The President's Effect on Partisan Attitudes
Jacobson, Gary C., Presidential Studies Quarterly
In rapid succession the 2006, 2008, and 2010 elections reaffirmed the common belief that a president's standing with the public has a powerful impact on the electoral fortunes of his party's candidates. (1) The connection between the president's job approval ratings and aggregate election outcomes has been thoroughly researched and appears firmly established, but the processes forging the connection have received comparatively little attention and remain somewhat obscure. The guiding premise of the research reported here is that the president influences election results in substantial part by shaping popular attitudes toward and beliefs about his party and by altering its appeal as an object of personal identification.
In a previous article that focused largely though not exclusively on the G. W. Bush administration (Jacobson 2009a), I documented several ways in which presidents affect their party's popular image and support. This article broadens the analysis by adding observations from the Obama administration as well as an expanded set of data from the Clinton and Bush administrations. I also perform additional tests of the president's influence on affective and cognitive reactions to his party using American National Election Studies (ANES) data, and I examine data covering the last 11 administrations to gauge the president's influence on perceptions of his party's competence in handling national problems. I find that presidents strongly affect how their parties are evaluated, perceived, and adopted as objects of identification, which, in turn, helps to account for the president's influence on the electoral fates of his party's candidates. I also find that opinions of Barack Obama have, so far, had an even larger effect on attitudes toward his party than did opinions of his predecessors, including G. W. Bush, with consequences that speak to an important controversy in the literature on mass partisanship.
Why Presidents Matter to Parties
Intuitive reasons for expecting a president to influence the public standing of his party come easily to mind. The president is his party's dominant public face. His words and actions articulate and define his party's current principles and objectives. Judgments about his competence in managing domestic and foreign affairs inform assessments of his party's competence in such matters. The components of a president's supporting coalition, and the interests he favors while governing, help to define the party's constituent social base and thus appeal as an object of individual identification. People's affective reactions to the president, whatever their source, inevitably color their feelings about the other politicians in his coalition. Every president thus shapes public attitudes toward his party as well as beliefs about who and what it stands for and how well it governs when in office; insofar as the party label represents a brand name, the president bears prime responsibility for the brand's current image and status.
These observations suggest several testable hypotheses:
1. Evaluations of the president's party's will reflect his standing with the public; popular ratings of the president's party will vary with popular ratings of the president's job performance.
2. Affective reactions to the president's party will vary with affective reactions to the president.
3. Perceptions of the president's ideology will shape perceptions of his party's ideology.
4. Evaluations of the president's job performance will influence judgments about his party's capacity to handle national problems.
5. Aggregate partisanship (or macropartisanship)--the proportion of Republicans or Democrats in the electorate--will vary with the popular success or failure of the president.
6. Individual assessments of the president's job performance will affect individual partisan identities.
This sixth hypothesis was strongly supported by analysis of changes between elections in presidential approval and party identification in ANES panel studies covering the Gerald Ford, G. H. W. Bush, Bill Clinton, and G.W. Bush administrations (Jacobson 2009a, 207-09). As yet no comparable panel study is available for the Obama administration, so this hypothesis is not tested further here. The focus, then, is on the first five.
The president's party is not evaluated in isolation from its rival, but it is not obvious, a priori, if or how evaluations of the president should affect attitudes toward the opposition party. It is certainly conceivable that opinions of the president would influence opinions of his opponents under conditions of partisan conflict: the better (worse) people think of a president, the worse (better) they might think of the party opposing him. But the contrary is also conceivable, with people viewing both parties positively when times are good or during rally events such as the one provoked by the terrorist attacks of September 11, 2001, and negatively when they are disdainful of the government's performance, as were most Americans after observing the debt ceiling showdown in the summer of 2011. The effects of a presidency on opinions of the opposition, then, should depend on the particular circumstances shaping opinions of the president.
Changes in mass and individual partisanship are, as usually measured, zero sum, so mass identification with rival party will be affected (in the opposite direction) as much by the president's standing as is mass identification with the president's party. But it is also conceivable that the president's effects on mass partisanship are asymmetrical; he may, for example, alienate some of his own partisans but without thereby rendering the other party more attractive. I explore this possibility in the section on macropartisanship.
