concurrently with European Parliament elections, or countries where the clarity of ..... Northern Irish parties (along with respondents from Northern Ireland, and.
The Economy and the Vote: Electoral Responses to Economic Conditions in 15 Countries
by
Wouter van der Brug, Cees van der Eijk (University of Amsterdam) and Mark Franklin (Trinity College, Hartford CT, and University of Amsterdam)
Copyright © 2000 by Mark N. Franklin, Cees van der Eijk and Wouter van der Brug
Paper prepared for the annual meeting of the American Political Science Association, Washington D.C., AugustSeptember 2000 0
ABSTRACT Past research has strongly suggested that there is a relationship between economic conditions and election outcomes, but attempts to find individual-level concomitants of these aggregate-level effects have yielded inconclusive findings. This research is the first to look for effects of variations in real economic indicators (growth in GDP, changes in inflation and unemployment) on individual-level party preference across both time and space. We investigate party preferences of voters in 15 countries that are now members of the European Union over a fifteen year period using survey data collected in the aftermath of the European Parliament elections of 1989, 1994 and 1999. These surveys relate to 42 separate electoral contests, a number large enough to provide considerable variation in economic conditions. We do find significant effects of these variations on individual party preferences, even when controlling for past vote and other individual-level independent variables. These effects vary in understandable ways as we remove countries where national elections were held concurrently with European Parliament elections, or countries where the clarity of government responsibility for policy was expected to be low. Economic effects are stronger than any except for issues or distance from parties in left/ right terms. However, it is clear that the economy is only one of the factors that decide elections. Depending on its salience in a given country at a specific time, it could decide the outcome of an election or it could prove less important than other, temporarily more salient, voter concerns.
1
Introduction Past research has strongly suggested that there is a relationship between economic conditions and election outcomes (e.g., Tufte, 1976; Christal and Alt, 1981; Hibbs, 1987; Fair, 1988; Lewis-Beck, 1988; Erikson, 1989; MacKuen, Erikson, and Stimson, 1992). However, efforts to find individuallevel concomitants of these aggregate level findings have yielded findings that are by no means conclusive (Fiorina, 1978; Kinder and Kiewiet, 1979; 1981; Lewis-Beck, 1988). Subjective views of economic conditions have been shown to affect voting choice, but these views have themselves been found to be heavily contaminated -- by voting choice and by economic expectations (Norpoth, 1996; Wlezien, Franklin and Twiggs, 1997; Nannestad and Paldam, 2000). The effects of objective economic conditions (variables like unemployment, inflation and economic growth) have never been unequivocally demonstrated at the individual level.
Of course, as has more than once been pointed out, there need be no individual-level relationship to correspond with aggregate-level findings (Kramer, 1983; Wlezien, Franklin and Twiggs, 1997). Government popularity, and changing electoral fortunes that result from changes in government popularity, could correlate with economic conditions because of common antecedent factors, or because of any number of other aggregation artifacts.
Nevertheless, so long as we find no definitive individual-level concomitants of aggregate-level findings the possibility remains that the aggregate-level findings are themselves the result of happenstance correlations that appear important only because of the small number of cases available for analysis at the aggregate level. The suspicion that previous evidence of economic voting might be spurious was fuelled when the apparent failure of British voters to unseat an incumbent government during bad economic times in 1992 was followed by their perverse insistence on replacing the same
2
government in a landslide defeat when economic times were excellent five years later.1 This dagger at the heart of the theory of economic voting needs to be accounted for.
The purpose of this paper is to establish the importance of objective economic conditions for individual-level voting decisions. We conduct the analysis across fifteen countries over three successive elections, in the process acquiring a large enough variation in economic conditions to make it feasible to measure their effects; but we employ individual-level data -- more than 36,000 individuals whose party preferences are shown to be affected by these varying conditions. Because we investigate the effects of objective economic conditions, we are not plagued by any problems of contamination between people's party preferences and their views of the economy. Because of the large number of different political systems included in the analysis, we are able to take account of systemic differences that impinge on the extent to which individual voters have mutually exclusive preferences for parties, as opposed to preferences that overlap, as we shall see. We can also see whether differences between systems in terms of clarity of government responsibility affects the extent to which governments are in fact held responsible for economic conditions, as has previously been suggested (Powell and Whitten, 1993).
Venue We have chosen as our research venue elections to the European Parliament conducted in all countries of the European Union in 1989, 1994 and 1999. This gives us a total of 42 economic contexts within which to evaluate
individual-level political choices. In 1989 and 1994 we have twelve
member states among which we distinguish 13 political systems because of the difference between Flanders and Wallonia. In 1999 we have fifteen member states in which we distinguish 16 political systems because of the same subdivision of Belgium (whose two different party systems yield different dependent variables).2 These 42 contexts provide considerable variation of economic 1
Similar instances that run counter to conventional wisdom are commonplace. Marsh and Mitchell (1999: xv), when discussing the Irish election of 1997, wonder “how a smoothly functioning and apparently popular government presiding over an economic boom could fail to be reelected” 2 Respondents from Northern Ireland, as well as respondents from East Germany in 1994 (so soon after German reunification) were excluded from our analyses, because normal patterns of credit and blame are not expected to operate in the same way in these contexts. 3
4
conditions (see Table 1). Unemployment ranges from 2% to 24%, inflation ranges from 1% to 14% and growth in GDP from 1% to 8.6%. We also get considerable variation in terms of which political parties are to be credited or blamed, ranging from governments that include a communist party to governments that include a far right party. Moreover, the 1989-99 time frame is one within which economic conditions vary considerably even for single countries, and government composition changes for many of the countries. This extent of contextual variation is unprecedented in academic studies of the effects of economic conditions on individual choices. Within our 42 contexts we investigate party preferences for a total of 36,002 individuals, generally about 1,000 individuals per context though with smaller samples for Luxembourg and the two Belgian subsystems.
