Resumé

Economic issue effects ("sociotropic" effects) are found to eclipse "pocketbook" effects on the vote for shifting Danish governments in cross-section surveys of the general elections of 1987, 1990 and 1994. Voter stands on the economic issue, Denmark's economic conditions, are in turn affected by left-right orientation and by images of the competing governments as managers of the economy and in the aggregate differ markedly from real economic growth.

Introduction

The interest in economic voting is a product of the 1980s and 19905, so far as mass surveys are concerned. With regard to European election studies the 1960s were dominated by an interest in social cleavages and in particular class voting, while in the United States, psychological group attachments and particularly party identification became the leading model for understanding the nature of the electoral choice. In the course of the 1970s attention shifted toward issue voting, as social or psychological group membership no longer seemed to provide an adequate framework for understanding the observed changes in elections and party systems.

Economic voting, which studies the impact of economic conditions on the vote (especially the vote for governments), is part of a third wave in survey analysis. First observed in aggregate data analysis, it has encountered some difficulties in translating its findings to the level of the individual voter and in relating to the frameworks established by the previous waves of research. The lessons from earlier experience with micro-level voting data tend to be forgotten in the enthusiasm over new findings. For example, economic voting has had to rediscover the classical problems of macro-level versus micro-level data and of cross-section studies versus overtime studies. Its proponents also had to abandon, or at least relax, those rational choice assumptions that prevented them from recognizing that economic factors,

---

*Ole Borre, Department of Political Science, University of Aarhus, Universitetsparken, DK-8000 Aarhus C, Denmark.

like other factors affecting behavior, must operate through the mind of the
voter and therefore unavoidably become deflected by the voter's preconceived
political beliefs.

Thus, attempts to trace the macro-level relationship between economic conditions and election outcomes to the micro level remained unconvincing until Kinder & Kiewiet (1979) showed that controlling for party identification, people reacted not to changes in the content of their pocketbooks but rather to perceived national economic conditions and government competence. It is perhaps unfortunate that Kinder & Kiewiet christened their effect "sociotropic," as they were not studying the impact of different national economic conditions but different judgments about the same conditions. Cross-section surveys such as those used by Kinder & Kiewiet can only observe variations in judgments about economic conditions and not changes in these conditions, as pointed out by Kramer (1983). In order to isolate the effects of "real" change in economic conditions from variations in judgment, measurement error, and personal and political factors influencing individual incomes, either panel studies or pooled cross-section surveys from different time points are needed. The aggregated income can then be observed to change over time and to cause changes in support for the government.

Such studies have been conducted by Marcus (1988) in the United States and by Nannestad & Paldam (1997) in Denmark. They apply a quite different conceptualization of sociotropic effects. If people estimate their own incomes correctly, the aggregated income computed from a survey will closely mirror the national income, which is a real sociotropic variable. The pocketbook effect now becomes the effect of changes in the individual income relative to the average income, whereas the sociotropic effect becomes the effect of changes in national income controlled for the individual's income. Obviously the pocketbook effect will dominate in such a design. One might even expect the sociotropic effect to sometimes become negative because people who do not participate in the general income rise may react against their government out of envy, or "relative deprivation."

It is clear that sociotropic voting means two different things in these two approaches and should be called by different names. In the present study we concentrate on the effect observed by Kinder & Kiewiet. Since sociotropic voting in this sense generally prevails over self-interested voting, as noted in Lewin's review (1991), we anticipate that in an election the state of the national economy may be a powerful issue, whereas the size of the pocketbook depends on so many non-political factors that few people will be able to see the government's hand in it. One may add that parallel findings from a broader set of relationships suggest that rational expectations are in general not very helpful in predicting micro-level political behavior. For example, David Sears and his colleagues showed that Whites' opposition to

school busing does not depend on having school-age children and living in neighborhoods where busing occurs or is planned (Sears, Hensler & Speer 1979). A common finding is that environmentalist attitudes are most widespread among the well-educated, although they tend to live and work in clean surroundings. In direct connection with personal economic conditions, Grofman & Muller (1973) showed that protest behavior occurs not only among those reporting a drop in household income (the deprived) but at least as much among those reporting an increase; Brody & Sniderman (1977) argued that an ethics of self-reliance kept American respondents from blaming crisis effect on their government; and Thomassen (1990) found in a panel study of Dutch, West German, and American voters that the 1979 oil crisis caused little deprivation and did not harm the support for their governments.