Testing Hypothesis 1: Opinions of Presidents and Parties
To test the first hypothesis, that the president's standing with the public shapes his party's standing with the public, I created two data sets. One was assembled from the NBC News/Wall Street Journal polls, which regularly ask respondents to evaluate various leaders and institutions, including the president and the two major parties, as very positive, somewhat positive, neutral, somewhat negative, and very negative. For analytical purposes, I use as variables the proportion of respondents with very or somewhat positive views of the president and of the parties; the results are not sensitive to this choice. (2) The second data set was assembled from a variety of surveys that asked both the standard presidential approval question and a question about whether the respondent had a favorable or unfavorable view of each parry. (3) The analysis focuses on the proportion approving of the president's job performance and the proportion with a favorable view of each party.
Detailed graphic perspectives on the relationship between opinions of the president and his party for the Clinton, G. W. Bush, and Obama administrations appear in the Appendix (Figures Ala through A3f). The figures depict the across-time and cross-sectional relationships between the aggregate measures of presidential and party evaluations in the two data sets. For the across-time figures using the second data set, I use lowess smoothing to summarize the trends from the rather noisy data (a product of the diverse survey sources). A third pair of figures displays the cross-sectional relationships broken down by the respondent's party identification for each administration. (4)
Tables 1 and 2 report the OLS regression estimates of the cross-sectional relationships between aggregate evaluations of presidents and their parties depicted in the figures. In every case we observe the expected positive relationship, in most cases with the slope quite precisely estimated (p < .001). On average for all respondents, a 10-point difference in positive opinions of the president predicts a 4.7-point difference in positive opinions of his party; a 10-point difference in presidential approval predicts a 4.4-point difference in party favorability. The relationships are not identical across administrations, however; the coefficients for the Obama administration are larger than those for the Clinton and Bush administrations, although the difference is statistically significant (at p < .05) only in the "approve/favorable" data set. (5)
Of course, partisanship powerfully influences evaluations of the parties as well as of presidents, but we also observe clear linear relationships between opinions of presidents and their parties within partisan categories (both phenomena are readily apparent in the "e" and "f" figures for each administration in the Appendix). Although the regression results suggest some variation in these relationships among party identifiers (for example, Democrats display largest coefficients during the Obama administration), the overall pattern indicates that partisans of all persuasions adjust their opinions of the president's party as they adjust their opinions of the president in roughly the same way.
At the aggregate level, then, public evaluations of presidents and their parties varied together in all three administrations. The individual-level connection between opinions on presidents and parties was also very tight. For example, about 90% of respondents who had positive or negative views of the president and his party expressed consistent opinions, viewing both positively or both negatively (Table 3). Similarly, individual-level consistency in presidential approval (disapproval) and a favorable (unfavorable) opinion of the president's party was high, averaging 82% among respondents with an opinion of both (Table 4). Even independents, who tend to hold parties in rather lower regard than presidents, were consistent in their opinions 86% of the time on the first set of questions, 76%, in the second set.
Not surprisingly, consistency among partisans diverges at the extremes: It is highest among the president's partisans when the president is most popular and among opposition partisans when the president is least popular; and it is lowest among the president's partisans when the president is least popular and among opposition partisans when the president is most popular. Thus, there are limits to both the positive and negative effects of presidents on the popular status of their parties.
Public assessments of the president's party vary positively with assessments of the president. What of the opposition party? OLS estimates comparable to those reported for all respondents in Tables 1 and 2, but with opinions of the opposition party as dependent variables, suggest that the relationship varies with the conditions affecting opinions of the president (Tables 5 and 6). During the Clinton administration, opinions of the Republican Party varied negatively with opinions of Clinton in both data sets, with coefficients of equal magnitude (though with opposite signs) to those for the Democratic Party. A further parsing of the data suggests that the Republicans' attempt to impeach Clinton for lying about his dalliance with Monica Lewinsky drove the relationship between presidential approval and party favorability. Clinton received his highest approval ratings, and the Republican Party its lowest ratings, in late 1998 and early 1999, just as Republicans were trying to impeach him, and it was only from that period onward that we observe a strong negative relationship between approval of Clinton and favorable opinions of the Republican Party (Table 6). For good reason (a booming economy, low unemployment and inflation), most people were happy with the condition of the country during Clinton's second term and did not want him removed from office; the Republican-controlled Congress persisted in trying to expel him anyway, and the party's favorability ratings reflected the unpopularity of the move. However, only before the impeachment period was the incidence of positive views of Clinton negatively related to the incidence of positive views of the Republican Party (Table 5).