It might be objected that European elections constitute an inappropriate venue for studying the electoral impact of economic conditions. After all, these are not 'real' elections in which the allocation of government power is at stake (van der Eijk and Franklin, 1996). It has been argued that such secondorder elections are merely beauty contests in which people 'vote with the heart' for parties that they would not necessarily support had government power been truly at stake (Reif and Schmitt, 1980; Reif, 1984). Our own previous work on these electoral contexts shows their character to be more complex (van der Eijk, Franklin and Marsh, 1996). Nevertheless, precisely because they are not about holding a European government accountable, their outcomes do appear to be structured by national considerations and concerns. Though the future of a national government is not at stake in most countries, still the vote is interpretable in terms of government popularity -- much like the monthly polls that in most countries that ask about the standing of the government -- and such data are often employed to test the idea that government popularity responds to fluctuations in the real economy. Moreover, to the extent that these elections are not about holding government parties accountable for what happened on their watch, we would expect to find less powerful effects of economic conditions. So any test, using these data, of the conventional wisdom that "it's the economy, stupid" should err on the side of conservatism. If we found such effects in second-order elections, we should be confident of also finding them (if we had appropriate data) in first-order elections. Unfortunately appropriate data at the individual level in first-order elections are scarce. 5
As it happens, in seven of the contexts that we study, European elections occurred coterminously with national elections.3 This enables us to check whether economic effects on party preferences in European Parliament elections differ from what would found in national elections.
Theory How do we expect the party preferences of voters to respond to economic conditions? In the first place, on the basis of literature cited above, we expect government parties to be held accountable for good or bad economic times; whereas we do not necessarily expect an equivalent impact on preferences for parties that do not hold (or share in) government power.4 Moreover, we do not necessarily expect all government parties to be held equally accountable. Parties that do not need to share government responsibility (single-party governments), as in Britain during our period, evidently take all credit and all blame for what happens on their watch. Parties that share government power in a governing coalition could be expected also to share in the credit or blame. We might have found that they are all equally held responsible as if they were the only government party, or it might turn out that the amount of credit and blame assigned to parties is distributed among them in proportion to their strength in the coalition (the share of seats that they contribute).5 In what follows we allow for both possibilities.
We look at three aspects of economic conditions: unemployment, inflation, and growth of national product. The first two of these have sometimes been combined into a so-called 'misery index' but we consider them separately in this research.6 The conventional wisdom would lead us to expect negative effects of unemployment and inflation on preferences for government parties, and positive 3
Greece, Ireland and Luxembourg in 1989; Luxembourg in 1994; Belgium (2 contexts) and Luxembourg in 1999. The extant literature has not investigated the effect of economic conditions on parties that are not part of the government. Logically, effects on such parties could be nil or such effects could be opposite in sign to those for governing parties. If we were restricted to data that measured government popularity by votes for governing parties, it would not be possible to distinguish between these two eventualities. In our analyses we allow for either possibility. 5 This is to all intents and purposes the same as the share of ministerial portfolios allocated to each party (Browne and Franklin, 1973). 6 The reason these two independent variables are sometimes combined is because the small number of contexts generally available for aggregate-level analyses necessitates parsimony in the number of independent variables. With 42 contexts we do not have as strong a need to combine variables that might be found to have separable effects. Moreover, the effects of inflation and unemployment that we find are sufficiently different as to call into question the validity of any joint measure. 6 4
effects of economic growth. Particularly with regard to inflation, our own expectations for the period that we study are not so clear. No West European country has recently experienced crippling hyperinflation. Moreover, virtually all of them have systems in place that adjust earnings and pensions regularly to inflation rates. This eliminates many of the negative effects on individuals commonly associated with high levels of inflation. Finally, the increasing ubiquity of individual debt for mortgages and consumption purposes might lead people to evaluate moderate levels of inflation rather more favorably than is commonly assumed. Research on the effects of economic factors on government approval in West Germany (Falter, 198?) has also cast doubt on the conventional wisdom that inflation would hurt governing parties.
In order to test these various expectations, the common practice of using 'party voted for' or 'voted for incumbent party' as the dependent variable is inadequate. Doing so makes it impossible to assess whether or not economic conditions have any direct impact on preferences for non-government parties. It also would hamper the analysis of the effects of economic conditions on government parties, because there would be no straightforward way to control for various factors that impact on voters' choices, as will become clear below (for a more elaborate discussion of this reasoning, see van der Eijk and Franklin, 1996: 334-338). The dependent variable we use instead is the electoral utility that would accrue to an individual from voting for a specific party. We obtained these utilities from each respondent for each party in each political system at each election that we study. Past research has shown them to be related to actual party choice in a straightforward and almost deterministic manner: voters choose the party that yields them most utility. The theory of voting that underlies these analyses is a two-stage one. Voters choose parties to vote for on the basis of utility which, in turn, is formed on the basis of factors that are commonly employed in voting studies (sociodemographic background, issues, ideology, etc.). The theory has been shown to be empirically valid (van der Eijk and Franklin, 1996; Tillie, 1995; Oppenhuis, 1995; Burden, 1997),7 and transforms the objective of explaining party choice into one of explaining the utilities that voters get from voting for one party rather than another. As a side-benefit of measuring utility for each party we are able to
7
compare these utilities for each voter across the parties that compete for his or her vote, thus providing us with far more analytic power than is provided by measuring party preference with the conventional dependent variable, 'party voted for.' That variable does not permit us to distinguish between the utilities associated with parties that were not voted for.
