The 2000 U.S. Presidential Election: Can Retrospective Voting be Saved?
by MORRIS FIORINA, SAMUEL ABRAMS, and JEREMY POPE
British Journal of Political Science, April 2003, volume 33, issues 2, pages 163-187
Abstract
According to a portrait of elections wide]y held in academic political science, election outcomes depend on the fundamentals', especially peace and prosperity. Al Gore's election showing in 2000 runs counter to the preceding interpretation of elections. Objective conditions pointed to a comfortable victory, if not a landslide, but Gore's narrow popular vote margin fell well below the expectations held by many political scientists. This article attempts to account for Gore's under-performance via detailed analyses of National Election Studies surveys. We find that Gore's often criticized personality was not a cause of his under-performance. Rather, the major cause was his failure to receive a historically normal amount of credit for the performance of the Clinton administration. Secondary contributors were the drag of Clinton's personal affairs and Gore's decision to run to the left of where Clinton had positioned the Democratic party. Quite possibly these three factors are logically related: failure to get no rmal credit reflected Gore's peculiar campaign, which in turn reflected fear of association with Clinton's behaviour.
Full Text
In the wake of the 1998 Monica Lewinsky scandal John Zaller argued that the movement of President Clinton's job approval ratings during and after the scandal had serious implications for theoretical accounts of American mass behaviour. After plunging when the scandal broke, Clinton's approval ratings recovered, then rose to a level higher than before the scandal, where they held to the end of his term. The American public seemingly resisted every attempt by the media or by Republicans to frame the president's behaviour as anything more than a personal sexual impropriety. According to Zaller,
The tradition of studies on economic and retrospective voting, which maintains that the public responds to the substance of party performance, seems strengthened by the Lewinsky matter. On the other hand, the tradition of studies that focuses on the mass media, political psychology, and elite influence ... seems somewhat weaker ... However poorly informed, psychologically driven, and 'mass mediated' public opinion may be, it is capable of recognizing and focusing on its own conception of what matters. (1)
Zaller approvingly cites Key, Kramer, Mueller, Fiorina, Rosenstone and Brody, scholars who view public opinion and electoral politics as driven by the 'bottom line' -- especially peace and prosperity -- the so-called 'fundamentals' of common political parlance. (2)
'Bottom line' scholars of our acquaintance read Zaller's article with some satisfaction -- but not for long. In November 2000, for the fifth time in a decade, an American election jolted prevailing scholarly interpretations. (3) For 'bottom line' scholars the 2000 election should have been a runaway victory for Al Gore. Indeed, at the American Political Science Association meetings held just two months before the election a panel of seven forecasters gave Gore from 53 to 60 per cent of the two-party popular vote, with an average prediction of 56 per cent. (4) But one hardly needed a statistical forecasting model to predict a Gore victory in 2000. The United States had enjoyed a lengthy economic boom that had seen unemployment fall to levels previously considered impossible without sparking inflation, and burgeoning budget surpluses had replaced 'two hundred billion dollar deficits as far as the eye can see.
Elections are not only about economics, of course, but the news on other fronts was positive as well. The crime rate had been falling for several years, welfare reform was an apparent success and American soldiers were not in harm's way. Moreover, there was no significant divergence between traditional objective indicators and subjective perceptions as there had been in 1992; presidential approval hovered around 60 per cent, economic confidence was at 70 per cent and satisfaction with the state of the country was above 60 per cent. (5) Nor, despite the subsequent economic slowdown, did any of these indicators begin to decline until after the election. (6) Under such conditions the election was a foregone conclusion, or so the portrait of elections advanced by 'bottom line' scholars suggested. (7)
But somehow Al Gore failed to receive the full benefit of these favourable circumstances. To be sure, he won the popular vote by a tiny margin, but his share of the two-party vote (~ 50.2 per cent) was 3-9 points lower than forecast, making the election a cliff-hanger when many forecasters and pundits expected a landslide. (8)
Thus, whether the standard is sophisticated statistics or common sense, the puzzle is why Al Gore was not swept into the White House by a wave of grateful retrospective voters. This article attempts to account for Gore's under-performance. In the next section we discuss various hypotheses that have been offered. Some of these reflect arguments that economic, retrospective or 'bottom line' voting occurred less commonly or less strongly in 2000 than in earlier elections. Other hypotheses identify factors in 2000 that might have overridden the operation of traditional retrospective voting. In the third section of the article we bring American National Election Studies (NES) data to bear on the hypotheses. Some we can evaluate directly, others only indirectly, but the data support three hypotheses about Gore's under-performance: (1) he failed to receive a historically typical allotment of credit for the performance of the Clinton administration; (2) Gore abandoned Clinton's centrist strategy and waged a neo-popul ist campaign ('I will fight for you!'); (3) voters' disapproval of Bill Clinton as a person finally showed up in their behaviour. In the conclusion we discuss the possibility that these findings are causally connected. Interestingly, we find little indication that the personalities of the candidates played any important role in the outcome.
WHY DID AL GORE FALL SHORT OF EXPECTATIONS?
In the aftermath of the election journalists, pundits, politicians and political scientists offered explanations for Gore's feat of snatching defeat from the jaws of victory. These can be incorporated under four general hypotheses. The first asserts that the 2000 election was different in one or more important respects that weakened the impact of traditional 'bottom line' considerations such as the state of the economy and the administration's performance. The second and third hypotheses allow that traditional 'bottom line' considerations might have operated normally -- in Gore's favour -- but identify other anti-Gore factors that overwhelmed the effects of the so-called fundamentals. Finally, there is a more complex hypothesis that partly overlaps the others. Some have suggested that a factor peculiar to 2000 -- 'Clinton fatigue' -- produced what one might call 'moral retrospective voting' that directly overwhelmed the fundamentals, indirectly weakened their impact through Gore's campaign choices, or both. W e provide a brief elaboration of the logic and circumstantial evidence underlying these hypotheses.
Hypothesis 1: The Fundamentals Were Weaker in 2000
According to the works cited by Zaller, elections have a heavy retrospective component, reflecting the electorate's judgement of whether incumbents have performed acceptably on the fundamental dimensions of social welfare -- maintaining prosperity and keeping the peace. (9) By that standard the Clinton administration was a solid success. The most common explanation for Gore's underperformance is that the electorate did not give him a normal allotment of credit for the administration's performance, although the reasons offered differ.