A possible explanation for the curious discrepancy between the results using these two measures is that after the scandal broke, people tended to separate their assessments of Clinton's job performance from assessments of him as a person (Jacobson 2000); the proportion approving of Clinton's job performance rose, but the proportion with positive opinions of him fell (see Figures Ala and Alc in the Appendix). A negative opinion of Clinton's character would not necessarily affect opinions of the opposition party one way or the other. (6)
These relationships also varied across the Bush administration. During Bush's first term, evaluations of the Democratic Party were unrelated to opinions of Bush. As I noted in the earlier article, Bush achieved his highest approval ratings following the terrorist attacks of September 11, 2001, when his initial response to the crisis won overwhelming bipartisan support from leaders and citizens alike. Under those conditions, positive views of the president would not be expected to inspire negative views of the opposing party (Jacobson 2009a, 183). During his second term, Bush's ratings fell in lockstep with increasing public disaffection with the Iraq War. Most Democratic leaders had by then become strong critics of the war, and under these conditions, it is not surprising that evaluations of Bush and the Democratic Party would be inversely related.
During the Obama administration, opinions of the Republican Party improved as the president's standing declined during his first three years in office, but the effect is significantly smaller compared to that for the president's own party. More generally, the relationship between aggregate evaluations of the two parties across these surveys is rather weak and varies in sign, indicating that no common dynamic automatically connects the two. For example, the correlations between opinions of the two parties in the "positive" (NBC News/Wall Street Journal) series were -.21 during the Clinton administration, .21 during the Bush administration, and -.18 during the Obama administration.
Testing Hypothesis 2: Affective Reactions to Presidents and Parties
Aggregate data from the three most recent presidencies suggest a strong and consistent relationship between evaluations of presidents and their parties, and a more tenuous and variable relationship between views of the president and the opposition. For another perspective on these relationships covering a longer time period, and to test the second hypothesis, that affective reactions to the president shape affective reactions to his party, I turned to the ANES cumulative data file and two ANES panel studies. (7) The ANES studies regularly ask respondents to rate their feelings toward the president and the parties on a 100-point thermometer scale, with 0 being coldest, 50 neutral, and 100 warmest. (8) Table 7 reports the OLS regressions of the parties' thermometer ratings on the president's ratings, controlling for the respondent's party identification (measured by the standard ANES 7-point scale (9)) for years in which the questions were asked. I estimated the models with year fixed effects, but the results are virtually identical without this control. The results indicate a strong linear relationship between thermometer ratings of presidents and their parties. Ratings of presidents are unrelated to evaluations of the opposing party, which are dominated by the respondent's party ID. These results hold for each election year separately, although the coefficient for each president tends to be largest in years when a president is seeking reelection (results not shown). (10)
These cross-sectional results are replicated in analyses of the ANES panel studies covering the Nixon-Ford (1972-1974-1976) and Clinton (1992-1994-1996) administrations. In the Nixon-Ford panels, respondents were asked to rate "Republicans" and "Democrats," rather than the parties specifically. Nonetheless, the results in Table 8 indicate that respondents changed their affective ratings of the president's fellow Republicans in response to changes in their ratings of the president (controlling for their initial ratings of both and for their current party identification). The president's lagged thermometer also had a significant positive effect on his party's thermometer rating. The estimated relationship is stable across two- and four-year intervals. In contrast, affect toward Democrats was unrelated to affect toward these presidents. The estimated effects of changes ratings of the president on ratings of his party were larger for the 1992-1994-1996 panel study, presumably because the questions referred to the parties specifically rather than to partisans (Table 9); a 10-degree change in Clinton's thermometer score predicts about a 5-degree change in scores for the Democratic Party. Again, changes in feelings toward Clinton had comparatively little effect on feelings toward the opposition party, although all signs on the coefficients are appropriately negative, and one meets the p < .05 standard of statistical significance.
[FIGURE 1 OMITTED]
Figure 1 provides a final perspective on the relationship between thermometer ratings of presidents and their parties by plotting their averages among voters in the 14 ANES studies for which these measures are available. Note that at this maximum level of aggregation, the structure of the relationship is similar (a slope of about 0.5) to that observed at lower levels of aggregation in Tables 7-9. No matter how they are viewed, then, the ANES data support the conclusion that affective evaluations of presidents shape affective evaluations of his party. In contrast, presidents exert little apparent influence on affective reactions to the opposing party.
