Resumé

This article deals with the relationship between leader approval and party support. Despite an unmistakable trend towards stronger "personalization" of politics, the relationship is seen as conditional, dependent upon the particular parties and leaders as well as the political system that surrounds them. The author tries to make explicit the daunting specification and measurement problems involved in this type of analysis. A simple version of the Granger causality test and simultaneous equation models based on Norwegian panel and cross-sectional data identify leaders who appear uncompromising and adamant as the most successful vote getters.

Introduction

In stark contrast to political journalism, West European electoral studies have downplayed the importance of leader charisma. The popularity of the leader is seen as indistinguishable from that of the party (Bean & Mughan 1989, 1165), or, alternatively, the party is held to overshadow the leader in the competition for voters' attention (Stewart & Clarke 1992, 447). However, with the manifest "personalization" of politics there are many examples that leaders can help or hinder the fortunes of their parties (Harrison & Marsh 1994, 290). Indeed, several recent studies conclude that voter evaluation of party leaders significantly affects party choice (see e.g. Bean 1993, 112). Single country studies also show a strengthening of leadership effects over time (Lanoue & Headrick 1994, 196).

The purpose of this article is to examine whether leadership effects can be detected in Norwegian vote behavior. Since leadership effects are shaped by the characteristics of the political system (see Bean 1993, 120; and below), the small, homogenous and centralized Norwegian system should provide ample reason to reject the null hypothesis of no leadership effects. Yet, as in most other non-American analyses, Scandinavian electoral studies concentrate more on the parties than on their leaders.' Analyses of leadership

---

* Tor Midtbø, Department of Comparative Politics, University of Bergen. Christiesgate 15. N-5007 Bergen, Norway.

effects have been few, sketchy and impressionistic. The purpose of this article is therefore twofold. First, it tries to systematize the theoretical and methodological questions that pertain to the relationship between the leader and the party. Second, it seeks to develop a sounder statistical foundation for the analysis. To identify the causality structure, a simple version of the Granger causality test based on a two-wave panel for the 1989 and 1993 parliamentary elections is presented. Since party support and especially leader popularity are not fully captured by the long-term perspective in a panel model, a nonrecursive, cross-sectional model for the 1993 Storting election is estimated as well.

Theoretical Background

The impact of leader images must be seen in a context where the traditional electoral models are no longer adequate (Brown et al. 1988, 729-732). The stabilizing influence of party identification and social background has weakened (Dalton et al. 1984, 13), and new and profound social cleavages are yet to emerge (see e.g. Clarke & Dutt 1991, 905). As a result, the literature is now less preoccupied with long-term vote behavior and more focused on short-term change and volatility. Issues in general, and economic issues in particular, are held to be increasingly important for the vote choice (Lewis 1986, 315). However, theories of issue voting do not fully explain the variation in voting behavior. In an early review article, Paldam (1981, 181) concludes that macro-economic variables explain roughly one third of the variation in party popularity, but that the influence is unstable. That conclusion still seems valid. The limited explanatory success of macro-economic variables must also be seen in connection with the tendency of convergence between the social democratic and non-socialist parties which makes party choice less dependent upon ideological leanings (Brown 1992, 545). In addition, the spatial theory of voting (Downs 1957, 114-142), on which many issue models seem to build, has recently been challenged by a directional theory (see e.g. Macdonald et al. 1991). Here the intensity of preferences and the underlying idea of bounded rationality are core elements (Iversen 1994a, 163). Voting is seen as driven by emotions and affected by political symbols, and not by "sterile," rational considerations (Iversen 1994b, 46-47).

In sum, the number of core voters declines, the dealignment process accelerates, and the ideological cleavages fade, which means that fewer cues guide voters' decision making calculus (Lanoue & Headrick 1994, 193). In this vacuous situation, leadership factors can be expected to make up for some of the unexplained variations in the traditional models. Theoretical innovation is needed, however, since models based on social background, party identification and issues all fail to disentangle the intimate relationship

between the popularity of the parties and their leaders. Recently, however, attempts have been made to identify and systematize the conditions that structure the relationship between leader images and party support.² Harrison & Marsh (1994, 291-293) distinguish between global, contextual, partyspecific and individual effects, and that classification is used as the point of departure here, too.

Global Effects

With the improvement and expansion of communication channels, a "new political technology" (including polling, media appearances, PR, etc.) has replaced the party organization as the main link between political leaders and voters (Ginsberg & Stone 1986, 71-75). This development - which partly, and only partly, is a result of the "Americanization of electioneering" (Field 1994, 58-59) - has given the party leader a much more visible role than before (Bean & Mughan 1989, 1175; Clarke & Stewart 1995, 155-156). In particular, television acts as a catalyst for the effect of leader images on party support (Nadeau & Mendelsohn 1994, 224). That "TV makes you or breaks you" is only a slight exaggeration nowadays,³ and the politicians are fully aware of this. Though not entirely representative of the party (see Bjørklund 1991, 289), political campaigners focus exclusively on candidate attributes and "electability" (Norris et al. 1992, 496). Media specialists are often hired to improve the appearance of party leaders, a job that, inter alia, includes staging "photo ops" and "walkabouts" (Farrell & Wortman 1987, 297-318).

