It is easy to understand the widespread assumption that leaders' personalities must influence election results in presidential systems such as those of the United States, France, and Russia. In those countries, voters select individuals to exercise wide‐ranging executive powers, whether it be to wage war, appoint the senior judiciary or dissolve the legislature. These sorts of decisions are very likely to be heavily influenced by the personal qualities and quirks of the decision maker. It might therefore be supposed that it is only natural for voters to take into account the personal traits of presidential candidates—their intelligence, honesty, political acumen, and sheer likeability—when deciding for whom to vote.

The assumption that leaders' personalities matter in parliamentary systems such as Britain's, however, is less obviously correct. British prime ministers are not directly elected by the electorate as a whole. Like every other member of Parliament, they owe their membership of the House of Commons to the seventy thousand or so electors in their own constituency. They owe their position as prime minister to election by their party as its leader and, once in office, to the continuing support of their senior cabinet colleagues and parliamentary party.¹ Formally, at least, British government is party government: a collective enterprise in which the prime minister is merely primus inter pares. Margaret Thatcher's ignominious departure from office in 1990 well illustrates a prime minister's fundamental dependence on cabinet and party.

In recent years, however, most informed observers of British politics, with the important exception of academic political scientists, have taken it as axiomatic that the personalities of party leaders strongly influence the way people vote and therefore the result of elections. Some make the point that a (p. 71 ) prime minister's control of the cabinet agenda, coupled with his extensive powers of appointment, has replaced cabinet government by prime ministerial government.² Others go further and refer to the “presidentialization” of British politics and, in particular, British elections: the parties and media, they argue, have come to organize and depict election campaigns as a contest between the Labour and Conservative leaders.³ The personalities of party leaders have come to be thought of as looming every bit as large in the consciousness of British voters as they do among the voters of the United States, France, and Russia. This view of British politics is not confined to foreign commentators, tempted to construe British parliamentary elections in more familiar presidential terms. Nor is it confined to saloon bar discussions or the ruminations of tabloid journalists.⁴ Writing in 2000, one highly respected commentator, Hugo Young of the Guardian, expressed the firm belief that there is an identifiable personality component in the vote:

In modern [British] politics, nothing matters more than the leader. We have a parliamentary system but a presidential impulse. This takes a certain view about power, but a determining one about style. Think of the Tories and you get William Hague: bald, struggling, robotic, Yorkshire. Think of Labour, and there's only one face in front of it and one mind behind. Almost the entire apparatus of party presentation is devoted to the daily manicuring, weekly shaping and permanent controlling, down to the finest detail, of the impression these leaders make on the voting public.⁵

Communications advisors take it as axiomatic that leaders' physical appearance, the clothes they wear, their accent and their tone of voice all help to create positive or negative impressions. Many politicians apparently agree and are therefore willing to undergo “makeovers” in order to improve their appeal.⁶ Moreover, a failure on the part of a leader to project the right image, it is supposed, can have disastrous consequences. Neil Kinnock attributed Labour's defeat in 1992, at least in part, to voters' assessments of him as a leader: “One of the reasons [voters] eventually put their crosses by the Conservative candidate was this innate feeling among a relatively small number of people that they couldn't see me as prime minister. It's just there in the biochemistry, as it were. It's a pity but it's a fact of life I recognise.”⁷ In a similar vein William Hague attributed the Conservative Party's defeat in June 2001 to his personal failure “to persuade sufficient numbers that [he was] their (p. 72 ) alternative prime minister.” Indeed, his subsequent resignation was based, at least in part, on the belief that the Conservatives needed a leader who could “command a larger personal following in the country.”⁸

Many ordinary voters share this belief. One member of a Guardian focus group in late 1998 expressed the view that “Labour consists of one person—Blair. He's able to sway people with his oratory. But the people around him haven't a clue what to do.”⁹ The belief that there is an identifiable personality component of the vote is, as Anthony King suggested in Chapter 1, part of the culture, seldom questioned, and completely taken for granted.¹⁰

The remarkable tenacity of people's belief in leadership effects can, at least in part, be attributed to a failure to distinguish between direct and indirect effects. However, certain features of the British political system also serve to sustain the belief that leaders' personalities matter. The bear pit that is Prime Minister's Question Time, broadcast live on television from the House of Commons since 1989, provides voters with a ready opportunity to compare the cleverness, quick‐wittedness and knowledge of the party leaders.¹¹ Moreover, it could be argued that, since there is often no other benchmark by which to judge leaders, voters not unnaturally compare leaders with each other.¹²

The media further encourage the personalization of politics by focusing on the leaders, especially during election campaigns. Newspapers and television alike report elections as if they were gladiatorial combats between two generals rather than battles between two armies.¹³ This focus partly reflects journalists' fascination with politicians as individuals; but it may also reflect a belief that their readers or audience have a deep‐seated need for human interest stories to stimulate their interest in politics. Moreover, the point–counterpoint style of reporting, whereby the speeches, comments, or (p. 73 ) sound bites of one party leader are immediately followed by those of another emphasizes the tendency to look upon an election as effectively a choice between alternative prime ministers.¹⁴