Kinder & Kiewiet's positive findings about the impact of national economic conditions and the government's economic competence, as perceived by the voter, were echoed in Lewis-Beck's study (1988) of five European countries (with France as a possible exception, cf. Lewis-Beck 1983). That study controlled for respondents' self-placement on a left-right scale, thereby making a determined effort to clean the economic effects of spurious effects due to partisan and ideological reference group influence. Thus the assessment of economic trends is brought into line with the accustomed model of issue voting, and the economic issue finds its natural place among other issues. And, of course, the finding of a positive effect is precisely what might be expected of an issue that plays a leading role in most elections. Prior to almost any election the government and opposition quarrel over how much the economy has improved or deteriorated, and whether the government should be blamed or might have done better. Objective economic indicators may certainly be expected to sway opinions one way or the other, but so may the declared goals of the government, memories of previous governmental performance, and statements about what the opposition would have done if they had been governing. In short, national economic conditions should be viewed as a political issue even if it does not feature the neatness of opposed policy positions that lies at the core of the issue voting model. We shall term these perceived national economic conditions, or retrospective evaluations, the economic issue for brevity.

The objective of this article is to asses the effect of the economic issue on the vote for the government, using the Danish electoral surveys of 1987, 1990 and 1994. The question looming in the background is whether economic voting is gradually replacing old-style positional issue voting, just as issue voting eclipsed social cleavage voting a decade earlier. This is the perspective raised by the findings of Kinder & Kiewiet for USA, by Lewis for the major European countries, as well as by Aardal. Listhaug, Holmberg, Gilljam and others for Norway and Sweden. In the case of

Denmark, Nannestad & Paldam find that the sociotropic effect is much smaller than the pocketbook effect. However, their initial design (Nannestad & Paldam 1993; 1995) did not represent the economic issue in the form of the simple retrospective item on national economic conditions. Their later design (Nannestad & Paldam 1997) uses a pooled cross-section design with 28 time points covering the period 1986-92. From their findings Nannestad & Paldam argue that a vanishing sociotropic effect is what should be expected in a welfare society where the state is made responsible for the incomes of individual citizens.

As we noted above, pooled cross-section studies are superior if one is estimating what the Kramer school terms the "true" sociotropic effect, undistorted by perceptual noise or bias. However, the "bias" that is eliminated in this way is precisely the economic issue, the center of the present analysis. Far from eliminating the economic issue, we want to see to what extent it affects voter decisions separately, controlling for other variables along the lines of Kinder & Kiewiet's and Lewis-Beck's model. In Section 2 this model, which controls for pocketbook variation and partisan predispositions, is discussed. Section 3 offers details about the economic variables for the three elections. Section 4 then presents the regression analysis and hence our conclusion concerning the existence of economic issue effects in Danish voting behavior. Finally, in Section 5 we make an attempt to dissect positions on the economic issue into some of their constituent beliefs.

Analytic Model

In order to assess how the economic issue affects the vote, it seems logical to appoint the vote for government our dependent variable. Now, some importance attaches to whether our results are stable across governments and elections. This can be judged because we make use of three election surveys from the general elections of 1987, 1990 and 1994. The economic situation varied characteristically between these three elections, and moreover, the governments changed. Up to the 1987 election the government was the socalled bourgeois four-clover government consisting of Conservatives, Liberals, Center Democrats, and the Christian People's Party. Up to the 1990 election it consisted of Conservatives, Liberals, and Radical Liberals. And up to the 1994 election it consisted of Social Democrats, Radical Liberals, Center Democrats, and the Christian People's Party. We suggest that if the effects we find are fairly equal at the three elections, they must be considered more robust than if their magnitude depends on the election or the color of government.

Among the independent variables we include items on whether the
personal economy of the voter has improved or deteriorated, and whether the

voter believes the country's economy has improved or deteriorated. Furthermore, because of the obvious risk that these economic assessments are biased by social, partisan and ideological predispositions, we need to include control variables for these predispositions or, as we shall call them, identifications. Ideological self-placement is particularly important because we must realize that attitudes to issues usually are clustered in various dimensions. By including only one issue (in this case the economic issue), that issue will take over most of the effect of the whole dimension of which it is a component.