This process is perhaps best captured by the notion of political marketing (ibid., 298-299) and the consumer theory of voting (see e.g. Heath et al. 1992, 208-209). In contrast to the traditional orthodoxy that childhood socialization influences a person's political party identification and ultimately determines the vote choice, voting is compared with the purchase of consumer goods.⁴ Similar to the way marketing strategies affect consumer behavior, political marketing is held to sway hesitant voters in an increasingly volatile electoral market. Akin to the motivation behind consumer behavior, the way the political product is advertised and wrapped is seen to affect voting more than the inner qualities of the product itself. To some extent therefore, political marketing requires the politicians not only to sell ideas but to sell themselves as well (Newman 1996, 654). Indeed, it is only from this perspective we can fathom the bizarre media events during elections campaigns, where "Mrs Thatcher puts on Wellington boots and tramps around the farm; Mr Kinnock goes to the bakery and wears a starched white hat" (Butler 1988, 72), and why Norwegian party leaders "meet the very old, the newborn, fish in fishing-wells and wear work clothes that they never before have worn."⁵ In short, as the media in general, and television in particular, monopolize the channels of political information, the party leader

acts frequently as the central messenger. And the relationship between the messenger and the message is an intimate one. Tellingly, media attention is almost as intense when the Norwegian finance minister files his annual personal tax form as when he presents his annual national budget (Hansson & Teigene 1995, 136). Thus, voters are not only expected to sanction the message (which can be traced back to the party manifesto) but the messenger as well.

Contextual Effects

Context affects all political parties, but not all of the time in all places. Esaiasson (1985, 106-107) identifies some of these factors. First, there is the constitutional condition. Leadership effects are expected to be stronger in presidential systems than in parliamentary systems, simply because in the former system there is a stronger focus on the candidate. Second, historical factors come to play where leader effects should be weaker for parties molded by a history of stable and continuous democracy than for parties born in new democracies. Third, in systems where parties are firmly embedded in the cleavage structure, leader effects are expected to be more feeble than in systems that allow vote maximizing party behavior independent of for example class and religion. Fourth, the electoral system itself may be considered a contextual factor. Majority systems generally focus more on the local candidate than systems with proportional representation. On the other hand, the national party leader should attract more attention in centralized systems. A fifth condition pertains to the prevailing leadership culture. The stronger the infusion of authoritarian values, the more important leadership effects are expected to be.

Based on these criteria alone, it is difficult to make clear predictions as to the strength of Norwegian leadership effects. On the one hand, Norway has a parliamentary system based on proportional representation and a stable party system. Both factors should weaken leadership effects. On the other hand, the political system is centralized, small and transparent, and the cleavage structure has started to fade here as elsewhere. The egalitarian aspect of the Norwegian social democratic culture should prevent effects caused by blind leader loyalty. However, in recent years, while Brundtland was still prime minister, the media and the voters alike started treating her with remarkable subservience. This phenomenon, which will be discussed further below, suggests that the Norwegian culture contains a dose of hierarchical elements as well.

Party-Specific Effects

The relationship between leader images and party support is stronger for

some parties than for others. One distinction here is the one between the opposition and the government (see Harrison & March 1994, 292-303). Since the policies of the opposition are necessarily hypothetical and government policies are real, opposition leaders might play a more important role as advertisers than the prime minister. On the other hand, the latter has to defend and explain government policies on a day to day basis, something that evidently attracts media attention. Note also that the visibility of the opposition leader should, ceteris paribus, be a negative function of the number of parties in the political system: As the number of opposition leaders increases, so does the probability of being ignored by the media.

In addition to the government-opposition dichotomy, there is a related distinction between small and large parties. In a British analysis, Nadeau & Mendelsohn (1994, 222) observe that the small parties fail to attract media attention and that they lack government potential. Since such parties are both invisible and irrelevant, leadership effects are held to be insignificant. This description is more apt for Britain than it is for Norway, though. A Norwegian non-socialist government cannot be established without participation by, or at least the explicit support from, the small centrist parties. Also, in Norway the size of the parties is usually negatively related to the charisma of their leaders. As a matter of fact, it has been argued that there are not only charismatic leaders but "charismatic parties," too (Harmel & Svåsand 1993, 67). These are parties very closely tied to, and often founded by, the party leader. In Norway, the Progress Party fits this definition rather well. This brings us to a third dichotomy between old and new parties. Within a context of "frozen cleavages," a charismatic party leader is required to make up for the initial electoral handicap. Old parties have core voters, new parties do not. Whereas established parties can choose leaders based on tenure and party loyalty, personal charisma and voter appeal are the tools new parties must use to pass the electoral threshold.

Individual Effects

If leaders influence party support, then the next question is how they influence it. Despite the large menu of leader qualities to choose from.⁶ it appears that perceived competence, responsiveness and integrity are particularly decisive for the vote choice (Stewart & Clarke 1992. 448-449: Bean 1993, 121). These are qualities politicians themselves like to highlight. For example, the 1983 West German electoral campaign featured billboards of CDU leader Kohl saying that "Dieser Kanzler shafft Vertrauen" ("this chancellor inspires confidence"). In the 1981 Irish election, the voters were informed that Fitzgerald actually "listens" (Farrell & Wortman 1987,303-312).

Leadership qualities can be tied to Weber's (1971, 5) traditional authority, or habitual compliance, and charismatic authority due to the extraordinary personal qualities of the leader. As for tradition, a "longitudinal incumbency" factor has been identified (Norris et al. 1992, 500). The point is simply that the longer politicians remain active in politics, the larger their crowd of loyal followers will be. When it comes to charisma, an analysis inspired by the directional theory of voting shows that politicians who present clear and intense policy alternatives tend to attract more votes than those who are afraid to leave "the empty center" (Iversen 1994b, 46). It appears, then, that a popular leader is one who leads, rather than one who is being led.

Norwegian Party Leaders⁷ and the 1993 Storting Election

The 1993 Storting election was very much affected by the controversial question of Norwegian membership in the European Union, a question decided by referendum the following year. The issue made some leaders more visible than others. In particular, Anne Enger Lahnstein from the Center Party assumed the role of "general" of the movement against EU membership. Her party finished as the undisputed winner of the 1993 election with an increase from 6.6 percent of the votes in 1989 to 16.7 percent in 1993. The Center Party even replaced the Conservatives as the largest opposition party in parliament. As Figure 1 shows, Lahnstein was a popular leader. Whether this popularity is causally related to the success of her party will be explored further in the empirical analysis below.

