disapprove should be more likely to vote for the opposition’s candidates. The literature
on tests of this hypothesis using survey data reports mixed results (Jacobson 2001b), as
do the equations shown in Table 8. The table presents results of a simple logit model of
voting behavior in each midterm election since 1974. The model estimates the effects of
presidential approval on the House vote, controlling for the respondent’s party
identification and ideology and the candidate’s incumbency status.
variables have been adjusted so that all of the expected signs are positive.
[Table 8 here]
The results indicate that the impact of assessments of the president has varied
from election to election. Although the coefficient on approval always has the expected
positive sign, it is significant in only four of the eight election years. There is, however,
no doubt about its importance in 2002, when the coefficient on approval is largest in the
series. Unlike regression coefficients, logit coefficients are not open to easy intuitive
interpretation, because the estimated relationships are nonlinear.
the estimated effects of presidential approval on the House vote across a range of
hypothetical initia1 vote probabilities for a selection of our midterm elections (Figure 5).
The figure demonstrates that presidential approval can have quite dramatic effects on the
probability of voting for the president’s party’s candidate in House elections; in 2002, for
example, the probability of voting for the Republican candidate was as much as .32
higher for approvers compared to disapprovers.
[Figure 5 here]
Why does the impact of presidential approval vary so widely across midterm
elections? The pattern of coefficients in Table 5 suggests that it matters more in elections
in which the president, his performance and policies, were the focus of campaign politics.
The 1974 election was held in the shadow of Watergate, and Gerald Ford’s controversial
21
The dependent variable takes a value of 1 if the respondent voted for House candidate of president’s
party, 0 if for candidate of the opposition party. The independent variables have been adjusted so that all
of the expected signs are positive:
Presidential approval: 1 if respondent approves of president’s performance, 0 if disapproves;
Party identification: 7-point scale, where 1 equals strong partisan of the opposition party, and 7
equals a strong partisan of the president’s party;
Ideology: 7-point scale, where 1 equals an extreme conservative during a Democratic
administration or an extreme liberal during a Republican administration,
and 7 equals an extreme liberal during a Democratic administration or an
extreme conservative during a Republican administration;
Incumbency: 1 if incumbent is of the president’s party, -1 if incumbent is of the opposition
party, 0 if there is no incumbent or two incumbents in the race.
22
In logit equations, the substantive effect of any independent variable on the probability that dependent
variable takes a value of 1 depends on the initial probability set by the other independent variables’
coefficients and values.
19
Convention