a variable
which counts the number of times an executive has held elections. I add this variable and
an interaction between this variable and economic growth to the set of models already
estimated above (cf. table 1). The key estimates from this new set of models, namely the
estimates attached to the interaction between growth and tenure, is reported in the first
three columns of table 9. As can be seen from table 9 the interaction remains negative,
it has the same size, and in two of the three specifications it is statistically significant
(p > 0.1).
Table 9: Can the findings be explained by election timing?
(1)
Economic growth
0.38+
(0.21)
Economic Growth * Tenure -0.04
(0.03)
2
R
0.25
RMSE
6.96
Observations
327
# of terms
Fixed term
(2)
(3)
(1)
(2)
(3)
+
∗
0.38
0.42
-0.26
0.66
0.72∗
(0.21) (0.20) (0.42) (0.82) (0.18)
-0.06+ -0.08∗ 0.04
-0.05 -0.05∗
(0.03) (0.02) (0.03) (0.06) (0.01)
0.43
0.71
0.69
0.88
0.91
6.46
5.18
5.40
4.03
2.31
327
327
45
45
45
Standard errors in parentheses
+
p < 0.10, ∗ p < 0.05
The second way I disentangle election-timing and tenure is by restricting the sample
of elections to the five countries in which election dates are fixed. In these countries the
executive cannot time the election, and accordingly, any relationship found between time
in office and the importance of the economy cannot be attributed to election timing. The
key estimates of the models, namely those attached to the interaction between growth and
tenure, is reported in the last three columns of table 9. As can be seen from table 9, the
interaction effect remains negative in the two most demanding specifications (model two
and three), and in the third model, which includes the leader fixed effects, the interaction
is also statistically significant (p > 0.05).
Taken together, the fact that across both types of control for election timing, the
interaction between economic growth and tenure remains negative and statistically significant in the most demanding specification (model 3), seems to suggest that the results
36
laid out in table 1 were not the result of strategic election-timing. As such, the results
imply that voters are in fact less likely to punish and reward the executive party for
economic conditions as time in office increases.
Appendix E: Two-step models of individual-level data
Another way to examine whether there is an interaction between time in office and national economic perceptions, is to estimate a multilevel model which allows for a random
slope with respect to national economic perceptions across the different surveys, and then
examine whether the size of the survey-sepecific slopes are related to the tenure of the
incumbent party at the time of the survey.
To do this I estimate a set of multi-level logit models of the probability of voting
for the executive party with the full set of individual level controls, omitting time in
office and instead allowing the effect of national economic perceptions to vary across
surveys. Specifically, I estimate three different models; one with survey random effects,
one with leader fixed effects and one with survey fixed effects. For each of these models
I obtain sixty different logit coefficients, which represent the effect of national economic
perception in the individual surveys. Finally, I plot the logit coefficients obtained in the
three different models and the tenure of the incumbent at the time of the survey in figures
0
5
10
Time in office
15
2
−2
0
Logit coefficient
4
2
−2
0
Logit coefficient
2
0
−2
Logit coefficient
4
4
3.
0
5
10
Time in office
15
0
5
10
15
Time in office
Figure 3: Random slope of NEP plotted with 95 pct. confidence intervals. From left to
right the models used for plotting include random effects, leader fixed effects and survey
fixed effects. Uniformly distributed random boise added to the horizontal placement of
the dots. Linear fit (95 pct. CI).
As can be seen from these figures there is a negative relationship between time in
office and the size of the logit coefficients. OLS regressions of time in office on the logit
37
coefficients reveal that the negative relationship is statstically significant (p < 0.05, using
country clustered standard errors). As such, this alternative way of estimating the effect
of time in office on the economic vote gives the same baisc result as that identified above.
Appendix F: Consequenses for electoral accountabilty
In table 10 below we examine whether tenure has adverse effects on economic growth.
We find no evidence of this, as the effect of tenure is statistically insignificant.
In table 11 we examine whether tenure has adverse consequences for voters perception
of the national economy (NEP). We find no evidence of this.
Table 10: Effect of tenure on economic growth
(1)
-0.03
(0.03)
X
Tenure
Time fixed-effects
Country fixed-effects
Leader fixed-effects
R2
RMSE
Observations
(2)
-0.04
(0.04)
X
X
0.31
2.77
327
0.41
2.73
327
(3)
0.09
(0.23)
X
X
0.40
3.05
327
Standard errors in parentheses
+
p < 0.10, ∗ p < 0.05
Table 11: Effect of tenure on national economic perceptions
Tenure
R2
RMSE
Observations
(1)
(2)
(3)
-0.01
(0.01)
-0.02∗
(0.01)
-0.01+
(0.00)
43,771
43,771
43,771
Standard errors in parentheses
+
p < 0.10, ∗ p < 0.05
38