By far the most popular leader in the 1993 election was Gro Harlem Brundtland. She resigned as leader of the Labor Party in November 1992, and the position was given to Torbjørn Jagland. However, voters undoubtedly perceived the leadership of the Labor Party, and indeed of the country as a whole, to lie firmly in the hands of the Prime Minister. Brundtland was the first woman prime minister in the Nordic countries, and the youngest prime minister ever to be appointed in the history of Norwegian parliamentarism. Except for Einar Gerhardsen, she is the longest serving prime minister in the post-war period. In contrast to other current and previous Norwegian politicians, she has also acquired an international reputation (Hansson & Teigene 1995, 207-208). Brundtland is known for attempts to rise above everyday politics (Jenssen 1991, 91), and at times her role seems more congruent with that of a president. She has been criticized by the media and the opposition alike for her silence on issues concerning actual (and unpopular) government policies, while being very accommodating

when interviewed about official duties and on her many travels abroad,
These interviews portray her as a statesman first and a politician second.

The "Brundtland factor" was frequently mentioned in editorial columns. The day after the election, one newspaper wrote: "There can be little doubt that Gro Harlem Brundtland's strong position as prime minister has been the greatest advantage for her party."⁸ This argument must be seen in relation to developments during the campaign. Only three months before the election, the opinion polls showed that support for the Labor Party was at a dismal 27 percent of the votes (Valen 1994, 168-171). From June to September, the support increased by 10 percentage points. A large part of this increase was due to a tactical move by the new leader of the Labor Party, Torbjørn Jagland. He declared that a poor election result would mean the replacement of the Labor government, probably by a Conservative minority government. The question of governance boiled down to a choice between Kaci Kullman Five from the Conservatives and Gro Harlem Brundtland as the future prime minister.⁹ This was an uneven match. Apart from the fact that there was no parliamentary support for a Conservative government, Brundtland was simply much more popular than Five. Figure 1 shows that Brundtland received an average score of 71.1 percent in the electorate as a whole

compared with only 45.8 percent for Five.10 In fact, Five is one of only two
leaders who is less popular than their party.

As previously mentioned, directional theory identifies intense preferences, tenure and charisma as prerequisites for high leader approval. Apparently, the leader of the Progress Party, Carl I. Hagen, possesses all three qualities. Since 1978, he has remained in charge of a party that, at least by Norwegian standards, has adopted an extreme right-wing orientation. Hagen has been described as the most charismatic contemporary leader in Norway, and "many of the party's voters and members continue to wonder whether 'Hagen's party' could survive without him" (Harmel & Svåsand 1993, 79-82). Nonetheless, Figure 1 suggests that Hagen, though more popular than his party, is less popular than the other leaders. This does not necessarily mean that Hagen is an electoral liability. A low score may reflect high intensity, in the sense that the respondents who approve of Hagen give him disproportionally high scores, while those who disapprove give him disproportionally low scores. Since the latter group is larger than the former, the average score will automatically be low.

The inverse should also apply. A high leader score does not in itself increase the probability of electoral success. This is illustrated by the leaders of the Christian Peoples' Party, Kjell Magne Bondevik, and the Socialist Left Party, Erik Solheim. As for the latter, despite a drop from his sympathy level in the previous election, Solheim was more popular with the average voter than the leaders of the Liberal Party, the Conservative Party and the Progress Party. Yet, the Socialist Party only received 7.9 percent of the votes."

Not all leaders from small parties are popular. The lack of leadership charisma has been a persistent problem for the Liberal Party, and, apart from Hagen of the Progressive Party, its leaders have received the lowest sympathy scores in all three Storting elections since 1985. The Liberal Party only got 3.6 percent of the votes in the 1993 election, although it was roughly as popular as the Socialist Left Party, the Christian Peoples' Party, the Conservative Party, and overwhelmingly more popular than the Progress Party.

Despite these differences in party and leader sympathy levels, there are common global effects that should tie all the parties and their leaders closely together. In fact, the probability of leadership effects for any of the parties should be higher in the 1993 Storting election than in any previous election. Volatility was record high (Aardal 1994, 173), and party popularity changed dramatically during the campaign (Valen 1994, 169). This was also the election when the Norwegian Broadcasting Corporation lost its monopoly on campaign coverage. As commercial television channels joined the competition, viewers were exposed to political debates and presentations to an extent that was historically unprecedented (Aardal 1994, 177). In this media-driven campaign, the party leaders were at the center of attention.¹²

Methodology and Model Specification

According to the conventional wisdom of electoral research, variation in the assessment of party leaders does not affect party choice when other factors are controlled (Clarke & Stewart 1992, 151). However, such control is only one aspect of model specification. Just as omitted variables yield biased and inconsistent results, so does misspecification of causal relationships (Johnston 1984, 440). It is common knowledge that the higher the sympathy for a party, the higher the sympathy for its leader (see e.g. Valen & Aardal 1983, 32). This positive correlation can have as many as six different sources. First, a general observation is that "the relationship between leaders and followers is always double-sided and interdependent" (Putnam 1976, 134). A reciprocal relationship between leader approval and party support may therefore have to be specified. The second alternative, which prevails in the literature (see e.g. Bean and Mughan 1989; Bean 1993; Stewart and Clarke 1992), focuses on what leaders do for their parties rather than on what parties do for their leaders. Leader approval is seen as a cause of party support. The third alternative, on the other hand, is that a popular party can make up for lack of leadership charisma, and an unpopular party can stifle the appeal of its leader. In this perspective, party support causes leader approval, and not the other way around.

