REGIONAL ECONOMIC VOTING:
Russia, Poland, Hungary, Slovakia, and the Czech Republic,
1990-1999
Joshua A. Tucker
Princeton University
The National Council for Eurasian and East European Research
910 17th
Street, N.W.
Suite 300
Washington, D.C. 20006
TITLE VIII PROGRAM
ii
Project Information*
Principal Investigator: Joshua A. Tucker
Council Contract Number: 819-17g
Date: November 10, 2005
Copyright Information
Scholars retain the copyright on works they submit to NCEEER. However, NCEEER possesses
the right to duplicate and disseminate such products, in written and electronic form, as follows: (a)
for its internal use; (b) to the U.S. Government for its internal use or for dissemination to officials of
foreign governments; and (c) for dissemination in accordance with the Freedom of Information Act
or other law or policy of the U.S. government that grants the public access to documents held by the
U.S. government.
Additionally, NCEEER has a royalty-free license to distribute and disseminate papers submitted
under the terms of its agreements to the general public, in furtherance of academic research,
scholarship, and the advancement of general knowledge, on a non-profit basis. All papers
distributed or disseminated shall bear notice of copyright. Neither NCEEER, nor the U.S.
Government, nor any recipient of a Contract product may use it for commercial sale.
* The work leading to this report was supported in part by contract or grant funds provided by the National Council for
Eurasian and East European Research, funds which were made available by the U.S. Department of State under Title
VIII (The Soviet-East European Research and Training Act of 1983, as amended). The analysis and interpretations
contained herein are those of the author.
iii
Executive Summary
Two models are presented for predicting the effect of cross-regional variation of
economic conditions on cross-regional variation in election results in post-communist countries.
The Referendum Model predicts that Incumbent parties will perform better in areas of the
country with better economic conditions, while the Transition Model predicts that New Regime
parties will perform better in areas of the country where the economy is stronger and Old Regime
parties will perform better where the economy is weaker.
Using an original data set of regional level economic, demographic, and electoral
variables, it is demonstrated that across 20 national presidential and parliamentary elections from
Russia, Poland, Hungary, Slovakia and the Czech Republic, there is substantially stronger
empirical support for the Transition Model. Moreover, the effect of the economy on Incumbent
parties is largely conditional on their status as New Regime parties, Old Regime partiers, or
neither type of party. These findings should be of interest to students of elections and voting in
post-communist countries, those concerned with economic voting more generally, and scholars
interested in the methodological challenges of conducting comparative analysis. Moreover, they
have interesting implications for current debates regarding the relationship between elections and
representation.
1
Introduction
Since the collapse of communism in the former Soviet Union and Eastern Europe, the
field of post-communist studies has found itself at an important juncture. On the one hand, the
field has a long tradition of analysis based on the intricacies of the communist experience, one
that placed it outside the bounds of most avenues of research in mainstream political science. On
the other hand, many of the fundamental political realities of societies undergoing nascent
democratic transitions – some succeeding and some failing – are anything but unique. Indeed,
mainstream political science has much to say about democratic transitions and, ultimately,
political behavior in a democratic society. Thus one of the overriding questions the field must
answer is how to draw on the vast reservoir of outstanding scholarship from mainstream political
science without losing sight of the peculiarities of what is now the post-communist experience.
One field that is clearly ripe for making a contribution to a wider audience is the study of
economic voting in post-communist countries. To date, though, most analyses have examined
only a single election (Gibson and Cielecka 1995; Colton 1996; Mason and SidorenkoStephenson
1997; Powers and Cox 1997) or a single country (Wade, Groth and Lavelle 1994;
Bell 1997).1
As a result, they generally focus on the question, “How has the economy affected
the vote for the parties that contested this particular election?” This is not to say that the authors
do not have a priori expectations, but they are by and large couched in the particular
circumstances of that country and tend to remain more implicit than explicit (although see
Powers and Cox 1997). Although one can certainly find similarities in these expectations across
articles, the authors are less concerned with the generalization of the implications of their
1
Although see Pacek 1994; Fidrmuc 2000a, 2000b; Harper 2000; Tucker 2000, 2001; Duch 2002 for more
comparative analyses. Even from this list, though, only Fidrmuc (2000a, b) and Tucker (2000, 2001) analyze more
than four elections.
2
findings than with what has been learned about the effect of the economy on that particular
election. The same holds for the role of the economy in other general studies of voting in postcommunist
countries.
In this paper, I move beyond country-specific approaches to understanding the effect of
economic conditions on election results by searching for broader, more general models that can
be applied in any post-communist country. Moreover, I take the additional step of directly
linking the models I propose to the existing literature on the effects of economic conditions on
election results in established democracies, while at the same time modifying them to take
account of the realities of the post-communist context. In doing so, the goal is to make a
theoretical contribution to both our understanding of the interaction between economic factors
and electoral outcomes in post-communist countries and the general economic voting literature.
More specifically, two theoretical frameworks for hypothesizing about the effect of
economic conditions on election results in post-communist countries that are directly tied to both
the existing mainstream literature as well as the realities of the post-communist experience are
proposed. The Referendum Model is similar to the main strand of the economic voting literature
in proposing that incumbent parties should perform better in parts of the country where economic
conditions are stronger. The Transition Model, however, takes its cue from a secondary strand of
the existing literature and suggests instead that economic conditions affect a party’s electoral
fortunes based on the party’s relationship to the transition: Old Regime parties are expected to
enjoy more electoral success in parts of the country where economic conditions are worse while
New Regime parties should do better in parts of the country where economic conditions are
stronger.
3
In an effort to provide a systematic test of these claims, I employ a broadly comparative
framework in this paper that takes two forms. First, the paper examines election results from
twenty separate presidential and parliamentary elections that took place that took place between
1990 and 1999 in Russia, Poland, Hungary, Slovakia, and the Czech Republic. Thus none of the
patterns identified can be said to be a product of a particular set of circumstances peculiar to a
given country at a given time, or even in a given country over time.
Second, the study looks for concentrations of economic winners and losers and their
effects on election results not at the national level – which would have yielded but one
observation of economic conditions and election results per election – but rather at the regional
level. In this manner, I am able to examine on a party by party basis whether or not individual
parties that contested all twenty of the elections performed better in areas of the country where
economic conditions were better or where economic condition were worse.
At the same time, I employ a method of analysis that allows for comparison of these
party by party results with one another, allowing me to search for patterns of support across the
entire set of cases contained in the study. Empirical tests are conducted on an original database
containing regional level macro-economic data and election results from twenty national
parliamentary and presidential elections, making this the one of the most thorough study of
economic voting in post-communist countries to date, as well as the only comparative study to
include both Russian and Eastern European cases. Moreover, in expanding the study of
economic voting to the regional level, this paper joins a small but growing number of papers in
illustrating another level of analysis at which we can observe the effect of economic conditions
on election results.2
2
Moving the study to the regional level of analysis also allows us to greatly increase the amount of data that we can
bring to bear on the topic; at the same time, though, it is important to note that because this study takes place at a
4
The result of the empirical analysis is greater support for the predictions of the Transition
Model than the Referendum Model. The Transition Model also has a significantly wider scope,
allowing us to make predictions about a wider range of cases. Moreover, the effect of the
economy on Incumbent parties is largely conditional on their status as New Regime parties, Old
Regime partiers, or neither type of party.
The rest of this paper proceeds as follows. In the next section, I introduce the Transition
Model and the Referendum Model and outline the underlying theoretical arguments for both.
After a short section touching on methodological considerations, the results of empirical tests
conducted across all twenty elections are presented. This is followed by a more focused
examination of the results from two particularly informative elections, the 1997 Polish and 1998
Hungarian parliamentary elections. The paper concludes with a discussion of the implications of
these findings.
Post-Communist Models of Economic Conditions and Election Results
In this section, I introduce two models that allow us to make predictions about the effect
of economic conditions on election results in post-communist countries. The models are
designed with intent of grounding the study of voting in post-communist countries in the existing
literature on economic voting. The reader will note that all of the hypotheses developed below
refer to aggregate-level relationships between economic conditions and election results, as
opposed to micro-level models of individual vote decisions.3
As such, it is important to note that
different level analysis, the results should be seen as supplementing as opposed to testing findings from others levels
of analysis. For much more on the method of analysis, including its benefits and drawbacks, see Tucker 2004, ch. 3.
5
I am not directly testing micro-level propositions about individual behavior, but rather
relationships between aggregate-level indicators. The hypotheses were designed, however, to be
consistent with micro-level propositions about the individual behavior of voters, as will be
explained in detail below.
A Referendum Model
Setting aside the peculiarities of the post-communist world for a moment, consider an
electoral environment in which voters possessed no information about how parties were likely to
behave vis a vis the economy once elected to office. Although such a claim may seem far
fetched, there are numerous reasons why voters could have no information in this regard: the
election could be contested by parties that had never held office before, either because the
election is taking place in a new democracy or following a major shake up of the party system in
a more established democracy; voters could lack the will or the ability to make judgments about
the relationship between a victory for certain parties and the likely effect on economic
conditions; or there could be so much uncertainty about the current economic situation that even
the most informed citizens would be hard pressed to predict future economic developments.4
Even in such a low information environment, though, we would expect voters to be aware
of the parties that make up the government. And in the absence of information about differences
between the political parties contesting the election, the only way we could expect the economy
3
I chose to rely upon aggregate data because it facilitates more seamless comparisons across multiple countries and
time periods; reliance on survey data would either introduce problems with comparing results across different
surveys in different languages or would greatly reduce the number of elections analyzed.
4
See Aidt 2000 for a theoretical argument concerning the question of whether we ought to expect citizens in
established democracies to expend effort to acquire enough information about the state of the economy to judge the
competence of the government.
6
to have an effect on election results would be through the vote for or against incumbent parties.
To generate this prediction, we need to make two additional assumptions, neither of which ought
to be particularly controversial. First, we assume that people concerned with the state of the
economy are more likely to want to vote the government out of office than people who think the
economy is performing well. Note that this assumption is different from a retrospective
evaluation of government performance or competence based on evaluating the impact of the
current government on economic conditions during its tenure in office (a point that I return to in
more detail later in this section). Instead, it simply proposes that the more dissatisfied an
individual is with the state of the economy, the more likely he or she will be to choose to cast a
vote for a party that is not currently in power. Similarly, the more satisfied an individual is with
the state of the economy, the more likely he or she will be to cast a vote in favor of an incumbent
party as opposed to an unknown “other” option.5
The second assumption is that the worse economic conditions are in a particular area of
the country, the more likely we should be to find more people dissatisfied with the state of the
economy in that area. This affect can work both directly and indirectly. Consider
unemployment as an example. If I have lost my job and can not find work, then I am likely to
think that the economy is performing poorly. The higher the unemployment rate, the more
people there are who are likely to be unemployed and share these beliefs. Thus the higher the
unemployment rate, the more people there are who are likely to be dissatisfied with the state of
the economy because they are out of work.