Testing Hypothesis 3: Perceptions of Party Ideologies
The third hypothesis, that perceptions of the president's ideology will influence perceptions of his party's ideology, is also testable with ANES data. In 16 of the ANES surveys taken since 1972, respondents have been asked to place themselves, the president, and the parties on a 7-point liberal conservative scale. (11) Table 10 displays estimates of the effect of the president's perceived location on his party's perceived location, controlling for the respondent's self-location, party identification, and the interaction between the latter two variables (necessary because partisans, especially Republicans, tend to place the other party and its politicians at more extreme locations than do the party's own adherents). The results indicate that, other things equal, a 1-point difference in the ideological location attributed to the president predicts a little more than a 0.5-point difference in the location attributed to his party. A Democratic president's perceived ideology has no significant effect on where respondents locate the Republican Party, but the more conservative respondents believe a Republican president to be, the more liberal they perceive the Democratic Party to be. (12)
Analysis of the two panel studies recapitulates these results; respondents whose perceptions of the president's ideology change between elections tend to change their perceptions of his party's ideology in the same direction. Conditional on the initial placement of the president and his party, a 1-point change in the location attributed to the president produces about a 0.6-point change in the location attributed to his party (Table 11; the exception is for the 1972-74 period, when the shift was between presidents, from Nixon to Ford): lagged president L/C placement also similar effect on party placement (average slope, 0.56). The effect on perceptions of the opposition party's location was much smaller but consistently in the negative direction; if the president is perceive as moving in one direction, the rival party is perceived as moving on the opposite direction on this scale.
The ANES no longer cover midterm elections, so to include the Obama administration in this part of the analysis, I estimated the models depicted in Table 10 using the comparable data from the Cooperative Congressional Elections Study (CCES) surveys taken in 2009 and 2010. In both studies, placements of the Democratic Party strongly reflected placements of Obama, while placements of the Republican Party varied only slightly with placements of the president (Table 12). Obama appears to have had an even larger impact on assessments of his party's ideology than previous presidents (about 0.75 points for every 1-point difference in the president's attributed location), but because the ANES and CCES samples and procedures are so different, (13) it remains uncertain whether these larger coefficients are attributable to Obama or the survey. The 2012 ANES survey may provide more information on this question.
To conclude this section, Figure 2 plots the mean ideological placements of presidents and their parties for the 16 election years for which ANES data are available. The slope for Democratic administrations is comparatively flat, a consequence of a pair of observations involving Jimmy Carter, who was considered about a 0.5-point more conservative than his party (observations for 1978 and 1980, the two Democratic points furthest to the right on the figure). Otherwise, and taken as a whole, the evidence amply supports the conclusion that perceptions of the president's party's ideology vary directly with perceptions of the president's ideology. (14)
[FIGURE 2 OMITTED]
Testing Hypothesis 4: Party Competence on the Nation's Most Important Problem
To test the fourth hypothesis, that evaluations of the president's performance influence perceptions of his party's competence, I assembled data from 180 surveys, covering presidents from Harry Truman through Obama, that asked respondents what they considered to be the nation's most important and which, if either, party they thought would handle it better. (15) I created a measure of relative party competence for each survey by subtracting the percentage of respondents rating the opposition party as handling it better from the percentage rating the president's party as handling it better. Figure 3 displays the relationship between this measure and presidential job approval. As hypothesized, perceptions of relative party competence vary positively with presidential approval. The figure also shows that respondents have tended to favor the Democrats on this question, for responses to it are deeply colored by partisanship, and the Democrats held the advantage in aggregate partisanship for most of this period (although one considerably diminished since the Reagan administration).
[FIGURE 3 OMITTED]
The regression results reported in Table 13 provide a more detailed depiction of this relationship across presidencies. (16) In every administration, evaluations of the relative capacity of the parties to handle the most important national problem have varied directly with presidential job approval, with estimated regression slopes showing considerable consistency across administrations. A summary model estimating the relationship for all observations (with administration fixed effects) indicates that a difference of 1 point in presidential approval produces a difference of 0.65 points in assessments of the president's party's relative competence. The coefficient is quite precisely estimated, with a 95% confidence interval that ranges from 0.53 to 0.78. The estimated coefficient for the Obama administration is about three times as large as the average for the other administrations, but it is based a mere six observations, so these results should be considered exceedingly preliminary and very unlikely to be sustained when additional observations become available. (17) Partisan differences on this question are typically very large (on average, 64% consider their own party the more competent, and only 6% consider the other party the more competent in this regard), but even within partisan categories, presidential approval strongly predicts assessments of relative party competence.