In line with this discussion, the model for our analysis looks as follows:

Here, Gov stands for a vote for one of the parties in government (coded 1) as opposed to voting for an opposition party (coded 0). Pers stands for the voter's personal economic condition, coded 1 if improved, 0 if unchanged, and —1 if deteriorated. Coll stands for the voter's perception of the country's economic condition, coded in the same way.

The main control variable is Ident, for which we are facing several options. Kinder & Kiewiet controlled for the direction of party identification, a design criticized by Kramer on the ground that party identification may itself be affected by economic conditions. There is some evidence from Fiorina's (1981) study of retrospective voting that party identification sums up the voter's previous impressions of Democratic and Republican governments. In the case of Denmark it is difficult to separate the direction of party identification from the vote, hence if the vote changes the direction of party identification will also change or be abandoned (Borre & Katz 1973). Controlling for party identification in that case may mean to over-control and in consequence reduce the effect of economic issue voting along with other short-term effects.

Controlling for left-right ideology seems to avoid the problem of reverse causality, as it seems odd to argue that voters shift toward the left or right because they think the country's economy is improving or deteriorating. Controlling for ideology is also attractive because it makes our results roughly compatible with the larger study by Lewis-Beck. However, we do not code left-right placement in a simple linear way. Danish governments are often center-left or center-right governments, which means that they may have their strongest support in the political center. Therefore, the effect of placing oneself to the left is not the opposite of that of placing oneself to the right. The two first governments in 1987 and 1990 were governments of the right but included small center parties while excluding the rightmost Progress Party. The 1994 government was left of center but excluded two left-wing parties.

At least, therefore, left and right identification must be coded separately, and we shall do it in the simplest possible way as two dummy variables. For this purpose the self-placement scale containing ten positions was divided into three intervals: left = I^4, center = 5-6, and right = 7-10. We therefore split the Ident variable into two dummy variables, Left indicating placement to the left (M on a ten point scale) and Right indicating placement to the right (7-10 on that scale). Hence both variables are coded 0 for voters in the center (positions 5 or 6). Finally we add a third variable, Class, since several studies and a preliminary analysis of our regression model showed that class identification is an important control variable. For 1990 and 1994 this variable will be scored +1 for those who identify with the working class, —1 for those who identify with the middle class, and 0 for those who do not identify with any class. For 1987, where class identification was not asked, it will be scored as occupational class +1 for workers and —1 for salaried employees or self-employed.

In controlling for class identification and ideological self-placement we also hope to control for positional issues to which the perceived economic situation may be related. For example, a voter who votes socialist because he or she wants welfare services to be expanded may be inclined to argue that the national economy has improved ("so that we can afford it"). In all likelihood, however, such people already sign up on the left or in the working class, so that we do not need to explore the relationship between the economic issue and other issues in detail.

If our regression equation is interpreted in the simplest way, as giving estimates by ordinary least squares (OLS) methods, the constant b₀ will be an estimate of the government's vote for respondents who are coded oon the independent variables. That is, respondents who perceive no change in their own or the country's economy have no particular class or ideological or partisan predisposition, and are neutral on other issues. The other b's will be estimates of how much the government's vote deviates from this neutral position among respondents who react positively or negatively to the economic items, who have some type of identification, or who are leftist or rightist on some other issue. This ease of interpretation may be thought to outweigh the problems connected with a dichotomous dependent variable such as our Gov, and we shall apply OLS method throughout. In so doing we are also better able to compare our results with those of Kinder & Kiewiet, Lewis-Beck, and the Swedish studies, all of which use OLS.

Economic Conditions and Support for Government

Before plunging into the regression analysis it is advisable to take a look at
the critical economic variables Pers and Coll. The coding and distribution of

these two variables are seen in Table 1. It is observed that in 1987 slightly more people believed their personal economy had improved over the past two or three years than believed their economy had deteriorated, 25 to 22 percent. However, a lot more respondents believed the national economy had deteriorated than believed it had improved, 40 percent against 25 percent.

In 1990 and 1994 a five-way item was used for both variables. It appears
that in 1990 opinions were balanced as to the personal economy; but the
predominant opinion was that the national economy had improved over the

past three years. 58 percent believed it had improved against only 17 percent who believed it had deteriorated. By 1994 the evaluation of especially the personal economy was positive: 26 percent believed it had improved, but only 14 percent believed it had deteriorated. In regard to the national economy the opinion was more even, as 29 percent believed it had improved against 25 percent who believed it had deteriorated.