Fourth, there might be an indirect relationship between leader images and party support. This possibility is easily overlooked. As an example, based on three key variables, i.e. leader sympathy, party sympathy and actual vote choice, Macdonald et al. (1995, 463) find that Norwegian party sympathy has a stronger effect on vote choice than candidate sympathy. As long as the voters vote for a party and not for a leader, this is hardly surprising. Party sympathy is a direct and immediate cause of vote choice (the more you like the party, the higher the probability that you vote for it), but it is also part of the vote choice (the party you like is the party you vote for). Whereas party sympathy and vote choice to some extent are different aspects of the same phenomenon, the distinction between leader sympathy and vote choice is more clear-cut. However, we would expect indirect effects by which leader sympathy affects party sympathy that in turn affects vote choice. In other words, the positive correlation between party sympathy and vote choice is partly a result of a positive correlation between leader and party sympathy. This point is pursued further in the empirical analysis below.

The fifth alternative is that voters fail to distinguish between the party and its leader. If you like the party, you tend to like the leader as well, and vice versa. In this view, one factor is not the cause of the other, but the two factors are part of the same phenomenon (Clarke & Stewart 1995. 156). The sixth alternative is the occurrence of a spurious relationship (Hudson 1984. 89; Goodhart & Bhansali 1970, 69-70). From the economic voting literature

it may be hypothesized that the popularity of the. leader and the party are both affected by government responsibility (see e.g. Goergen & Norpoth 1991). A prolonged period in office usually wears out the popularity of both. There is also the more subtle argument that in times of economic depression, the government selects unattractive candidates, since it expects to lose the coming election anyway (see Lewis-Beck 1988, 30). If this argument is valid (which it probably is not), unpopular politicians in an unpopular government are both the result of a poor economy.

Since the relationship between leader images and party support cannot be decided upon a priori, model specification in general, and causality specification in particular, has to be determined on empirical grounds. In Norway, no regular time series data for leader popularity is available, and the data menu is restricted to two-wave panel and cross-sectional survey data. As a first step in the empirical analysis, a panel model is combined with the notion of Granger causality. Since causes tend to appear before effects, a correlation between party sympathy in one election and leader sympathy in the previous election, or alternatively, between leader sympathy in one election and party sympathy in the previous election, should provide some information about the underlying causative process. The test is based on two equations. In the first equation, party sympathy is regressed on lagged party sympathy and lagged leader sympathy. In the second equation, leader sympathy is regressed on lagged leader sympathy and lagged party sympathy. To put it simply, the point is to decide whether either lagged leader sympathy in the first equation and/or party sympathy in the second equation are significantly different from zero. The technical details of this approach are relegated to Appendix 1.

A study of the relationship between leader images and party support should distinguish between short-term and long-term effects. For example, a time series analysis of monthly opinion poll data could reveal whether a good performance by a leader in a televised debate boosts party support in the subsequent months. However, according to Brown (1992, 562), leader charisma has to mature over a long period, and a change of impact during a short election campaign is not probable. Therefore, a popular leader in one election may pave the way for a successful party only in the next election (Esaiasson 1985, 110-118). A panel analysis, where the interval is defined by the period between one election and the next, captures such long-term effects. Still, in a turbulent political climate, a popular leader in one election could easily be an unpopular leader in the next election. The interval between two elections could be too long to capture this change (Page & Jones 1979, 1079; Asher 1983, 27-28). Therefore, a contemporaneous model, which ipso facto comprises both long-term and short-term effects, is also included in the empirical analysis.

The above discussion suggests that if the data are cross sectional, a single equation model for the relationship between leader images and party support would be inappropriate. Recursive models (where ßy # 0 implies that ßy_s = 0) do not allow for the identification of causal relationships (Asher 1983, 11). Also, a reciprocal relationship between leader images and party choice means that the error term will be related to the explanatory variables, and OLS (or LOGIT or PROBIT for that matter) will produce biased and inconsistent parameters (Berry 1984, 61-62). As long as the relationship is positive, the bias will also be positive and the estimated leadership effect inflated. Due to this reciprocity, a nonrecursive model is needed, and in standard econometrics, 2SLS (Two Stage Least Squares) is commonly preferred.¹³ In such a model, the crucial task is to find variables that are genuinely exogenous, that is, variables that have a direct causal effect in one equation but no direct effect in other equations. In our particular model, we have to find exogenous variables that affect leader approval directly but party support only indirectly, and vice versa. Since the identification process is tedious, the selection of variables is discussed in Appendix 2. At this point, suffice it to say that to control for the salience of the EU membership issue, three (over-identified) equations were used to capture voter behavior in the 1993 Storting election. The nonrecursive causal model is specified as follows (with intercepts omitted):

The model needs three final clarifications. First, leader sympathy is measured on a scale from 0 to 100. To identify a potentially reciprocal relationship, having two commensurable measures is advantageous. Consequently, the assumption is that party sympathy, which is measured on a similar scale, is a valid indicator of actual vote choice. This is not a strong assumption. As previously mentioned, Macdonald et al. (1995, 463) find party sympathy to be a strong predictor of actual vote choice in Norway. Their conclusion is supported in a simple bivariate LOGIT analysis presented in Appendix 3. Second, since the nonrecursive model is estimated with a limited information technique, each equation is estimated separately without considering other structural equations in the system.¹⁴ This simplifies analysis inasmuch as the equation for EU sympathy can be excluded from the estimation.¹⁵ Third, in contrast to some previous studies (see e.g. Valen & Aardal 1983, 36), no distinction is made between a leader's popularity among voters of a particular party and the popularity of the same leader among voters from other parties. Instead, all the respondents in the survey are included in the analysis. In an era of strong dealignment and volatility, parties and party leaders need to attract voters from a wide spectrum. An analysis of the relationship between leader images and party support should therefore include both core and floating voters. Indeed, it is precisely the dealignment effect that is expected to make leadership effects important.