Variant 1: Vice-presidents do not get credit. Some argue that if Clinton had been able to run for re-election, he would have won easily, but as vice-president, Gore was not able to claim credit as effectively as Clinton could have done. Consistent with this argument, Lewis-Beck and Tien report a significant (post-hoc) improvement in their 2001 forecast after incorporating an interaction between economic conditions and the absence of an (elected) incumbent president in the race. (10) Other forecasters report contrary results, however. (11)
While commonly asserted, the premise of the vice-presidential variant of Hypothesis 1 conflicts with scattered poll evidence suggesting that Clinton would not have won again. (12) Furthermore, a vice-president running in 1988 apparently did not make that election look particularly different from others. Incumbent Vice-President George Herbert Walker Bush did succeed a popular president, forecasting models based on the 'fundamentals' were not far off the mark, (13) and detailed survey analyses concluded that Bush won largely on the strength of voter evaluations of the Reagan administration. (14)
Variant 2: 'What have you done for me lately?' During the campaign various commentators suggested that the good economic times had gone on for so long that voters were taking them for granted. Economic voting studies traditionally presume that voters' time horizons are short, (15) and there are suggestions in the economic voting literature that voters are more sensitive to changes in economic condition than to their levels. Thus, the very fact of sustained prosperity may have worked against the Democratic candidate -- too many voters remembered nothing but the present good times and consequently discounted them. (16) This variant of the hypothesis suggests that economic retrospective evaluations should be a less important determinant of the vote for Gore than for candidates not running after such a sustained period of prosperity.
Variant 3: Entrepreneurs and Alan Greenspan got the credit. In early 1999 a Rasmussen poll asked, 'Who is most responsible for the strong economy, American businesses, Federal Reserve Chairman Alan Greenspan, President Clinton, Congressional Republicans, or Congressional Democrats?' (17) Forty per cent of the respondents credited business and 22 per cent Greenspan, compared to 12 per cent for Clinton, and a total of 25 per cent for Clinton or either congressional party. Thus, Americans in 2000 may simply have believed that the condition of the country was not strongly affected by who held the presidency. Such perceptions are reasonable enough. The country is in transition to a more global, information-based economy, and the political system appears to be mostly in a reactive mode. Fiscal policy ceased to be an important tool for managing the economy decades ago, and in the years leading up to the election monetary policy czar Greenspan received a great deal of positive, if not adulatory, press.
Unfortunately, there is no long time-series of identical survey questions about responsibility for the economy. Lacking these we cannot directly test the hypothesis that Americans today believe that politicians affect the bottom line less than Americans of earlier eras believed they did. There is a short NES time-series that provides some support for the hypothesis, but it was discontinued in 1996. (18) Again, the indirect implication of this variant of the hypothesis is that economic evaluations were less important in the decisions of voters in 2000 than in earlier elections.
Variant 4: Al Gore did not try to take credit. Given that forecasting models do not include campaign variables, forecasters are often accused of claiming that 'campaigns don't matter', On the contrary, most forecasters as well as proponents of 'bottom-line' politics more generally hold a more complex view in which the campaigns are themselves dependent variables. (19) At least at the presidential level both candidates have access to plenty of money and talent; thus, they will run campaigns as good as the background conditions permit. Running a good campaign is easy for candidates who have peace and prosperity behind them, as did Ronald Reagan in 1984 and Bill Clinton in 1996. But even with talent and money, running a good campaign is far more difficult if the opposition has peace and prosperity on their side, a predicament that faced Walter Mondale in 1984 and Robert Dole in 1996. Popular accounts often assume that the quality of the campaign is purely a function of the brilliance of the candidate and the qua lity of his personality. For example, Republicans have charged that Dole lost the 1996 election because he ran a terrible campaign. No doubt Dole should bear some of the blame, but the reverse causal argument is probably equally plausible: his campaign was terrible because he had no chance of winning.
Some attribute Gore's under-performance in 2000 squarely to a poor campaign strategy. In particular, until late in the campaign Gore seemed to make little attempt to claim credit for the successes of the Clinton administration. Given the conditions that prevailed in the country one would have expected a campaign organized around classic retrospective themes like 'You've never had it so good!' 'Eight more years!' and 'Are you better off today than you were eight years ago?' Instead, Gore declaimed that 'This is an election about the future not the past', in effect throwing away his trump cards. Thus, voters may not have given Gore credit for peace and prosperity because -- unusually and surprisingly -- he made little attempt to claim credit. (20)
In sum, the general hypothesis that the operation of the fundamentals was muted in 2000 rests on a number of different bases. While our analyses shed some light on these underpinnings of the basic hypothesis, we cannot conclusively differentiate among them. All predict the same general finding: the weight that American voters placed on the economy and the performance of the administration was lower in 2000 than in previous elections.
Hypothesis 2: Al Gore Was an Especially Unattractive Candidate
Perhaps the fundamentals exerted their normal effect in 2000 but the boost they gave to Gore was overwhelmed by some negative consideration -- such as Gore's allegedly unattractive personality. This is the second most common explanation for Gore's failure to win the presidency. During the campaign there was much discussion of Gore's personal characteristics and mannerisms. He was depicted as arrogant and a know-it-all. He continually reinvented himself. He was the kind of boy who reminded the teacher she had not assigned homework. And he was a 'serial exaggerator'. So widespread were these negative characterizations that Gore's persona in the first debate was the subject of a famous Saturday Night Live skit that his own campaign advisers made him watch before the second debate -- apparently with counter-productive consequences. (21) In contrast, George Bush was portrayed as a likeable fellow whose comfort in his own skin made Gore's unnaturalness all the more striking. Thus, a common explanation for Gore's po or electoral showing is that a significant portion of the electorate simply did not like him, bottom-line considerations notwithstanding.
Most forecasting models contain no candidate variables, (22) and some research indicates that the importance of candidate personality is exaggerated. (23) After all, the United States did elect Richard Nixon twice, as well as George Bush the father. These men had some strengths but few would argue that their personalities were among the more noteworthy. (24) Still, survey studies of voting behaviour show candidate evaluations to be an important determinant of voting decisions as well as a major source of change across elections. (25) To be sure, candidate evaluations incorporate experience, policy stands and personal capabilities as well as personal attractiveness, but with a little effort we can construct reasonable measures of the latter from the NES. We will use these to determine whether voter evaluations of Gore the person were abnormally low.
Even if evaluations of Gore's personality were not extraordinarily negative, however, there are variations of Hypothesis 2 identifying other paths that allow for Gore's persona to produce his poor showing.
Variant 1: Oprah and MTV. Some argue that personality is more important today than it was in the past. American presidential candidates now appear on television talk shows, such as The Oprah Show, The Tonight Show with Jay Leno and The Late Show with David Letterman, and non-traditional cable networks such as MTV, leading some to fear that American politics has entered a new era where the candidates are evaluated more as entertainers than as executives. This line of argument suggests that even if evaluations of Gore were not unusually low, voters weighted them more heavily when deciding how to vote. Such a change in voting behaviour should show up in unusually large coefficients for candidate characteristics in voting models estimated with 2000 data as compared to those for earlier elections.
Variant 2: Media bias. During the campaign some Democrats complained that the media were unfairly hard on Gore, dwelling almost obsessively on his perceived personal shortcomings, particularly his exaggerations. Indeed, a preliminary study of media coverage during October 2000 found that Gore received more negative coverage than Bush, although other studies report contrary findings. (26) This media variant of the personality hypothesis gains credence from evidence that forecasting models went awry when the senior Bush was defeated in 1992 in part because the media created a perceptual picture of the economy that was significantly more negative than the real economy. (27) The media variant of the Gore personality hypothesis suggests that support for Gore should have been lower among avid consumers of the media, other things being equal.