It is legitimate to ask how people come to have different perceptions about what should, in theory, be the same phenomenon. It is clear that "the country's economy" or "Denmark's economic situation" is a multidimensional concept which voters can fill with employment, inflation, real income growth, trade balance, or other forms of content. Still, it is amazing how poorly the two variables are related at the aggregate level, considering that the interview questions about the personal and national economic conditions were phrased analogously. By multiplying the percentages in Table 1 by the codes indicated, we find that the mean Pers shifts only slightly from +0.03 in 1987 to +0.01 in 1990 and +0.08 in 1994, whereas the mean Coll shifts dramatically from -0.15 in 1987 to +0.225 in 1990 and +0.015 in 1994. If we consult the official statistics in order to decide which series is more correct, the mystery merely becomes deeper. With regard to September 1987, the pessimism expressed in the Coll value is understandable. During 1987 the growth in real GNP was almost zero, down from the level of 4 percent in the previous years (Økonomiministeriet 1997, 4). With regard to September 1990 it is more difficult to understand the optimism: the real growth was one-half percent in 1989 and rose merely to H percent in 1990 (ibid.). Finally, with regard to December 1994, the perceptions of most voters simply contradict the facts, as the economic growth was A\ percent in 1994, up from almost zero in 1992 (ibid.). This positive development appears to have been suppressed in the public discourse.

As an issue of major importance, the assessment of the economic trend and present situation is likely to have an ideological component. When the government is right of center, those who identify with the political right will be inclined to take a more favorable view of the economy than left-wing

voters. When the composition of the government shifts to the left, right-wing voters will suddenly discover economic problems that they did not notice earlier, whereas their opposite numbers on the left wing will tend to ignore or understate these problems. We can see how this operates in Table 2.

In 1987 the right wing exhibited an opinion balance of +30 in their assessment of the economy, compared to the left wing's -47. In 1990 almost the same ideological difference prevailed, the opinion balance soaring to +74 on the right wing and adjusting to a fairly neutral level of +7 on the left wing. Thus, at least everybody agreed that the situation had improved between 1987 and 1990. After the change of government from right-ofcenter to left-of-center in 1993, however, the right wing shifted to a negative view of the economy, an opinion balance of —19, whereas the left wing continued its upward course to an opinion balance of +28. What is especially remarkable in view of the real development we have just discussed is the

amount of misperception on the right wing. The perceptions on the left wing,
at least of the economic issue, is much more in line with the growth statistics.

Turning now to the behavioral effects, Table 3 indicates the government vote at each position of the economic variables. In 1987 the government received 40 percent of the vote among those whose personal economy had improved, and only 19 percent in the opposite group of respondents. The differences between these groups were somewhat smaller in 1990 and especially 1994: in 1990 the government received 40 percent among those who were somewhat better off and only 21 percent among those who were much worse off, in 1994 a different government received 49 percent among those who were somewhat better off and only 35 percent in the "somewhat worse" group.

Thus a crude relationship exists for the personal income variable. However the differences in the first column are dwarfed by those in the second column concerning the collective income variable. In 1987, 61 percent of those who perceived economic growth voted for the government, but only 18 percent of those who perceived a decline. In 1990 the government vote varied from a high of 58 percent among those who believed the economy had improved a lot to 6-8 percent among those with the opposite opinion; in 1994 the government vote - this time mainly Social Democratic voters - varied from 59 to 3 percent according to the perceived economic trend.

The fact that the vote varies much more by national economic perceptions than by personal economic conditions indicates a sizable issue effect, but of course it does not rule out an effect of the latter, i.e. a pocketbook effect. Such a pocketbook effect may be small but direct, that is, uncontaminated by other variables; or it may be indirect, as when respondents consult their own pocketbooks as one source of information when asked to assess the national economy. The latter possibility suggests itself by a sizable correlation between the two economic variables (r = 0.29 in 1987, 0.20 in 1990, and 0.26 in 1994). But before drawing conclusions about the order of the two variables, we should investigate how much of the variation in Table 3 is really due to partisan predispositions from before the period the respondents are asked to consider - that is, the past two or three years. Probably a lot, if we can believe their vote recalls. A study of the 1987 election found that of those who believed their own economic situation had improved, 48 percent had voted for the government or a supporting party already in the previous election three years earlier, whereas of those who believed their own economy to be worse, only 23 percent had done so (Borre 1989, 294). Thus it is very likely that even the content of one's pocketbook, or changes in that content, are judged so as to appear consistent with one's previous vote (or consistent with identifications affecting both vote and perceptions), and we must anticipate that controlling for this perceptual bias will reduce the economic effects in the regression analysis below.