The parties themselves encourage this same tendency by fighting leader‐centered campaigns. In the postmodern campaign, leaders are not simply the mouthpieces for the party's values and policies but have become part of the message. Their personality is managed and manufactured by the party organization to project an image of the party they lead. Attempts to create the impression that William Hague, the former Conservative leader, was a down‐to‐earth, no‐nonsense man of action, fit to be prime minister—as distinct from being a precocious “anorak” and “something of a prat”—included the wearing of a baseball cap during a visit to a theme park, a much photographed ascent of Ben Nevis with his wife, and his claim to have occasionally drunk as many as fourteen pints of beer a day as a teenager.¹⁵ Parties and the media are locked into an interdependent relationship. Television craves pictures. The parties, craving exposure for their leaders, provide them.

The common belief that leaders' personalities influence the outcome of British elections appears to be corroborated by survey evidence. A MORI poll conducted in August 1999 revealed that an impressive 98 percent of respondents recognized a picture of Tony Blair and 97 per cent knew that he was prime minister. Fully 84 percent recognized William Hague and 84 percent, again, knew that he was leader of the Conservative opposition. Recognition of secondary leaders such as Gordon Brown (45 percent) and Michael Portillo (42 per cent) was far fainter. In contrast, the British Election Study, the major academic survey of opinion at each general election, revealed that the proportion of the electorate who associated even the most distinctive policies with the correct party was often far lower. Only 36 percent correctly identified the Liberal Democrats as the party “most in favour of proportional representation,” and only 32 percent identified the SNP as the party “most in favour of independence for Scotland.”¹⁶ While some voters evidently find it difficult to assess politicians' policies, they undoubtedly find it much easier to assess their personalities.

Additional evidence for the electoral importance of voters' evaluations of leaders is seemingly provided by questions that ask respondents how they would react if a named politician were to lead a party or if the existing party leader were replaced by some unspecified individual. The answers to these questions seem to show that such changes could make a significant difference to the party's electoral prospects. Certainly, quasi‐experimental “trial heat” (p. 74 ) evidence is pored over by commentators and the parties themselves when selecting their leader. Responses to such questions influenced the decision of Conservative MPs to replace Margaret Thatcher with John Major in November, 1990.¹⁷ In the autumn of 1999 similar evidence was used to suggest that the Conservatives could not improve their standing with the electorate by the simple expedient of replacing William Hague with Michael Portillo.¹⁸

Although findings like these suggest that leaders' personalities can have an effect on political preferences, such evidence must be treated with caution. For one thing, the questions are hypothetical: the respondents have no stake in the outcome and cannot necessarily predict their own behavior. For another, such questions rarely provide any indication of the magnitude, as distinct from direction, of any shift in voting behavior. Knowing that “if Charles replaces David then voters are more likely to support party Y” indicates that support for Y is likely to rise, but not necessarily by how much. Finally, it is far from clear what thought‐processes go through respondents' minds when responding to such a question. One process might be, “If Alice rather than Bernard leads party X, but nothing else changes, then I would vote for X.” However, another process might be, “If Alice rather than Bernard leads party X, then she will change its policies on tax, trade unions, and defense. If these things happen then I will vote X.” This second thought process presupposes the sort of indirect effects discussed in Chapter 1, and so the question does not help us to isolate the unique effect of the leader's personality. In short, it is difficult to extract meaning from responses to seemingly simple survey questions.¹⁹ From the perspective of, say, a member of Parliament deciding whether or not to support the incumbent leader, the sorts of hypothetical survey questions outlined above can be informative and provide possible guides to action. Yet from the perspective of a political scientist trying to estimate the direction and magnitude of leadership effects, such evidence is of more dubious value.

The consensus that leaders' personalities matter is often self‐reinforcing. Journalists assume that personalities matter and concentrate on them when reporting politics. What is more, they commission opinion polls on voters' evaluations of the leaders. Because so much media coverage is devoted to leaders' personalities and polls, voters, politicians, and commentators alike assume that personalities must matter. This consensus, however, excludes academic political scientists. The long‐established view in British political (p. 75 ) science was that leaders counted for little in elections: what mattered was what the parties stood for, not who stood for the parties. Analysts of voting behavior emphasized the formation and impact of party loyalty. Most voters, they argued, coped with the complexity and remoteness of politics by forming an enduring psychological attachment to a particular party.²⁰ This party identification was rooted in the enduring features of their everyday lives and, once established, rarely changed. Party identification heavily conditioned voters' responses to politics in general, such that most voters rarely formed attitudes or made judgments at odds with their long‐established identification. Labour identifiers, for example, overwhelmingly preferred Harold Wilson to Sir Alec Douglas‐Home as prospective prime minister in 1964 simply because he was “their” leader. The causal arrows, it was assumed, ran from parties to leader evaluations rather than the other way round. Moreover, in those rare instances where preferences relating to party and leader diverged, party usually trumped leader. The small minority of Conservative identifiers who happened to prefer the Labour leader overwhelmingly voted Conservative; similarly, the small minority of Labour identifiers who happened to prefer the Conservative leader overwhelmingly voted Labour.²¹ Party identification, rooted in enduring group loyalties, was a far more reliable guide to voting behavior than the ephemeral appeal or nonappeal of the current party leader.

