George W. Bush has provoked the widest partisan differences in popular evaluations of a president's performance ever observed in the nearly seven decades that the surveys have been asking the job approval question. (1) In a new book explaining why Bush has become such a divisive figure, I argue that, while the polarized political atmosphere he inherited was a contributing factor, Bush's leadership style, agenda, tactics, strategies, and policy decisions (most notably, to invade Iraq) are the main reasons he has leapt far ahead of his closest competitors, Ronald Reagan and Bill Clinton, to become the most polarizing president on record (Jacobson 2006).
It is clear from Bush's example that presidents do much to shape how they are perceived by their own and the other party's identifiers and therefore how far they polarize the public. This should be true of other politicians as well. Hard-line ideologues, for example, ought to provoke more divergent reactions than moderate centrists. Politicians who lead or stand loyally with their parties on the fundamental issues that divide Democrats and Republicans ought to be more polarizing than party mavericks or those who keep low national profiles and busy themselves delivering local benefits. Context and agendas should also matter. Politicians holding office at the national level, where the parties have become sharply divided across a range of issues, should find it harder to attract bipartisan approval than those serving at the state level, where the agenda (not to mention the local political ethos) may not encourage sharp partisan divisions. And, because they can cultivate personal connections with a larger share of their electorates, politicians serving relatively small, homogeneous populations should find it easier to reach across party lines than should those serving large, heterogeneous populations.
Data to test these hypotheses, and, more generally, to compare partisan differences in evaluations of President Bush to those for other prominent elected officials, have recently become available. Beginning in May 2005, SurveyUSA, a polling firm whose main clients for its political surveys are local news media, has been conducting monthly statewide polls in all fifty states. Their automated telephone surveys ask samples of approximately six hundred respondents in each state whether they approve of the performance of G. W. Bush, the state's governor, and both of its senators, as well as questions about each respondent's party identification, ideology, religious service attendance, and demographic characteristics (age, education, sex, race, and in some states, region). In some months they also ask respondents' positions on policy issues. The aggregate results, including breakdowns of the approval questions by all of the respondents' other characteristics, are posted on the Internet shortly after the surveys are taken. (2) Obviously, such data are potentially of great value for addressing a variety of interesting questions about state-level public opinion and elections--if they are of sufficient quality to be reliable. I thus examined the data carefully for internal and external consistency as well as intuitive plausibility and, as I describe in the Appendix, they passed all of the tests very satisfactorily. For the analyses that follow, I use the data produced by the nine monthly surveys taken from May 2005 through January 2006.
Comparing Performance Approval Ratings
I began by comparing the means and distributions of job approval ratings across states for the three offices. I expected greater dispersal for senators and governors than for the president, if only because I was comparing evaluations of a single individual to those of 50 (governors) or 100 (senators) separate officeholders. Figure 1, which displays the kernel density plots of approval levels across states for all nine monthly surveys, confirms this expectation, but barely in the case of senators. Evaluations of Bush have the narrowest distribution and the lowest mean value. This was, of course, a period when the president's approval ratings had fallen to their lowest levels of his administration to that date; had the surveys been taken two years earlier, his average would have been about 13 points higher, and three years earlier, 26 points higher. (3) Senators enjoyed the highest average approval ratings during the period under examination, and their ratings are only slightly more dispersed than those of the president. On average, a majority of the governors' constituents also approve of their performance, but with considerably more variance across states than observed for the national politicians.
[FIGURE 1 OMITTED]
Next, I examined the extent to which partisans are divided in their assessments of these officials. Figure 2 shows that partisan differences in evaluations are on average more than twice as wide for Bush as for senators or governors, and their variation across states is much smaller. Republicans and Democrats differ in their approval of Bush's performance by 43-83 percentage points, with an average of 66 points. Views of senators and governors are far less polarized, and the degree of polarization far more variable; indeed, a few receive higher approval ratings from the opposing party's identifiers than from their own. In only 8 of the 900 possible comparisons were respondents more polarized in their evaluations of a senator than of the president (4); in no state was the average partisan gap across the nine monthly surveys wider for any senator than for Bush. In none of the 450 comparisons were any state's respondents more widely divided by party over the governor's performance than over Bush's.
[FIGURE 2 OMITTED]
Figures 3-5 present more detailed comparisons of the distribution of partisan approval ratings across the three offices using a common scale. Figure 3 displays the widely separated modal statewide evaluations of Bush offered by Republicans, Democrats, and Independents. There is almost no overlap between partisan categories; Bush's worst showing among Republicans in any of these 450 surveys (56 percent approving) is 24 points higher than his best showing among Democrats (32 percent approving). There is also little overlap between evaluations of Independents and partisans. As in national polls, approval among Independents is on average much closer to that of Democrats than to that of Republicans.