Regression Analysis

The first set of regression equations, shown in the upper section of Table 4, contains only the two independent variables Pers and Coll, because we are interested in how much the subsequent control procedure matters. On the face of it, these uncontrolled effects suggest a large amount of economic issue voting, whereas "pocketbook" economic voting vanishes. So far these results are quite in line with those of Lewis-Beck (1990, 56), who found insignificant effects of personal economy and significantly positive effects of national economy in all five countries in his study. They differ, however, from those found in a cross-section study by Nannestad & Paldam (1995), but that study conceptualized "sociotropic" voting differently. As discussed in a critique by Hibbs (1993, 21) of an earlier version, Nannestad & Paldam measured the national economic, or sociotropic, factor by means of items

asking whether the respondent worried about certain social problems, rather than by the simple retrospective item on the country's economy. In a reply (Nannestad & Paldam 1994) the authors claimed that they had tried five different measures of sociotropic voting, including the simple retrospective one, without success. Therefore we can safely say that our findings in Table 4 are new in Danish research.

Concerning the first section of Table 4, our uncontrolled issue effects appear to be roughly similar to those found in the Swedish election studies of 1982 (Holmberg 1984), 1985 (Holmberg & Gilljam 1987), 1991 (Gilljam & Holmberg 1993) and 1994 (Gilljam & Holmberg 1995). From the 1982 election Holmberg concludes (p. 159) that the majority "followed their economic-political evaluations whether their pocketbooks had become lighter or heavier in later years" (my translation). Four elections later, and in the middle of a Swedish economic crisis, Gilljam & Holmberg conclude: "The voters' judgments about how their own personal economy improved show a quite negligible direct relationship with party choice and party change" (my translation). These quotations might as well have referred to the Danish findings in Table 4.

For Norway, Miller & Listhaug (1984) found an effect of personal economy, controlling for party identification, left-right ideology, and social class, in the 1981 election but not in the 1969 or 1977 elections. However, these surveys did not include the retrospective judgment about the country's economy; therefore it is possible that the pocketbook effect only emerges because the "sociotropic" issue effect is left out of the model. For the 1985 election the economic issue effect can be assessed from the fact that 65 percent of those who believed that the country's economy had improved voted bourgeois, but only 12 percent in the opposite group who thought the economy had changed to the worse (Aardal & Listhaug 1989, 148). The difference is roughly similar to those in the second column of Table 3 and thereby underscores the similarity between Norway and Denmark. In the case of the 1985 election, Aardal & Listhaug controlled for former vote and found both pocketbook effects and issue effects; that is, people whose own economic situation had deteriorated tended to shift away from the bourgeois government. However, as we shall see in a moment, it is not a good idea to control for former vote, neither in panel nor cross-section studies. Our conclusion would be that pocketbook voting may have emerged in Norway in the course of the 1980s, but from the studies of Miller, Aardal & Listhaug we cannot tell with certainty.

This article is not concerned with the direct effects of social class and ideology on the vote, only with the extent to which these control variables reduce the effects of the economic variables. Section 2 of the table shows what happens when we impose controls for social position. In 1994, evidently part of the effect of a country's economy is taken over by working

class identification, by far the strongest of the social identifications. Thus there seems to be a class bias in voter perceptions about the economic trend. This causes the variance explained to rise by an average of four percentage points in the three surveys. Further experiments, the results of which will not be shown, suggested that two other social variables, gender and educational level, had a significant effect on these perceptions, but not enough to visibly reduce the variance further.

Now, in modern "dealigned" electorates, controlling for social class or other social attributes is inefficient: when class no longer can be equated with partisan predisposition, this means that within each class, those disposed in favor of the government will be likely to claim that the economy has improved, whatever the real economic trend is. It is much more efficient to control for left-right identification, as done in the second section of Table 4. We immediately observe a vast change in the parameters of the model. The effect of a country's economy is cut to half its size and the variance explained more than triples. Controlling for ideology in this fashion, the effect of the country's economy drops to 0.10 in 1987, 0.12 in 1990, and 0.20 in 1994. Part of its effect is taken over by the ideological variables, hence there is a strong ideological bias in voter judgments about the state of the country's economy. Of the two ideological variables, left identification exerts a negative effect even on the 1994 government, which means that its support drops on the left side. Right identification exerts a positive effect on the 1987 and 1990 governments but a negative effect on the 1994 government. The Social Democratic government with three small center parties therefore entailed that it was seen as a center government and obtained its highest vote in the center. The constant term of 0.73 in 1994 means that the government vote is estimated to be 73 percent among center voters who identified with neither left nor right and who did not take sides on the other variables.