Hypothesis 3: Al Gore Was Too Liberal
Somewhat overlooked in popular and journalistic discussions of Gore's under-performance were the ideological positions staked out by the candidates. The reason probably lies in the common survey finding that voters tended to agree more with Gore than with Bush on the issues -- education, health care, the environment and so forth -- generally issues on which the Democrats have an advantage in public opinion. Thus, taking account of the candidates' positions apparently only adds to the puzzle of Gore's under-performance.
However natural, the preceding interpretation may be a misreading of the electorate's perceptions of the candidates. Many scholars and pundits think that Bill Clinton's signal political contribution was to bring the Democratic party back to the centre -- to convince traditionally-oriented middle-class voters they had nothing to fear from electing a Democratic president. In contrast, regardless of his positions on specific issues, many commentators saw Gore as running further to the left than had Clinton, a charge common among sympathizers of the Democratic Leadership Council. (28) Particularly in view of Zaller' s finding that ideological moderation is an important correlate of the post-war presidential vote, we think it worthwhile to take a close look at voter perceptions of the candidates' positions in 2000. (29) Perhaps Al Gore lost support because of his ideological positioning. In the VNS exit poll, for example, more voters opined that Gore was 'too liberal' (43 per cent) than that Bush was 'too conserva tive' (34 per cent).
Hypothesis 4: The 2000 Election Result Was an Anomaly Caused by Clinton Fatigue
Why might Gore have decided not to claim credit for the performance of the Clinton administration and run as a neo-populist instead? During the campaign there were suggestions in the media that psychological factors were at work. Gore wanted to be his own man (recall his widely-noted assertion to that effect in his acceptance speech at the Democratic convention), and win or lose on his own with no help from Clinton, (30) perhaps because Clinton had displaced Al Gore Sr as the object of Gore Jr's resentment. (31) Psychologists are better qualified to evaluate such hypotheses, but such interpretations run counter to the common perception that Gore would do anything to be president. (32)
A more political explanation is the belief in Gore campaign circles that disapproval of President Clinton's personal behaviour was a serious threat to the vice-president's propects. (33) Going into the election the one negative element in the public's perception of the state of the nation was the belief that the country was morally on the wrong track, whatever the state of the economy or world affairs. (34) According to some insiders, anything done to raise the association between Gore and Clinton would have produced a net loss of support--the impact of Clinton's personal negatives would outweigh the positive impact of his job performance on support for Gore. (35)
Thus, Hypothesis 4 suggests that a previously unexamined variable played a major role in 2000 -- the retiring president's personal approval. Analyses of presidential voting customarily include the job approval of the president, but personal variables typically are limited to those of the contending candidates. So, when the president is not himself a candidate, his personal approval is not included in the analysis.
One reason why analysts may not think to include such a variable is that personal and job approval for most presidents track reasonably closely, so that job ratings stand in for personal ratings. But a high correlation is not invariably the case. Jimmy Carter's 1980 ratings were the converse of Clinton's 2000 ratings: high personal and low job ratings. Reflecting these observations we estimate 2000 voting models that include a measure of President Clinton's personal standing as well as his job approval. If the former is a significant determinant of the vote, it may have overridden the traditional performance variables. More subtly, if coefficients on the performance variables are lower than in past elections, that may have been the price of the Gore campaign's attempt to minimize the negative effect of Clinton's personal ratings.
ANALYSIS: THE 2000 PRESIDENTIAL ELECTION
The empirical analysis that follows utilizes data from the American National Election Studies. (36) This compilation of academic surveys contains a wealth of data on the past thirteen US presidential elections (1952-2000), although item availability limits our analysis to the period 1972-2000 for the most part. The NES surveys are far more detailed than most commercial surveys and many important variables are measured in the same way over time, permitting temporal comparisons that are impossible when using commercial surveys and exit polls.
In evaluating the hypotheses we consider both the distributions of such important variables as presidential approval, ideological distance and candidate evaluations, and the weights Americans assigned to such variables when casting their votes in 2000. Specifically, we estimate voting models that include party identification, (37) perceptions of the economy, presidential performance, ideology and candidate evaluations. (38) Tables 1-3 report the estimations, where for ease of exposition and interpretation only significant coefficients are reported. (39) For reasons explained below, the vote equations abstracted in these three tables differ in their treatment of ideology. The equations in Table 1 do not include an ideological variable (the least missing data), the equations in Table 2 include the voter's ideological self-placement (more missing data), and the equations in Table 3 include a measure of candidate ideological proximity (the most missing data). (40) Inclusion of the proximity measure takes some exp lanatory power away from the Republican candidate variables, the economic condition variables and presidential performance, but we think our major conclusions are robust, whether we rely on Table 1, 2 or 3. We say more about the specific variables included in the equations and how they were operationalized as we evaluate each substantive hypothesis.
Hypothesis 1: Did the Fundamentals Operate Differently in 2000?
Yes, it appears that they did. Although the direct link between economic perceptions ('economy better') and the vote was on the high side of the historical range, the link between the far more important presidential performance variable and the vote was noticeably weaker in 2000. Figure 1 graphs the relationship, showing the average coefficients taken from the models reported in Tables 1-3. Even with myriad control variables, the association between performance ratings and the vote is large and highly significant in all elections, but the coefficient in 2000 is lower than in the other seven elections. Absent any measure of ideology (Table 1) or in the presence of ideological self-identification (Table 2), the impact of presidential performance in 2000 is significantly (p <0.05) lower than in the three elections in which an incumbent won a landslide re-election (1972, 1984, 1996), while the 95 per cent confidence intervals for the elections of 1976, 1980, 1988 and 1992 overlap that for 2000. Thus, the evidence for the vice-presidential hypothesis is mixed. Votes for Gore were significantly less closely tied to administration performance than votes for Clinton in 1996, but not significantly less so than the votes for incumbent presidents Carter in 1976 or Bush the elder in 1992. However, setting statistical significance aside, the 1988 and 2000 coefficients are the six smallest of the twenty-four reported in Tables 1-3.