Data and methods The data for this study are taken from the European Election Studies conducted immediately following elections to the European Parliament of 1989, 1994 and 1999 (Schmitt and Mannheimer, 1991; Marsh and Norris, 1996; van der Eijk, Franklin and Schmitt, 1999). These were cross-section surveys using random samples from the electorates of each of the member states of the European Union. Approximately 1,000 respondents were interviewed in each country.8 The dependent variable -- electoral utility for each party -- is derived from a set of questions, one for each party, that ask how likely it is (on a scale of 1 to 10) that the respondent 'would ever vote' for the party concerned.9 The independent variables of primary interest -- measures of unemployment, inflation
7
In the 1994 European elections, for example, the percentage of respondents in each country choosing to vote for the party which offered them the most utility was never less than 93% and averaged 96% (Schmitt and Thomassen 1999: 168). 8 These studies were designed and organized by Pilar del Castillo (Madrid) Roland Cayrol (Paris), Cees van der Eijk (Amsterdam), Mark Franklin (in 1989 at the University of Strathclyde, in 1994 at the University of Houston, and in 1999 at Trinity College, Hartford CT), Renato Mannheimer (Genova), Colette Ysmal (Paris) and Hermann Schmitt (Mannheim) who co-ordinated the efforts of the group. Additionally, Manfred Kuechler (New York) participated in the first of these research efforts, while Michael Marsh (Dublin) and Erik Oppenhuis (Amsterdam) joined the team while analysis of the 1989 findings was in progress. Wouter van der Brug joined the team during preparations for the 1999 study. The studies consisted of independent cross-sectional surveys fielded in each member country of the EC/EU immediately after the EP elections. The questionnaires, administered in the language of each country, in 1989 and 1994 formed part of the European Omnibus Surveys which also contained the regular Eurobarometer (EB) surveys of the Commission of the EC. Each wave involved interviews with some 12,500 respondents divided into independent national samples of some 1,000 respondents each. This number was lower for Luxembourg (about 300) and higher for the United Kingdom where an additional sample of 300 was drawn from Northern Ireland. In 1994 another additional sample was drawn from the former East Germany. In 1999 the study was based on a stand-alone survey conducted by telephone interviewing. In this article we focus on data from the (post-election) wave of each study which, in 1989, was largely funded by a grant to Mark Franklin from the British Economic and Social Research Council, and in 1994 was largely funded by grants to Hermann Schmitt and Cees van der Eijk from the German and Dutch National Science Foundations. The 1999 study was funded largely by the Dutch NWO, the Spanish National Science Foundation, the Universities of Mannheim and Amsterdam, and Trinity College, Hartford. The 1989 and 1994 studies are available for secondary analysis from ICPSR at the University of Michigan, and from various European data archives. The data from the 1999 study will be deposited in 2001.The studies are extensively documented on the European Elections Studies Web Site (http://www2.trincoll.edu/~mfrankli/EES.html). This site not only describes the surveys, but also contains an extensive list of publications emanating from these studies. 9 Note that the question does not relate specifically to European Parliament Elections, but to elections of any type. 8
and economic growth -- are taken from Eurostat/OECD statistics (see note to Table 1) and imputed to respondents residing in the appropriate country at the appropriate time. The inflation and unemployment variables are deviation scores relating to the average, over the previous five years, for each country. The measure of economic growth (change in gross domestic product, percent) was used in its raw form.10 In order to allow for the possibility that economic conditions affect opposition parties as well as government parties, each of these variables is taken twice, interacted first with a dummy variable registering government party status and second with a dummy variable registering opposition party status.
Additional control variables were taken from the surveys in order to provide us with a properly specified model. Previous research (Oppenhuis, 1995; van der Eijk and Franklin, 1996) tells us that the controls we need to properly specify party preference include social class and religion, approval of European unification, closeness to each party in left-right terms in each country (left-right proximity), and closeness to each party in terms of issues salient at each time-point in each country (issue proximity). In addition, controls were added for party choice at the previous national election in each country,11 and two interactions: left-right proximity with perceptual agreement of party positions,12 and issue proximity with perceptual agreement of party positions (see footnote 12). The exact operationalization of these variables is somewhat complex for two reasons. One is the stacking of the data for analysis (see below); the other is the need to generate variables that are
10
All analyses presented in this paper were also conducted with different specifications of the economic indicators. For example, we tried absolute levels (in percentages) of unemployment and inflation as well as change in unemployment and inflation in the year(s) preceding the elections, and lagged effects. None of these alternative specifications would have led to different overall conclusions, although all of them yielded less definitive findings than those presented here. 11 This variable, which is not customary in cross-sectional research designs, is included to pick up any effect of stable but otherwise unspecified individual-level characteristics impinging on party preferences in each country (eg. education or post-materialism). Most aggregate analyses (whose findings we are trying to confirm at the individual level) employ a lagged version of the dependent variable as a predictor. Our procedure gives us maximum comparability with such studies. 12 Perceptual agreement of party positions is a measure of the extent to which respondents in each country agree as to the location of their political parties in left-right terms (van der Eijk, 1998). In earlier research we found that the effect of left-right proximity varies across political systems, and that this variation can be accounted for by the extent of perceptual agreement. The higher the perceptual agreement the stronger the effect of left-right proximity on party preferences (Oppenhuis, 1995). Conversely, the effect of proximity to parties in terms of issues declines as perceptual agreement about parties' left-right location increases. 9
comparable across countries. Both of these requirements, and their practical implications, have been discussed at length elsewhere (van der Eijk and Franklin, 1996; Schmitt and Thommassen, 1999; van der Brug, Tillie and Fennema, 2000). In practical terms, the result is a set of independent variables measured on the same 10-point scales as the dependent variable, making for easy interpretation of coefficients. One control used in previous research that could not be used here was a control for size of party. When the dependent variable is party utility, one important determinant of the utility of voting for that party is the likelihood that it will exercise government power. In practical terms this means that voters prefer large parties, other things being equal. We could not include this variable in the analyses reported here because of multicolinearity between it and government/ oppositeion party status. Evidently, larger parties are more likely to be included in the government, whereas the smallest parties are generally opposition parties.
In order to properly attribute credit and blame, we needed to distinguish (as already mentioned) government from opposition parties. In both cases, there was a possibility (see above) that credit or blame would be apportioned to parties proportionately to their size rather than equally. We tested this idea by defining two variables, government share and opposition share, that took on values of 0 for parties not in the government/opposition, or a value equal to the contribution of that party to total parliamentary support for that government or opposition. In practice it was discovered that government parties share equally (no matter their size) in voters' minds the credit or blame that comes from economic conditions. The same was true of opposition parties. Interactions with government or opposition share add nothing to variance explained. Consequently those variables play no further part in this story.
Since our dependent variable (party utility on a ten-point scale) is interval in character, we use ordinary least squares regression analysis.13 As we have this utility for each party, some data 13
An analysis of the same variable, subjected to a logistic transformation, yields results that are substantively the same as those described in the present paper. However, the transformation produced a drop in variance explained of 22 percent, indicating that the assumption of linear effects more closely approximates the characteristics of these data. For an extended discussion of the measurement characteristics of these variables, which reinforces this conclusion, see Tillie (1995). Even though countries are repeated at three elections, the samples drawn at each time-point are new ones, guaranteeing independence of most variables across time. Nevertheless, multicolinearity for each country at each time 10
manipulation is required in order to treat them simultaneously. This is accomplished by 'stacking' the data matrix for each country(Stimson, 1985, Tillie, 1995; Oppenhuis, 1995; van der Eijk and Franklin, 1996; van der Brug, Tillie and Fennema, 2000) so that party-respondent becomes the unit of analysis and there are as many cases for each respondent as there are parties for which utilities were measured (6.5 per country, on average). These stacked data matrices are pooled across countries and time, with the addition of relevant identifiers, yielding a grand total of 237,620 records. Finally, the data are weighted in two stages: first to reflect in each country the actual outcome of each election, and second to reduce the effective N for each country and time-point to an equal magnitude (857), in order to maximize variance in contextual variables.14 This number of cases in each of 42 contexts yields a weighted N of 36,002 (cf. van der Eijk and Franklin, 1996; van der Brug, Tillie and Fennema, 2000).15
Findings Table 2 shows findings for three models. The first contains indicators of the economic conditions that we investigate in this paper. The second contains a set of controls deriving from earlier research (see above). Finally, Model C takes both sets of variables together. Each model includes a control prior party choice in national elections (see footnote 10).