5
Note that these are probabilistic and not deterministic statements. So the claim is not that all people dissatisfied
with the economy will choose to vote against incumbent parties, but only that people dissatisfied with the economy
will be more likely to do so than people who are satisfied with the state of the economy.
7
However, the effect also works indirectly. Even if I am not unemployed, higher
unemployment rates may mean that more people I know are unemployed. I may also see more
unemployed people loitering on the streets, and I may be more afraid that I will become
unemployed.6
All of these factors could lead me to the same conclusion as someone who is
unemployed: the economy is not performing well. Now it is doubtful that everyone who is
employed will feel this way, and perhaps it is just a small minority that will come to this
conclusion. Nevertheless, the higher the unemployment rate, the more people we are likely to
find sharing this belief.7
With these assumptions in hand, the following aggregate level prediction falls out nicely.
In areas of the country where economic conditions are stronger, incumbent parties should enjoy
more electoral success than in areas of the country where economic conditions are weaker. This
holds because we expect worse economic conditions to lead to a greater concentration of voters
dissatisfied with the state of the economy and consequently fewer voters inclined to support
incumbent parties than in areas of the country where economic conditions are better. Where the
6
Indeed, Paldam and Nannestad 2000 argue that for every person who is unemployed, an estimated 100-300 people
will know that person and “will notice and be concerned”. And their estimate does not even include the people who
don’t know the unemployed directly but notice their presence. They also find a strong degree of knowledge about
the issue of unemployment throughout the population. See Paldam and Nannestad 2000, 371-3.
7
It is important to note that this argument sets aside the debate about whether citizens actually perceive economic
conditions correctly. A number of scholars have noted variation in the perception of similar economic
circumstances across different voters – see for example Goidel and Langley 1994; Holbrook and Garand 1996; Duch
2000 – although in the Danish (Paldam and Nannestad 2000) and British cases (Sanders 2000), authors have found
that perceptions of the economy at the micro level actually match well to conditions at the macro-level. In some
ways, though, the cross-regional approach insulates the analysis from these kinds of concerns. Even if there are
systematic misperceptions by voters of the state of the economy, as long as these misperceptions are randomly
distributed across geographic regions the analysis should not be affected even if, as Hetherington 1996 found in the
United States, perceptions are more important in affecting vote choice than actual macro-economic conditions. We
should still expect more voters to have a negative perception of the state of the economy in areas of the country
where actual economic conditions are worse than in areas of the country where economic conditions are better even
if the assessment of the economy does not map perfectly from actual conditions to perceptions.
8
economy is better, we expect to find more voters who are likely to be satisfied with the state of
the economy and thus more voters likely to support the incumbent parties.8
The same aggregate level hypothesis can also be arrived at if it is believed that voters use
a two-step process in making their vote choice whereby they first decide whether or not to
support an incumbent party and only then make the decision of which opposition party to support
(if the decision is not to support an incumbent party). If this is the case, then regardless of the
level of information voters possess about political parties, voters who are more satisfied with the
state of the economy should still be more likely to support incumbent parties than voters who are
dissatisfied with the state of the economy, and, consequently, based on the same assumptions
described above, that incumbent parties should perform better in areas of the country where
economic conditions are stronger than in areas where economic conditions are weaker. This
interpretation also presumes that voters will focus first and foremost on whether to punish or
reward the incumbent party for the current state of the economy irrespective of their level of
information or believes about alternative vote choices. And even if their preference for some
alternatives are greater than others, there should still be a pattern of more support for incumbent
parties in areas of the country where economic conditions are better and less support where they
are worse.
8
It is important to note that this is not a relative deprivation argument. Voters are not posited to be voting against
the government because they perceive that their region is being treated poorly in comparison to other regions.
Instead, the logic is that in the absence of information about different parties, voters who are more concerned about
the state of the economy will be more likely to want to see a change in the government than voters who are less
concerned about the state of the economy, and, on the aggregate-level, there should be more voters concerned about
the state of the economy in areas of the country where economic conditions are worse.
9
While this particular formulation of the voting choice process stands in marked contrast
to traditional spatial models of voting,9
the logic is not that far off from the theories underlining
most cross-national models of the effect of economic national election results, which inevitably
focus on the vote for incumbent parties independent of the options presented by other parties.10
Moreover, the idea of voters first making up their minds about whether or not to vote for
incumbent parties is attractive in the transitional context because it emphasizes the only concrete
information voters have: the knowledge of which parties are currently incumbents. If voters
have beliefs about what other parties are likely to do when they are in office but are at all
uncertain about these beliefs, one could imagine that voters who are pleased with the state of the
economy might not want to take the risk of being wrong about economic effects of electing a
different party to office. Conversely, voters who are unhappy about the state of the economy
might be much more willing to risk replacing the current governing parties with other parties,
even if they run the risk of being wrong about their assessment of these other parties. While
such a calculus is not necessarily peculiar to transition countries, the levels of uncertainty that
force voters to fall back on a yes/no assessment of the incumbent parties are probably more
likely to be found in transitional countries than in established democracies.
Before turning to an alternate set of hypotheses, I want to emphasize that although the
hypothesis looks very similar to standard economic voting hypotheses insofar as it predicts that
better economic conditions will lead to more votes for incumbent parties, the argument is
theoretically distinct from the standard retrospective economic voting models. The argument
presented above is not one in which voters use changes in the economy since the previous
9
See in particular Downs 1957.
10
Although see Anderson 2000 for an example of a cross-national study that considers the availability of alternative
options.
10
government took office as a way to measure the competence of the existing government, but
rather one in which uncertainty about parties and policy-making leads economically satisfied
citizens to be more likely than economically dissatisfied citizens to want the current government
to remain in office. Thus the prediction is that incumbent parties ought to enjoy greater electoral
success in areas of the country where the economy is stronger, as opposed to hypotheses that
focus on either changes in vote totals or in economic conditions since the previous election.
At the same time, it is also useful to note the similarities between the Incumbency
Hypothesis proposed above and arguments common to the economic voting literature. Although
the focus here is not on using economic conditions to assess the competence of the current
government, both approaches do rely on the basic argument that dissatisfaction with economic
conditions can lead voters to be more likely to want to “throw the bums out”. At the end of the
day, both approaches suggest that the worse economic conditions are, the less likely incumbent
parties are be favored by voters. Furthermore, in some ways such an approach is a direct
descendent of what Fiorina calls the “pure” retrospective approach of V.O. Key, based on
“retrospective voters – unaffected by party ties, ideologies, and future issue expectations – who
regard elections as referenda on the state of the country".11
If ever there was an opportunity to
locate such pure retrospective voters, it would seem to be in the uncertain environment of postcommunist
elections.
As the Incumbency Hypothesis suggests the economic conditions affect election results
solely by increasing or decreasing the likelihood that voters will support incumbent parties, the
model has the feel of a referendum on the current government. For this reason, I have labeled
this approach the Referendum Model. The obvious shortcoming of the Referendum Model is
that it offers analysts no leverage over the question of where voters who are dissatisfied with
11
See Fiorina 1981, 36.
11
incumbent parties are likely to turn in casting their vote. This is not a large concern in the
context of a two party system, where a vote against the incumbent by definition means a vote for
the opposition party. As the number of parties contesting an election increases, however, models
that make predictions solely based on incumbency status are likely to suffer from an increasingly
narrow scope and an inability to comment on the relationship between economic conditions and
the vote for large number of parties. So the next question to consider is how to motivate a model
of the effect of economic conditions on election results that can move beyond this limitation.
A Transition Model
Although most studies of economic voting tend to concentrate on how economic
conditions affect the vote for incumbent parties, there is another strand of the literature. This
second strand makes predictions about the effect of economic conditions on the vote for parties
based on their partisan alignment (Lewis-Beck and Bellucci 1982; Kiewiet 1983; Bellucci 1984;
Host and Paldam 1990; Powell and Whitten 1993). The basic thrust of the argument is that
certain economic conditions – generally higher unemployment – are likely to increase the vote
for left wing parties, while other economic conditions – generally higher inflation – are likely to
increase the vote right wing parties. The logic underlying the argument is that voters are
concerned with specific economic problems and have a belief about which parties would be
better suited to solving those particular problems. In the context of the advanced industrialized
democracies, it is generally assumed that right wing parties are more likely to favor control of
inflation at the expense of rising unemployment while left wing parties are the opposite. Thus,
an electorate concerned about inflation would be more attracted to right wing parties, while an
electorate scarred by unemployment would be more likely to vote for a left-wing government.
12
Transporting a simple partisan based approach to the post-communist world is
problematic because of the pervasive levels of uncertainty in any new democracy (Bunce and
Csanadi 1993). It is difficult to apply a simple left-right classification scheme to parties in
transition countries, both because it is hard to know where parties stand and because they can
often change their positions (Kitschelt 1992; Colton 1998). Moreover, even if analysts could
come up with a compelling schema, it would be a stretch to assume that voters would be able to
associate parties of a certain partisan persuasion with specific policies as closely as voters can in
established democracies. This is a function of the proliferation of new parties, the presence of
new economic realities, and voters’ lack of experience in seeing different parties in power.
Moreover, due to the enormity of the economic transition from a centrally planned economy to a
market economy, it is unlikely that voters would be motivated by concerns over specific
economic conditions as opposed to the state of the overall economy.
With these concerns in mind, I make the following two assumptions to craft a model that
is based on the partisan approach to economic voting but is appropriate for the post-communist
context. First, I assume that voters are aware of a party’s relationship vis à vis the transition.
More specifically, I assume that voters can identify the parties most closely associated with the
prior regime, the Old Regime parties, and that voters can identify the parties most closely linked
to the newly emerging democratic and capitalist world, the New Regime parties.