To ram the point home, political scientists often cited those elections when the party with the more popular leader nonetheless lost the election—those of 1945, 1970, and 1979—as evidence that leaders counted for next to nought alongside the records, policies, and images of their parties.

Until the 1980s political scientists regarded the issue as settled and proceeded to ignore it.²² This assumption in turn limited the data that were collected on voters' assessments of leadership traits. Indeed, the main vehicle for research on voting behavior in Britain, the British Election Study, contained very few items that specifically referred to the party leaders. Little attempt was made to determine which personality traits or broad categories of traits were (p. 76 ) of greatest theoretical relevance.²³ “The popular lore about the role of party leaders in British elections and the bulk of the political science literature on the subject [were] thus like ships that [passed] in the night.”²⁴ The main corpus of work on voting behavior simply ignored an explanation of voting behavior that those who were not political scientists regarded as being obvious.

However, by the mid 1980s the conventional academic wisdom about the electoral insignificance of party leaders began to wane. More studies, adopting a variety of analytic strategies, were undertaken. Bean and Mughan formulated a thought experiment strategy (“How would people have voted if the Labour and Conservative parties exchanged leaders but everything else about them remained the same?”) to estimate the effect of leaders' personalities on voting behavior and concluded that leaders' personalities could have a very considerable effect on vote decisions.²⁵ Yet research by Crewe and King based on a similar approach produced far more modest effects.²⁶ The formal popularity functions approach has produced similarly discrepant results. For example, Clarke and Stewart concluded from their analysis of monthly opinion poll data that leader evaluations are a primary determinant of party choice,²⁷ whereas Sanders and his colleagues, using the same data, claim that the impact of a change of party leader on that party's electoral support rarely lasts more than a month or two.²⁸ Improved‐prediction strategies, based on (p. 77 ) regression analysis of single surveys, have similarly failed to produce a consensus: Stewart and Clarke concluded that in the 1987 election “public reactions to the leaders had sizeable effects on electoral choice,” ²⁹ whereas Bartle, Crewe, and King, found that “the addition of further information about voters' evaluations of specific leadership traits adds little to our ability to predict how any given individual will vote.”³⁰ Anthony King's introduction to this volume provides further detail about the inability of political scientists to agree on the true impact of British party leaders' personalities on voters' party choice.

There is thus nothing like consensus in political science about the role of party leaders in motivating the voting choices of individual electors. Interestingly, however, there is a substantial measure of agreement on the broader—and politically more germane—question of the net effect of individuals' leadership preferences on overall election outcomes. The almost universal view is that, whatever effects party leaders' characteristics may or may not have on individuals, these effects, at least in Britain, are only very seldom both on such a large scale and so skewed in their direction as to determine which party actually wins.

For example, Graetz and McAllister in their study of the October 1974, 1979, and 1983 elections concluded that, although the personalities of the party leaders made some difference to individual voters' decisions in 1983, they made no net difference to the overall outcome of any of the three elections.³¹ The contribution of leaders to election outcomes, they say, is typically “more marginal than decisive.”³² Stewart and Clarke do not dispute the same point as regards 1987.³³ As regards 1992, Crewe and King conclude that, if leadership effects helped anyone, it was the losing Labour Party rather than the winning Conservative Party; logic dictates that these effects cannot have been decisive in determining the overall outcome.³⁴ With respect to the 1997 general election, Bartle, Crewe, and King claim that Blair's personal ascendancy over Major (which undoubtedly existed) made only a modest net contribution to Labour's victory, which was on a prodigious scale.³⁵

Only two of the eleven British general elections fought since 1964 (and for which survey data are available) stand out as elections in which leaders' (p. 78 ) personalities may have been decisive. One was 1964 itself, when the Labour leader, Harold Wilson, was vastly preferred by voters to the Conservative leader, Sir Alec Douglas‐Home. The other was the election of February 1974 when the Labour leader, still Harold Wilson, was preferred by a wide margin to the Conservative leader, Edward Heath. Both elections were distinctive for their exceptionally close result. In 1964 only 0.7 percentage points separated the two main parties and Labour secured a majority of just four seats; and in February 1974 a mere 0.8 percentage points separated the two parties, and neither secured a parliamentary majority. Crewe and King are not alone in suggesting that, at these two unusually close elections, leadership preferences on their own may well have tipped the balance; but they are not alone either in concluding that “while leadership effects exist and may on occasion be electorally decisive, they are seldom on a large scale and are not decisive very often.”³⁶