The signs and magnitudes of the effects of control variables are largely as found in previous research (van der Eijk and Franklin, 1996; van der Brug, Tillie and Fennema, 2000) and require little expli-
point (economic indicators are correlated 1.0 across the respondents for each country) suggests the need for robust standard errors, which will be provided in a later iteration of this paper. Note that such a change will not affect the strength of effects, merely the outcome of significance tests. 14 The actual number of parties for which utility-scores are measured, is 297. However, in 1999 utility-scores were obtained for two Danish parties - the JuniBevaegelsen and the Folkebevaegelsen mod EU - that only participate in European elections and do not attempt to win any seats in national elections. In our analyses of economic voting, we expect the electoral fortune of parties to depend upon the state of the national economy. It was therefore decided to exclude these two parties from the analysis. Northern Irish parties (along with respondents from Northern Ireland, and also from East Germany in 1994) had already been excluded (see footnote 2). 15 We equalize the N in each country in order to give equal weight to independent variables measured at the context (country-time) level. The final N is the same as the number of respondents in the original surveys (36,002), giving conventional interpretability to significance tests. Extensive diagnostic tests were conducted in order to ensure that no artifacts of stacking or pooling the data had been introduced. In particular, after all analyses were complete, residuals 11
Table 2 Three models explaining party preference MODEL A
MODEL B
Econ. Factors + Prev. Vote Independent variables Previous national vote Class Religion L/R distance Issues EC-approval L/R * perc.agreement Iss * perc.agreement Govt pty * GDP Govt pty * unemplmnt Govt pty * inflation Opp pty * GDP Opp pty * unemplment Opp pty * inflation Weighted N R2
B 1.022
SE 0.009
Beta 0.537 ***
0.236 -0.007 -0.087 -0.004 0.053 0.087 36,002 0.312
0.012 0.013 0.019 0.010 0.009 0.010
0.127 *** -0.003 -0.023 *** -0.003 0.032 *** 0.043 ***
Individual-level factors + Previous national Vote B SE Beta 0.730 0.010 0.384 *** 0.349 0.028 0.059 *** 0.287 0.027 0.051 *** -0.386 0.007 0.280 *** 0.425 0.020 0.102 *** 0.250 0.033 0.037 *** -0.189 0.067 0.010 *** -1.359 0.203 0.030 ***
36,002 0.388
MODEL C Individual-level factors + Prev. Vote + Econ. factors B SE Beta 0.744 0.010 0.391 *** 0.350 0.028 0.059 *** 0.294 0.027 0.052 *** -0.360 0.007 -0.268 *** 0.429 0.020 0.103 *** 0.257 0.032 0.038 *** -0.160 0.067 -0.011 *** -1.330 0.200 -0.031 *** 0.190 0.013 0.102 *** -0.020 0.014 -0.009 -0.080 0.020 -0.021 ** 0.003 0.010 0.002 0.052 0.009 0.031 *** 0.059 0.010 0.030 *** 36,002 0.405
________________________________________________________________________________________________ Note: Significant at * 0.05,
**
0.01; *** 0.01.
cation.16 The large coefficient for the effect of prior party choice indicates that this variable does stand in for important effects coterminous with our unit of analysis, which otherwise would have been missing from our prediction equation. As already explained, we included this variable as a 'catch-all' insurance policy, in case we had omitted independent variables of major importance (which it is clear that we did).
Turning to the economic variables, the first thing to say about their effects is that their impact is somewhat unexpected. While inflation both hurts government parties and boosts opposition parties,
were analyzed to assess whether there were significant interactions with country and/or time-point (cf van der Eijk and Franklin, 1996). No such significant interactions were found. 16 The larger effect (in terms of beta-coefficients) of some independent variables implies that these variables (left-right distance in particular) have more impact than the economy on individual party preferences. Some of the independent variables are quite static over time (eg. religion). Such variables are unlikely to be responsible for changing party fortunes, at least in the short run. Others are likely to be more volatile over time, yet to cancel out over any large number of respondents when their effects on party choice is considered (see below). Economic factors, on the other hand, are unusual in that they can change quite radically over short periods of time and have consistent effects across individuals. Thus changes in economic performance can generate non-trivial effects in the aggregate even though effects on each individual voter may be small. 12
unemployment and economic growth have effects that are asymmetric. Economic growth provides a boost for government parties but has no effect on opposition parties, while unemployment provides a (smaller) boost for opposition parties but has no effect on government parties. Because the effects of inflation are felt on both government and opposition parties, the two coefficients should be added when considering the effects of inflation on the probability that the government will be supported. All effects have signs that are consistent with theory, except for the (insignificant) effect of GDP growth on opposition parties in Model C. On the other hand, the effect of this variable on government parties is the strongest of any economic effect. Government parties gain from an expanding economy (which is, after all, the most direct manifestation in our data of what was meant by the Clinton spin doctors when they invented the phrase "it's the economy, stupid"). And this effect is a large one. It is exceeded in Table 2 only by the effects of left/right distance and of previous vote. A change in economic growth of 5 percent is enough to shift the average voter almost one point (5 x 0.19 from Model C) on a ten point scale in terms of their probability of supporting a government party (we will return to the question of whether this would cause an actual change in party choice). A similar 5 percent change in inflation together with the same change (in the same direction) in unemployment would, in combination, cause a change in party preference of equal magnitude, even in the absence of any variation in economic growth rate.17
The second thing to say about these findings is to emphasize that they do show effects on preferences for opposition parties that are logically consistent, if not actually expected. Though economic growth has no significant effect on opposition parties, inflation benefits them about as much as it hurts
17
government parties
while
unemployment
uniquely
benefits
opposition
parties.
