average, presidents now fundraise in more states than they used to, but presidents still

travel a great deal to states in which they don't attend fundraising events.

**Assessing Factors Related to Fundraising and Other Presidential Travel**

tion and

nd

nalysis

itudinal, with repeated observations of states over time, I

use a fi

To assess more effectively the relationships between presidential atten

electoral factors, we turn our attention to regression analysis. Because the dependent

variable, the number of non-fundraiser-related presidential public events held in each

state in each year, is a count of an event, which only takes on positive, integer values a

is not normally distributed, regression analysis using ordinary least squares would yield

inefficient, inconsistent, and biased estimates. Instead I use maximum likelihood

techniques to estimate a negative binomial regression model, a type of regression a

for over-dispersed

count data.

Because the data are long

xed effects negative binomial model.

This model takes into account that there

likely is unobserved heterogeneity among the states in the study – that is, that they vary

in ways that are not measured by the variables in the model. The fixed effects estimator

calculates coefficients by looking at variation within each state but not across states, so

unobserved heterogeneity among states does not affect the model’s estimates. Table 6

presents the results of two fixed effects negative binomial models, the first with non-

2

I.e., the variance is greater than the mean.

3

See Cameron, A. Colin and Pravin K. Trivedi. 1998. *Regression Analysis of Count Data*. Cambridge:

Cambridge University Press; and Long, J. Scott. 1997. *Regression Models for Categorical and Limited *

Dependent Variables. Thousand Oaks, California: Sage Publications.

4

This model was run using STATA’s xtnbreg function, which estimates negative binomial models for

longitudinal data.

Doherty

15