Number of councilors that are women. The number of
councilors that are women is distributed as a count variable;
therefore, to estimate the effects of the independent variables on
this count we use the negative binomial model (see Long, 1997 sec.
8.3). The negative binomial model is used rather than the Poisson,
because this data is over-dispersed (the variance of the number of
women councils conditional on the controls is greater than the
mean number of councilors). Zero-inflation was not a problem in
this data, so there was no need to complicate the model with other
forms of count data. In the model estimation, we also include an
exposure variable – the total number of councilors on the council –
to account for the fact that the number of women can never exceed
the total number of councilors. We also use probability weights
that are inversely proportional to the number of user groups
sampled in a given country so that having more user groups for
certain countries gives less weight to those countries. The results
from this analysis are reported in