Testing Hypothesis 5: Macropartisanship
The fifth hypothesis, that the relative proportions of Republicans and Democrats in the electorate will vary with the popular success or failure of the president, is well supported in the literature on macropartisanship, although the dynamics of the relationship remain in dispute (Erikson, MacKuen, and Stimson 1998; Green, Palmquist, and Schickler 1998; MacKuen, Erikson, and Stimson 1989). The hypothesis was also supported for the Bush administration by results reported in my earlier article (Jacobson 2009a). Figure 4, which plots the monthly averages for the president's party's share of partisans in the Gallup series (with leaners included as partisans) against presidential job approval for the three most recent administrations, displays the relationship in its simplest form. The distribution of aggregate partisanship is clearly related to presidential approval; the relationship is tighter for the Bush and Obama administrations than for the Clinton administration (compare the [R.sup.2]s), while the slope is steeper for Obama (.32, compared to .17 for both Bush and Clinton).
[FIGURE 4 OMITTED]
In light of the theoretical stability of party identification (Green, Palmquist, and Schickler 2002) and prior work on macropartisanship documenting the autoregressive structure of this relationship (Erikson, MacKuen, and Stimson 1998; Green, Palmquist, and Schickler 1998), the OLS results accompanying the data in Figure 4 require further elaboration. I thus estimated two additional models, an OLS model with macropartisanship lagged one month and an ARMA (error correction) model with one-lag autoregressive and moving average terms. I also performed these analyses on the Gallup series excluding independent leaners and on the CBS News/New York Times party ID series (which includes leaners as partisans). The results appear in Table 14. In every case, current presidential approval is estimated to have a significant impact on macropartisanship. The shape of the relationship is similar for the Clinton and Bush administrations, with coefficients ranging from .09 to .13 in the OLS models and from .13 to .19 in the error-correction models. For the latter, we observe large autoregressive and moving average parameters, indicating considerable "stickiness" in mass partisanship. For these presidents, then, a 10-point change in approval would be expected to produce initially somewhere between a 1- and a 2- point change in macropartisanship, with further effects accumulating at a diminishing rate over time.
The estimated effects of approval on aggregate partisanship during the Obama administration are substantially larger than for the earlier administrations, and the autoregressive and moving average terms are all insignificant except for the ARMA model for the CBS/New York Times data. Aggregate partisanship as measured by Gallup evidently experienced little stickiness in responding to the decline in Obama's approval ratings during his first couple of years in office. As is evident from the full Gallup party ID series for the last 12 years displayed in Figure 5, the gains in mass partisanship won by the Democrats during Bush's second term had vanished by the 2010 midterm election.
These findings speak to a debate in the macropartisanship literature that could not be addressed in my earlier article, which covered only the Democratic surge during the Bush administration. Erikson, MacKuen, and Stimson "expect the equilibrium value of partisanship (macro and micro) to represent the cumulation of its various causes, with no 'forgetting' " and hence conclude that "while partisanship changes only slowly, its changes are permanent rather than transitory" (1998, 904-05). Green, Palmquist and Schickler argue, on the contrary, that while short-term forces may knock individual voters out of their long-term partisan identities, "at the individual level, voters return quickly to their long term attachments. However, voters tend to respond in the same way to short-term events, which may cause the overall distribution of party identification to lurch in one direction before returning to its long-term equilibrium" (2002, 59). The swift and seemingly frictionless reversion, during the first two years of Obama's presidency, of the aggregate partisan balance back to where it had been in 2004, erasing the wide advantage that the Democratic Party had accrued during the long downward slide in Bush's approval ratings during his second term, certainly looks more like a return to equilibrium than a process producing the slow, but permanent changes anticipated by Erikson, MacKuen, and Stimson (1998).
[FIGURE 5 OMITTED]
The patterns for the Obama administration are instructive in a second way: the coefficients estimated in the Gallup equations that do not treat leaners as partisans are only about half as large as those for equations that treat leaners as partisans. Moreover, if we run separate equations in which the proportions of Democratic and Republican identifiers (excluding leaners) are dependent variables, Obama's approval ratings affect only the proportion of Democrats, with coefficients of .25 (standard error, .07) and .29 (.03), respectively, for the OLS equation ARMA equations. The equivalent estimates for the proportion of Republicans are tiny and insignificant at -.02 (.04) and .01 (.05). For the Clinton and Bush administrations, the comparable coefficients were of roughly equal size for each party (with, of course, opposite signs), and all were statistically significant by at least the p < .05 standard. For these administrations, the two Gallup measures of aggregate partisanship are highly correlated, at .88 for Clinton, .93 for Bush. For the Obama administration, the correlation is only .78, and thus the shared variance between the two measures is only about 60%.