Empirical Analysis 16

The Granger causality test is restricted to the four parties with the same leader in both the 1989 and the 1993 elections, i.e. the Socialist Left Party, the Labor Party, the Christian Peoples' Party, and the Progress Party. Wald tests (see e.g. Green 1991, 128-130 for an explication) in Table 1 identify Brundtland and Hagen, the leaders of the Labor Party and the Progress Party, as the vote getters. In both cases, the chi-squared statistic exceeds the critical value.

Both Hagen and Brundtland are experienced politicians with intense and uncompromising leader styles. These are qualities that the directional theory identifies as vote magnets (Iversen 1994b, 70). Whereas Brundtland provides strong political leadership, Hagen represents a clear sense of ideological

direction. Note that even if Bondevik receives more sympathy from the average voter than Hagen, the former has no significant impact on party sympathy. The reason is probably that whereas many voters either love or hate Hagen, many voters remain indifferent to Bondevik. Moreover, and in contrast to the Progress Party, the Christian Peoples' Party attracts deferential followers (Strøm 1993, 431). For them, the ideological and religious profile of the party should weigh heavier than the mundane qualities of a party leader. The Socialist Left Party does not seem to derive any electoral advantage from their party leader either. This may be because the party (together with the Conservatives) attracts a disproportionately large number of highly educated voters (Valen et al. 1990, 75-78). Education is positively related to political interest, and voters interested in politics presumably concentrate more on the issues than on the leaders (Esaiasson 1985, 114).

Apart from the Labor Party, where the coefficient (not shown) on closer inspection turns out to be negative,¹⁷ the analysis rejects the hypothesis of a long run causal influence from the respective parties to their leaders. The result from the pooled version of the Granger test only adds to this conclusion: A popular leader in one election increases the popularity of his/ her party in the next election. However, a high level of party sympathy in one election does not make much of a difference in the evaluation of the leader in an election four years later.

Because Granger tests based on short lag lengths cannot always provide
definitive empirical answers, a 2SLS analysis is used as a complementary

econometric device (see Tables 2 and 3). A comparison of the unstandardized coefficients across the equations in Table 2 shows that all party leaders, except Dørum, the leader of the Liberal Party, contribute positively to their party's fortunes. In contrast to the other parties, the Liberal Party did not receive any boost from its leader. The leadership coefficient is even negative, albeit insignificant.

On the other end of the spectrum, we find Lahnstein, the leader of the Center Party. With an unstandardized coefficient of 1.3, she had the largest impact on party support of all the party leaders. According to the table, the image factor is so large that it almost obliterates the impact of the EU issue itself, which is large, but insignificant. This finding suggests that the voters had some difficulties separating the messenger from the message. It also underscores the

caveat issued by Valen (1994, 173) that the effect of the EU controversy,
though important, was channeled through "complex relationships."

According to Table 2, Solheim did well in the 1993 election. His party, however, did not. Compared with the 1989 election, the Socialist Left Party lost 2.2 percentage points of the votes. The political agenda, and not Solheim's personality, appears to have been the culprit. The Socialist Left Party was overshadowed by the almost dogmatic posture of the Center Party as an anti-EU party. In addition, the environmental issue, which was important in the 1989 election, no longer received the same attention (Aardal & Valen 1995, 19). Nor did the signaling of a more cooperative attitude towards the governing Labor Party improve the party's situation (Valen 1994, 177). The Socialist Left Party failed to sell its ideas, and the offsetting effect of a congenial party leader was not enough. On the other hand, the support might have been even weaker without a popular leader.

Considering the result from the long-term model, the weak leadership effect for Hagen in Table 2 is surprising. However, compared with the previous parliamentary election, the 1993 election was a failure for the Progress Party. From 13 percent in 1989, the vote share for the Progress Party declined to 6.3 percent of the votes in 1993. Hagen was almost invisible during the election campaign, and when he finally attracted media attention, he managed only to embarrass his party. The leader of a party whose very existence builds on a diehard fight against "Taxes, Rates and Public Intervention" (as spelled out in the initial party slogan), publicly declared that this was not the time for tax relief.

If we take into account a booming economy and a divided opposition, the "Brundtland effect" may have been exaggerated. Indeed, according to the unstandardized regression coefficients in Table 11, the leaders of the Center Party and the Socialist Left Party attracted more votes than she did. However, the real importance of Brundtland's appeal was that it stifled the competition from the main opposition party, the Conservatives. In fact, if we take the unstandardized regression coefficients as a standard of comparison, Five stands out as the leader with the weakest mobilizing capacity, not counting Dørum. A percentage point increase in sympathy for Five increases party sympathy on average by 0.4 points. The equivalent figures for Lahnstein, Solheim and Brundtland are 1.3, 1.2 and 0.9, respectively.

On the face of it, just as leader sympathy dominates the equation for party sympathy in Table 2, party sympathy dominates the equation for leader sympathy in Table 3. However, these effects consist of two elements. Part of the effect is due to reciprocity, but another part is a result of a common factor. The relative importance of these two factors is impossible to determine in a cross-sectional analysis. Therefore, based on this table we cannot say that, for example, the effect of party sympathy on leader sympathy is larger than the gender effect. However, we can establish

whether leader sympathy affects party sympathy more than party sympathy affects leader sympathy. This is done by comparing the standardized coefficients for leader sympathy in Table 2 with the equivalent coefficients for party sympathy in Table 3.¹⁸ There we can see that Solheim, Brundtland and Lahnstein contributed more to their party popularity than the other way around. The effects for the Christian Peoples' Party and their leader, Bondevik, are roughly of the same size. On the other hand, the Liberal Party, the Conservatives and the Progress Party contributed more to their leaders' popularity than vice versa. Tellingly, these three parties, but especially the last two, lost heavily in the election.