British political science has been slow to assess the influence of leaders' personalities on voting choice and election outcomes and has generally, although not always, reached negative conclusions. In the following section, we revisit the issue by examining the apparent influence of leaders' personalities in a recent general election: that of 1997. We focus on this general election for several reasons. First, the 1997 British Election Study comprised three separate studies: the cross‐sectional study (XBES), the campaign study (BECS), and the British Election Panel Study (BEPS). We can, therefore, use these studies to gauge just how sensitive our estimates of the impact of leadership traits are to different data‐sources. Second, our experience suggests that many people had strong intuitive beliefs about the effect of leaders at that election; clearly many people strongly believed that comparative evaluations of Major and Blair played a major role. Third, preliminary experiments with vote models for earlier elections have highlighted major data omissions in earlier studies.

The Effect of Leaders' Personalities in 1997

Caveats: The Importance of the Models' Assumptions

The models outlined in Figure 3.1 are used to estimate the effect of a wide range of variables both on individual vote decisions and on the aggregate election outcome in 1997. However, all the estimates derived from such models are subject to a degree of uncertainty that arises from the models' assumptions.⁴⁰ By far the most important of these assumptions is the imputed causal order (p. 80 )

Fig. 3.1 Assumed Casual Order In Vote Models

among the explanatory variables. Those variables that are assumed to be located early in the causal sequence will be allocated some explanatory credit for their assumed effect on other explanatory variables located between them and the vote decision.⁴¹ It follows that, if these assumptions are violated, then the assumed prior variable will have allocated to it some explanatory credit that properly belongs to something else. The models outlined in Figure 3.1 contain two important assumptions: (1) that the causal arrows flow from long‐term predispositions (party identification and ideological positions) to short‐term or contemporary political preferences (policy preferences and evaluations of political objects), and (2) that, while party and leadership images are caused by the same variables, they do not in turn “cause” each other.

Early studies of voting behavior suggested that most voters formed a psychological attachment to a party. It was assumed that this identification was enduring and was modified only in response to large social changes or slowly accumulating political experiences. However, there is increasing evidence that party identification—at least as traditionally measured in Britain by the BES—is itself responsive to many of the very short‐term factors, such (p. 81 ) as evaluations of the economy, that it was hitherto assumed to cause.⁴² Controlling for party identification when assessing the effect of these variables may, therefore, conceal part of their causal impact, since party identification has already changed in response to that variable. We are concerned that the traditional measure of party identification may well conceal some effect of contemporary political variables, and so we engage in some sensitivity testing to assess the apparent effect of leadership evaluations controlling for ideological positions only and omitting the—potentially flawed—measure of party identification. These findings are then reported in footnotes to this chapter in order not to distract from the main lines of the argument.

Similarly, the relationship between party images and leadership images is an ambiguous one. It could be argued that voters largely view the world in terms of partisan stereotypes and, therefore, that party image causes leadership image. This assumption is challengeable since we find that party and leadership images do diverge. Alternatively, it could be argued that because party leaders represent relatively enduring elements in the political environment (typically remaining in place for between seven and ten years) and because they have extensive influence over party organization, party images bear the imprint of their leaders. Our own view, for what it is worth, is that the precise relationship between leadership and party images cannot be fully determined given the limited data available. This is why we prefer, as the Americans say, to “take the fifth” on this issue. Our bloc recursive model assumes that party and leader images are located at the same stage within our model. However, we again engage in some sensitivity testing to assess any potential biases. These, too, are reported in the footnotes to this chapter.

Apart from concerns about causal structure, we are also exercised by the problem of omitted‐variable bias. Ideally our data sets need to be as comprehensive as possible if we are to arrive at valid estimates of the effect of leadership traits. However, all three surveys omit a number of theoretically important prior variables, such as voters' ideological positions, perceptions of policy differences between the parties, retrospective evaluations of governmental performance or prospective evaluations of the parties' likely performance in office.⁴³ Unfortunately, not one of the three BES data sets contains (p. 82 ) items that record voters' evaluations of the Conservatives' performance in office—variables that might be thought to condition evaluations of both John Major and Tony Blair. It is obvious that the omission of variables like these means that we run some risk of omitted‐variable bias. Yet by using all available component parts of the BES we should be able to provide an indication of the likely magnitude of leadership effects in 1997.⁴⁴

Another element of uncertainty in our analysis is the assumption that the effect of explanatory variables is uniform across the various subgroups of the electorate. In reality, the effect of leadership traits may well vary between, say, those with high and low levels of political awareness or between those with extreme and moderate ideological positions.⁴⁵ While these issues are interesting, we regard them as being of somewhat secondary importance to our main goals and we do not explore them here. A final source of uncertainty relates to our use of ordinary least squares (OLS) rather than logistic or probit models.⁴⁶ Again, although we recognize that the statistical assumptions embedded within OLS models are problematic, we regard these issues as being of somewhat secondary importance compared with the problems of causal order.