The trend data plotted in Figure 6 point to the explanation. While the Democratic Party lost adherents among nonleaned identifiers, the Republican Party did not pick up a commensurate share of new nonleaned identifiers. Rather, the independent category absorbed almost all of the net Democratic losses; it also therefore necessarily registered almost all of the net Republican gains. (18) According to the Gallup series, then, the "shocks" to partisanship delivered first by the Bush administration and then by the Obama administration have had the effect of unmooring adherents from each party in succession. A similar increase in the proportion of independents (including leaners) during the Obama administration also appears in the CBS News/New York Times, ABC News/Washington Post, and Pew surveys. The trends since the 2010 election suggest that the unseemly partisan disputes over taxes, spending, and deficits that brought the U.S. government to the brink of paper insolvency in the summer of 2011 not only earned near-universal popular disdain, (19) but also reduced people's inclinations to identify themselves with either party. Still, the proportion of pure independents in the Gallup series has not increased at all and remains near the 10% mark, where it has been for most of the past decade. Thus, the movement away from the parties has simply swelled the ranks of the independent leaners. And because, as voters, these leaning partisans are only slightly less loyal to their party's candidates than weak partisans, (20) the impact of presidential evaluations on macropartisanship with leaners included among the partisans is what counts on election day.
[FIGURE 6 OMITTED]
This brings us back to the original motive for this study, for the president's most important proximal impact on his party is measured when voters go to the polls. (21) The findings reported in this article point to several complementary pathways through which the president's electoral influence might be registered. Foremost is, of course, the effect presidents have on mass partisanship, for party identification remains, by far, the most important determinant of voting in general elections, increasingly so over the last three decades (Jacobson 2013a, 128). The president's influence on his party's image and standing with the public may also make a significant difference. The equations in Table 15, which estimate the effects of three party-related variables on the net shift in the number of House seats won by the president's party between elections, provide evidence of such effects. The available data on the party variables cover different sets of elections, so I estimated three pairs of equations. The first equation in each pair includes one of the party variables, the president's party's "exposure"--the share of seats it currently holds compared to its average over the previous eight elections (22)--and a control for the party of the administration. The second adds the president's approval rating (percentage of respondents approving in the last Gallup Poll taking before the election) and the change in real income per capita over the election year, both standard variables in models estimating aggregate outcomes of House elections (Jacobson 2013a, 166-68; Tufte, 1975).
The results indicate that relative party competence in dealing with the nation's most important problem (defined as in Table 13) has a substantial and statistically significant effect on the fortunes of the president's party in House elections. (23) The effect is reduced by about half when presidential job approval and the economy's performance are taken into account, but it remains significant. The coefficient suggests that a 10-point difference is worth about 8 House seats; the predicted difference in the seat swing between years at the extremes on this measure is 30 seats.
The difference between voters' average thermometer ratings of the president's party and the rival party is also strongly related to the seat swing. (24) Again, with presidential approval and the economy controlled, the effect is less than half as large but remains substantively and statistically significant (at p < .05). A 10-degree difference in average party thermometer ratings translates in to an 18-seat difference in the swing; the difference between the highest and lowest values of this variable translates into a 29-seat difference in the swing.
The coefficients estimating the effects of ideological distance--measured as how much closer or more distant the average respondent locates president's party compared to the rival party with respect to the respondent's own self-placement on the 7-point scale (described in Table 10)--display the appropriate negative signs (the more distant the average respondent, the worse for the president's party), but the coefficients in this case are indistinguishable from zero. In the full equation, the difference between the ideologically closest and most distant electorates is estimated (unreliably) to account for a 16-seat difference in the seat swing.
In every equation, presidential approval and the economy's performance are estimated to have large effects, similar those reported in the literature explaining aggregate House election outcomes (cited in footnote 1). I estimated a combined model with all three party variables and all controls, but with only 16 available observations and 8 independent variables, several subject to multicollinearity, the equation is too unstable to warrant presentation and discussion. Still, it is worth noting that joint effect of the three party variables was statistically significant (p = .016), and their inclusion increased the explained variance in the president's party's House seat swing from 75% to 90% (based on the adjusted [R.sup.2]s).
Obviously, more research and analysis will required before we can say anything conclusive about the relative importance of and linkages between the various pathways through which presidents shape their parties' electoral fortunes. Still, these results give considerable support to the idea that the president's performance in office has important indirect as well as direct effects on his party's electoral fortunes.