Though we should not put too much emphasis on the results for the control variables, the impact of gender on leader and party sympathy is interesting. With Solheim as the only anomaly, the support for the female leaders is consistently higher among women than among men. Whereas Table 2 suggests that the gender variable influences party sympathy only to a very limited extent, Table 3 shows that women tend to sympathize with women, and men with men, independent of ideological color. For example, in Table 3, both Brundtland and Five are more popular in the female part of the electorate. Apparently, the "gender gap" is just as pronounced for the evaluation of leaders as for the evaluation of parties.

As an overall assessment based on both the Granger test and the 2SLS analysis, Brundtland emerges as the leader who contributed the most to party sympathy in the 1993 election. Today, all Norwegians immediately recognize the name "Gro" not only as the first name of the Norwegian former prime minister but as something distinct from, and above, everyday party politics (Hansson & Teigene 1995, 121). It has not always been like this. As opposition leader in the beginning of the 1980s, her temper and assertive attitude made her almost an electoral liability. In fact, on the sympathy barometer for the 1981 and 1985 elections, Brundtland was the only leader, apart from the leaders of the Liberal Party, whose sympathy level was lower than that of the party. However, whereas the support for the Labor government - partly due to the well-known voter disillusion mechanism - has been characterized by more ebbs than flows during the last decade, the support for the prime minister has increased steadily. Brundtland transcended the setbacks and policy failures of her government, and during the last part of her tenure, she seemed to enjoy a "teflon effect," in many ways similar to that of former US president. Ronald Reagan.

Conclusion

At first glance, the relationship between leader images and party support
appears theoretically trivial. To say that those who like the leader also like

the party, and vice versa, sounds more like a truism than a building block for theoretical analysis. Not only that, but as we have seen, the relationship between leader and party presents huge empirical and statistical obstacles to be vanquished as well. Measurement without theory is bad; ambiguous measurement without theory is even worse. Yet, the point is not whether the relationship between leadership and party support is trivial or ambiguous, but whether it exists or not. If it does, the immediate implication is that established voter models are misspecified. Omitted relevant variables yield biased coefficients and an overestimated disturbance variance (Johnston 1984, 260-262). Thus, the traditional models will present a distorted view as to how, and to what extent, social background, party identification and issues affect the choice of party.

Methodological problems aside, the fact that those directly affected by the phenomenon, the politicians, are almost absorbed by the question of leader appeal cannot be inconsequential. Even if electability has not always motivated the choice of Norwegian party leaders, "youth and charisma" are the new criteria that have superseded seniority and party loyalty as selection criteria (Strøm 1993, 344—345). In Norway, as elsewhere, television has contributed a lot to this development. Needless to say, it is the leader and not the party who is on television. This simple fact can explain why the Granger test identified the flow of causality as running from leader to party, and not the other way. Thus, literally speaking, the messenger explains the message. To some extent, television also establishes a symbiosis between leader and party, a symbiosis reflected by the dominance of global effects in the cross-sectional model above. With one exception, all party leaders contributed positively to their party's fortunes. However, the empirical analysis also identifies leader and party-specific effects. In particular, Brundtland, Hagen and Lahnstein are identified as the main vote getters. All three do very well on TV. In addition, even if their political styles and rhetoric are different, they share a willingness to take a clear stand on important political issues. And according to the directional theory, vote seekers should adopt distinct policy positions.

The theoretical triviality of the matter can be disputed as well. Leader images as an explanation of vote behavior avoids much of the determinism of the old sociological models, and it does not need the strong rationality assumption that encumbers models of issue voting. In fact, a scale of rationality may be constructed by which voting is motivated either according to who the candidate is, what the candidate says, or what the candidate does. From the perspective of democratic accountability, the voters' choice of parties should be affected by what politicians actually achieve in terms of policy formulation and implementation. However, rhetoric and appearances may be more important than policies. An innocent example of the perceived impact of appearances is the British woman MP who wears a lot of "red and pink" so that her constituents can easily spot her in the televised

parliamentary debates (Norris et al. 1992, 502). Politicians can also attract attention by using different kinds of rhetorical styles, where they appeal to logic or facts, but also to morals, myths and feelings. Market research by the parties themselves has discovered a political "primitiveness" in voter attitudes (Farrell & Wortman 1987, 395). Even the economic voting literature embraces the notion of "affective voting," which means that vote choice is a result of passionate assessments instead of careful and rational evaluation (see Lewis-Beck 1986, 318). Thus, it might be that "[j]ust as consumers don't take the time to read labels on products, neither do citizens take the time to listen carefully to what politicians say" (Newman 1996, 661). For Brundtland that should have been an advantage. Her speech was often complicated and her argumentative style had been characterized as "insufferable linguistic breathing exercises" (see Hansson & Teigene 1995, 129). So, if voters do not listen, it is perhaps because there is nothing worth listening to.

---

NOTES

1. In one of the few explicit analyses on the topic, Granberg & Holmberg (1988, 88-107) adhere to the established view that in Sweden the "principal actors" are the parties and not their leaders. Basedon Norwegian data, Valen & Aardal (1983, 32-42), Aardal & Valen (1989, 31-35), and Macdonald et al. (1995, 463) reach the same conclusion. In an early and thorough study, Esaiasson (1985, 118-119) finds leadership effects, albeit weak ones, in two Swedish Riksdag elections. His findings have received further support in recent Swedish electoral studies (see Gilljam & Holmberg 1995, 196; Gilljam & Holmberg 1993, 98-109, and the references cited therein).

2. The rather opaque socio-psychological analyses of leadership effects are not considered here. See Brown et al. (1988, 732-736) for a discussion.