Old Regime parties are defined as the political party or parties that are directly linked to
the party that ruled during the previous non-democratic regime. Two sets of parties fall under
this general rubric. The first is the legal successor party to the previous ruling party. The second
group includes all splinter parties that have split from the legal successor party in order to remain
what is perceived as a closer ideology of the old ruling party. In the case of post-communist
13
countries, this is an easy distinction to make, as the splinter party would declare that it wants to
remain truer to communist principals than the legal successor. The important point is that Old
Regime parties are visibly linked to the previous regime, either through the people that make up
the party, the ideology it continues to publicly embrace, or its legal obligations.12
The primary goal of coding New Regime parties is to arrive at a set of parties that are
most likely to appeal to voters primarily through their connection to, association with, and in
general support for the transition itself. To identify these parties, I use what amounts to a two
step process. The first step is to trace out the descendents of the initial opponents of the existing
Communist regime in each of the countries. The second step is to then eliminate any parties that
attempt to appeal to voters on grounds that are separate and distinct from the transition. In
practice, this primarily means excluding parties that attempt to appeal to voters on the basis of
their identity or profession, which excludes religious, ethnic, nationalist, agrarian or trade union
parties from the category of New Regime parties.13
The purpose of this is to make the category as “pure” as possible by limiting it to parties
whose primary association in the mind of voters is likely to be to the transition away from the
Communist political and economic systems, which forms the basis for the logic underlying the
New Regime hypothesis. While the coding rules are certainly subjective to a certain extent, the
list of New Regime parties that can be found below in Table 1 is one that I believe most scholars
would agree seems to be the correct list.14
12
One should also note that “Old Regime” is not synonymous for “leftist”. There are examples of dominant leftist
parties that are not Old Regime parties, such as the Czech Social Democratic Party, as well as Old Regime parties
that have moved closer to the center, such as the Democratic Left Alliance in Poland.
13
One could argue that agrarian parties appeal to voters on the basis of interest as opposed to identity in the more
classic sense of the word. Nevertheless, agrarian parties clearly target a specific sub-section of the population based
on what they do and where they live, and as such possess the same special appeal to particular citizens as, for
example, an ethnic party has for members of that particular ethnic group.
14
Knowledge of whether a party is an Old Regime party, New Regime party, or neither is
all that is required of voters in this model.15
They are not required to have knowledge of the
specifics of parties’ economic platforms, the likelihood of the parties keeping promises regarding
these platforms, or the ability of parties to implement such policies if elected. They are only
required to be able to identify the parties most closely connected to the transition.
The second assumption of the model is that, instead of assuming that voters are guided by
concerns with specific economic problems, we assume that they have general concerns about the
economy; they react to it either being “bad” or “good”. As both assumptions are motivated by
the presence of simultaneous economic and political transitions, we refer to this model as the
Transition Model. The Transition Model therefore produces two hypotheses. The New Regime
Hypothesis predicts that New Regime parties will do better where economic conditions are
better, and worse where economic conditions are worse. Conversely, the Old Regime Hypothesis
predicts that Old Regime parties will benefit where economic conditions are worse, and will do
worse where economic conditions are better.
Both predictions stem from a conception of the transition as representing a sharp break
from the past, complete with new opportunities and dangers. The New Regime Hypothesis is
based on a fairly straightforward logic. Voters who are enjoying more benefits from economic
reforms and have avoided more of the pain from economic reforms are likely to want the reforms
to continue. Irrespective of who is currently in power, there is a good chance that voters will
think that New Regime parties are likely to continue pursuing policies that ensure the transition
to a full fledged market economy. As we should find more voters who are enjoying the benefits
14
By definition, Old Regime parties can not become New Regime parties; the two categories are mutually exclusive.
15
Readers are invited to see Tucker 2004, which goes in to many of these coding decisions in great detail.
15
of the transition where the economy is performing better, we expect New Regime parties to
likewise perform better where the economy is stronger.
The logic underlying the Old Regime Hypothesis is a bit subtler. If a voter has not
benefited from reform – has not enjoyed the new prosperity and is suffering greater pain – then
she is likely to want to stay away from parties that are associated with reform for the same reason
described in the preceding paragraphs. Having eliminated New Regime parties as an option,
however, the voter still has many other parties to choose between in casting her vote. While she
may not have much information about most of those parties, she does know something about the
Old Regime parties. When they were in power, there were none of the new problems associated
with the reforms, such as unemployment, inflation, and non-payment of wages. Thus if the voter
is particularly concerned about these problems, then she knows at least one type of party that in
the past – at least on the surface – dealt with these problems successfully.
Even if a return to the past is not desired, the voter still is likely to have more confidence
in the Old Regime parties to address these issues than any of the other non-New Regime parties.
And the worse the economic conditions are, the more likely she is to act on these concerns when
she casts her ballot. By no means should we expect everyone to select Old Regime parties if
they are unhappy with the current path of reforms, but in the aggregate, it is likely that more
people should decide to vote for Old Regime parties when economic conditions are worse.
Likewise, if a voter were enjoying more of the benefits from the new economy, then she would
be less motivated to turn to an Old Regime party. Thus we arrive at the Old Regime Hypothesis,
which predicts greater success for Old Regime parties when economic conditions are poor and
less when they are strong.16
16
The Referendum and Transition Models should not be confused with the debate between retrospective and
prospective voting (Fiorina 1981; Kiewiet 1983; Kiewiet and Rivers 1985; MacKuen, Erikson and Stimson 1992).
16
Data and Methods: Analyzing 20 Post-Communist Elections
In this article, twenty national elections from the Czech Republic, Hungary, Poland,
Russia, and Slovakia are analyzed. These five countries were chosen for a number of reasons.
First, all five of the countries have enjoyed relatively free and fair elections.17
Second, all
embarked on some type of economic reform program. Although the reforms have differed in
term of pace, scope, and results, it is clear that all five countries now have substantially different
economies as compared to communist era. Third, the sample also includes elections from both
the former Soviet Union and East Central Europe, as well as both presidential and parliamentary
elections, in order to make as strong a case as possible for the generalization of the findings. In
particular, including parliamentary and presidential elections allows us to include both the vote
for individuals (in presidential elections) and parties (in parliamentary elections) in the analysis.
A final consideration is that all five countries report disaggregated election results and economic
conditions using the same sub-national units, thus facilitating the construction of the data base.
Included in the study are all elections for national office – parliamentary and presidential
– that took place in each of these countries between 1990 and 1999. This excludes initial
parliamentary elections that were contested largely between two forces, the communists and a
broad anti-communist umbrella coalition; these elections also often included restrictions on
participation and competition for certain seats. Thus the study begins with the first true multiparty
parliamentary elections that occur in each of the countries, in addition to all directly
Both models could be consistent with either retrospective or prospective approaches. Although the Referendum
Model is clearly based on past performance, a voter may be acting out of concern for the future in trying to keep the
incumbent parties out of power. Likewise, although the Transition Model is predicated on choosing preferred
parties for the future, the expectation of what these parties may or may not do is taken from the past.
17
Russia is the one exception, with questions persisting about the 1993 parliamentary elections. Nevertheless,
despite the likely presence of some fraud, it is doubtful that the overall outcomes were affected greatly. See McFaul
1996; Myagkov, Ordeshook and Sobyanin 1997; White, Rose and McAllister 1997. Moreover, eliminating this
election from the analysis has no effect on the overall conclusions.
17
TABLE 1. CODING OF PARTIES BY ELECTION
Country Date Incumbent New Regime Old Regime
Czech R. 1992 ODS, ODA, OH ODS, ODA, OH LB
Czech R. 1996 ODS, ODA, KDU ODS, ODA KSČM
Czech R. 1998 ODS, KDU, US ODS, US KSČM
Hungary 1990 MSzP, MSzMP MDF, SzDSz, FIDESZ MSzP, MSzMP
Hungary 1994 MDF, FKgP, KDNP MDF, SzDSz, FIDESZ MSzP, MP
Hungary 1998 MSzP, SzDSz
MDF, SzDSz, Fidesz-
MPP
MSzP, MP
Poland 1990 Cimoszewicz Mazowiecki, Wałęsa Cimoszewicz
Poland 1991
UD, KLD, PSL, PL,
PC, WAK
UD, KLD SLD
Poland 1993 UD, KLD, KKW, PL UD, KLD SLD
Poland 1995 Wałęsa
Wałęsa, Kuroń,
Gronkiewicz-Waltz
Kwaśniewski
Poland 1997 SLD, PSL AWS, UW SLD
Russia 1991 Ryzhkov, Bakatin Yeltsin Ryzhkov, Bakatin
Russia 1993 VR, PRES VR, PRES KPRF
Russia 1995 NDR NDR, DVR KPRF, KTR-SS
Russia 1996 Yeltsin Yeltsin Zyuganov
Russia 1999 Unity Unity, SPS KPRF, KTR-SS
Slovakia 1992 HZDS, KDH, ODU ODU SDL’
Slovakia 1994
DU, KDH, SV,
HZDS, SNS
DU, DS, KSS, ZRS, SV
Slovakia 1998 HZDS, SNS, ZRS SDK KSS, ZRS, SDL’
Slovakia 1999 Meciar Schuster None
*L = Legislative, P = Presidential. Full names of parties and candidates can be found in Appendix I.
18
contested presidential elections. For two-round presidential elections, results from the first round
are analyzed; for mixed legislative electoral systems (Hungary and Russia) results from the party
list proportional representation vote are analyzed. Table 1 lists the elections included in the study
and the coding of the candidates/parties (which, for the sake of simplicity, we will refer to simply
as parties for the remainder of the article) for each election.18
To test the three hypotheses, we need to estimate the effect of economic conditions on the
electoral fortunes of each of the eighty-eight parties included in the study. To generate these
estimates, an original database of regional-level electoral, economic, and demographic data is
utilized.19
More specifically, for each party listed in Table 1, the effect of economic conditions
on that party’s election results is estimated using the share of the vote received by the party as
the dependent variable, measures of macro-economic conditions as independent variables, and
demographic indicators as control variables, thus resulting in 88 separate analyses (one for each
of the parties). It is important to note that although regional-level data is used in the analysis –
e.g., the amount of the vote received by Party A in Region 1 – these are all results from national
elections disaggregated to the regional level.
18
In article of this length, it is clearly not possible to detail every coding decision. For this reason, a separate web
based appendix justifying all of the coding decisions will be made available to readers upon request and will be
posted on the web. Please note, though, that no parties with fewer than 2% of the national vote were included in the
analysis except in the one instance of an incumbent party (the AZR in the 1998 Slovak presidential election) that
failed to secure 2% of the vote in the election following its time in office. The reader should also note that the
existence of two incumbents in the 1991 Russian presidential election was due to the unique circumstance whereby
two candidates ran representing the more reformist (Bakatin) and hard line (Ryzhkov) wing of the ruling Communist
Party of the Soviet Union (White et. at. 1997).