Having briefly examined the inherent limitations of our models, we proceed to estimate the effect of leadership traits in 1997 using all the data available to us from the British Election Study.

(p. 83 ) The Relationship Between Leaders' Personalities and Vote in 1997

By the time of the 1997 general election John Major and Tony Blair had been leaders of their parties for seven and three years respectively. The Conservatives' surprising victory in April 1992 was attributed to voters' doubts about Labour's taxation policies and overall economic competence but also to the personal appeal of Major, the then new Conservative leader, and to deep‐seated skepticism about Neil Kinnock's qualifications to be prime minister.⁴⁷ Six months later, the standing of both the Conservative government and Major as prime minister had plummeted. Formal econometric modeling of aggregate opinion poll data suggests that two events contributed above all to the massive lead that Labour established and sustained through to the 1997 election (and after).⁴⁸ The first was “Black Wednesday” in September 1992, when Britain hurriedly withdrew from the Exchange Rate Mechanism (ERM). The humiliating circumstances and manner of Britain's withdrawal destroyed the Conservatives' reputation for economic competence and reopened their internal divisions on Europe, which then dominated the rest of the parliament.⁴⁹ In the wake of the ERM disaster Labour enjoyed a clear lead in the polls of around 17–20 percent, having been level‐pegging with the Conservatives before. The second crucial event was the election of Blair as leader of the Labour Party in July 1994. In the following five months, Labour's lead over the Conservatives leapt to an unprecedented 33 points and barely subsided from that level until shortly before the election.

The electoral impact of Blair's election as Labour leader could be thought of as powerful evidence of a leadership effect. However, as Anthony King stresses in Chapter 1, Blair's assumption of the Labour leadership produced other consequences: the abolition of Clause 4 of the Labour Party constitution, which ditched Labour's commitment to public ownership; the further weakening of Labour's links with the trade unions; renewed and strenuous efforts to shed its tax‐and‐spend image; and a much tougher position on crime.⁵⁰ All these changes, but in particular the abolition of Clause 4, represented the abandonment of conventional socialism, and all were pushed by Blair against considerable opposition within his party. These changes had a major indirect effect on the outcome of the election but are not what most (p. 84 ) commentators mean when they talk about Blair's impact on the vote. The remainder of this chapter therefore focuses exclusively on the direct effects that can be uniquely attributed to the rival leaders' personalities.

The British Election Cross‐Section Study (XBES)

The XBES offers the best evidence currently available on leadership effects. It comprises a large sample (N 2,906), asks more questions than any other survey, and has particularly good coverage of voters' partisan predispositions.⁵¹ However, as a postelection survey it may be subject to a degree of pro‐Labour rationalization. It has other quirks too. It does not contain any retrospective evaluations of governmental performance, nor prospective evaluations of party performance. These variables may influence both evaluations of leaders' personalities and the vote. Their omission is likely to inflate estimates of the effect of leadership traits on the vote.

Moreover, most theories of leadership effects suggest that it is comparisons of the rival party leaders' traits that matter.⁵² Unfortunately, in 1997 respondents were asked to evaluate the specific traits of only one leader, Tony Blair. This omission is slightly mitigated by the inclusion of two general evaluations of party leaders. In the first question respondents were asked: “Who would make the best prime minister, Tony Blair, John Major, or Paddy Ashdown?” In the second, respondents were asked: “How good a job do you think that [John Major/Tony Blair] is doing/would do as Prime Minister?” We constructed a simple synoptic measure of the two leaders by subtracting evaluations of Blair from those of Major. We assume that responses to both these general questions are influenced by voters' assessments of specific leadership traits. Our models allocate these variables to the penultimate bloc preceding the actual vote decision.⁵³

(p. 85 ) Table 3.1 displays respondents' evaluations of Blair across six individual traits in the summer of 1997. It can readily be seen that evaluations of the young Labour leader were almost uniformly positive. Over 80 percent rated him favorably in terms of strength, caring, decisiveness, and the ability to listen to reason. The only slight question marks against him related to whether he stuck to his principles and would keep his promises. However, even here many respondents merely raised a question mark by responding “don't know” rather than by giving a negative response. On the “best prime minister” question, Blair enjoyed a massive 42‐point lead over Major (60 points to 18 points).⁵⁴ On the face of it, therefore, Blair's overwhelming popularity at the time of the 1997 election must have contributed something to the result. But did it?