It might be thought that the effects of objective economic conditions will have been attenuated in our model by the inclusion of so many control variables, many of which could themselves reflect responses to economic conditions. Most importantly, our issue variable is constructed from a battery of individual questions several of which relate directly to the economy. However, when we remove these issue variables from the equation, the effects of objective economic conditions are unchanged. This rather unexpected finding supports suggestions that perceptions of the economy are too contaminated by subjective circumstances to be useful, as already suggested in recent research (Norpoth, 1996; Wlezien, Franklin and Twiggs, 1997; Nannestad and Paldam, 2000). 13
Implications for party choice What implication might a one-point change in party utility have for party choice? The translation of utilities into party choice is straightforward. Voters choose the party that yields them the highest utility. Depending on the magnitude of the gap in utilities between different parties, a quite small change in utility for one party could change its position from most preferred to second most preferred, or the other way around. Whether or not such changes in rank at the top of the preference scale occur frequently is an empirical question. In some systems a strong preference for one party virtually rules out also having a strong preference for a different party. In such systems even quite large shifts in economic conditions will change few votes. On the other hand, there are systems in which large numbers of voters hold two or more parties in very high regard. In such systems small changes in economic conditions may translate into sizeable shifts of votes between parties. Table 4 Effects on party choice of an 0.5 point shift in utility for their most preferred party, in 1999 Column A (range 1) % of electorate with 2 or more parties within 1 point of highest utility score
Column B (range 0)
B + 1/2 A % of electorate with % of electorate who would change 2 or more parties at their party choice in response to highest utility score a 1/2 point shift in party utility
Country Austria 20.2 18.1 B-Flanders 21.7 19.0 B-Wallonia 19.8 19.2 Britain 22.0 17.8 Denmark 28.7 23.5 Finland 28.7 21.5 France 34.6 27.4 Germany 17.5 16.1 Greece 23.5 20.7 Ireland 29.9 23.8 Italy 31.4 24.2 Luxembourg 19.3 14.7 Netherlands 28.9 20.3 Portugal 16.0 14.6 Spain 14.4 13.4 Sweden 24.9 20.4 ----------- ------------------------- ------------------Source: European Election Study 1999
28.3 29.9 29.1 28.8 37.9 35.9 44.7 24.9 32.5 38.8 39.9 24.4 34.8 22.6 20.6 32.9 ------------------------------
Table 4 shows, country by country, the proportion of the electorate for which differences in utilities between the most preferred and second most preferred parties falls within various ranges in 14
1999.18 Of these, the range of zero (column B) is given to respondents who have two parties that are equally most preferred. Any effect on utilities, however small, would serve to break such a tie. Moreover, half of those with a difference in observed utilities of one point could change their vote as a consequence of a change in utilities of half a point.19 The final column of the table adds this proportion of column A to column B in order to show how many people might be expected to change their vote as a consequence of a change in GDP of 5%.
Everyone else (those not included in Table 4) has differences between these two utilities of two or more. Such respondents are unlikely to change their vote on the basis of changes in economic conditions, at least in terms of the range of difference in economic conditions observed in our countries during the time period under study (see Table 1). Note, however, that in every country at least 20 percent of voters fall in the right-most column of Table 4. In France, Italy and Ireland, the equivalent percentages are twice that great. Even more impressively, in all but three countries at least 15 percent of voters fall in Column B -- meaning that they would alter their vote as a result of any change at all (in the right direction) in the utility they gain from their most-preferred or second most-preferred party. Irrespective of how this would translate into gains or losses for particular parties in these countries, the numerical conditions for coalition formation would certainly be altered.20
Concurrent national elections As we mentioned in the introduction, there is some question as to whether European Elections provide adequate contexts within which to study the effects of the economy on the vote. Included 18
The extent to which parties tie for first place, or fall within one point of first place, does vary from election to election. However, these estimates are very similar to those made for 1989 in van der Eijk and Oppenhuis (1991). 19 This is due to the lack of resolution of a ten point scale, which can result in people who are, for example, only 0.1 points apart, being respectively rounded up or down to the nearest whole number, yielding an observed difference of 1.0. Assuming utility to be a continuous variable with roughly uniform distribution in the intervals between integer scale values, it follows that half of those falling between any two integer values have a real (latent) difference of no more than 0.5. 20 In principle it is possible to project potential gains and losses for each party, or for all parties in the government. However, this would be a major undertaking that has to be the subject of future research. Note that the even though we have said that effects of economic conditions are unusual in being the same on all voters, not all voters will necessarily perceive the same parties as deserving credit or blame. Table 4 thus cannot be translated directly into changes in party strength. It represents the maximum effect of a given shift in utilities on election outcomes if no canceling out occurred. 15
Table 5 Comparing findings including and not including concurrent national elections
Previous national vote Class Religion L/R distance Issues EU-approval L/R*perceptual agreement Issues * perceptual agreement Government party * GDP Governt party* unemployment Government party * inflation Opposition party * GDP Oppostn party * unemployment Opposition party * inflation Weighted N R2
Model C All countries (base) B Se beta 0.744 0.010 0.391 *** 0.350 0.028 0.059 *** 0.294 0.027 0.052 *** -0.363 0.007 -0.268 *** 0.429 0.020 0.103 *** 0.257 0.032 0.038 *** -0.155 0.067 -0.011 *** -1.329 0.200 -0.031 *** 0.190 0.013 0.102 *** -0.020 0.014 -0.009 -0.080 0.020 -0.021 ** 0.003 0.010 0.002 0.052 0.009 0.031 *** 0.059 0.010 0.030 *** 36,002 0.405
Model D (All except countries with concurrent national elections) B Se beta 0.741 0.011 0.386 *** 0.333 0.031 0.056 *** 0.287 0.029 0.052 *** -0.371 0.008 -0.273 *** 0.420 0.023 0.096 *** 0.265 0.037 0.038 *** -0.230 0.073 -0.016 *** -1.071 0.232 -0.024 *** 0.141 0.015 0.074 *** -0.070 0.015 -0.031 *** -0.128 0.022 -0.035 *** 0.002 0.012 0.001 0.058 0.010 0.037 *** 0.052 0.012 0.026 *** 36,002 0.397
_______________________________________________________________________________________ Note: Significant at * 0.05,
**
0.01; *** 0.01.
in our dataset, however, are seven contexts where national elections were held concurrently with European elections. In those seven contests the question that gives rise to party utilities was asked very shortly after respondents had in fact made their choice of party in a real national election. Since the question is not specific to European elections (see footnote 9), these seven contexts (Ireland, Greece and Luxembourg in 1989, Luxembourg in 1994. Flanders, Wallonia and Luxembourg in 1999) allow us to see whether the typical European election provides a context that is significantly different. If these seven contexts are removed from the analysis, R2 from Model C in Table 2 falls only 0.8 percent to 0.397. Table 5 repeats Model C from Table 2 and adds a new model (D) which does not contain the seven contexts concerned.21 Comparing the two models we do see differences (in particular the effect of economic growth on government parties is less and the effect of inflation
21
We cannot analyze separately the seven countries in which national elections were held concurrently with European elections because the small number of separate contexts truncates the variance in the economic indicators (which are only measured once for each context). This results in many possible values of these variables being unrepresented. So any estimates generated for just these seven contexts would be grossly unreliable. 16
on government parties is greater). But these differences would not lead us to substantively different conclusions. European elections are good surrogate venues for studying the effects of economic conditions, even when they are not held in conjunction with national elections.