19
The number of regions per country is Hungary (20), Slovakia (38), Poland (49), Czech Republic (76), and Russia
(79). While it would be desirable to have as many observations as possible, the extent of disaggregation of the
voting data is limited by the availability of commensurate economic data.
19
In order to control for any socio-demographic patterns of party support, a constant set of
control variables are included in each regression: the percentage of workers that work in
industry; the percentage of workers that work in agriculture; the percentage of the population
living in urban areas; the percentage of older citizens in a region, and the log of the population of
the region.20
While these variables may not capture every socio-demographic basis for party
support, they do touch on most of the larger themes in the literature, including rural-urban splits,
agricultural and industrial labor patterns, and center-periphery distinctions. They also capture
features that have been highlighted as being important for nascent post-communist cleavages,
including the propensity of older voters to support communist successor parties and urban voters
to prefer more liberal parties. The key point to note is that by including these control variables,
all economic effects identified by the statistical analysis should be interpreted as being
independent of any of these underlying socio-demographic explanations for party support.
Traditional analyses of the effects of economic conditions on election results in
established democracies have usually relied on measures of changes in income or economic
growth. In the post-communist context, however, unemployment and inflation have been
arguably the most drastic economic developments and the ones to which we might think voters
would be most likely to respond. In an ideal world, therefore, we would include all four of these
variables in any economic voting analysis.
20
Following convention, population is logged to account for the extreme outliers of major cities. Please note as well
that the proportion of non-ethnic Slovaks is also included as a controlling variable in the Slovak election due to the
presence of a Hungarian minority party that captured close to 10% of the vote in these elections. Earlier robustness
tests had suggested that not including this control variable made the results less stable, although more recent tests
have not seen this pattern repeated. Nevertheless, including the control variable seemed prudent just in case.
Including similar control variables in other countries had no effect on any of the results.
20
Inflation, however, does not lend itself to a cross-regional study, as the most substantial
variation in inflation is usually over time as opposed to cross-regional within a single country in
a single time period. Moreover, it is very difficult to find inflation figures disaggregated to the
regional level; indeed, I could only locate such data for 4 of the 20 elections included in the
study. For these reasons, I use change in wages, growth, and unemployment as the three base
economic variables in every regression analysis.21
In addition to these three variables, I also relied on three other economic variables that
were intended to capture additional indicators of transitional economic success (or failure):
foreign direct investment; foreign trade; and the non-payment of wages (or wage arrears).22
As
with inflation, I was only able to find data on these variables for some of my cases, therefore
they are included whenever they are available. As my rule for whether or not to include these
supplemental variables – whether or not the country reported the data at the regional level –
should not in any way be correlated with likelihood of economic conditions effecting the vote for
incumbents, New Regime, or Old Regime parties, including these extra variables when available
should not bias my findings.
But in order to alleviate any doubts, I also reran all of the analyses using only the three
base variables that were available for every elections and, as will be discussed later in this paper,
found practically the exact same overall results. It is also instructive to recall that the hypotheses
that are being tested with this data do not make specific predictions about the effect of individual
21
The one exception is the 1991 Russian presidential election, for which I was unable to locate any unemployment
data. It is also impossible to find GDP growth disaggregated at the regional level early in the decade. For this
reason, GDP growth is used when available; when it is not available, industrial growth is used instead. In no case is
industrial growth used when GDP growth is available.
22
Wage arrears were a seriously problem in Russia, although not in any of the East Central European cases.
21
variables, but rather about areas of the country that are enjoying more or less economic success.
Therefore any variables that can help touch upon how “good” the economy is doing in a region
should be useful for such an analysis.
The actual regression analyses are conducted using a model appropriate for multi-party
data (Tomz, Tucker and Wittenberg 2002). Details of the model can be found in Appendix II,
but the basic idea is that it is designed to take account of the fact that aggregate electoral data is
compositional: all vote proportions are bounded between 0 and 1, and the sum of all vote
proportions must equal 1. It is primarily distinguished from OLS in that the dependent variable
is modeled as the log of the ratio of the vote for the party in question relative to a base category
(the vote for one or more of the other parties) and that all of the equations for any given election
are estimated simultaneously using a seemingly unrelated regression (SUR) model.
Empirical Results
Although space prohibits reporting the regression results for all twenty elections, consider
as an example the 1996 Czech Parliamentary elections.23
Two parties that competed in the
election are coded as both Incumbent and New Regime parties, the Civic Democratic Party
(ODS), and the Civic Democratic Alliance (ODA). There is one additional incumbent party, the
Christian Democratic Union (KDU), and one Old Regime party, the Communist Party of
Bohemia and Moravia (KSCM). Table 2 presents the results of regressing the vote for each of
these parties on our base economic variables (unemployment, change in income, and growth)
and our five demographic control variables (the percentage of the population employed in
23
The full set of regression results are available upon request and will be included as a web-based appendix with
future versions of the paper.
22
agriculture, the percentage employed in industry, the percentage of elderly citizens, the
percentage of the population living in urban areas, and the population of the region).
These regression results as presented in Table 2 can tell us something interesting about
the effect of the economy on these four parties. But it is unclear how useful they are for the
larger task of assessing whether or not each case provides support for the hypotheses we are
TABLE 2: ESTIMATED COEFFICIENTS (STANDARD ERRORS) OF EFFECT ON PARTY VOTE FOR
THE 1996 CZECH PARLIAMENTARY ELECTIONS*
ODS ODA KDU KSCM
Unemployment Rate
-.087
(.013)
-.113
(.021)
-.087
(.033)
.029
(.013)
Change in Income
.001
(.007)
.008
(.011)
.004
(.018)
.000
(.007)
Industrial Growth
-.001
(.001)
.002
(.002)
-.006
(.003)
.001
(.001)
Percent Agriculture
-.016
(.006)
-.016
(.009)
.082
(.015)
.003
(.006)
Percent Industry
-.005
(.003)
-.005
(.005)
.018
(.007)
-.009
(.003)
Percentage Elderly
.036
(.013)
.052
(.020)
.000
(.031)
.007
(.012)
Percent Urban
.000
(.002)
-.000
(.002)
-.012
(.004)
-.003
(.001)
Log Population
.050
(.047)
.099
(.073)
.806
(.115)
-.069
(.045)
Constant
-1.266
(1.038)
-4.675
(1.628)
-11.418
(2.564)
-.531
(1.008)
R-Squared 0.61 0.51 0.60 0.36
N 76 76 76 76
*Models estimated using seemingly unrelated regression (SUR) with logistic transformation of dependent variable.
Source: Okresy Ceske Republiky v roce 1996 & 1995, Vekove Slozeni Obyvatelstva Ceske Republiky v roce 1996,
Volby do Poslanecke Snemovny Parlamentu Ceske Republiky v roce 1996 and Aktuality CSU: 1996 Duben, Kveten -
1 .cast (Latest News of the Czech Statistical Office: April, May 1996 - Part I). Praha, Czech Republic, Cesky
Statisticky Ura
23
testing, or how we would compare these results with those found from another election. This
would be a difficult enough task if we employed a model such as OLS, whereby we would have
one intuitively logical set of coefficients and standard errors for each of the 88 parties being
analyzed. But as the multi-party election model described above is dependent on a one party (or
group of parties) functioning as a base category, regression analysis will produce a different set
of coefficients and standard errors depending on which of the parties are included as the base
party in exactly the same way as a multi-nomial logit analysis. Thus the correct way to interpret
the coefficient for unemployment on the ODS in Table 2 is not as the effect of unemployment on
the vote for the ODS, but rather as the effect of unemployment on the vote for the ODS relative
to the composite “other” category (the vote for all parties besides the ODS, ODA, KDU, and
KSCM; see Appendix II for details).
Therefore, if we choose to rely on the coefficients and standard errors of economic
variables for assessing the empirical support for our hypotheses, we would actually need to rotate
through every party except one as the base party in turn in order to get a complete set of
coefficients and standard errors.24
And with an average of four economic variables per analysis
and four and a half parties per election over twenty elections, we would end up comparing close
to 2,000 coefficients and standard errors in an effort to make our overall assessments of the
empirical support of the different hypotheses. Moreover, with a log-transformed dependent
variable, none of these coefficients would have any real intuitive meaning. Thus in order to
analyze the our hypotheses comparatively, we need a much more compact estimate of the effect
of economic conditions on the election results for each party that can be compared across cases.
24
We would not need to rotate every party through because we can calculate the final party from the other
regressions, as a pairs of parties (one in the regression, one in the base) will also produce the exact opposite
coefficients with the same standard errors by mathematical definition.
24
An appropriate measure to use in this case is a first difference that calculates the change
in the predicted share of the vote for each party when demographic variables are held constant
and economic conditions are varied to simulate a standardized shift from a “bad” economic
situation to a “good” economic situation. To do this in practice, I calculate two estimates for the
expected vote of the party in question: one with economic variables at their 10th
percentile level
and another with economic variables at their 90th
percentile level. In both cases demographic
variables are set at their means. Negative indicators, such as unemployment, are shifted from the
higher value to the lower value, and vice versa for positive indicators, such as change in
income.25
When the expected vote in the “bad” economic situation (e.g. high unemployment,
low wage growth) is subtracted from the expected vote in the “good” situation (e.g. low
unemployment, high wage growth), the result is a single measure of the overall impact of better
economic conditions on the vote for that particular party, which is exactly what is needed to
analyze the Incumbency, New Regime, and Old Regime hypotheses. Moreover, not only do first
differences have the advantage of producing a single, intuitive measure of the effect of economic
conditions on the electoral fortunes of the party in question, they are also robust by mathematic
definition to which party is included as the base party for the purpose of the regression analysis.
There is, however, one important shortcoming in comparing first differences for different
parties. In moving from a set of betas with standard errors to a single point estimate of a first
difference, however, we run the risk of losing information regarding our uncertainty in the
quantity of interest being reported. Fortunately, this can be avoided by using stochastic
simulation to simulate an entire distribution of first differences.26
Such an approach has the
25
For example, if there were 100 regions, we would use the unemployment level from the region with the 10th
highest rate in one case and from the 10th
lowest rate in the second case.
25
added benefit of allowing us to quantify not only our best estimate of the first difference (the
mean value of the simulations) but also our level of uncertainty surrounding that estimate (the
standard deviation of the simulations). Moreover, we can easily generate confidence intervals
surrounding substantively meaningful quantities of interest.