As Danish elections go, how important is the economic issue for their outcomes? The conclusion from Table 4 is that the "pure" economic issue - whether the national economy has improved or deteriorated during this government - exerts a considerable effect on the vote for that government. The average b2b₂ value of 0.14 in Table 4, section 3, means that those who believe the economy has improved a lot give the government on the average a 28 percent higher vote than those who believe the economy has deteriorated a lot. But fortunately for the stability of elections, these extreme optimists and pessimists are in minority. In practice, as we gathered from Table 1, the mean assessment varied rather modestly between +0.22 in 1990 and —0.15 in 1987. A difference of 0.37 in the Coll variable from one election to the next, when multiplied by 0.14, implies that the economic issue can be expected to produce an effect of about five percentage points in the vote cast for the government. Thus, a government may be hurt markedly by an

unfavorable image on the economic issue, but certainly other issues may be
stronger at any individual election.

We shall argue that these estimates of economic issue effects, obtained by controlling for class and ideology, are superior to those obtained by controlling for either party identification or respondent's recall of the former vote. Party identification is available for 1990 and 1994 but not for 1987. As noted, controlling for party identification runs a risk of overcontrolling. Table 5, first section, shows how this works. Among identifiers of the government parties almost everybody votes for the government; among identifiers of the opposition parties practically no one does so. The economic issue effect thereby becomes limited to the non-identifiers, where indeed it appears quite impressive: 55 percent of those who believed the economy had improved voted for the government compared to only 26 percent of those who believed the economy had deteriorated. The reason why we should suspect such a result of underestimating the issue effect is that it seems unlikely that the economic issue makes no impression whatsoever on party identifiers. Seeing the economy improving may well have caused some voters to acquire an identification with the government parties or to abandon an identification with an opposition party; and the reverse will be the case among those who think the economy has deteriorated.

If, instead, we control for respondents' recall of former vote, we face the usual problem that recalls are biased toward present partisan inclinations and therefore underestimate the change. But there is a much more serious objection to controlling for former vote: the propensity of switching away from a party will be higher in categories where the party is in minority than where it is the majority party. This is also the case in our data in section 2 of

Table 5. In the "better" group, 21 percent of the opposition voters switched to the government, whereas 11 percent of the government voters switched to the opposition. However, this apparent triumph for the government only entailed that it managed to keep its 58 percent of the voters in that group. In the "worse" group, only 5 percent of the former opposition voters switched to the government parties, whereas wholly 31 percent of the former government voters switched to the opposition. Again, this may seem like a strong issue effect superficially, but our enthusiasm is dampened on discovering that the net result in that group is a decline from 22 to 20 percent in support for the government. In short, controlling for former vote means slipping size effects into the regression model, confounding the issue effects one is looking for.

Dissecting the Economic Issue

So far, the simple issue of the economic trend has exhibited a consistent though limited effect on the vote for the government. But we have no clues as to the underlying reasoning that produces such an effect. Our data offer a few possibilities of digging deeper into the issue by probing the government's image as a competent master of the economy. Other studies, notably Lewis-Beck's, have found that directing the issue more clearly toward the government's economic performance or competence improves its effect. This is natural: many people may agree that their own economy or the country's economy has improved, but disagree that this is an effect of the government's policy; others may perceive that their own or the collective economy has deteriorated, but without blaming the government. In the case of the national economy we may attempt to remove these "fatalists" from our regression, expecting that in consequence the noise in the Coll variable should decline and the b2b₂ coefficient should rise. In the case of the personal economy there will be "fatalists" who think that powers stronger than the government rule their pocketbook as well as "individualists" who think that their own skills and diligence do so. We expect both types of voters to be at least partly eliminated when the respondents are cued to direct their attention toward the government's performance.