Whether statistically different or not, the coefficients suggest voting differences that are quite important. In particular, simulations using the equations reported in Tables 1 and 2 indicate that had Gore received the same boost from Clinton's performance in 2000 that the senior Bush did from Reagan's performance in 1988, he would have received about 8 more percentage points of the vote. (41)
Figure 1 does not speak directly to the other variants of Hypothesis 1, but some additional analysis does. Pondering President Clinton's high approval ratings some commentators wondered whether they were somehow artificial, reflections of popular resentment of congressional Republicans or a salacious media, for example, rather than genuine approval of substantive performance. On the contrary, Clinton's job ratings reflect economic substance about as much as those of previous presidents. Table 4 reports statistical associations between voters' perceptions of the economy and their evaluations of the president's job performance, controlling for voter partisanship. (42)
The relationships are quite variable -- Nixon's ratings in 1972 and Carter's in 1980 probably reflected foreign affairs more than presidential ratings did in other years, and some insignificant coefficients probably reflect too few cases - it would be hard to argue that the economy had improved in 1980, although a few respondents did. But perceptions of economic conditions are associated with presidential performance judgements in each of the elections, with negative perceptions somewhat more consistently important than positive ones. The 2000 coefficients are a bit smaller than some earlier ones in the series, but the probabilities derived from them do not suggest a noticeably smaller association between economic conditions and presidential performance ratings in 2000. For example, the third column of the table shows the estimated difference in the probability that an independent voter approves of the president's performance depending on whether she thought the national economy had improved or deteriorated. In 2000 the estimated impact of 'better' versus 'worse' economic perceptions on independents' presidential performance ratings is 0.22; the average of the preceding five estimates using the same item is 0.26. Thus, voters in 2000 did not think that politicians were irrelevant to economic performance, nor had they experienced good times for so long that they could imagine no alternative. Americans recognized that times were good and gave Clinton a normal allotment of credit for them. (43)
In sum, the fundamentals still mattered a great deal in 2000, but not as much as in earlier elections. Clinton's job ratings reflected the state of the economy to about the same degree as the ratings of previous presidents, but his job ratings affected the presidential vote less than job ratings did in all the earlier elections. Here alone is a sufficient explanation for the forecasters' Waterloo in 2000: their longitudinal models assume the fundamentals exert a constant effect overtime, an assumption that apparently failed in 2000. While we hesitate to assert conclusively why it failed, the process of elimination points the finger towards Gore's campaign, although there is some weak evidence that vice-presidents get a weaker boost from the fundamentals than presidents.
Hypothesis 2: Was Al Gore Especially Unlikeable?
Estimating the independent impact of the candidates themselves is the most difficult task in this analysis. The reason is that the candidates are at least in part empty vessels into which voters can pour their hopes and fears. That is particularly true of newcomers to the national scene such as George W. Bush. Moreover, voters' judgements of the candidates are functions of the positions the candidates adopt and the records they have compiled. But since most of the critical commentary on Al Gore focused on his personality, or more broadly, on a set of negative personal attributes independent of his government experience and his policy stands, we need valid measures of the candidates' personal attributes in order to investigate this explanation for Gore's sub-par showing in 2000.
To construct such measures we turned to a set of open-ended questions about what voters like and dislike about the candidates. The NES series includes these questions for every election since 1952. (44) Voters answer in their own words and their responses are recorded verbatim, to be coded into broad categories at a later date. We took the list of coding categories (which had grown to more than 700 by 2000) and divided them into two. The first category - personal attributes - includes all comments about the candidates' inherent characteristics - intelligence/stupidity, arrogance/humility, sincerity/insincerity, honesty/dishonesty, morality/immorality and so forth. (45) The complementary category is more heterogeneous and includes all the responses that seem in any way acquired by the candidates over the course of their careers -- their experience, records, issue positions, group allegiances and so forth. For want of a better term we refer to this dimension by the acronym ERIG -- experience, records, issues an d groups. Given our interest in voters' feelings about the candidates as people, a clearly-defined first category is our primary concern. (46)
Figures 2-5 summarize a fascinating history of Americans' judgements of the presidential candidates in the second half of the twentieth century. We briefly discuss the entire sequences both because of their inherent interest and as a way of putting the 2000 race in a broader context.
Figure 2 compares public perceptions of Republicans on the personal and ERIG dimensions. (47) There are two clear patterns. First, on balance voters generally feel positively about Republican candidates: only Goldwater in 1964 and Reagan in 1980 received a net negative rating. Secondly, in ten out of thirteen elections Americans rate Republican candidates less positively on the Personal dimension than on the ERIG, or Experience etc., dimension. In fact, in election after election Americans feel either neutral or ambivalent about Republican presidential candidates as people, scoring them very close to zero. (48) Surprisingly, this was true even for Eisenhower in 1956 and Reagan in 1984, both now warmly remembered in personal terms. As these observations indicate, pundits' judgements after the fact often differ from what Americans felt in the months leading up to the election. In particular, it is striking that aside from Goldwater the Republican viewed most negatively on the personal dimension was Reagan in 19 80. Popular history tends to forget that many Americans were quite nervous about entrusting this sabre-rattling former movie actor with the presidency.
And what of 2000? Bush's reputed likeability does show up in the data. By a very small margin Bush is the highest rated Republican in the half-century on the personal dimension -- higher than Eisenhower in 1956 and Reagan in 1984. (49) In absolute terms the population still rates him very close to neutral, but for a Republican, he was well liked. Note however, that on the ERIG dimension, Bush is lower than any Republican except Goldwater, although still close to the neutral point.
Figure 3 reports analogous figures for Democrats. Here, patterns are not apparent. Democratic candidates are evaluated negatively more often than Republicans, and they are rated more highly as people about as often as they are rated more highly on the ERIG dimension. Overall, McGovern in 1972 was the weakest Democrat in the series, but Bill Clinton in 1996 set the low water mark on the personal dimension. As for Al Gore, he does not particularly stand out. He is rated very near the neutral point in both categories, although on the personal dimension he bests only Clinton in 1996, McGovern in 1972 and Humphrey in 1968. Overall, Gore was not as well liked personally as most of his Democratic predecessors, but there is little indication in the data that Americans found him particularly unlikeable.
Figure 4 rearranges the data to contrast Republican and Democratic candidates on the personal dimension. Americans tend to like Democrats at least as much as Republicans. The biggest Democratic edge occurred in 1980, but although Americans liked Jimmy Carter as a person considerably more than they did Ronald Reagan, that was not nearly enough reason to re-elect him. Similarly, the largest Republican edge occurred in 1996, but Americans did not dislike Bill Clinton enough to elect Robert Dole, whom they disliked too,just not as much. Such findings should reassure those who fear that presidential politics has degenerated into a personality contest. (50)
Figure 4 also shows that John Kennedy did not win in 1960 because of his personality -- Americans liked Nixon about as much. (51) Nor did Dukakis lose in 1988 because voters thought he looked silly riding around in a tank. Perhaps he did, but Dukakis was rated ever so slightly more positively than George Bush the elder. As for 2000, George Bush the younger does have a slight edge over Al Gore, the first time a Republican was rated both positively and higher than the Democrat since Nixon in 1972.
Figure 5 contrasts Republicans and Democrats on the ERIG dimension. Republican candidates have a clear edge in this category, achieving a higher rating in eight elections, and essentially tying in two more. But Al Gore is one of only three Democrats in a half-century to have an edge in this category and the first since Carter in 1976. Conversely, Bush was the first Republican since Nixon in 1968 to be rated clearly lower on the ERIG dimension. (52)
In sum, Americans in 2000 did not register strong feelings about the candidates, either individually (a bare majority gave both Bush and Gore neutral ratings) or in the aggregate. True, Americans did not like Al Gore personally as well as they liked many of his Democratic predecessors or, of greater relevance, quite as much as they liked George Bush. But the differences are small, and -- unusual for a Democrat -- Gore bested Bush on the experience and record dimension.