Throughout this analysis, I have treated the causal arrow as running from presidents to parties. Causality also flows in the opposite direction, of course, for partisan predispositions profoundly influence Americans' perceptions and evaluations of political actors and events. (25) More specifically, party-induced biases and party images shape beliefs about and reactions to future presidents as soon as they arrive on the national stage and continue to affect assessments of their performance in office--and to a quite remarkable extent for recent presidents (Jacobson 2013b). However, my working assumption has been that for each administration, these initial reactions can be considered the starting point, a product of the existing configurations of public attitudes toward and beliefs about parties that are subsequently updated in response to developments during the president's years of service.
The evidence presented here is uniformly consistent with the proposition that, during their time in office, presidents move people to update their attitudes toward and beliefs about the president's party. These relationships offer a plausible explanation for the observed effects of the president's popular standing on the electoral fortunes of his party's candidates, for they share not only a brand name but, to a considerable extent, the reputation and image it conveys. Insofar as the data allow comparisons across administrations, they suggest that all presidents at least back to Truman have affected their parties in broadly similar ways.
Among the three most recent presidents, Obama has had the strongest apparent impact on his party's standing. This may, in part, represent a regression toward the mean after the disequilibrium provoked by his predecessor, Bush, but the idea that Obama's impact on his party has been greater than that of previous presidents is backed by evidence from the 2010 midterm election. In 2010, individual-level consistency between presidential approval and the House and Senate vote--voters opting for the president's party's candidate if they approved, for the rival party if they disapproved--was the highest yet measured, as was the correlation between the president's vote share in districts and states in the previous presidential election and his party's House and Senate candidates' vote shares in the midterm election. (26)
Finally, it is intriguing to discover that estimates of the structures of the relationships analyzed here fall into two broad patterns. For party positivity, approval, feeling thermometer scores, perceived ideological location, and relative competence, a 1-point difference in the presidential variable predicts about a 0.5-point difference in the party variable (the coefficients for these models generally fall between 0.4 and 0.6). Such consistency in estimated effects across diverse measures hints at a common dynamic and perhaps origin for these relationships. For macropartisanship, however, the estimated presidential effect is notably smaller, as is should be if partisanship is correctly characterized as a component of a person's identity and thus less responsive to the short-term influences emanating from the current administration.
[FIGURE A1a OMITTED]
[FIGURE A1b OMITTED]
[FIGURE A1c OMITTED]
[FIGURE A1d OMITTED]
[FIGURE A1e OMITTED]
[FIGURE A1f OMITTED]
[FIGURE A2a OMITTED]
[FIGURE A2b OMITTED]
[FIGURE A2c OMITTED]
[FIGURE A2d OMITTED]
[FIGURE A2e OMITTED]
[FIGURE A2f OMITTED]
[FIGURE A3a OMITTED]
[FIGURE A3b OMITTED]
[FIGURE A3c OMITTED]
[FIGURE A3d OMITTED]
[FIGURE A3e OMITTED]
[FIGURE A3f OMITTED]
Abramowitz, Alan I. 1985. "Economic Conditions, Presidential Popularity, and Voting Behavior in Midterm Congressional Elections." Journal of Politics 47 (February): 31-43.
Abramowitz, Alan I., and Jeffrey A. Segal. 1986. "Determinants of the Outcomes of U.S. Senate Elections." Journal of Politics 48 (May): 433-39.
Ansolabehere, Steven. 2011. Cooperative Congressional Election Study, 2010: Common Content. [Computer File] Release 1: May 17. Cambridge, MA: Harvard University [producer]. http:// projects.iq.harvard.edu/cces (accessed August 2, 2012).
Barrels, Larry M. 2002. "Beyond the Running Tally: Partisan Bias in Political Perceptions." Political Behavior 24 (June): 117-50.
Campbell, James E. 1993. The Presidential Pulse of Congressional Elections. Lexington: University of Kentucky Press.
Erikson, Robert S. 1990. "Economic Conditions and the Congressional Vote: A Review of the Macrolevel Evidence." American Journal of Political Science 34 (May): 373-99.
Erikson, Robert S., Michael MacKuen, and James A. Stimson. 1998. "What Moves Macropartisanship? A Response to Green, Palmquist, and Schickler." American Political Science Review 92 (December): 901-21.
Green, Donald, Bradley Palmquist, and Eric Schickler. 1998. "Macropartisanship: A Replication and Critique." American Political Science Review 92 (December): 883:99.