3. Leadership focus is further intensified by so-called horse-race journalism where newspapers use opinion polls to declare "losers" and "winners" from the televised political debate the night before (Bjørklund 1991, 286-288).

4. These two perspectives are not entirely incompatible. Consumers may for instance buy the same product over and over again because of habit.

5. Verdens Gang, September 13, 1993.

6. Brown et al. (1988, 737-738) identify as many as eleven leader attributes.

7. In Norwegian politics, the definition of "party leader" is not entirely unambiguous (Strøm 1993, 322). There are several leadership positions within each party, and the formal power accruing to the party leader differs between parties and over time. Yet, it is easy to identify the key figures of the 1993 election.

8. Dagbladet, September 14, 1993.

9. After Kåre Willoch retired as the Conservative leader, the formal power shifted from the parliamentary leader to the party chair. In 1993, the leading figure of the Conservative Party was therefore Kaci Kullman Five, and not the parliamentary leader. Anders Talleraas.

10. A public opinion poll published in a Norwegian newspaper also showed that 54 percent of the voters saw Brundtland as a competent politician, while only 11 percent thought the same of Five {Verdens Gang September 14, 1993).

---

10. A public opinion poll published in a Norwegian newspaper also showed that 54 percent of the voters saw Brundtland as a competent politician, while only 11 percent thought the same of Five {Verdens Gang September 14, 1993).

11. This mismatch was particularly evident for the previous leader of the Socialist Left Party, Hanna Kvannmo. Although she was the most popular leader in the electorate, she presided over a party that received a meagre 5.5 percent of the votes in the 1985 election (Aardal & Valen 1989, 29-35).

12. In fact, a leading Norwegian newspaper complained that "[v]oters gradually grow tired of seeing the same party leaders reappearing in political debates time and time again. .. The next election campaign should be conducted by the parties and not the TVchannels" {Aftenposten, September 12, 1993).

13. The small sample properties of 2SLS are not fully determined, but the estimator seems generally less biased compared to OLS (Berry 1984, 67). The main problem with the 2SLS model, in addition to finding appropriate instruments, is built-in multicollinearity. However, in the analysis below, a closer inspection did not detect any multicollinearity problems.

14. By contrast, full information techniques estimate all the parameters for all the equations simultaneously (given that they all are identified). The latter techniques are more efficient than 2SLS but more sensitive to errors in model specification (Berry 1984, 62).

15. The instruments for the third equation are, of course, included in the estimation of the two other equations.

16. The data sets from the 1989 and 1993 Norwegian Election Studies - as well as data from previous election studies referred to in the text - were made available from the Norwegian Social Science Data Service (NSD). The author assumes full responsibility for the analysis and the interpretation of the data.

17. The "wrong" sign may reflect the fact that the sympathy score for Brundtland has increased tremendously over the years.

18. For a critical discussion on the utility of standardized coefficients, see e.g. Asher (1983, 49); Hanushek & Jackson (1977, 78-79).

ACKNOWLEDGEMENTS

Earlier versions of this article benefited from the criticism of Gunnar Grendstad, Ola Listhaug,
Kristin Strømsnes, and three anonymous reviewers.

REFERENCES

Aardal, B. 1994. "The 1994 (sic) Storting Election: Volatile Voters Opposing the European
Union," Scandinavian Political Studies 17, 171-180.

Aardal, B. & Valen, H. 1989. Velgere, Partier og Politisk Avstand. Oslo: Statistisk
Sentralbyrå.

Aardal, B. & Valen, H. 1995. Konflikt og Opinion. Oslo: NKS-forlaget.
Aftenposten, September 12, 1993.

Asher, H. A. 1983. Causal Modelling. Quantitative Applications in the Social Sciences.
Newbury Park: Sage.

Bean, C. 1993. "The Electoral Influence of Party Leader Images in Australia and New
Zealand," Comparative Political Studies 26, 111-132.

Bean, C. & Mughan, A. 1989. "Leadership Effects in Parliamentary Elections in Australia and
Britain," American Political Science Review 83, 1165-1179.

Berry, W. D. 1984. Nonrecursive Causal Models. Quantitative Applications in the Social
Sciences, 37. Newbury Park: Sage.

Bjørklund, T. 1991. "Election Campaigns in Postwar Norway (1945-1989): From Party
to Media-Driven Campaigns," Scandinavian Political Studies 3, 279-302.

Brown, J. A. 1992. "The Major Effect: Changes in Party Leadership and Party Popularity,"
Parliamentary Affairs 45, 545-564.

Brown, S. D., Lambert, R. D. Kay, B. J., Curtis, J. E. 1988. "In the eye of the Beholder: Leader
Images in Canada," Canadian Journal of Political Science 21, 729-755.

Butler, D. 1988. "The 1987 general Election in Historical Perspective," in R. Skidelsky, ed.,
Thatcherism. Oxford: Basil Blackwell.

Clarke, H. D. & Dutt, N. 1991. "Measuring Value Change in Western Industrialized Societies:
The Impact of Unemployment," American Political Science Review 85, 905-920.

Clarke, H. D. & Stewart, M. C. 1992. '-Partisan Change in Britain, 1974-83." in Denver, D. &
Hands. G., eds.. Issues and Controversies in British Electoral Behaviour. New York:
Harvester Wheatsheaf.

Clarke, H. D. & Stewart, M. C. 1995. "Economic Evaluations. Prime Ministerial Approval and
Governing Party Support: Rival Models Considered," British Journal of Political Science
25, 145-170.

Dagbladet. September 14. 1993.

Dalton, R. J., Beck, P. A. & Flanagan, S. C. 1984. "Electoral Change in Advanced Industrial
Democracies," in Dalton, R. J., Flanagan, S. C & Beck. P. A., eds., Electoral Change in
Advanced Industrial Democracies. Princeton: Princeton University Press.