As for the weights voters gave to candidate qualities, Tables 1-3 show that they were not unusually high in 2000. On the contrary, they were on the low side relative to other elections. Thus, there is no support for the Oprah variant of Hypothesis 2. (53) Nor did we find support for the anti-Gore media hypothesis. We checked to see if high media consumers were more likely to evaluate Gore negatively than low media consumers. They were not. And while evaluations of Gore's person were significantly lower among those interviewed after the first debate, evaluations of his experience, record and positions were similarly lower. (54) All in all, despite the attention it received in the popular press and the late night television shows, it appears that the candidate factor in 2000 was a wash. (55) Gore's personality does not account for his poor showing.
Hypothesis 3: Did Gore Run Too Far Left?
We constructed a general measure of voter distance from the candidates on the standard liberal--conservative scale. (56) Figure 6 plots the comparative closeness of the in-party and out-party nominees to the voting public, where a positive number indicates that the average voter places herself further from the in-party than the out-party. In 2000 the average voter saw Gore as 0.7 of a scale position more distant than Bush. (57) This may not seem like a large number in absolute terms, but it is the largest ideological disadvantage any candidate registered in the eight elections in the series, larger even than Dukakis suffered in 1988 at the hands of the Bush Sr campaign or McGovern suffered in 1972 at the hands of the Nixon campaign. Apparently the laundry list of campaign issues on which voters preferred Gore to Bush -- social security, environment, education and so forth -- did not capture the subjective processes voters used to position the two candidates.
The statistical analyses reported in Table 3 estimate the weight voters give to each point of relative ideological distance to be small, but significant. The coefficients are generally similar from election to election, but, interestingly, the 2000 coefficient is the smallest in the series. Thus, while Gore's perceived ideological distance is the largest in the series, the weight voters attached to it is the smallest, resulting in less electoral damage than otherwise would have occurred. Still, simulations based on the Table 3 equations suggest that had Gore been as close to the average voter in 2000 as Clinton was in 1996, his vote would have increased by approximately 4 percentage points, all else being equal. (58) He still would have under-performed, but almost certainly he would have won the presidency.
Hypothesis 4: Were Americans Tired of Clinton?
Earlier we noted that Gore did not get as big a boost from Clinton's high job performance ratings as the senior George Bush did in 1988 from Reagan's ratings. One possible reason is that Gore ran the wrong campaign. But in order to draw such a conclusion, one would need evidence that a closer association with Clinton would have boosted Gore's showing. There is no way to provide direct evidence -- the campaign was what it was, but the indirect evidence suggests that the Gore campaign did have reason to fear that their candidate might suffer from close association with Clinton.
Our conclusion reflects a closer comparison of the 1988 and 2000 elections. To the equation abstracted in Table 1 we added another variable -- a measure of the personal ratings of the retiring president. Numerous polls have shown that while Clinton's job approval stayed quite high throughout the two years preceding the election, his personal ratings -- never as high to begin with -- were much lower. (59) Unfortunately, the NES studies ask the likes/dislikes questions only about the candidates, not about retiring presidents, so we cannot measure how voters evaluated the retiring presidents in a manner comparable to how they evaluated the candidates. In fact, our options for rating retiring presidents are quite limited.
Although admittedly imperfect, we have utilized a set of so-called 'affect' questions that ask the voter whether the president has ever made her feel angry, hopeful, afraid or proud -- two positive emotions, two negative ones. These items elicit emotional -- even visceral -- reactions to the president and tap people's feelings about the presidents as people. (60) We combine voter responses to give them net scores for President Reagan in 1988 and President Clinton in 2000. (61) Figure 7 contrasts the two.
A majority of the 1988 sample evaluated Reagan positively on this index, with the most positive rating being the one most commonly observed. In contrast, the 2000 sample rates Clinton much lower -- a bit more than a third positive -- with the most positive rating being the one least frequently observed. Did such ratings affect the decisions to support the vice-presidents of these presidents? The first two columns of Table 5 show that the answer is a clear 'Yes'. Over and above the influences already considered, emotional reactions to the retiring president are highly significant in a statistical sense. The estimated weights that voters attached to Reagan's and Clinton's personal ratings are very similar, but Reagan's actual ratings were much more positive. Simulations based on these equations suggest that had Clinton's personal ratings in 2000 been as positive as were Reagan's ratings in 1988, Al Gore's vote would have been approximately 3-4 percentage points higher, all else being equal. (62)
The third column of Table 5 investigates this matter a bit further. A new item on the 2000 NES questionnaire asks the respondent whether, since 1992, the moral climate of the country has got better, got worse or stayed about the same. As shown in Figure 8 the distribution of responses was quite skewed: less than 5 per cent of the sample thought it had become better and 45 per cent thought it had become worse. (63) Adding these sentiments to the equation appears to pick up something beyond that incorporated in Clinton's job and personal ratings and implies that Bill Clinton's legacy exacted a penalty on support for Al Gore. (64)
What of the Gore campaign's contention that had they tried to associate Gore more closely with the economic successes of the Clinton administration, they would have simultaneously linked themselves more closely with the president's moral failings? We have no way of estimating what the trade-off might have been, but the implication of Table 5 is that such fears were not imaginary. Disapproval of Bill Clinton's personal behaviour hurt Al Gore. Still, our personal view is that Gore already was suffering from the fallout of Clinton's behaviour and would have lost no more had he tried to emphasize the positive accomplishments of the administration. Had Gore captured just one more small state -- Arkansas, Missouri, Nevada, New Hampshire, West Virginia -- that he narrowly lost, Florida would have been irrelevant. Even if a closer connection with Clinton had lost popular votes, had it led to victory in one more state it would have meant the presidency. (65)
SUMMARY: AL GORE'S UNDER-PERFORMANCE IN 2000
A hirsute Al Gore re-emerged on the political scene in the summer of 2001, setting off a flurry of speculation about his plans. Judgements about his presidential prospects or lack thereof hinged heavily on the pundit's beliefs about Gore's loss. Did Gore barely top 50 per cent of the two-party vote when a large number of forecasting models had him topping 55 per cent because of his wooden personality? His populist campaign? The drag of Bill Clinton's personal escapades? No one can answer that question with certainty, but the preceding analyses enable us to go beyond punditry and parlour talk. To sum up our results:
Al Gore the person. Of the various explanations for Gore's poor showing, only the one that focuses on negative voter reaction to his personality finds little support in the data. Although Gore was not particularly popular, that alone hurt him little, if at all. Evaluations of both candidates were largely neutral within the population -- both in the aggregate and individually, and there is no indication that voters attached any more weight to personality factors than they usually do.
Ideology. Consistent with the interpretations of the Democratic Leadership Council, Gore's decision to run from the left exacted a vote penalty. Our estimates based on Table 3 suggest that Gore's decision to abandon Clinton's centrist approach and campaign as a born-again populist cost him about 4 percentage points of the vote, easily enough to lose the election, all else equal.
