15

Interestingly there seems to be a bimodal distribution among the middle categories of

education with the LAST POLL coefficients and standard errors closely approximating

one another for those with both a High School diploma and those with a 4-year degree

and dropping in the middle category for those with some college education. There is no a

priori theoretically grounded explanation for this distribution. Regardless, in the case of

education it seems pretty clear that our expectations seem to have been on the right track

for establishing the groups for whom we may have expected a contagion effect. Like the

“follow politics” sub-groupings the R-Square values increase as education increases.

Thus the better educated an individual is, the better we can explain the variance in a

respondent’s evaluation of the candidates, given the predictors in the model.

Results – Vote Intention

Good feelings alone are not adequate to prove that bandwagons operate during

national election cycles. Given that there is evidence that people feel more positively

about candidates who are doing well in the polls, what remains to be seen is that they also

express an increased disposition to support that candidate come the actual election date.

The setup for the models for vote intention is essentially the same as for candidate

favorability with the only differences being the dependent variable and the estimation

technique. Instead of favorability, the dependent variable is the two party vote intention

for either Democrat John Kerry (coded 1) or Republican President George W. Bush

(coded 0). As the dependent variable is no longer a scale but a dichotomy, instead of

using Ordinary Least Squares regression estimation, logistic regression is used.

[Insert Table 4 here]