For example, in the empirical analysis to follow, the crucial concern is whether we can be
confident that the party in question either benefits from better economic conditions (Incumbent
and New Regime parties) or is hurt by better economic conditions (Old Regime parties). To
assess this claim, we will want to know how confident we are that the first difference is greater
than zero (in other words, that better economic conditions increase the expected vote for the
party), or, in the case of the Old Regime parties, less than zero. If 90% of the simulated first
differences are greater than zero, then we can claim with 90% confidence that the party in
question is helped by stronger economic conditions; the same holds in the reverse direction if
90% of the simulated first differences are less than zero. In this manner, calculating estimated
probability distributions of first differences allows us to measure exactly what we need to test our
hypotheses: how confident we should be that better economic conditions translated into better (or
worse) election results for the party in question.
For ease of interpretation, these simulations can be plotted graphically, resulting in an
estimated probability distribution of the first difference. Figure 1 plots these estimated
distributions for the previous example, the 1996 Czech parliamentary elections.
26
More technically, we draw 1,000 betas from the sampling distribution of the parameter estimates. From each set
of betas, a separate first difference was calculated. The simulations were performed using Clarify 2.1 (Tomz,
Wittenberg and King 20003). For a full treatment of the approach, see King, Tomz and Wittenberg 2000).
Moreover, stochastic simulation of first differences also produces unbiased estimates of the first differences
regardless of base parties in the same manner as the calculation of first differences without stochastic simulation.
While the estimated distribution may vary across repeated runs, all variance will be due to the unbiased variance in
the simulation procedure and not to the choice of base party. Put another way, repeatedly running simulations with
the same base party will generate no less variation across sets of simulations than running repeatedly running
simulations rotating the base party.
26
From Figure 1, we find that this particular election provides clear empirical support for
the New Regime and Old Regime hypotheses, and mixed empirical support for the Incumbent
Hypothesis. We are very confident that both New Regime parties, the Civic Democratic Party
(ODS) and the Civic Democratic Alliance (ODA), do better where economic conditions are
better and worse where economic conditions are worse, and that the one Old Regime party, the
Communist Party of Bohemia and Moravia (KSCM), does better when economic conditions are
FIGURE 1: ESTIMATED PROBABILITY DISTRIBUTION OF EFFECT OF ECONOMIC CONDITIONS ON NEW REGIME, OLD
REGIME, AND INCUMBENT PARTIES IN 1996 CZECH REPUBLIC PARLIAMENTARY ELECTIONS
Civic Democratic Alliance (ODA)
Density
Predicted Change in % Vote
-10 -5 0 5 10
0
Christian Democratic Union -CPP
Density
Predicted Change in % Vote
-10 -5 0 5 10
0
Civic Democratic Party (ODS)
Density
Predicted Change in % Vote
-10 -5 0 5 10
0
Communist Party of Bohemia and Moravia (KSCM)
Density
Predicted Change in % Vote
-10 -5 0 5 10
0
100%
Positive
60.6%
Positive
99.7%
Positive
1.2%
Positive
27
worse and worse when economic conditions are better (as approximately 99% of the simulations
predicted a decrease in the vote for the party as economic conditions improve). The evidence in
support of the Incumbent Hypothesis is not quite as strong. As noted above, we are confident
that the Incumbent ODA and ODS are helped by better economic condition. The evidence
provided by our analysis of the vote for the Christian Democratic Union (KDU), the third
incumbent party is very weak. While our best guess at the effect of a positive shift of economic
conditions (the mean of the simulations) is positive, almost 40% of the simulations show that the
party’s fortunes would decline as economic conditions improved. Thus we have very little
confidence that this example provides support for the Incumbency Hypothesis.
Comparative Analysis
Rather than display the 84 additional estimated distributions from the other 19 elections
and continue to discuss each individually, Table 3 concisely summarizes our level of confidence
in all of the empirical evidence for all of the hypotheses.
TABLE 3. NUMBER OF CASES WITH 90% CERTAINTY OF PREDICTED EFFECT BY MODEL
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 25 21 20
90% Confident in Wrong Direction 1 1 8
Total Cases 40 29 49
Percentage with 90% Confidence in
Predicted (Wrong) Direction
63% (3%) 72% (3%) 41% (16%)
Table 3 displays a count of the number of parties for which we are 90% certain that
moving from a bad economy to a good economy produces the type of effect – positive or
negative – that the hypotheses predict should be present. As explained above, the percentage of
28
positive first differences is used to measure the level of certainty that the effect is as predicted by
the model. Thus for Incumbent and New Regime parties, we are 90% confident that the effect is
as predicted if at least 90% of the predicted first differences are positive; for Old Regime parties,
the same threshold is achieved if no more than 10% of the predicted first differences are positive.
The table also lists the number of cases for which 90% of the simulations are in the opposite
direction from what is predicted; these are the cases for which the models appear to be clearly
wrong. We identify these cases in the same manner as the cases where there is 90% confidence
of the effect, only with the signs reversed.
Table 3 suggests strongly that there is more empirical support for the Transition Model
than the Referendum Model. Observe first the Old Regime Hypothesis, where we have strong
confidence that the empirical data supports the hypothesis in 72% of the cases (21 out of 29).
The empirical support for the New Regime Hypothesis is not quite as strong, but we still have
strong confidence that the empirical data supports the hypothesis in almost two-thirds (25 out of
40, or 63%) of the cases. Moreover, for both the New Regime and Old Regime Hypotheses there
is only one example of a case where we have strong confidence that the empirical data actually
supports the opposite from what was predicted by the hypothesis.27
By comparison, we only find strong empirical support for the Incumbent Hypothesis in
slightly more than 40% (20 out of 49) of the cases. This is a rather startling finding if one stops
to consider it. After all, the Incumbent Hypothesis tends to dominate the way most analysts of
27
For the Old Regime Hypothesis, this is a small Russian far-left communist movement, the “Communist Workers
of Russia – For the Soviet Union” bloc in the 1995 Russian parliamentary elections. For the New Regime
Hypothesis, this is perhaps one of the most difficult coding decisions in the entire dataset, Rudolph Schuster in the
1999 Slovak presidential elections. Schuster is coded as a New Regime candidate due to his being the candidate of
choice of the Slovak government that came to power in 1998 and was led by a New Regime Party, the Slovak
Democratic Coalition, despite the fact that Schuster’s own party, the Party of Civic Understanding, was not coded as
a New Regime party. So Schuster could easily not have been coded as a New Regime candidate, which would have
left the New Regime Hypothesis without a single case where there was strong empirical support for a party
performing in the opposite as predicted by the hypothesis. For more, see Tucker 2004, ch. 5.
29
the politics of economic reform think about elections in transition countries. And yet, we now
find that in more than half of the cases for which we have empirical evidence, it is impossible to
state with 90% confidence that this relationship does exist. In addition, the evidence suggests
that the exact opposite from what the Incumbent Hypothesis predicts occurs with a 90% level of
certainty in almost one-sixth (8 out of 49, or 16%) of the cases.
Robustness Tests
TABLE 4. TRANSITION MODEL HAS MORE EMPIRICAL SUPPORT REGARDLESS OF THRESHOLD
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 63% 72% 41%
95% Confident in Predicted Direction 50% 59% 33%
66% Confident in Predicted Direction 73% 83% 49%
Majority in Correct Direction (>50%) 80% 90% 63%
Total Cases 20 29 49
Moreover, these findings are robust in a number of important ways. Table 4
demonstrates that our conclusion about the superiority of the evidence supporting the two
Transition Model hypotheses is in no way dependent upon the choice of our threshold for
“strong” confidence in the predicted effect. The evidence from the table is remarkably clear. No
matter which of these thresholds were adopted, the conclusion would have been the same: there
is more empirical support for the Transition Model than for the Referendum Model. In addition,
30
our ordering of the empirical support for the two Transition Model hypotheses would also have
remained the same, as in each case we would have concluded that there was more empirical
support for Old Regime Hypothesis than the New Regime Hypothesis.
The final line of the row of the table – noting the number of cases where at least a
majority of the simulations were in the correct direction – is worth noting because this is
essentially what we would have come up with had we just calculated first differences and then
looked to see whether these first differences were positive or negative without attempting to
assess our level of uncertainty. Two observations can be drawn in this regard. First, we still
would have come to the same overall conclusion as to the superiority of the Transition Model to
the Referendum Model. At the same time, we would have significantly overestimated the degree
of empirical support for all the hypotheses, claiming support in quite a number of cases where
although our “best guess” would be in the correct direction, there would be so much underlying
uncertainty in that guess that claiming it as strong evidence would be unwarranted.28
The overall results also hold up nicely to respecification of the original economic voting
models. In an attempt to test the robustness of the findings, I re-analyzed every election in four
different ways by: (1) adding an extra explanatory variable to measure poverty (2) moving at
least one additional party out of the “other” category and including it as a separate dependent
variables (3) changing the size of the economic shock from the 10th
to 90th
percentile to the 20th
to 80th
percentile and (4) as mentioned earlier, reducing the number of economic variables to
only the three base variables (unemployment, growth, and change in income).
28
In a sense, this would be akin to running a regression and looking only at the sign of a coefficient without
considering its standard error.
31
As wage arrears were so important in Russia and a non-issue elsewhere, I also re-ran the
Russian cases using the base economic variables plus arrears. While each change led to some
differences in the results for individual parties, in no case did any of the overall conclusions
changes. As demonstrated in Table A1 (found in Appendix III), no matter which of these
specifications are used, the overall conclusions remain the same: there is much stronger
empirical support for the Transition Model then the Referendum Model; we always have the
most support for the Old Regime Hypothesis, then the New Regime Hypothesis, and then the
Incumbency Hypothesis; and we never have more than half of the cases providing support for the
Incumbency Hypothesis.
It is also interesting to note that the findings hold up across intuitively relevant
subsections of the dataset.29
Table A2 (also in Appendix III) demonstrates that if I had chosen to
include only parliamentary elections, or only East Central European elections, or only parties
that had received at least 10% vote, we would still find much stronger support for the Transition
Model than the Referendum Model.30
Finally, we can also subject these results to a limited out of sample test. When I
originally started this project, I was working with only the 14 elections that took place between
1990-96 due to the fact that the data necessary to analyze the remaining elections were not yet
available. I presented these findings in a series of conference papers (e.g., Tucker 1999a; Tucker
29
Another interesting finding in this regard is that there is practically no difference in the degree of support for
reformed Old Regime parties (e.g., the Democratic Left Alliance in Poland or the Hungarian Socialist Party) than
there is for unrepentant Old Regime parties (e.g., the Communist Party of the Russian Federation or the Communist
Party of Bohemia and Moravia in the Czech Republic); see Tucker 2004, chapter 6 for details.