Clinton fatigue. Consistent with the views of the Progressive wing of the Democratic Party, the personal transgressions of Bill Clinton finally came home to roost in 2000. Many Americans who judged Clinton's job performance positively looked on his personal behaviour with distaste, and many of them voted for Bush, perhaps contributing to the unusually weak relationship between the fundamentals and the vote. Holding everything else constant, estimates based on Table 5 suggest that Clinton fatigue, broadly defined, probably cost Gore directly 3-4 percentage points of the vote, easily enough to lose the election.
The fundamentals. Judgements of presidential performance mattered less in 2000 than in earlier elections. The weaker effects do not appear to be a result of the sheer length of the good times, nor of Alan Greenspan getting all the credit. Gore's vice-presidential status may have played some role, but in the end the choices of the Gore campaign seem most prominent. Other things being equal, the weaker relationship made a difference of about 8 percentage points as compared to the 1988 election when another vice-president followed a successful president. The weaker impact of the fundamentals turned a prospective landslide into a virtual tie.
Total. Re-run the election statistically correcting all of Gore's mistakes or misfortunes: Gore receives the same degree of credit for the economy and the administration's performance as G. H. W. Bush got in 1988; Clinton's personal ratings are as positive as Reagan's were in 1988, and Gore is as close to the voters as Clinton was in 1996. If all these conditions hold, then Gore wins the election in a landslide, with 58 per cent of the vote. (66)
Political science forecasting models performed poorly in 2000 because the traditional effects of the fundamentals were muted, because the models omitted an important variable -- Bill Clinton's personal ratings -- that worked in opposition to the fundamentals, and because Gore ran a neo-populist campaign that was not optimal given the views of the electorate. From the standpoint of an election forecaster everything that could go wrong did.
The great question that we cannot answer is why Gore ran the campaign that he did. The most parsimonious explanation of our findings is that the fundamentals were muted because Gore deliberately declined to claim credit for prosperity and chose instead to campaign in a way that minimized his association with Clinton. If so, and his fears of taint from Clinton's sleazy behaviour were justified, then there is little need to revise the forecasting models in any major way. Analysts need only add a presidential personal rating to their models. (Given that personal and performance ratings rarely have the kind of disjunction evident in 2000, a dummy variable for 2000 may suffice.)
By contrast, if future research shows that the effects of the fundamentals were muted in 2000 because voters have shifted away from classical retrospective voting based on national and international conditions towards moral retrospective voting based on character and ethics, then the models will continue to go awry. We think the former possibility is the more likely of the two, but data from elections to come will settle the issue.
[TABLES AND FIGURES OMITTED.]
NOTES
(1.) John Zaller, 'Monica Lewinsky's Contribution to Political Science', PS: Political Science and Politics, 31 (1998), 182-9, p. 186.
(2.) Zaller, 'Monica Lewinsky's Contribution to Political Science', p. 185.
(3.) The 1992 election brought unified Democratic government in the same year that Fiorina characterized Republican-headed divided government as the normal state of affairs in modern American politics (see Morris Fiorina, Divided Government (New York: Macmillan, 1992), p. 2). The Republican takeover of Congress in 1994 upset longstanding interpretations of House elections that presumed a more-or-less permanent Democratic majority (see Gary Jacobson, 'The Persistence of Democratic House Majorities', in Gary Cox and Samuel Kernell, eds, The Politics of Divided Government (Boulder, Cob.: Westview, 1991), pp. 57-84). The 'status-quo' election of 1996 (see Everett Ladd, '1996 Vote, the "No Majority" Realignment Continues', Political Science Quarterly, 112 (1997), 1-28) produced an unprecedented pattern of divided government - the first Democratic president ever elected with Republican congressional majorities, contradicting arguments like Jacobson's that voters naturally preferred Republican-headed divided control over the opposite (see Jacobson, 'The Persistence of Democratic House Majorities', pp. 68-71). The Republican loss of House seats in 1998 undercut one of the hoariest generalizations in American politics - that the party of the president loses seats in the mid-term election. All in all, the past decade has been quite hard on empirical generalizations about American elections.
(4.) Christopher Wlezien, 'Why Gore (Probably) Lost', Slate, 8 November 2001.
(5.) Gallup Organization News Releases, 2000.
(6.) 'Economy,' Public Perspective (November/December 2000), 40-1; and 'Polity Watch', Public Perspective (November/December 2000), 42-3.
(7.) For explanations of and excuses for the 2000 performance of forecasting models, see recapitulations in PS: Political Science and Politics, March 2001, and in the May 2001 issue of American Politics Research.
(8.) Larry Bartels and John Zaller, in 'Presidential Vote Models: A Recount', PS: Political Science and Politics, 34 (2000), 9-20, make the unique argument that the average of forty-eight forecasting models was reasonably accurate in 2000. They reach this conclusion by constructing a class of models that predicts a Gore loss of 2-4 points under the conditions prevailing in 2000. These models that predict a Gore defeat partly balance out the models that predict a comfortable Gore victory. Given that their argument does not pass the straight-face test, we can only surmise that the Bartels and Zaller piece is a brilliantly executed spoof of election forecasting.
(9.) See e.g. V. O. Key Jr. The Responsible Electorate (New York: Vintage, 1966); Morris P. Fiorina, Retrospective Voting in American National Elections (New Haven, Conn.: Yale University Press, 1981).
(10.) Michael Lewis-Beck and Charles Tien, 'Modeling the Future: Lessons from the Gore Forecast', PS: Political Science & Politics, 34 (2001), 21-3. The 'elected' qualification assigns Gerald Ford to the non-incumbent candidate category in 1976.
(11.) Douglas Hibbs, 'Bread and Peace Voting in US Presidential Elections', Public Choice, 104(2000), 149-80; and Brad Lockerbie, 'Election 2000: An Afterthought', American Politics Research, 29 (2001), 307-12, p. 310. Re-estimation of Hibb's Model with the 2000 results included still does not find a significant effect for vice-presidents (analysis by Jeremy Pope and Shawn Treier, Stanford University, June 2001.)
(12.) Renshon cites various polls reporting that majorities would not vote for Clinton if he could run again, and majorities were 'glad to see him go' (Stanley Renshon, 'The Polls: The Public's Response to the Clinton Scandals, Part 2: Diverse Explanations, Clearer Consequences', Presidential Studies Quarterly, 32 (2002), 412-27, pp. 422-3.
(13.) Ray Fair, 'Predicting Presidential Elections: The Polls Versus Fundamentals' (unpublished paper, 9 November 1988). Indeed, it is the senior author's recollection that forecasting models gained prominence precisely because they predicted a Bush win during the spring of 1988 while Dukakis maintained a lead in the trial heats until after the summer nominating conventions.
(14.) Paul Abramsom, John Aldrich and David Rohde, Change and Continuity in the 1988 Elections (Washington, DC: Congressional Quarterly Inc., 1990); and Merrill Shanks and William Miller, 'Partisanship, Policy and Performance: The Reagan Legacy in the 1988 Election', British Journal of Political Science, 21 (1991), 129-97.