Downs, A. 1957. An Economic Theory of Democracy. New York: Harper & Row.

Esaiasson, P. 1985. "Partledareffekter även i Sverige? - Partiledarpopularitetens betydelse for
valresultatet i Sverige ur ett jämförande perspektiv." Statsvetenskapliq Tidsskrift 2.
105-122.

Farrell. D. M. & Wortman. M. 1987. "Party strategies in the electoral market: Political
marketing in West Germany. Britain and Ireland," European Journal of Political Research
15.297-318.

Field, W. 1994. "On the Americanization of Electioneering.'" Electoral Studies 13. 58-63.

Freeman. J. 1983. "Granger Causality and Time Series Analysis of Political Relationships.
American Journal of Political Science 27, 327-358.

Gilljam. M. & Holmbere. S. 1993. Vciljarne inför 90-talet. Stockholm: Fritzes.

Ginsberg. B. & Stone. A. 1986. "The New Politics and the Meaning of Elections." in B.
Ginsberg & Stone, A. eds., Do Elections Matter? New York: M.E. Sharpe.

Goergen, C. & Norpoth, H. 1991. "Government Turnover and Economic Accountability."
Electoral Studies 10, 191-207.

Goodhart. C. A. E. & Bhansali. R. J. 1970. "Political Economy." Political Studies 1. 43-106.

Granberg, D. & Holmberg, S. 1988. The Political System Matters. Social psychology and
voting behavior in Sweden and the United States. Cambridge: Cambridge University Press.

Green, W. H. 1991. Econometric Analysis. New York: MacMillan.

Hansson, S. & Teigene. I. H. 1995. Makt og Mannefall. Historien om Gro Harlem Brundtland.
Oslo: Cappelen.

Hanushek, E. A. & Jackson. J. E. 1977. Statistical Methods for Social Scientists. Orlando:
Academic Press.

Harmel. R. & Svåsand. L. 1993. "Party Leadership and Party Institutionalization: Three Phases
of Development." West European Politics 16. 67-88.

Harrison. M. J. & Marsh. M. 1994. "What Can He Do for Us? Leader Effects on Party
Fortunes in Ireland." Electoral Studies 13. 289-312.

Heath, A. R., Jowell. R. & Curtice. J. 1992. "Policy and Ideology in the 1983 Election." in D.
Denver & G. Hands, eds.. Issues and Controversies in British Electoral Behaviour. New
Harvester Wheatsheaf.

Hudson, J. 1984. "Prime Ministerial Popularity in the UK: 1960-81." Political Studies 32,86-97.

Iversen, T. 1994a. "The Logics of Electoral Politics. Spatial. Directional and Mobilizational
Effects," Comparative Political Studies 27. 155-189.

Iversen, T. 1994b. "Political Leadership and representation in West European Democracies: A
Test of Three Models of Voting." American Journal of Political Science 38. 45-74.

Jenssen. A. T. 1991. Verdivalg. Ny Massepolitikk i Norge. Manuscript. Trondheim: Dep. of
Sociology.

Johnston, J. 1984. Econometric Methods. Third edition. New York: McGraw-Hill.

Lanoue, D. J. & Headrick, B. 1994. "Prime Ministers. Parties and the Public. The Dynamics of
Government Popularity in Great Britain." Public Opinion Quarterly 58. 191-209.

Lewis-Beck, M. S. 1986. "Comparative Economic Voting: Britain. France. Germany. Italy."
American Journal of Political Science 30, 315-346.

Lewis-Beck, M. S. 1988. Economics and Elections. Ann Arbor: The University of Michigan
Press.

Macdonald, S. E., Listhaug, O. & Rabinowitz, G. 1991. "Issues and Party Support in
Multiparty Systems," American Political Science Review 85, 1107-1131.

Macdonald, S. E., Listhaug, O. & Rabinowitz, G. 1995. "Political Sophistication and Models
of Issue Voting," British Journal of Political Science 25, 453^83.

Nadeau, R. & Mendelsohn, M. 1994. "Short-term Popularity Boost Following Leadership
Change in Great Britain," Electoral Studies 13, 222-228.

Nadeau, R., Niemi, R. G. & Amato, T. 1996. "Prospective and Comparative or Retrospective
or Individual? Party Leaders and Party Support in Great Britain," British Journal of
Political Science 26, 245-258.

Newman, B. I. 1996. The Role of Marketing in American Politics. Paper presented at the 1996
annual conference of the PSA in Glasgow.

Norris, P., Vallance, E. & Lovenduski, J. 1992. "Do Candidates Make a Difference? Gender,
race, Ideology and Incumbency," Parliamentary Affairs 45, 496-517.

Page, B. I. & Jones, C. C 1979. "Reciprocal Effects of Policy Preferences, Party Loyalties and
the Vote," American Political Science Review 73, 1071-1089.

Paldam, M. 1981. "A Preliminary Survey of the Theories and Findings on Vote and Popularity
Functions," European Journal of Political Science 9, 181-199.

Putnam, R. D. 1976. The Comparative Study of Political Elites. Englewood Cliffs: Prentice Hall.

Stewart, M. C. & Clarke, H. D. 1992. "The (Un)lmportance of Party Leaders: Leader Images
and Party Choice in the 1987 British Election," Journal of Politics 54, 447-470.

Strøm, K. 1993. "Competition ruins the good life. Party leadership in Norway," European
Journal of Political Research 4, 317-347.

Valen, H. 1994. "Norway: A Storting Election in the Shadow of the EU," Electoral Studies 13,
169-179.

Valen, H. & Aardal, B. O. 1983. Et valg i perspektiv. Oslo: Statistisk Sentralbyrå.

Weber, M. 1971. Makt og Byråkrati. Translated by Dag Østerberg. Oslo: Gyldendal Norsk
Forlag.