18
STATA computed R
2
for each regression. I then tested the statistical significance of the R
2
from
each of the auxiliary regressions using an F-test. From Gujarati (1999: 324),
F = [R
2
/(k-1)]/[(1-R
2
)/(n-k)], where n is the number of observations and k is the number of
explanatory variables including the intercept. The null hypothesis is R
2
= 0. Table 6 summarizes
the results of significance tests of R
2
in each auxiliary regression. The results of the individual
regressions are in the Appendix.
Table 6
Significance of R
2
in Auxiliary Regression
auxiliary regression
dependent variable
Value of R
2
Value of F
Is F significant?
ideology
education
newspaper
television news
.004
.029
.051
.036
2.37
25.87
46.55
32.35
Yes
Yes
Yes
Yes
R
2
in each of the auxiliary regressions was significant, indicating that multicollinearity is
a problem. Gujarati (1999) explains that except in the case of perfect multicollinearity,
multicollinearity does not mean that the estimators are biased or inefficient. Gujarati (1999)
explains that multicollinearity may result in a high R
2
and few significant coefficients. However,
in my main model, R
2
is not high (.004). Gujarati (1999) suggest that one possible method for
correcting for multicollinearity is to drop the variable causing the multicollinearity from the
model. I cannot drop ideology from the model, despite it’s collinearity with the other three
explanatory variables because it is the main explanatory variable of interest in this model.
Dropping it from the model would result in specification error (Gujarati 1999: 331). Instead, I
estimated the model after dropping television news from the model. The following table
summarizes the results.
Convention