(15.) For discussion, see Hibbs, 'The Bread and Peace Model'. Hibbs's results indicate that voter time horizons are longer than usually assumed.
(16.) Consistent with this suggestion, an August Gallup poll reported that 52 per cent of Americans remembered 1992 economic conditions as excellent or good, whereas on average only 12 per cent of Americans queried in 1992 actually reported such positive views. See David Moore, 'Booming Economy No Advantage for Gore', Gallup Poll Release, 16 August 2000.
(17.) Public Perspective (June/July 1999), p. 39.
(18.) The NES item reads 'Would you say that the economic policies of the federal government have made the economy better, worse, or haven't they made much difference either way?' For the elections of 1984-96 the percentage answering 'Haven't made much difference' is 41 (1984), 57 (1988), 53 (1992) and 60 (1996).
(19.) Morris Fiorina and Paul Peterson, The New American Democracy, 2nd edn (New York: Addison, Wesley, Longman, 2002), pp. 305-6.
(20.) Divided control might have weakened the credibility of such a claim, but a majority of the thirteen elections used by the forecasters were instances of divided control and Helmut Norpoth concludes that the president's party bears responsibility for economic conditions whatever die pattern of institutional control. (Helmut Norpoth, 'Divided Government and Economic Voting'. Journal of Politics, 63 (2001), 414-35). Generally, the tendency among scholars who study economic voting in the United States has been to assume that responsibility falls more or less automatically on the party of the president and therefore is constant from election to election. The latter assumption is implicit, of course, in time-series models that estimate a time-invariant coefficient for economic conditions.
(21.) The consensus was that Gore overreacted to critical reviews of his behaviour in the first debate and behaved too passively in the second.
(22.) An exception is James Campbell's model described in 'Polls and Votes: The Trial-Heat Presidential Election Forecasting Model, Certainty, and Political Campaigns', American Politics Quarterly, 24 (1996), 408-33, which includes the Gallup Poll trial heat available at the time of the Labor Day holiday in late August or early September.
(23.) Anthony King, ed., Leaders' Personalities and the Outcomes of Democratic Elections (Oxford: Oxford University Press, 2002).
(24.) In fact, the country came very close to electing Richard Nixon in 1960 when he ran against the far more personable John F. Kennedy.
(25.) Donald Stokes, 'Some Dynamic Elements of Contests for the Presidency', American Political Science Review, 60 (1966), 19-28.
(26.) The Project for Excellence in Journalism monitored 1,149 stories from seventeen news publications, television programmes and websites, reporting that 24 per cent of Bush stories were positive compared to 13 per cent of Gore stories. Fifty-six per cent of Gore stories were negative, compared to 49 per cent of Bush stories (www.journalism.org/publ_research/campaign1.html, accessed 10 July 2001). In partial contrast, Johnston, Hagen and Jamieson report that Gore was treated more positively than Bush on the evening news shows from the Democratic convention until the beginning of October, but coverage was balanced thereafter. See Richard Johnston, Michael Hagen and Kathleen Hall Jamieson, 'Dynamics of the 2000 Presidential Campaign: Evidence from the Annenberg Survey' (paper presented at the 2001 Annual Meetings of the American Political Science Association).
(27.) Marc Hetherington, The Media's Role in Forming Voters' National Economic Evaluations in 1992', American Journal of Political Science, 40 (1996), 372-95.
(28.) E.g., Michael Kinsley, 'The Art of Finger-Pointing', Slate, 31 October 2000. In part, Gore's leftward shift may have been a response to Bradley's challenge from the left in the primaries, but Gore seemed to make less of an effort to return to the centre than Bush did after he lurched to the right to fight off the McCain challenge.
(29.) Zaller, 'Monica Lewinsky's Contribution to Political Science', pp. 185-6.
(30.) Ron Brownstein, 'Gore, Curiously, Fails to Take Credit for Policy Achievements', Los Angeles Times, 17 October 2000.
(31.) Such suggestions crossed the boundary into silliness at times. For example, 'At some level I think he's very angry about the way he was raised. This Oedipal dynamic lies at the heart of the campaign he has run. With big Albert dead, Gore has made Bill Clinton into a father figure, one who while ostensibly wishing him well is also causing him damage', quoted from Jacob Weisberg, 'Why Gore (Probably) Lost', Slate, 8 November 2000.
(32.) James Ceaser and Andrew Busch, The Perfect Tie (Oxford: Rowman & Littlefield, 2001), p. 118.
(33.) While Clinton's job approval ratings held steady at about 60 per cent during the January to August period when the scandal raged, the percentage of the population that believed Clinton was 'honest and trustworthy' declined 16 points to 31 per cent, the percentage believing Clinton 'shares your values' dropped 13 points to 37 per cent, and the percentage believing Clinton 'has high moral and ethical standards' fell 20 points to 19 per cent. See 'As the Scandals Have Persisted, Clinton's Standing on Matters of Integrity Has Plunged', The Public Perspective (October/November 1998), 24-5.
(34.) In the VNS Exit Poll 57 per cent of the respondents said that the country was on the wrong track in terms of its moral climate.
(35.) Personal conversation with Sam Popkin, public opinion analyst for the Gore campaign.
(36.) Data from the National Election Studies is available at www.umich.edu/~nes/.
(37.) Measures of voter party ID have a strong and statistically significant impact in all models estimated in this article; thus, we say no more about it. No one should be surprised that people who say they are Democrats (Republicans) are very likely to report that they approve the performance of a Democratic (Republican) president and generally vote for Democratic (Republican) candidates. As reported in the text, however, even controlling for voter partisanship, factors such as performance ratings, candidate evaluations and ideology have a significant impact on the vote.
(38.) All equations were estimated using CLARIFY (see Gary King, Michael Tomz and Jason Wittenberg, 'Making the Most of Statistical Analysis: Improving Interpretation and Presentation', American Journal of Political Science, 44 (2000), 374-61). Missing data were imputed via AMELIA (see Gary King, James Honaker, Anne Joseph and Kenneth Scheve, 'Analyzing Incomplete Political Science Data: An Alternative Algorithm for Multiple Imputation', American Political Science Review, 95 (2001), 49-69.
(39.) Complete statistics are available on http://www.stanford.edu~mfiorina. Given clear expectations about the signs of the coefficients, one-tailed tests are appropriate. Thus, the probit coefficients listed in the tables have z values greater than 1,6, approximately the 0.05 significance level.
(40.) Use of self-placements rather than proximities increases the average number of observations by about 80 per election. Omitting ideology entirely increases the average number of observations by about 270.
(41.) These calculations assume that the 2000 electorate attached the same weight to presidential approval as the 1988 electorate, holding everything else constant. This conclusion would not follow from Table 3, where the 1988 and 2000 presidential performance coefficients are equal, but as explained below, in that estimation ideological proximity (not included in Tables 1 and 2) puts Gore at a significant disadvantage relative to Bush.