Indeed, the findings from Model D are rather cleaner than the findings we get when countries holding concurrent national elections are included in the dataset (Model C): more of the findings are now significant at the 0.001 level, and the only non-significant finding has an even smaller coefficient (in the wrong direction) than it has in Model C. This difference reminds us that national elections are subject to all sorts of random influences (such as campaign errors and untimely scandals) that can affect the results in unpredictable ways. We have discovered before that European elections, being elections of lesser salience, are less subject to such influences (van der Eijk and Franklin 1996). Such elections might thus, paradoxically, be better venues for studying effects of the economy because they are less subject to the noise that would have caused findings from national elections to be less clear-cut.
Clear government responsibility Powell and Whitten (1993) have argued persuasively that certain types of government should show the effects of accountability more clearly than other types of government. In particular, single party governments in unitary systems have no way to pass the buck in the face of policy failures, and no difficulty taking credit for policy successes. In other systems accountability is less clear-cut. On this basis countries can be divided into two groups of clarity of government responsibility, as we have done for our 15 countries in Table 6, using Powell and Whitten's (1993) criteria.
Because we need the largest possible number of contexts in order to obtain reliable coefficients, in Model E of Table 7 we have eliminated the countries shown in Table 6 to have the least clarity of government responsibility (even though the number of contexts they represent is only marginally fewer), in order to be able to compare accountability in the remaining countries with the baseline presented by Model C (repeated in the left hand columns of Table 7). If the countries included in Model E do indeed see greater clarity of government responsibility, we would expect coefficients 17
Table 6 Number of electoral contexts for high clarity and low clarity countries Country High clarity Austria B-Flanders B-Wallonia Britain 3 Denmark Finland France 3 Germany Greece 3 Ireland 3 Italy Luxembourg 3 Netherlands Portugal 3 Spain 3 Sweden 1 ------------------ -------------------Total 22
Low Clarity 1 3 3 3 1 3
3 3
----------------20
having to do with issues and the economy to be significantly stronger in that model. With the number of cases at our disposal, we need a difference in betas of 0.15 or greater in order to be able to conclude that that difference is significant at the 0.05 level, and only three of the differences between the two models meet this criterion. Two of these are effects of the real economy (effects of economic growth on government parties and of unemployment on opposition parties are both significantly greater in countries with clearer government responsibility). Other effects of the economy are generally stronger, though not significantly so. Surprisingly, the effects of issues are not significantly stronger in these countries. The only other independent variable to show a significant difference in its effect is the interaction of left/right self-placement with perceptual agreement about the location of parties. This variable has significantly less effect in countries with clearer government responsibility, probably because these are generally countries with fewer political parties, but perhaps also because longstanding clarity of choices has had the result in these countries of causing parties themselves to clarify their policy stances. Obfuscation of party positions may well be less in such polities. Finally we should note that variance explained is higher in Model E, though not significantly so. 18
Table 7 Comparing baseline findings with effects for high responsibility countries
Previous national vote Class Religion L/R distance Issues EU-approval L/R* perceptual agreement Issues * perceptual agreement. Government party * GDP Governmt party* unemployment Government party * inflation Opposition party * GDP Opposition party * unemplmt Opposition party * inflation Weighted N R**2
MODEL C All countries (base) B Se beta 0.744 0.010 0.391 *** 0.350 0.028 0.059 *** 0.294 0.027 0.052 *** -0.363 0.007 -0.268 *** 0.429 0.020 0.103 *** 0.257 0.032 0.038 *** -0.155 0.067 -0.011 *** -1.329 0.200 -0.031 *** 0.190 0.013 0.102 *** -0.023 0.014 -0.009 * -0.080 0.020 -0.021 *** 0.003 0.010 0.002 0.052 0.009 0.031 *** 0.059 0.010 0.030 *** 36,002 0.405
MODEL E High clarity governments B Se beta 0.757 0.014 0.391 *** 0.313 0.041 0.050 *** 0.291 0.041 0.047 *** -0.363 0.010 -0.273 *** 0.500 0.032 0.113 *** 0.279 0.049 0.037 *** -0.072 0.088 -0.006 *** -1.462 0.316 -0.033 *** 0.242 0.016 0.144 *** -0.021 0.016 -0.010 -0.102 0.022 -0.034 *** 0.018 0.013 0.013 0.063 0.010 0.047 *** 0.055 0.012 0.034 *** 36,002 0.416
______________________________________________________________________________________ Note: Significant at * 0.05,
**
0.01; *** 0.01.
Discussion Our findings might surprise some readers, but only to the extent that they came to our work with expectations about responses to economic conditions that were derived from past literature that made simplistic assumptions (for example that unemployment must hurt government parties) that we did not have to make in order to conduct this study. We were ourselves somewhat surprised by the clear differences between effects on government and opposition parties. These differences might be happenstansical. Quirks in the data could perhaps determine which variables have their primary effects on which parties. Still, this is not likely given the structure of our data. The standing of opposition parties is in no way connected in these data to the standing of government parties. The two are not coded on the basis that credit and blame are logical concomitants. Each party was the subject of separate questions, without regard to the government or opposition status of that party. The standing of each party in the responses is not constrained by this status. So when we see clear 19
differences between the ways in which different variables have their effects we should begin by assuming that these differences are real, and seek a theoretical basis for the finding.
An interesting question for further research would be to relate this asymmetric pattern of effects of economic conditions for government and opposition parties to the ideological complexion of the respective governments in this period. Perhaps economic problems thought to be best dealt with by socialist parties will have benefited opposition parties in our findings, while economic gains thought likely to be the result of actions by conservative governments will have benefited government parties. It is certainly implausible to suppose that opposition parties get credit or blame for unemployment simply because they are in opposition.