We answer this question by presenting a series of tables that outline the effect of leadership traits on both individual voters and the aggregate election outcome. In our models, the vote is scored +1 if Conservative and −1 if Labour. All the explanatory variables are scored from +1 (the most pro‐Conservative opinion) to 1 (the most pro‐Labour opinion), with 0 representing a neutral or balanced opinion. In the following tables, we focus on the effect of leadership traits alone in order to avoid unnecessary distractions.

Table 3.2 reports our findings from the cross‐sectional data and suggests that not one of the six specific traits outlined in Table 3.1 had a visible impact on the vote in 1997. All the p‐values exceed our—statistically generous—p < 0.1 criteria and are statistically insignificant. Only the general evaluations of party leaders—the “best prime minister” and the “synoptic evaluation” item—distinguish Conservative and Labour voters, once controls are applied for prior variables. However, in both cases, the strong bivariate relationships documented in the first column are severely reduced once controls are applied

The coefficient of 0.09 reported in Table 3.2 for the “best prime minister” item represents the effect of a one‐unit movement in this explanatory variable on the vote. We designate this the apparent total effect (ATE) for that variable. Thus, controlling for prior variables, the effect of moving two units—or preferring Major (+1) to Blair (−1)—is to increase the predicted vote by 0.18 (2 × 0.09) and to make a Conservative vote more likely. The ATEs for party identification (0.83) and socialist laissez faire positions (0.27) are substantially larger.⁵⁵ This suggests that, once we know a voter's social characteristics, enduring partisan predispositions, policy preferences, retrospective evaluations of national conditions, and assessments of the parties, the addition of further information about their judgments of the party leaders adds little to our ability to predict how any given individual will vote.

What effect, if any, did those judgments have on the overall election result? To estimate the scale of the effect we need to combine information about the effect of the “best prime minister” items on individual voters with additional information about the extent to which opinion about the “best prime minister” is skewed in one party's favor. In 1997 voters overwhelmingly thought that Tony Blair would make the best prime minister. This pro‐Labour advantage is reflected by the mean of −0.37 for this variable in Table 3.2. However, our adoption of a multistage model implies that some portion of this mean value may be attributable to the already pro‐Labour rather than pro‐Conservative score on prior variables. This raw mean must, therefore, be adjusted to assess just how much more (or less) pro‐Labour it was than we could have expected given our knowledge of prior variables. This implies that

To estimate the impact of people's judgments of who would make the best prime minister on the election result, we multiply the ATE (0.09) by the adjusted mean score for the variable (−0.09). The resulting figure (−0.0081) represents our estimate of that variable's contribution to the aggregate election outcome.⁵⁷ Among Conservative and Labour voters in the XBES, 19 percent more claimed to have voted Labour than Conservative in the election (hence the mean score of −0.19). Since Labour received some 3.9 million more votes than the Conservatives and enjoyed what was in fact an 11.9 percentage point lead on election day itself we can calculate precisely how many votes Labour gained as a result of favorable evaluations of Blair. Our model suggests that Labour gained an additional 166,263 votes or just 0.5 percentage points as a result of direct leadership effects.⁵⁸ This amounts to 262 votes per constituency, enough to have tipped only four Conservative seats into Labour's hands.

However, Model 2 in Table 3.2 shows that, if we prefer the synoptic evaluation of leadership (which compares separate judgments of Blair and Major) as our summary indicator of leaders, it has a slightly stronger effect on the election result. Both the ATE (0.12) and the adjusted mean (−0.20) for this variable are larger and so make a bigger contribution—of −0.0240—to the overall result. In this case, therefore, our model suggests that Labour gained an additional 492,631 votes or 1.5 percentage points because of more favorable evaluations of Blair. This is equivalent to 777 votes per constituency, sufficient to deliver fourteen Conservative seats to Labour. Had Major and Blair been evaluated equally favorably, Labour's lead would have been cut from 11.9 to 10.4 points. These effects are significant but far from enough to alter the election result. It appears that the source of Labour's victory lies somewhere other than in evaluations of leaders' personalities.

It must be noted, however, that sensitivity testing suggested that omitting party identification makes an appreciable difference to our findings. When this (p. 88 ) is done, one of the specific leadership traits (keeping promises) becomes statistically significant. The ATE of 0.05, when combined with the adjusted mean of 0.63, makes a substantial contribution of −0.0265 toward Labour's plurality.⁵⁹ Equally, assuming that leader images were causally prior to party images also substantially alters our findings. When we do not control for party image, the “sticks to principles” item becomes statistically significant and makes a substantial (−0.0315) contribution to Labour's plurality.⁶⁰ These findings emphasize the sensitivity of our findings to specific assumptions about both measurement and causal order. However, we believe that—on the whole—the estimates contained in Table 3.2 are more plausible. Although others will undoubtedly disagree, we take the results of our sensitivity tests as representing the extreme upper‐limit estimate of leadership effects.⁶¹

The British Election Campaign Study (BECS)

However, relying on cross‐sectional data alone to estimate the effect of leaders' personalities would clearly be foolhardy, particularly when we have only noncomparative evaluations of the personality traits of one leader. To confirm our results, we examine evidence contained in the second wave of the BECS, which asked respondents about both party leaders. This study is far smaller (N = 1,592) than the cross‐section and unfortunately contains only one measure of voters' prior partisan predispositions: party identification as measured in 1996.⁶² It does not measure retrospective evaluations of the (p. 89 ) Conservative government's performance and contains few measures of respondents' policy preferences. However, unlike the cross‐section, it does include items on prospective party competence. Moreover, voters were asked to evaluate the party leaders before they voted, so the risk of postelection rationalization is reduced.