[FIGURES 3-5 OMITTED]
By comparison, the partisan distributions of senatorial approval are considerably less distinctive and much flatter (Figure 4). State-level approval is as low as 33 percent among a senator's own partisans and as high as 84 percent among the other party's partisans. Partisan differences are even more attenuated and variable among governors (Figure 5).
By this evidence, polarized partisan evaluation of elected leaders is not a systematic feature of contemporary American politics; Bush may be the most divisive president on record, but neither senators nor governors necessarily provoke anywhere near as much partisan discord as he does. Moreover, evaluations of Bush and these other politicians are not very strongly or systematically linked. Notice in Table 1 that state-level correlations between approval of Bush's performance and that of senators and governors tend to be quite low. The largest correlation, between Bush and Democratic governors, is unexpectedly positive across all partisan categories (an explanation is proposed below). Relationships between approval of senators and governors are typically somewhat stronger but still modest, with again the strongest relationship appearing across party lines (Democratic senator, Republican governor). The strongest relationship occurs between pairs of senators, and, in their case, the large negative correlation in partisan evaluations of senators of opposite parties suggests a considerable degree of polarization (the more partisans like their own party's senator, the less they like the other party's). The overall picture, though, indicates that state-level evaluations of top elected leaders are not strongly linked and must therefore be shaped by a somewhat divergent set of political forces. This conclusion is reinforced by the observation that the degree of state-level partisan polarization in ratings of political leaders is also rather weakly correlated across offices (see the lower section of Table 1). It makes sense, then, to examine the sources of partisan evaluations the three offices separately.
Table 2 lists G. W. Bush's average approval ratings by state across the nine monthly surveys in descending order along with his share of the major party vote in 2004. A glance at these data suggests an obvious source of variations in G. W. Bush's state-level approval ratings: they mirror almost perfectly his state-level support in the 2004 election, albeit at a discount because he had become noticeably less popular by May of 2005 than he had been at the time of his reelection. The regression results reported in the first equation in Table 3 confirm what the eye detects. Bush's average approval rating for the period May 2005 through January 2006 stood at approximately 79 percent of his 2004 major party vote share in the state in question. (5) The estimate is quite precise, with this single variable accounting for 93 percent of the variance in his approval ratings across states.
In light of this strong relationship, it is no surprise to find that the same set of state-level variables accounts for both Bush's vote in 2004 and his ratings in the surveys (the second and third equations). Both rise with the proportion of Republican identifiers in a state, fall with the proportion of Democratic identifiers, and rise as the state's ideological balance becomes more conservative. (6) These three variables predict both the vote and approval ratings with great accuracy (note the root mean square errors), and the equations confirm the decisive impact of partisanship and ideology in shaping popular reactions to Bush (Jacobson 2006). Of these factors, partisanship is clearly the most important, accounting by itself for 84 percent of the variance in the 2004 vote and 82 percent of the variance in subsequent state-level approval. If the 2004 vote is included in the approval equation (the fourth equation in Table 3), it dominates statistically, presumably because it contains information about the state's electorate beyond the distribution of partisans, conservatives, and liberals.
No other variables I examined, including region and state population, had any discernable effect on Bush's approval level, but there were some significant regional differences in the extent of partisan polarization on the approval question. Table 4 lists the states ranked in order of partisan differences in approval of the president.
The regression equations in Table 5 suggest some of the general factors shaping average approval ratings of states' Republicans and Democrats. Among Republicans, approval is higher in more conservative states and lower in more Democratic states, particularly those in the Northeast. (7) Bush's comparatively low standing among northeastern Republicans presumably reflects the remnants of moderate Republicanism more common in that region than elsewhere. Approval among Democrats is higher the better Bush ran in the state in 2004 but, with that variable controlled, lower in states that are more Republican and conservative and, regionally, in the Midwest. Combining these effects, polarization tends to be lower where Bush ran well in 2004 and in states that are relatively more Democratic and relatively less conservative; the Northeast is a bit less polarized (because of the lower enthusiasm for Bush of northeastern Republicans) and the Midwest a bit more so (because of the lower enthusiasm for Bush among midwestern Democrats). Overall, though, these equations indicate that most of the action is in the constant term--that is, that a wide partisan divide is characteristic of all the states.
As Figure 1 showed, senators generally enjoy higher approval ratings than the president but with greater variation across states. Table 6 lists the senators in rank order of the average approval rating they received over the nine monthly polls. Maine's Olympia Snowe and Susan Collins top the list, while John Cornyn barely edges Rick Santorum for last place. However, even the lowest rated senator is on average rated slightly higher than Bush during this period (his average was 43.0 …