--. 2002. Partisan Hearts and Minds; Political Parties and the Social Identities of Voters. New Haven, CT: Yale University Press.
Hibbs, Douglas A., Jr., Douglas Rivers, and Nicholas Vasilatos. 1982. "The Dynamics of Political Support for American Presidents among Occupational and Partisan Groups." American Journal of Political Science 26 (May): 312-32.
Jacobson, Gary C. 2000. "Public Opinion and the Impeachment of Bill Clinton." In British Elections and Parties Review, Vol. 10, eds. Philip Cowley, David Denver, Andrew Russell, and Lisa Harrison. London: Frank Cass, 1-31.
--. 2007. "Referendum: The 2006 Midterm Congressional Elections." Political Science Quarterly 122 (Spring): 1-24.
--. 2009a. "The Effects of the George W. Bush Presidency on Partisan Attitudes," Presidential Studies Quarterly 39 (June): 172-209.
--. 2009b. "The 2008 Presidential and Congressional Elections: Anti-Bush Referendum and Prospects for the Democratic Majority." Political Science Quarterly 124 (Spring): 1-30.
--. 2011. "The Republican Resurgence in 2010," Political Science Quarterly 126 (Spring): 27-52.
--. 2013a. The Politics of Congressional Elections. 8th ed. New York: Pearson.
--. 2013b. "Presidents, Partisans, and Polarized Politics," Can We Talk? The Rise of Rude, Nasty, Stubborn Politics, eds. Danial Shea and Morris Fiorina. New York: Pearson, 100-22.
Johnston, Richard. 2006. "Party Identification: Unmoved Mover or Sum of Preferences?" Annual Review of Political Science 9 (june): 329-51.
Lodge, Milton, Kathleen M. McGraw, and Patrick Stroh. 1989. "An Impression-Driven Model of Candidate Evaluation." American Political Science Review 83 (june): 399-419.
MacKuen, Michael, Robert S. Erikson, and James A. Stimson. 1989. "Macropartisanship." American Political Science Review 83 (December): 1125-42.
Oppenheimer, Bruce I., James A. Stimson, and Richard W. Waterman. 1986. "Interpreting U.S. Congressional Elections: The Exposure Thesis." Legislative Studies Quarterly 11 (May): 227-47.
Rahn, Wendy M. 1993. "The Role of Partisan Stereotypes in Information Processing about Political Candidates." American Journal of Political Science 37 (May): 472-96.
Rudalevige, Andrew. 2001. "Revisiting Midterm Loss: Referendum Theory and State Data." American Politics Research 29 (January): 25-46.
Snyder, James M., Jr., and Michael M. Ting. 2002. "An Informational Rationale for Political Parties." American Journal of Political Science 46 (January): 90-110.
Tufte, Edward R. 1975. "Determinants of the Outcomes of Midterm Congressional Elections." American Political Science Review 69 (September): 812-26.
--. 1978. Political Control of the Economy. Princeton, NJ: Princeton University Press.
GARY C. JACOBSON
University of California, San Diego
(1.) For analyses of these three elections, see Jacobson (2007, 2009b, 2011); a review of the effects of presidential candidates and presidents on congressional elections may be found in Jacobson (2013a, chap. 6); examples from the large literature on the president's electoral influence include Abramowitz (1985); Abramowitz and Segal (1986); Campbell (1993); Hibbs, Rivers, and Vasilatos (1982); Oppenheimer, Stimson, and Waterman (1986); Rudalevige (2001); Tufte (1975, 1978); for something of a dissenting view, see Erikson (1990).
(2.) Other variables constructed from these response categories, such as using the proportion positive of those with a positive or negative opinion or creating a five-point scale and taking its average, are very highly correlated with this measure and would produce the same substantive results.
(3.) Data sources include Gallup, ABC News/Washington Post, CBS News/New York Times/CNN/Time/ Los Angeles Times, Pew Research Center for the People and Press, and Quinnipiac polls. Survey results were taken from the sponsor's web sites and from the Roper Center archive, University of Connecticut; to account for house effects in these surveys, I estimate models using survey sponsor fixed effects.
(4.) For consistency across surveys, independents who lean toward a party are treated as independents, not partisans, in these data.
(5.) Estimated by interacting an administration dummy with the presidential variable.
(6.) The relationship was also stronger for the Democratic Party before (b = .55, r= .71) than after (b = .31, r = .46) the impeachment period in this data set.
(7.) ANES Sources: the American National Election Studies (http://www.electionstudies.org) TIME SERIES CUMULATIVE