In our analyses we have not taken into account differences in personal economic conditions: whether or not someone is unemployed or gets automatic income compensation for inflation, for example. On the other hand, the strongest effect of economic variables is that of economic growth which is quintessentially a sociotropic variable. The effects of economic growth are often not directly perceptible to most individuals, but are touted as good news on television and in the press. Voters evidently respond to the news of good times whether or not they themselves feel those good times. Inflation is a similar variable. It is generally touted as bad news, though its effects on individuals are not so clear, as we mentioned earlier. Its powerful effects (second only to those of economic growth when benefits for opposition parties are added to costs for government parties) suggest once again that what voters respond to is the news about inflation, rather than the effects of this variable on their lives. Unemployment levels are, of course, no less touted in the media, and in addition this is a variable whose effects are directly and unambiguously felt by certain individuals. Yet this variable has the least powerful effects of any of our economic variables, and its effects apply uniquely to opposition parties. This does not suggest that people are voting their pocketbooks.
Does the economy matter at the individual level? Our answer differs according to whether we consider party utilities or party choice. There is indeed an effect of economic conditions on preferences for parties, whether they be in government or opposition. This effect in our data is 20
essentially the same (to within sampling error) in all countries that we study once we have taken account of institutional differences that appear to make it easier to hold governments accountable in some systems than in others (cf. Powell and Whitten, 1993). When we consider effects on party choice (Table 4) we also find that economic conditions matter. This effect is not the same across countries, however. Indeed, the extent to which voters are presented with more than one acceptable electoral option determines the extent to which small changes in preferences could result in different party choices. This difference between countries happens (not coincidentally) to accord with differences in the clarity of accountability of government parties, thus accounting for Powell and Whitten's findings (and for our own in this regard).
Though our findings hold true across all the countries that we study, we do need to be cautious in generalizing from these findings to economic or political circumstances much beyond the range of those we study. Our universe consists of developed European economies during the 1990s and late 1980s. In spite of all their differences, these countries experienced in this period real economic growth, relatively mild inflation, and unemployment which (while of political concern in some countries) was nevertheless low in global perspective and accompanied by effective social safety nets. As it happens, in the contexts that we study, our measures relating to growth in GDP and inflation are correlated (r=0.2). This is a historical circumstance that is certainly not logically necessary, so we are unable to say what effects economic conditions would have in the absence of this circumstance. Moreover, as we have already pointed out, this was a period during which mainly right-wing governments were in power. Variations in any of these circumstances could result in findings that were different in detail from those we have reported.
Nevertheless, the effects we find of the economy on individual party preferences vindicates past studies which themselves yielded findings that were not entirely consistent (Tufte, 1976; Christal and Alt, 1981; Hibbs, 1987; Fair, 1988; Lewis-Beck, 1988; Erikson, 1989; MacKuen, Erikson, and Stimson, 1992). Our own findings demonstrate not only that fluctuations in the real economy give rise to concomitant effects at the individual level, but that this happens by way of a process that does not impinge directly on the vote but only indirectly, by way of effects on the utility that 21
would accrue to people who vote for one party rather than another. So the variety of effects of the economy found in past studies are probably due in part to the fact that past studies did not distinguish between countries which actually differ from each other in the way in which political preferences are translated into votes. Such differences, we suggest, are due to variations in the extent to which voters are presented with choices that they see as shading into one-another by small gradations, rather than with choices that are seen as being as quite stark. Differences in economic effects on party choice do not appear to be due to differences in the way in which voters assess economic performance or to differences in the way in which they attribute credit and blame for that performance.
Conclusions The economy matters. But the economy is not the only thing that matters. Other issues are just as important, and the distance between parties and individuals in left/right terms is much more important. Taking account of these other effects greatly reduces the effects of the economy (compare Models A and C in Table 2) and, at the same time, indicates that there can be elections in which the economy would not dominate the outcome (as it clearly did not in Britain in 1992 or 1997). This also suggests that, in those countries where the economy has historically dominated electoral outcomes (as it apparently did in the United States for many years), this is still a contingent situation: contingent upon the economy being the most salient issue (which, in the 2000 U.S. Presidential election, it apparently is not). Our findings thus anticipate the hand-wringing that seems likely to occur after these U.S. elections (even if Gore wins) among political scientists who have hung their hats on the notion that the economy would always decide election outcomes there.
Not only are economic effects in general contingent on circumstances, but so also are the primacy of particular economic circumstances. In the countries that we have studied over the time-period for which we have data, the primary economic effects have been those of economic growth. We hardly believe that the same would have been true in an era of high inflation. Which economic variable is paramount, just as much as whether the economy is paramount, must depend on which social and economic problems are most salient for a given country at a given point in time. Even more impor22
tant is the question of whether small shifts in party preferences readily give rise to changes in party choice. The economy can only easily decide election outcomes in countries where this is true. In countries where voters perceive parties as starkly different, other factors than economic performance are much more likely to dominate.
Our findings not only point to the contingent nature of economic effects on election outcomes but also to unexpected subtleties in the ways in which these effects are felt. It is not just that governments are held accountable by voters, their popularity rising and falling in step with economic performance. Opposition parties also gain and lose from fluctuations in economic conditions, and one of these conditions (unemployment) has effects (in these countries over this time period) only on the standing of opposition parties, even while economic growth affects only the standing of government parties. Inflation appears to have symmetrical effects, hurting the government while it benefits opposition parties, and vice versa. These differences may be contingent upon which particular parties are in government and opposition, socialists being thought best able to deal with unemployment while more right-wing parties being thought best able to bring economic growth.
A challenge to past theorizing is provided by the strong implication of our findings that effects of the economy are sociotropic rather than pocket-book in nature. The idea that voters are myopic maximizers of personal advantage comes to us from rational choice theories that do not give much space to the idea that men and women are social animals who identify with the community in which they live and take pleasure from that community's successes even while lamenting its failures. Economic growth has become the lodestone of the new millennium. Its effects are felt by the society in general. The fact that individuals will vote for parties that appear to bring economic growth to their countries while shunning parties that fail to deliver these goods suggests that the apparent importance of pocket-book concerns in surveys that ask such questions is due to contamination of the responses by economic expectations and by reciprocal effects of party choice (Norpoth, 1996; Wlezien, Franklin and Twiggs, 1997, Nannestad and Paldam, 2000). On the basis of our findings we would assert that people are not choosing parties that will bring them personally the greatest economic benefits. People are not making their choice on that basis. When asked whether they think 23
their personal economic situation will improve under their preferred party, the easiest answer is 'yes,' but this does not mean that voters actually choose the party on that basis (or believe themselves able to judge which party will really have that effect). Instead, the effects of the economy appear to be felt in very much the same way as the effects of other variables: on the basis of how voters identify themselves and how they respond to widely shared triggers that can be stimulated by media reports.