30
The other two patterns – finding the most support for the Old Regime Hypothesis and support in fewer than half
the cases for the Incumbency Hypothesis hold up for the parliamentary and East Central European subsections of the
data. In the case of parties with at least 10% of the vote, however, the New Regime Hypothesis enjoys as much
empirical support (73% of the case) as the Old Regime Hypothesis (72% of the cases). We also find empirical
support for the Incumbency Hypothesis in just over 50% of the cases (11 out of 21) in this subset of the data,
although to get to this level or accuracy requires a great reduction in the scope of the model, with the total number of
parties for which the hypothesis makes predictions dropping from 49 to 21.
32
1999b), making the same basic overall point that there was more empirical support for the
Transition Model than for the Referendum Model. In a sense, then, the final set of 6 elections
could be considered an out of sample test for the findings from the original 14 elections. As
Table A3 (in Appendix III) demonstrates, examining just the post-1996 elections also leaves one
with the inescapable conclusion that there is more empirical support for the Transition Model
than the Referendum Model.31
Interacting the Models
As none of the hypotheses are able to make correct predictions in all of the cases, we can
perform a final empirical test of each model to assess the degree to which it can help explain the
shortcomings of the other model. For example, we can break down Incumbents into New
Regime Incumbent parties, Old Regime Incumbent parties, and Other (defined as neither New
Regime nor Old Regime) Incumbent parties. If the Transition Model has strong explanatory
power, then New Regime–Incumbents should predominate in the group where we are confidence
that the results are in the correctly predicted direction, as the New Regime Hypothesis predicts
the same effect as the Incumbent Hypothesis. By the same logic, Old Regime–Incumbents
should make up most of the group where we are confident that the results are in the opposite
direction from what was predicted by the model, as the Old Regime Hypothesis predicts the
opposite effect of the Incumbent Hypothesis. We can make similar predictions for New Regime
parties (expecting that we should find more support for the New Regime Hypothesis amongst
New Regime Incumbents than New Regime Opposition parties) and Old Regime parties
31
Although beyond the scope of the present paper, it is interesting to note that all three hypotheses generate stronger
empirical support from the post-1996 elections. In particular, we now find the strongest empirical support for the
New Regime hypothesis, which has strong empirical support in 10 out of 11 cases, with the one exception being the
aforementioned example of Rudolph Schuster from the 1999 Slovak presidential election.
33
(expecting that we should find more support for the Old Regime Hypothesis amongst Old
Regime Opposition parties than Old Regime Incumbents)
Panel 1 of Table 5 demonstrates that the Referendum Model adds very little to our
understanding of the shortcomings of the Transition Model. While there is slightly more
empirical support for the New Regime Hypothesis amongst New Regime Incumbent parties than
New Regime Opposition parties, which the Incumbency Hypothesis would lead us to expect,
there is also slightly more empirical support for the Old Regime Hypothesis amongst Old
Regime Incumbent parties than Old Regime Opposition parties, which is not what the
Incumbency Hypothesis would lead us to expect. In either case, though, the differences in
support are slight.
Conversely, as Panel 2 of Table 5 demonstrates, the Transition Model can account for
almost all of the differences found across Incumbent parties. When Incumbents are also New
Regime parties, we find strong support for the Incumbency Hypothesis in two-third of the cases
(14 out of 21). But when Incumbents are also Old Regime parties, we do not find any support
for the Incumbency Hypothesis; indeed, we find strong support for the opposite of what the
Incumbency Hypothesis predicts in almost 80% (7 out of 9) of the cases. For Incumbents that
are neither Old Regime nor New Regime parties, we find support for the Incumbency Hypothesis
in roughly one-third (6 out of 19) of the cases, which is basically half way between the finding
for the New Regime Incumbents and the Old Regime Incumbents.
Thus we arrive at a second major empirical finding: the effect of the economy on
Incumbent parties is almost completely dependent on their relationship to the transition. For
Incumbent parties that are also New Regime parties, we find strong support for the prediction of
the Incumbency Hypothesis across a wide number of cases. But for Incumbent Parties that are
34
TABLE 5. INTERACTION OF REFERENDUM AND TRANSITION MODELS
Panel 1. Transition Model: Breakdown of Certainty by Incumbency
New Regime Old Regime
Incumbent Opposition Incumbent Opposition
90% Confident in Predicted Direction 14 11 7 14
90% Confident in Wrong Direction 0 1 0 1
Total Cases 21 19 9 20
Percentage with 90% Confidence in
Predicted (Wrong) Direction
67% (0%) 58% (5%) 78% (0%) 70% (5%)
Panel 2. Referendum Model: Breakdown of Certainty by Transition Model Categories
Incumbents
New Regime Other Old Regime Total
90% Confident in Predicted Direction 14 6 0 20
90% Confident in Wrong Direction 0 1 7 8
Total Cases 21 19 9 49
Percentage with 90% Confidence in
Predicted (Wrong) Direction
67% (0%) 32% (5%) 0% (78%) 41% (16%)
also Old Regime parties, we find almost no evidence that these parties perform better in parts of
the country where the economy is stronger. And Incumbent parties that are neither New Regime
nor Old Regime parties perform are effected by economic conditions as expected some of the
time, but not nearly as often as the received wisdom would have had us expect and not nearly as
often as Incumbent New Regime parties.
35
Case Study Analysis
The previous section examined how the models faired compared to one another across all
twenty elections. However, another legitimate question to ask is how the models fair against one
another within specific elections. In particular, elections that feature Old Regime parties running
as the incumbent would lead to contrasting predictions from the Referendum and Transition
Models. Two elections of this type are the 1997 Polish and 1998 Hungarian parliamentary
elections, both of which involved a communist successor party running for re-election after
having ruled in coalition with one other party for almost the entire period of time since the
preceding election.
In Poland, the communist successor party was the Democratic Left Alliance (SLD),
which had come to power in 1993 and ruled in coalition with the Polish Peasant Party (PSL) for
almost the entire period leading up to the 1997 elections. The SLD’s primary opposition was the
Electoral Action Solidarity (AWS) coalition, a grouping of over 30 right wing parties and
movements under the leadership of the Solidarity Trade Union.32
Competing with both the AWS
and the SLD was the Union of Freedom (UW), which had been created by a merger in April
1994 of the Democratic Union and the Congress of Liberal Democrats, the two parties which had
lead most of Poland’s governments from 1989-93 and were most closely associated with
Poland’s original shock therapy economic reform program. Thus for this election, the SLD is
coded as an Old Regime party, AWS and UW as New Regime parties, and SLD and PSL are the
incumbent parties.
In Hungary, the main communist successor party was the Hungarian Socialist Party
(MSzP), which had been in power since 1994 in a coalition with the liberal Alliance of
Democrats (SzDSz). The MSzP’s primary competition in the election was FIDESZ-Hungarian
32
See Szczerbiak 1999, Appendix I, for a list of the parties comprising the movement.
36
People’s Party. (Fidesz-MPP) was distinguished from the AWS in Poland in two important
ways. First, Fidesz-MPP was not a new party. It had run in both the 1990 and 1994 elections as
a liberal anti-communist party of young people. Indeed, “Fidesz” was the acronym from the
party’s original name, the League of Young Democrats (Fiatal Demokraták Szővetsége).
Second, Fidesz-MPP did not attempt to consolidate the anti-communist successor vote by
forming an electoral coalition, but rather by establishing itself as the premier opposition party.
Seemingly stuck between the two was the SzDSz, a liberal party which had entered into coalition
with the MSzP and now found itself in the unenviable position of needing to both defend the
status quo and present itself as an alternative to the main party responsible for the status quo, its
coalition partner the MSzP. The Hungarian Democratic Forum (MDF), another New Regime
party that had been the lead coalition member in Hungary’s first post-communist government,
also competed in the election. Thus for this election, MSzP and SzDSz are coded as incumbents,
SzDSz, Fidesz-MPP, and the MDF are coded as New Regime parties, and MSzP is coded as an
Old Regime parties; the Workers Party (MP), a hard line communist spin off from MSzP, is also
coded as an Old Regime Party. In both cases, the Incumbency and Old Regime Hypotheses
present a contradictory prediction for one party: the SLD in Poland and the MSzP in Hungary.
Table 6 presents the results of our test of all three hypotheses using the data from these
two elections and reflects the general pattern of greater empirical strength for the Transition
Model than for the Referendum Model. Most importantly, note that the two Old Regime parties
that are also incumbent parties (the SLD in Poland and the MSzP in Hungary) both fail to
provide any support for the Incumbency Hypothesis, but do in fact provide strong support for the
Old Regime Hypothesis. Of the two remaining incumbent parties, it is only the New Regime
SzDSz that provides strong support for the Incumbency Hypothesis; the analysis of the Polish
37
PSL, which is coded as neither a New Regime nor an Old Regime party, reveals no systematic
effect for the economy in either direction, despite its status as an incumbent party. The nonincumbent
Old Regime (the MP in Hungary) and New Regime parties (the UW in Poland and the
MDF and Fidesz-MPP in Hungary) also all provide empirical support for the Old Regime and
New Regime Hypotheses, respectively. Thus even in elections featuring head to head tests
between the two models, we find much stronger support for the Transition Model.
TABLE 6. SUPPORT FOR HYPOTHESES IN THE 1997 POLISH AND 1998 HUNGARIAN ELECTIONS
Country Year Party
Confidence in
Predicted Effect
Incumbent Parties
Poland 1997 Democratic Left Alliance (SLD) 1.2
Polish Agricultural Party (PSL) 50.7
Hungary 1998 Hungarian Socialist Party (MSzP) 0.4
Alliance of Free Democrats (SzDSz) 100
New Regime Parties
Poland 1997 Electoral Action Solidarity (AWS) 96.2
Union of Freedom (UW) 93
Hungary 1998 FIDESZ-Hungarian Civic Party (FIDESZ-
MPP)
100
Alliance of Free Democrats (SzDSz) 100
Hungarian Democratic Forum (MDF) 99.4
Old Regime Parties
Poland 1997 Democratic Left Alliance (SLD) 98.8
Hungary 1998 Hungarian Socialist Party (MSzP) 99.6
Hungarian Workers Party (MP) 100
38
Conclusions and Implications
The very idea of democracy is dependent to a certain extent on voters using elections to
enforce some sort of control upon their elected representatives. Both the Referendum Model and
the Transition Model are based on the idea that voters are attempting to do that, albeit based on
different means of doing so. Seen in this light, the results from the post-communist countries
analyzed in this article are very encouraging. Patterns that are consistent with these types of
models have undoubtedly emerged, and emerged quickly, in these countries. There is clearly
some effect for economic conditions on Incumbent, New Regime, and Old Regime parties
consistent with what the models predicted.