Table 3.3 displays respondents' evaluations of both Blair and Major as reported in the second wave of the BECS study for all voters. The figures for Blair are not wildly out of line with those in Table 3.1 but are a little less impressive. The slight differences suggest that the postelection XBES study did capture both some degree of rationalization and something of the honeymoon that invariably follows an opposition election victory. The figures for Major suggest that he, too, was regarded as caring and as a man who stuck to his principles, although his net scores were considerably less favorable than Blair's. Moreover, the majority of respondents considered Major “not capable of being a strong leader” and “not decisive”—judgments that are easy to understand given the evident divisions within the Conservative Party during his premiership. Blair's lead over Major in terms of these traits ranged from 24 points for “sticking to principles” to as much as 72 points for “decisiveness.”

Table 3.4 summarizes the effect of the leadership traits and best prime minister item on the choice between Conservative and Labour using the BECS data. In this case, the bivariate (uncontrolled) relationships between specific traits and the vote are even stronger, presumably because BECS data capture the fact that in this survey the measure of personality traits is a product of judgments of both party leaders, not just Blair.

Again, however, the relationship between evaluations of the party leaders and vote is markedly reduced once controls are applied for prior variables. The traits “capable of being a strong leader,” “caring,” and “decisive” all achieve statistical significance at the p < 0.1 level, but the fourth trait “sticks/does not stick to his principles” did not achieve significance and was dropped from Table 3.4. However, the ATEs are, once again, small when compared with that for party identification (0.83 even though this latter variable is measured one

Yet, intriguingly, our findings suggest that—once allowance is made for the already pro‐Labour distribution of prior variables—voters' general evaluations of who would make the best prime minister actually result in a pro‐Conservative advantage (+0.15). This reduces the contribution of leadership effects to the aggregate outcome by 0.0285, suggesting that, although many voters thought Blair caring, decisive, and strong, assessments of John Major's ability to do the job of prime minister continued to benefit the Conservatives!⁶⁴ Leadership effects as a whole thus contributed 0.0174 or 1.7 percentage points to Labour's recorded plurality of 22 percentage points.⁶⁵ Yet again, leadership effects cannot be ignored in the 1997 general election, but they were hardly decisive. Had Major and Blair been evaluated equally favorably, Labour's majority would have been cut from 11.9 to 11.0 points, altering the outcome in just four seats.

The British Election Panel Study (BEPS)

Our final parcel of evidence comes from the 1992–97 BEPS. This data set contains evidence on voters' evaluations of prospective party performance and— (p. 91 ) at least in principle—can be used to unravel complex issues of cause and effect.⁶⁶ Yet, like most panel studies, it suffers from serious problems of panel attrition. Indeed, the final sample we have at our disposal has an N of just 1,573, compared with one of 2,855 for the first wave in 1992, and the sample underestimates Labour's actual plurality in 1997. Moreover, the BEPS survey contained the least number of relevant policy items of any of the three surveys, raising strong concerns about omitted variable bias. Added to that, there is an unknown element of panel conditioning that—if present—can distort the estimates of causal impact.

These caveats should be borne in mind when interpreting Tables 3.5 and 3.6, which analyze voters' evaluations of Blair and Major in the BEPS final wave. In this case, respondents were asked to judge the two leaders in terms of their “extremeness” or “moderation” and whether they were “good for one class” or “good for all classes” as well as in terms of their capacity for “strong leadership” and for “keeping promises.” Blair's lead on the “capable of being a strong leader” item is far higher here than in the case of BECS. The reason for this is not clear, but it may be partly the result of rationalization, since the 1997 interview took place after the election.

Only two variables are statistically significant: whether the leaders “keep their promises” and general evaluations of the leaders (in this case measured by the synoptic item, since BEPS did not include the “best prime minister” question). In both cases, the strong bivariate associations are substantially reduced once controls are added for prior variables; once again, the pro‐Labour bias is considerably smaller. Nevertheless, this analysis suggests that leadership effects contributed −0.0621 to Labour's plurality of −0.15. In other words, in the absence of voters' evaluations of the two party leaders, Labour would have lost 1,600,000 votes and its lead would have been cut by as much as 6.9 percentage points.