An even more pointed version of the economic issue occurs when respondents are asked to evaluate the relative competence of alternative governments. There are people who would blame the government for poor performance but who will not vote for the opposition because they believe the opposition would have performed even worse if it had governed. Consider the following three statements: (1) The economy is improving; (2) the government handles the economy well; and (3) the government handles the economy better than a different government would do. Logically, only (3) is

a rational argument for preferring the government to the opposition. To the extent that our respondents follow such a chain of reasoning, we should expect (1), the pure economic issue, to affect the government vote only because it provides empirical evidence for (2), the government's economic competence, and (3), performance; furthermore, the respondents' appreciation of (2), the competence of the present government, should affect the vote only to the extent that this moves the present government beyond the opposition (3).

Items pertaining to the government's competence were included in the 1994 survey and, on a smaller scale, also in the 1990 survey. Concerning statement (2), in 1990 the item read, "How do you view the bourgeois government's competence in solving the country's economic problems? In your opinion, does it do it very well, fairly well, fairly poorly or very poorly?" In the 1994 survey two items appeared: "How has the government's policy affected the economy of you and your family?" And "How has it affected the nation's economy?" Here, the responses allowed for five categories, from "very positively" to "very negatively."

Concerning statement (3), which compares the present government with an alternative government, the 1994 survey included two items, "Who do you think is best at solving the problems I shall now read to you, either the present government led by Social Democrats, or a bourgeois government? Who is best at solving the country's economic problems? . . . And which government do you think would give yourself the most money at your disposal?" In 1990 only the first item, about the country's economy, was asked.

In contrast to the simple economic item (Coll), these items direct the respondent's attention away from the past and toward the present and future. Perhaps more importantly, they focus on the choice between two alternative governments, mirroring our dependent variable. By cueing the respondents in this way we should not be surprised to find that these items outperform the simple retrospective item about the country's economic condition. Table 6 shows that this is indeed the result.

As we hypothesized, the effect of the simple retrospective item Coll on the government vote declines when the competence items enter the regression model. Compared with Table 4 its effect has been reduced from 0.12 to 0.03 in 1990 and from 0.20 to 0.06 in 1994. Next, the items on how the government has affected the economy have moderate effects in 1994 - b's are 0.06 for affecting the pocketbook and 0.08 for affecting the country's economy. The government competence item has a similar effect in 1990. However, the items comparing the two governments emerge as the strongest economic issue variables, with regression effects of 0.10 for pocketbook voting and 0.16 for economic issue voting.

These results are in line with our hypothesis that the simple retrospective item furnishes empirical evidence for present government competence, and that such competence in turn improves the government's image relative to that of the opposition. We have no evidence of how fast this process operates; but we may speculate that it takes more than a single election period to improve a government's image. In Table 1 we noted that the voters were surprisingly reluctant to trust the positive evidence about the country's economic situation in 1994. Further examination of the data reveals that the 1994 government, led by the Social Democrat Nyrup Rasmussen, was generally viewed as inferior to the preceding bourgeois government in "solving the country's economic problems generally": whereas only 22 percent thought the present government best, 50 percent thought a bourgeois government would do better. The new government fared better on the question of which government would give the respondents the most money at their disposal: here, 34 percent preferred the present government against 30 percent preferring a bourgeois government. Thus, one reason why the two items do not match may be that the public takes a longer view when evaluating the country's economic situation than when evaluating its own.

Conclusions

The economic issue, here defined as the assessment of the national economic situation, has been demonstrated to exert a considerable impact on the vote for Danish governments. The impact was found to be fairly constant over three parliamentary elections, held under different economic situations and called by different government coalitions. Hence the effect emerged for different values of the control variables; notably, controlling for ideological orientations was important because they strongly affect the respondent's position on the economic issue. In the average election those who believed the economy to be on the rise gave the government 14 percent more of the votes than those who saw no change in the country's economy, given their ideological orientation and class identification - while those who perceived the economy to be declining punished the government by the same rate.

Our findings for Denmark are in rough agreement with those obtained for other countries in previous studies which use the same analytical model, and which conclude that "sociotropic" effects tend to eclipse "pocketbook" voting. However, the term "sociotropic" has different meanings in the literature depending on the analytic design used; we therefore prefer to speak of economic issue voting. "Pocketbook" effects emerge as part of that issue only when survey respondents are cued by items that ask which government will give them the most money at their disposal. The most direct effects on the vote are obtained on items that ask the respondents to compare the government's and the opposition's competence in managing the national economy.