However, the data have also proved to be sufficiently rich to allow us to assess the
comparative value of the two models. Interestingly, the results of this comparison are fairly
clear. Across numerous tests, the data definitely generate more empirical support for the
Transition Model, and its predicted effects on New Regime and Old Regime parties, than for the
Referendum Model, which makes predictions about Incumbent parties. Of the three hypothesis,
we consistently find the strongest empirical support for the Old Regime Hypothesis, followed by
the New Regime Hypothesis, and then the Incumbency Hypothesis. Finally, the effect of the
economy on Incumbent parties is almost completely conditional on their status as New Regime
parties, Old Regime parties, or neither.
The Transition Model is also attractive in the context of the post-communist cases for
other reasons. In political science, we generally judge models according to three criteria:
accuracy, scope, and parsimony. The preceding section demonstrates that the Transition Model
produces more accurate predictions than the Referendum Model. The question of whether the
Transition Model or the Referendum Model is more parsimonious is an open one. In the
39
universe of models, both are very parsimonious, yielding very simple hypotheses about the
direction of economic effects (e.g., do good economic conditions help or hurt?) on certain types
of parties. As ascertaining the direction of economic effects is constant across both models, the
only way we can distinguish their degree of parsimony from each other is by focusing on the
coding decision that both require for identifying the parties for which the models make
predictions. In a stable democracy, we would most likely conclude that simply identifying
incumbents would be an easier task than coding parties in most any capacity. However,
identifying incumbents in new democracies, especially following the dissolution of umbrella
movements, can be tricky and time consuming. By comparison, most of the Old Regime coding
decisions appear fairly simple, although it is more difficult to make this conclusion about New
Regime parties. Probably the safest conclusion is to note that both models are fairly
parsimonious although they do involve coding decision that range from easy to difficult and that
it is difficult to call one more parsimonious than the other.
Scope, on the other hand, provide a much clearer distinction. The Transition Model has a
significantly wider scope than the Referendum Model, affording us the opportunity to make
predictions for 69 parties as opposed to 49 parties. Moreover, the Transition Model tells us
where to look for parties that are likely to benefit both from better economic conditions (New
Regime parties) and from worse economic conditions (Old Regime parties), whereas the
Referendum Model only offers predictions about which parties are likely to benefit from better
economic conditions (Incumbents).33
In a sense, therefore, the Transition Model has both wider
theoretical and empirical scope than the Referendum Model.
33
We could of course greatly increase the scope of the Referendum Model to the maximum extent possible by
adding an “Opposition Hypothesis” that predicted that all non-incumbents should benefit where economic
conditions were worse. Aside from the theoretical concerns inherent in such a prediction (why would we expect all
opposition parties to benefit equally from poor economic conditions?) doing so would significantly further decrease
40
These findings also have potentially interesting implications for the long standing debate
in political science about the relationship between elections and political accountability.34
Originally, the Downsian approach held that voters insured accountability by selecting
politicians and parties based on their proximity to one’s own preferences over issues (Downs
1957). Empirical analysis, however, cast doubt on the Downsian approach; as Ferejohn (1986)
succinctly noted, “the actual evidence for extensive issues voting is fairly weak” (p.7). An
alternative approach is retrospective voting, which suggests that voters can use a punishment
mechanism to hold politicians accountable for their behavior in office (Fiorina 1981; Ferejohn
1986). But retrospective voting itself has also come under attack, both from micro-level studies
of economic voting that suggest voters may be more interested in future economic conditions
(MacKuen, Erikson and Stimson 1992), as well as from more theoretically oriented studies that
present sanctioning-oriented voting as but one means of ensuring electoral accountability, and
perhaps not even the most important one at that (Fearon 1999).
The debate has been raised to prominence once again in the edited volume Democracy,
Accountability, and Representation (Przeworski, Stokes and Manin 1999). In the opening
chapters of the volume, the authors very clearly frame the debate as one between “Sanctioning
Representation” and “Mandate Representation.” In the case of the former, voters exert control
over politicians by threatening to punish them in subsequent elections, while the latter, voters
our assessment of the accuracy of the Referendum Model. Indeed, if we tested such an Opposition Hypothesis using
the parties included in Panel 2 of Table A1, we would find almost as many cases with strong empirical support in
the wrong direction (24%, or 18 out of 76 cases surpassing the 90% threshold) as in the correctly predicted direction
(36%, of 27 out of 76 cases). Note as well that if we exclude Old Regime parties from this sample, the proportion of
cases in the incorrect direction (31%, or 17 out of 56 case) would actually surpass the number of cases in the
correctly predicted direction (23%, or 13 out of 56 cases).
34
Although some would argue that it is a debate which has generally received insufficient attention. As Fearon
1999 notes, “the main justification for freely contested and regular elections is that these are supposed to bring about
democracy, or rule according to the will of the people . . . [but] the tradition of democratic theory contains
surprisingly little on exactly how elections are supposed to produce this result.”(p.82).
41
control politicians by voting for politicians whose policies they prefer (Manin, Przeworski and
Stokes 1999a, see as well the table on p.9 of Manin, Przeworski and Stokes 1999b).
Another way to interpret the results of this article, therefore, is as a novel test for
examining whether voters are behaving in accordance with the worlds of Sanctioning
Representation or Mandate Representation by moving away from the context of American
politics. While the analyses presented in this article do not test, nor do they claim to test, actual
individual level motivation for voting decisions, the two models are based on the idea of voters
behaving as if they are living in one of these two types of worlds. More specifically, the
Referendum Model – with its emphasis on reacting to the presence of incumbents – is consistent
with voters living in a Sanctioning Representation world, while the Transition Model – with its
emphasis on voters seeking out parties because of the economic strategies that might result from
their being in office – is consistent with the idea of voters living in a Mandate Representation
model.
Thus one way to interpret the results presented above would be to suggest that voters in
post-communist countries are not only capable of behaving in accordance with the predictions of
Mandate Representation, they actually seem to do so much more often than they behave in
accordance with the predictions of Sanctioning Representation. Thus in a part of the world that
we might expect to be a more “difficult” case for the Mandate Representation approach (voters
know less about parties and their policies; we might expect them to be more “angry” given the
tremendous upheaval in their lives during the transition to capitalism), we find evidence that
strengthens the argument for Mandate Representation at the expense of Sanctioning
Representation. Even if one thinks this may be stretching the results of the study too far, then at
the very least it certainly suggesting an interesting avenue for future research.
APPENDIX I: POLITICAL PARTY AND CANDIDATE ABBREVIATIONS
Abbreviation Party Name Country
AZR Association of Slovak Workers Slovakia
Bakatin Vadim Bakatin Russia
Cimoszewicz Wlodzimierz Cimoszewicz Poland
CSSD Czechoslovak Social Democratic Party Czech Republic
DS Democratic Party Slovakia
DU Democratic Union of Slovakia Slovakia
FIDESZ League of Young Democrats Hungary
Gronkiewicz-Waltz Hanna Gronkiewicz-Waltz Poland
HZDS Movement for a Democratic Slovakia Slovakia
ISP Independent Smallholders Party Hungary
KDH Christian Democratic Movement Slovakia
KDNP Christian Democratic People's Party Hungary
KDU Christian Democratic Union Czech Republic
KKW Catholic Electoral Committee "Fatherland" Poland
KLD Congress of Liberal Democrats Poland
KPRF Communist Party of the Russian Federation Russia
KSCM Communist Party of Bohemia and Moravia Czech Republic
KSS Communist Party of Slovakia Slovakia
KTR Communist Workers of Russia Russia
Kuron Jacek Kuron Poland
Kwasniewski Aleksander Kwasniewski Poland
LB Left Bloc Czech Republic
Mazowiecki Tadeusz Mazowiecki Poland
MDF Hungarian Democratic Forum Hungary
MSzMP (MP) Hungarian Socialist Workers Party (Workers Party) Hungary
MSzP Hungarian Socialist Party Hungary
NDR Our Home is Russia Russia
ODA Civic Democratic Alliance Czech Republic
ODS Civic Democratic Party Czech Republic
ODU Civic Democratic Union Slovakia
OH Civic Movement Czech Republic
PC Center Citizens Alliance Poland
PL Agrarian Alliance Poland
PRES Party of Russian Unity and Accord Russia
PSL Polish Peasant Party Poland
Ryzhkov Nikolai Ivanovich Ryzhkov Russia
SDL Party of the Democratic Left Slovakia
SLD Democratic Left Alliance Poland
SNS Slovak National Party Slovakia
SV Common Choice Coalition Slovakia
SzDSz Alliance of Free Democrats Hungary
UD Democratic Union Poland
US Union of Freedom Czech Republic
VR (VDR) Russia's (Democratic) Choice Russia
WAK Catholic Electoral Action Poland
Walesa Lech Walesa Poland
Yavlinsky Grigorii Yavlinsky Russia
Yeltsin Boris Yeltsin Russia
Zyuganov Gennadii Zyuganov Russia
APPENDIX II: THE STATISTICAL MODEL
•Vi = (Vi1, . . . , Vi (J-1)) is a vector of vote proportions of party j = ( 1, … , J-1) for each district i
(i = 1, … n)
•Yi is a vector of J-1 log ratios where Yij = ln(Vij/ViJ) for party j (j = 1, …, J-1) relative to party J
•Yiis multivariate normal with a mean µ and a variance of Σ.