These are the strongest aggregate effects that we find. However, it is worth underlining that the BEPS is the least comprehensive data set available and the omission of causally relevant variables almost certainly inflates the effect of evaluations of leaders' personalities on the vote. On balance, therefore, we are more comfortable with estimates from the two other surveys, especially as they are in line with previous studies of the 1997 general election.⁶⁷

Summary and Conclusions

The idea that many voters are influenced by their impressions of leaders' human qualities is profoundly important. It influences whom parties select to be their leaders, how they campaign, and the very nature of government itself. Leaders who are thought to be an electoral asset are likely to have more authority than those thought to be a liability. In the wake of New Labour's 1997 landslide, Hugo Young, for example, wrote:

This is a government in thrall to its own disproportionate triumph on May 1st and to the leader who produced it. Its collective membership permits him to run it as a personal fiefdom, consulting here and there with selected colleagues, running the show through an inner cabinet, not all of whose members belong to the real thing or have any other base than a Blair familiar. The Cabinet has taken further giant strides into the desert of irrelevance towards which Mrs. Thatcher propelled it. Nobody these days even talks about the Cabinet as a centre of power, or its meetings as occasions where difficult matters are thrashed out between people whose convictions matter to them.⁶⁸

But is voting behavior influenced by voters' impressions of the party leaders? Given the large number of assumptions required to estimate these effects, together with the data limitations explored above, our answer must necessarily be cautious. However, we believe that the best evidence from 1997 is that the (p. 93 ) effects of the party leaders' perceived personal traits were small. Given our information about a voter's social background, partisan predispositions, policy preferences, and evaluations of national conditions, their assessments of the attributes of party leaders added very little to our ability to predict how they voted. The seeds of Labour's landslide victory in 1997 were to be found much further back in the “funnel of causality”: in the events of September 1992 and the creation of New Labour. Similarly, preliminary analysis suggests that Labour's second landslide victory in 2001 was the result of a buoyant economy and the Conservative Party's failure to reassure voters that they would look after the public services rather than voters' markedly more favorable evaluations of Tony Blair than William Hague.⁶⁹ This is not to deny that the two party leaders—Blair in particular—had substantial indirect effects on the vote, indirect effects that can be partly attributed to aspects of their personalities.

The belief that leaders' personalities influence voters is deeply ingrained in British politics. Journalists believe it. Politicians believe it. Even voters themselves believe it—at least of other voters.⁷⁰ Unfortunately, no one cares what political scientists believe—only what they can demonstrate.⁷¹ Those who have adopted an improved‐prediction strategy have demonstrated that, in Britain at least, leaders have not had much of an impact on election outcomes net of prior variables. These conclusions are supported by evidence from other research that has adopted the alternative thought‐experiment strategy discussed in the introduction. The findings, summarized in Table 3.7, suggest that there have been only two recent general elections in Britain (1979 and 1983) where the aggregate vote shares received by the parties were more than marginally influenced by evaluations of the party leaders.⁷² And there are only three instances (the knife‐edge elections of 1964, February 1974, and 1992) where evaluations of party leaders may have “decided” the outcome, in the sense that the impact of party leaders was greater than the winning margin.⁷³ Moreover, the estimates on which Table 3.7 is based represent the upper limits of leadership effects since the researchers were unable to apply sufficient controls for other variables. Leadership effects may well matter, but not by very much and not very often. Our analysis of the 1997 election, (p. 94 ) thus perpetuates the long and persistent tradition of skepticism among British political scientists about the actual difference that party leaders' personalities make to the way people vote and to the outcomes of elections. Moreover, nothing revealed so far about the 2001 election gives us grounds for changing our view.

There will doubtless be some readers who remain convinced that leaders must have more direct effects than we suggest and who will assert this in the teeth of the evidence we offer. We have experienced at first hand the stunned disbelief, bordering on hostility, of a non‐academic audience on being told that the impact of Blair's and Major's personalities on the 1997 election was negligible.⁷⁴ We believe, however, that it is for others to demonstrate, rather than merely to assert, that the image of party leaders exerts a powerful and direct effect on voting.

Our conclusions hold only for the election we have studied, although they broadly repeat the conclusions of studies of previous British elections using different analytic strategies. These conclusions are radical and counterintuitive. By and large, and with relatively few exceptions, the personal appeal to the voters that some democratic politicians are said to possess does not in fact lead to the success of themselves or their party; the success of themselves or their party leads to their appearing to have personal appeal to the voters.

We do not declare that leaders' personalities have no effect on aggregate election outcomes at any time. It is quite possible to imagine circumstances in which leaders' personalities might be crucial in determining both individual vote decisions and the aggregate outcome. We can easily imagine a future election in which partisan loyalties and clear differences of policy and values between the major parties are much weaker and where the perceived personalities of the party leaders do play a much larger role in determining the outcome. But Britain does not appear to have reached that point yet.