Must carry rules
17
relatively prosperous areas. RETAIL, the amount of retail sales per capita in a system’s
county, is included to test this assumption.
22
A positive effect of RETAIL on the
dependent variable is more consistent with the anti-competitive argument.
Estimation
This study used binary regression models due to the binary nature of the
dependent variable, DROP.
23
Specifically, the logit model was chosen because its
coefficients allow for a simpler interpretation in terms of odds ratio.
24
A variety of model specifications were considered. The models and their
Maximum Likelihood Estimation (MLE) results are summarized in Table 3. As shown,
coefficient estimates of these models were largely consistent with each other. The three
statistics presented in the last three rows of the table measure the goodness-of-fit of each
model and do not show clear superiority of one model over another. The following
section focuses on the results of the full model, Model IV.
Results
Table 4 lists again the logit coefficients from Model IV and offers two
interpretations of the coefficients. The first one represents the marginal effects of the
independent variables on DROP while the second one represents their effects on the
change in the dependent variable’s odds.
25
22
To test the theory that cable systems deny carrying local stations to monopolize local advertising
revenues, it would be more appropriate to include variables that measure the amount of local advertising
sales each cable system had. However, not only was local advertising unpopular in the late 1980s,
consistent data for those systems that did sell local ads were unavailable.
23
It is well-known that it is problematic to use linear regression models when the dependent variable is
dichotomous (see Long, 1996, pp. 39-41).
24
Probit is the other popular model for binary regression analysis. The regression results from both probit
and logit are qualitatively similar even though the coefficients may be different (Long, 1996, p. 84).
25
The marginal effect represents the partial change in the probability of the dependent variable as a result
Convention