•Means can be estimated as a linear function of explanatory variables, whereby :
•Equations are estimated simultaneously for all parties j = ( 1, … , J-1) using a seemingly
unrelated regression
•For more, see Tomz, Tucker, Wittenberg (2002)
jijij βµ X=
APPENDIX III: SUPPLEMENTARY TABLES
TABLE A1. NUMBER OF CASES WITH 90% CERTAINTY OF PREDICTED EFFECT BY MODEL BY
DIFFERENT SPECIFICATIONS OF ORIGINAL ANALYSES
Panel 1. Including an Extra Independent Variables (Poverty)
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 23 22 18
90% Confident in Wrong Direction 2 0 10
Total Cases 40 29 49
Percentage with 90% Confidence in
Predicted (Wrong) Direction
58% (5%) 76% (0%) 37% (20%)
Panel 2. Including Additional Dependent Variables
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 24 20 20
90% Confident in Wrong Direction 1 1 7
Total Cases 40 29 49
Percentage with 90% Confidence in
Predicted (Wrong) Direction
60% (3%) 69% (3%) 41% (14%)
Panel 3. Alternative Economic Shock (20th
to 80th
Percentile)
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 25 19 20
90% Confident in Wrong Direction 1 1 7
Total Cases 40 29 49
Percentage with 90% Confidence in
Predicted (Wrong) Direction
63% (3%) 66% (3%) 41% ( 14%)
TABLE A2. COMPARING THE TRANSITION AND REFERENDUM MODELS
ACROSS SUBSETS OF DATA
Panel 1. Parliamentary Elections
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 21 18 18
90% Confident in Wrong Direction 0 1 7
Total Cases 32 24 43
Percentage with 90% Confidence in
Predicted (Wrong) Direction
66% (0%) 75% (4%) 42% (16%)
Panel 2. East Central European Elections
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 20 19 16
90% Confident in Wrong Direction 1 0 8
Total Cases 32 21 42
Percentage with 90% Confidence in
Predicted (Wrong) Direction
63% (3%) 90% (0%) 38% (19%)
Panel 3. Parties with at Least 10% of the Overall Vote
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 16 13 11
90% Confident in Wrong Direction 1 0 4
Total Cases 22 18 21
Percentage with 90% Confidence in
Predicted (Wrong) Direction
73% (5%) 72% (0%) 52% (19%)
TABLE A3. OUT OF SAMPLE TEST
Panel 1. Original 14 Elections (through 1996)
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 15 14 14
90% Confident in Wrong Direction 0 1 5
Total Cases 29 20 37
Percentage with 90% Confidence in
Predicted (Wrong) Direction
52% (0%) 70% (5%) 38% (14%)
Panel 2. Post 1996 Elections
Transition Model Referendum Model
New Regime Old Regime Incumbents
90% Confident in Predicted Direction 10 7 6
90% Confident in Wrong Direction 1 0 3
Total Cases 11 9 12
Percentage with 90% Confidence in
Predicted (Wrong) Direction
91% (9%) 78% (0%) 50% (25%)
REFERENCES
Aidt, Toke. 2000. “Economic Voting and Information.” Electoral Studies 19: 349-362.
Anderson, Christopher J. 2000. “Economic Voting and Political Context: A Comparative
Perspective.” Electoral Studies 19 (2/3): 151-70.
Bell, Janice. 1997. “Unemployment Matters: Voting Patterns during the Economic Transition in
Poland, 1990-1995.” Europe-Asia Studies 49 (7): 1263 -1291.
Bellucci, Paolo. 1984. “The Effect of Aggregate Economic Conditions on the Political
Preferences of the Italian Electorate, 1953-1979.” European Journal of Political
Research 12 (4): 387-401.
Bunce, Valerie and Maria Csanadi. 1993. “Uncertainty in Transition: Post Communism in
Hungary.” East European Politics and Societies 7 (2): 240-275.
Colton, Timothy J. 1996. “Economics and Voting in Russia.” Post-Soviet Affairs 12 (4): 289-
318.
Colton, Timothy J. 1998. "Ideology and Russian Mass Politics: Uses of the Left-Right
Continuum," in Elections and Voting in Post-Communist Russia. Matthew Wyman,
Stephen White and Sarah Oates, eds. Northampton, MA: Edward Elgar: 167-189.
Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper & Row.
Duch, Raymond. 2000. “Heterogeneity in Perceptions of National Economic Conditions.”
American Journal of Political Science 44 (4): 635-652.
Duch, Raymond. 2002. “A Developmental Model of Heterogeneous Economic Voting in New
Democracies.” American Political Science Review 95 (4): 895-910.
Fearon, James D. 1999. "Electoral Accountability and the Control of Politicians: Selecting Good
Types versus Sanctioning Poor Performance," in Democracy, Accountability, and
Representation. Adam Przeworski, Susan Stokes and Bernard Manin, eds. Cambridge,
England: Cambridge University Press: 55-97.
Ferejohn, John. 1986. “Incumbent Performance and Electoral Control.” Public Choice 50: 5-25.
Fidrmuc, Jan. 2000a. “Economics of Voting in Post-Communist Countries.” Electoral Studies 19
(2/3): 199-217.
Fidrmuc, Jan. 2000b. “Political Support for Reforms: Economics of Voting in Transition
Countries.” European Economic Review 44: 1491-1513.
Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. New Haven, CT:
Yale University Press.
Gibson, John and Anna Cielecka. 1995. “Economic Influences on the Political Support for
Market Reform in Post-Communist Transitions: Some Evidence from the 1993 Polish
Parliamentary Elections.” Europe-Asia Studies 47 (5): 765-785.
Goidel, Robert and Ronald Langley. 1994. “Media Coverage of the Economy and Aggregate
Economic Evaluations.” Political Research Quarterly 48 (2): 291-312.
Harper, Marcus. 2000. “Economic Voting in Postcommunist Eastern Europe.” Comparative
Political Studies 33 (9): 1191-1227.
Hetherington, Marc. 1996. “The Media's Role in Forming Voter's National Economic
Evaluations in 1992.” American Journal of Political Science 40 (2): 372-395.
Holbrook, Thomas and James Garand. 1996. “Homo Ecomomus? Economic Information and
Economic Voting.” Political Research Quarterly 49 (2): 351-375.
Host, Viggo and Martin Paldam. 1990. “An International Element in the Vote?” European
Journal of Political Research 18: 221-239.
Kiewiet, D. Roderik. 1983. Macroeconomics & Micro Politics: The Electoral Effect of
Economic Issues. Chicago: The University of Chicago Press.
Kiewiet, D. Roderik and Douglas Rivers. 1985. "A Retrospective on Retrospective Voting," in
Economic Conditions and Electoral Outcomes. Michael S. Lewis-Beck and Heinz Eulau,
eds. New York: Agathon Press, Inc.: 207-229.
King, Gary, Michael Tomz and Jason Wittenberg. 2000. “Making the Most of Statistical
Analyses: Improving Interpretation and Presentation.” American Journal of Political
Science 44 (2): 347-361.
Kitschelt, Herbert. 1992. “The Formation of Party Systems in East Central Europe.” Politics and
Society 20 (1): 7-50.
Lewis-Beck, Michael S. and Paolo Bellucci. 1982. “Economic Influences on legislative elections
in Multi-party Systems: France and Italy.” Political Behavior 4 (1): 93-107.
MacKuen, Michael B., Robert S. Erikson and James A. Stimson. 1992. “Peasants or Bankers?
The American Electorate and the U.S. Economy.” American Political Science Review 86
(3): 507-611.
Manin, Bernard, Adam Przeworski and Susan Stokes. 1999a. "Elections and Representation," in
Democracy, Accountability, and Representation. Adam Przeworski, Susan Stokes and
Bernard Manin, eds. Cambridge, England: Cambridge University Press: 29-54.
Manin, Bernard, Adam Przeworski and Susan Stokes. 1999b. "Introduction," in Democracy,
Accountability, and Representation. Adam Przeworski, Susan Stokes and Bernard Manin,
eds. Cambridge, England: Cambridge University Press: 1-26.
Mason, David S. and Svetlana Sidorenko-Stephenson. 1997. “Public Opinion and the 1996
Elections in Russia: Nostalgic and Statist, Yet Pro-Market and Pro-Yeltsin.” Slavic
Review 56 (4): 698-717.
McFaul, Michael. 1996. “Russia's 1996 Presidential Elections.” Post Soviet Affairs 12 (4): 318-
350.
Myagkov, Mikhail, Peter Ordeshook and Alexander Sobyanin. 1997. “The Russian Electorate,
1991-1996.” Post Soviet Affairs 13 (2): 134-66.
Pacek, Alexander. 1994. “Macroeconomic Conditions and Electoral Politics in East Central
Europe.” American Journal of Political Science 38 (3): 723-44.
Paldam, Martin and Peter Nannestad. 2000. “What do Voters Know about the Economy? A
Study of Danish Data, 1990-93.” Electoral Studies 19 (2/3): 363-391.
Powell, 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
37 (2): 391-414.
Powers, Denise V. and James H. Cox. 1997. “Echoes from the Past: The Relationship between
Satisfaction with Economic Reforms and Voting Behavior in Poland.” American Political
Science Review 91 (3): 617-633.
Przeworski, Adam, Susan Carol Stokes and Bernard Manin. 1999. Democracy, Accountability,
and Representation. Cambridge, U.K. ; New York: Cambridge University Press.
Sanders, David. 2000. “The Real Economy and the Perceived Economy in Popularity Functions:
How Much Do Voters Need to Know? A Study of British data, 1974-97.” Electoral
Studies 19: 275-294.
Szczerbiak, Aleks. 1999. “Interests and Values: Polish Parties and their Electorates.” EuropeAsia
Studies 51 (8): 1401-1432.
Tomz, Michael, Joshua A. Tucker and Jason Wittenberg. 2002. “An Easy and Accurate
Regression Model for Multiparty Electoral Data.” Political Analysis 10 (1): 66-83.
Tomz, Michael, Jason Wittenberg and Gary King. 2000. CLARIFY: Software for Interpreting
and Presenting Statistical Results. Version 2.1. Cambridge, Massachusetts,
http://gking.harvard.edu/.
Tucker, Joshua. 1999a. "It's the Economy, Comrade! A Model of the Effects of Economic
Conditions on Election Results in Transition Countries." Paper Presented at the 1999
Annual Meeting of the American Political Science Association, Atlanta, Georgia.
Tucker, Joshua A. 1999b. "Reconsidering Economic Voting: Party Type vs. Incumbency in
Transition Countries." Paper Presented at the 1999 Annual Meeting of the American
Political Science Association, Atlanta, Georgia.
Tucker, Joshua A. 2000. It's the Economy, Comrade! Economic Conditions and Election Results
in Post-Communist Russia, Poland, Hungary, Slovakia, and the Czech Republic from
1990-96. Harvard University, Cambridge, Massachusetts. (Doctoral Dissertation).
Tucker, Joshua A. 2001. “Economic Conditions and the Vote for Incumbent Parties in Russia,
Poland, Hungary, Slovakia, and the Czech Republic from 1990-1996.” Post-Soviet
Affairs 17 (4): 309-331.
Tucker, Joshua A. 2004. Transitional Economic Voting: Economic Conditions and Election
Results in Russia, Poland, Hungary, Slovakia, and the Czech Republic from 1990-1999
(book manuscript in progress). Princeton, New Jersey.
White, Stephen, Richard Rose and Ian McAllister. 1997. How Russia Votes. Chatham, New
Jersey: Chatham House Publishers, Inc.