REFERENCES

Aardal, B. & Listhaug, O. 1989. "Økonomi og stemmegivning," in Aardal, B. and Valen, H.
eds., Weigere, partier og politisk avstand. Oslo: Statistisk Sentralbyrå.

Borre, O. 1989. "Belønnes regeringen?" in Elklit, J. and Tonsgaard, 0., eds., To
folketingsvalg. Aarhus: Politica.

Borre, O. & Katz, D. 1973. "Party Identification and Its Motivational Base in a Multiparty
System: A Study of the Danish General Election of 1971," Scandinavian Political Studies 8,
69-111.

Brody, R. A. & Sniderman, P. M. 1977. "From Life Space to Polling Place: the Relevance of
Personal Concerns for Voting Behavior," British Journal of Political Science 7, 337-360.

Easton, D. 1965. A Systems Analysis of Political Life. New York: Wiley and Sons.

Gilljam, M. & Holmberg, S. 1993. "Planboksröstning," in Väljarna inför 90-talet'.
Stockholm: Norstedt.

Gilljam, M. & Holmberg, S. 1995. "Ekonomisk kris," in Vdljarnas val. Stockholm: Norstedt.

Grofman, B. B. & Muller, E. N. 1973. "The Strange Case of Relative Gratification and
Potential for Political Violence: The V-Curve Hypothesis," American Political Science
Review 67, 514-539.

Hibbs, D. A., Jr. 1993. Solidarity or Egoism? Aarhus: Aarhus University Press.

Holmberg, S. 1984. "Ekonomisk kris och väljarbeteende," in Väljare i forandring. Stockholm:
LiberFörlag.

Holmberg, S. & Gilljam, M. 1987. "Ekonomi och politik," in Väljare och val i Sverige.
Stockholm: Bonniers.

Kinder, D. R. & Kiewiet, D. R. 1979. "Economic Discontent and Political Behavior: The Role
of Personal Grievances and Collective Economic Judgments in Congressional Voting,"
American Journal of Political Science 23: 3, 495-527.

Kramer, G. 1983. "The Ecological Fallacy Revisited: Aggregate versus Individual-Level
Findings on Economics and Elections," American Political Science Review 11, 92-111.

Lewin, L. 1991. Self-interest and Public Interest in Western Politics. Oxford: Oxford
University Press, series on Comparative European Politics.

Lewis-Beck, M. 1983. "Economics and the French Voter: A Microanalysis," Public Opinion
Quarterly 47, 347-360.

Lewis-Beck, M. S. 1988, 1990. Economics and Elections. The Major Western Democracies.
Ann Arbor, Mich.: University of Michigan.

Lipset, M. S. 1960. Political Man. London: Heinemann.

Marcus, G. 1988. "The Impact of Personal and National Economic Conditions on the
Presidential Vote: A Pooled Cross-Section Analysis," American Journal of Political Science
32, 137-154.

Miller, A. & Listhaug, O. 1984. "Economic Effects on the Vote in Norway," Political
Behavior A, 301-319.

Nannestad, P. & Paldam, M. 1993. "Sammenhængen mellem samfundsøkonomi, privatøkonomi
og tilslutningen til regeringspartierne i Danmark 1990-91, og nogle problemer med at
måle dem korrekt," Politico 25, 189-206.

Nannestad, P. & Paldam, M. 1994. "Review of Douglas A. Hibbs, Solidarity or Egoism?"
Scandinavian Political Studies 17, 83-88.

Nannestad, P. & Paldam, M. 1995. "It's the Government's Fault!" European Journal of
Political Research 28, 33-62.

Nannestad, P. & Paldam, M. 1997. "From the Pocketbook of Welfare Man: A Pooled Cross
Study of Economic Voting in Denmark, 1986-92," British Journal of Political
Science 27, 119-136.

Sears, D. 0., Hensier, C P. & Speer, L. K. 1979. "White's Oppositon to 'Busing': Self-interest
or Symbolic Politics?" American Political Science Review 73, 369-384.

Thomassen, J. A. 1990. "Economic Crisis, Dissatisfaction, and Protest." Chapter 4 in Jennins,
M. K. and van Deth, J. W., eds., Continuities in Political Action. Berlin and New York:
Walter de Gruyter.

Økonomiministeriet 1977. Udsigterne for Danmarks økonomi. Copenhagen: Schultz Information.