Economic performance has effects that matter for voting choice. But, when measured at the individual level, these effects are subtle ones that are highly contingent on place and circumstance. Investigating how these effects differ over a wider range of countries and a longer time period should be high on the agenda for future research. Such investigations will require survey data -- preferably survey data acquired in the aftermath of national elections worldwide. The data now becoming available through the auspices of the Comparative Study of Electoral Systems shows much promise of permitting researchers to build on our findings.
24
Bibliography K. Alec Chrystal and James Alt. 1981. "Some Problems in Formulating and Testing a PoliticoEconomic Model of the United Kingdom" The Economic Journal 91: 730-736. Browne, Eric and Mark Franklin. 1973. "Aspects of Coalition Payoffs in Parliamentary Democracies" American Political Science Review. 67: 453-469. Brug, Wouter van der Tillie and Fennema, 2000. "Anti-Immigrant Parties in Europe: Ideological or Protest Vote?" European Journal of Political Research. In press. Burden, B.C. 1997. "Deterministic and Probabilistic Voting Models" American Journal of Political Science 41: 1150-69. Eijk, Cees van der. 1998. "Measuring Agreement in Ordered Rating Scales" in Meindert Fennema, Meindert, Cees van der Eijk and Huibert Schijf, eds. 1998. In Search of Structure: Essays in Social Science and Methodology. Amsterdem: Het Spinhuis. Eijk, Cees van der, Mark N. Franklin, et al. 1996. Choosing Europe? The European Electorate and National Politics in the Face of Union. Ann Arbor: University of Michigan Press. Eijk, Cees van der, Mark Franklin and Michael Marsh. 1966. "What Voters Teach Us About Europe-Wide Elections; What Europe-Wide Elections Teach Us About Voters" Electoral Studies 15: 149-166. Eijk, Cees van der, Mark N. Franklin, Hermann Schmitt, et al.1999. European Election Study 1999: Design, Implementation and Results. Cologne: Zentralarchiv fur Empirische Sozialforschung (in preparation). Erikson, Robert S. 1989. "Economic Conditions and the Presidential Vote" American Political Science Review 83: 567-573. Eurostat. (Continuous) Basic Statistics of the European Union. Brussels: Eurostat. Fair, Ray. 1988. "The Effect of Economic Events on the Vote for President: A 1984 Update." Political Behavior 10: 168-79. Fiorina, Morris P. 1978. "Economic Retrospective Voting in American National Elections: A Micro-Analysis" American Journal of Political Science 22: 426-443. Financial Times Survey, Country briefs, at http://www.ft.com/ftsurveys/country
25
Franklin, Mark N., European Elections Studies Web Site at http://www2.trincoll.edu/~mfrankli/ EES.htm Hibbs, Douglas A. 1987. The American Political Economy. Cambridge, MA: Harvard University Press. Kinder, Donald R., and D. Roderick Kiewiet. 1979 "Economic Discontent and Political Behavior: The Role of Personal Grievances and Collective Economic Judgments in Congressional Voting" American Journal of Political Science 23: 495-527. Kinder, Donald R., and D. Roderick Kiewiet. 1981. "Sociotropic Politics" British Journal of Political Science 11:129-161. Kramer, Gerald. 1983. "The Ecological Fallacy Revisited: Aggregate- versus Individual-Level Findings on Economics and Elections, and Sociotropic Voting" American Political Science Review 77: 92-111. Lewis-Beck, Michael S. 1988. Economics and Elections. Ann Arbor: University of Michigan Press. Nannestad, Peter and Martin Paldam. 2000. "Into Pandora's Box of Economic Evaluations: a Study of the Danish Macro VP-function, 1986-1997" Electoral Studies. 19: 123-140. 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 86:597611. Marsh, Michael and Paul Mitchell (eds.). 1999. How Ireland Voted 1997. Boulder (Co), Westview.Marsh, Michael and Pippa Norris. 1997. "Political Representation in the European Parliament" European Journal of Political Research 32: 153-164. Norpoth, Helmut. 1996. "The Economy" in Lawrence LeDuc, Richard G. Niemi and Pippa Norris, eds., Comparing Democracies: Elections and Voting in Global Perspective. Thousand Oaks CA: Sage. OECD, Main Economic Indicators, http://www.oecd.org/std Oppenhuis, Erik. 1995. Voting Behavior in Europe: A Comparative Analysis of Electoral Participation and Party Choice. Amsterdam: Het Spinhuis.
26
G. Bingham Powell, Jr. and B. Guy Whitten. 1993. "A Cross-National Analysis of Economic Voting: Taking Account of the Political Context" American Journal of Political Science, 37: 391-414. Reif Karlheinz and Hermann Schmitt. 1980. "Nine Second-Order National Elections: A Conceptual Framework for the Analysis of European Election Results" European Journal of Political Research 8: 3-44. Reif, Karlheinz, ed. (1984) European Elections 1979/81 and 1984. Conclusion and Perspectives from Empirical Research. Berlin: Quorum. Schmitt, Hermann and Renato Mannheimer. 1991. "About Voting and Non-Voting in the European Elections of June 1989". European Journal of Political Research 19: 31-54. Schmitt, Hermann and JacquesThommassen, eds. 1999. Political Representation and Legitimacy in the European Union. Oxford: Oxford University Press. Stimson, James. 1985. "Regression in Time and Space: A Statistical Essay" American Journal of Political Science" 29: 914-47. Stimson, James. 1992. Public Opinion in America: Moods, Cycles and Swings. Boulder CO: Westview. Tillie, Jean. 1995. Party Utility and Voting Behavior. Amsterdam: Het Spinhuis. Tufte, Edward. 1976. Political Control of the Economy. Princeton, N.J.: Princeton University Press. Wlezien, Christopher, Mark N. Franklin and Daniel Twiggs. 1997. "Economic Perceptions and Vote Choice: Disentangling the Endogeneity" Political Behavior 67:453-469.
27