(42.) We begin in 1972 because that is the first year that the NES includes the standard presidential performance item. From 1972 to 1980 the economic performance item referred to the trend in business conditions during the past year. From 1980 to 2000 the economic performance item asked about the condition of the nation's economy during the past year. Both items were included in 1980 and as the table shows the later item seems somewhat more strongly related to presidential performance judgements. Given the near decade-long length of the US expansion, the references to during the last year' seem less than optimal, but the item continues to have considerable explanatory power, even controlling for partisanship.
(43.) A total of 39 per cent of respondents said that national economic conditions had got better, and 17 per cent (mostly Republicans) that conditions had got worse.
(44.) The series consists of four questions. It begins, 'Now I'd like to ask you about the good and bad points of the two major candidates for President. Is there anything in particular about (Republican candidate) that might make you want to vote for him?' The interviewer then probes for up to five responses. The negative variant of the question follows next: 'Is there anything in particular about (Republican candidate) that might make you want to vote against him?' Analogous questions enquire about the Democratic candidate. while a respondent conceivably might offer a total of twenty (four times five) comments, in practice large majorities confine themselves to one or two comments in response to each question.
(45.) For the numerical codes placed in the 'personal' category, see Appendix A to the website version of this article.
(46.) The authors conducted two independent codings, one in early 2001 before the 2000 study was released and another in the autumn of 2001. The agreement was substantial. In the first coding ninety-three of the approximately 700 codes were chosen as personal qualities. In the second ninety-seven codes were selected. Eight codes new in 2000 were added and four general codes selected earlier were dropped ('people have confidence (don't have confidence) in him', and 'just like (dislike) him'.)
(47.) The points in the graphs are the averages across all voters who were asked the questions that year. Each positive comment gets plus one point and each negative comment minus one point. The voter's score is the sum of the positives and negatives.
(48.) A zero score for a voter can reflect either no comments at all or an equal number of positives and negatives. Similarly, an average near zero across all voters can reflect any mix of neutral feelings among some voters along with a near balance of voters who like and dislike the candidate.
(49.) Admittedly, this finding seems implausible. One complication in using the likes/dislikes questions is that voters surely respond with whatever considerations are most easily accessible in their memories. For a successful incumbent like Eisenhower or Reagan, performance or policy responses are readily at hand, but for a newcomer like Bush, voters have little to go on besides impressions of his personality.
(50.) Over the time series there is a dramatic increase in the proportion of responses that fall in the personal category, from about 15 per cent in the 1950s to more than 60 per cent in recent elections. At first glance, this appears to be striking evidence in support of those who claim that the modem media, especially television, have perverted politics. Television covers politics in highly personal terms, and voters apparently have begun to take the cues and do likewise when asked about the candidates (see Sam Abrams, 'The 21st Century American Voter: Image, Information, and Presidential Appraisal in the Digital Age' (Honors thesis, Stanford University, 2002). As the examples in the text suggest, however, voters are perfectly willing to defeat candidates whom they like and elect those whom they do not like. Moreover, the statistical analyses reported in Tables 1-3 show no increase in the weight voters give to personal qualities when making their choices.
(51.) 'If the eventual account given by the political histories is that Nixon was a weak candidate in 1960, it will be largely myth' (Stokes, 'Some Dynamic Elements of Contests for the Presidency', p. 22).
(52.) These findings are consistent with those of Johnston, Hagen and Jamieson, 'Dynamics of the 2000 Presidential Campaign' (see their Figure 10), who conclude (using different data and measures) that Bush outscored Gore on character, whereas Gore outscored Bush on competence.
(53.) As an aside, regular viewers of Oprah (an item included on the 2000 NES survey) actually gave higher personal ratings to Gore than did non-viewers.
(54.) The tracking polls showed that Gore's lead over Bush began to drop before the first debate. The fact that both his personal and ERIG evaluations were lower among those interviewed after the first debate suggests that his somewhat obnoxious performance was not the critical factor.
(55.) Clearly there is much more that can be said about this subject, but our preliminary investigations have found little reflection in the data of various popular claims about the candidate factor in 2000.
(56.) Proximity measures are calculated in the usual way from seven-point liberal--conservative scales on which respondents place themselves and the candidates. The scale is calculated by taking the absolute value of the voter's distance from the incumbent party candidate and subtracting the absolute value of the voter's distance from the out-party candidate (a positive value indicates the voter is closer to the out-party). Proximity measures are problematic on two grounds. First, because scores are calculated only for voters who place themselves and both candidates, a significant number of voters are omitted from the estimations. Secondly, the questions offer an incentive for the voter to rationalize her answer: if she likes Al Gore and intends to vote for him, is she not tempted to place him closer to herself than Bush whatever the reality? In recognition of these problems, Tables 1 and 2 report two sets of alternative models, one using voter self-placements alone instead of proximities, thus lessening the rationalization possibility, and another omitting ideology entirely, greatly increasing the number of usable cases.
(57.) In 2000 a NES question format experiment resulted in less than half the sample having ideology and proximity scores comparable to those in previous years. The subsamples were randomly selected, but the smaller n (approximately 425 in the estimations that include the proximity measure) results in noisier data for 2000.
(58.) We obtain this estimate by substituting the means of the 1996 candidate proximity values within each of the seven party-identification categories for the actual 2000 values.
(59.) See footnote 33 above.
(60.) Evidence that responses to these items tap something beyond voter partisanship and performance evaluations comes from the fact that the coefficients on party identification, presidential approval and economic conditions are only marginally diminished by the addition of this measure to the equation.
(61.) The president received +1 for each positive emotion indicated by a voter, -1 for each negative emotion. So, each voter's score lies between -2 and +2 inclusive. Again, zero can reflect either a balance of positive and negative emotions or no emotions.
(62.) This estimate is obtained by substituting the means of 1988 Reagan ratings within the seven party-identification categories for the actual 2000 Clinton ratings, reversing the scale so that strong Democrats in 2000 liked Clinton as much on average as strong Republicans in 1988 liked Reagan, and so on for the other categories.
(63.) Sceptics might object that this item is merely a proxy for whether the voter likes Clinton. Note, however, that the item does not mention the retiring president. Moreover, analysis indicates that the item behaves as one would expect if taken at face value. Controlling for voter partisanship and Clinton job ratings, belief in moral deterioration relates significantly to various indicators of religiosity and moral traditionalism. While Democrats naturally are less likely to believe that the country is morally rotten than Republicans and independents, insignificant interactions between party identification and these measures indicate that religion and traditional morality had an across-the-board effect. For details, see Appendix B to the website version of this article.
(64.) To be fair, Bill Clinton's personal transgressions probably were not the only source of beliefs that the moral state of the country was in decline.
(65.) Ironically, it might have led to a reversal of the actual verdict -- a popular vote victory for Bush, but an Electoral College majority for Gore.
(66.) Taken all together the considerations do not predict a historically unprecedented victory for Gore because the effects are not additive. Declining marginal impact is inherent in probit models. Moreover, the effect of ideological proximity combined with presidential performance (Table 3) only equals the independent impact of performance (Tables 1 or 2).