in that category when the variable equals 0. For example, the sex
variable is 0.130 for the category indicating the wife is responsible
for attending meetings. This implies that households headed by
women are 13 percentage points more likely to have a wife as
being the one responsible for attending meetings than households
headed by men.
The remaining estimates are actual marginal effects and the
standard errors for all estimates are calculated via the delta
method. Note that since the categories are mutually exclusive that
the marginal effects across all categories sum to zero. That is, if a
variable increases the probability of being in any one category it
must decrease the probability of being in at least one other
category. (Although rounding errors may preclude the results
reported here from summing precisely to 0). The final column of
Table 4 indicates the value from a likelihood ratio test with a null
hypothesis that the variable has no effect in each of the outcomes.
A significant value indicates that the variable has overall statistical
significance at the signified level given that the variable is
distributed x2 with three degrees of freedom.
The bottompanel of Table 3 reports summary statistics including
the number of observations used for the estimation, the loglikelihood
value, and overall model significance (x2) from the null
model (i.e. without any explanatory variables). We have also
included a number of alternativemeasures of model fit included the
predicted count R-squared, which is simply the percentage of
correctly predicted categorizations, assuming the category predicted
is that one which has the highest estimated model
probability. The McFadden R-squared value is one minus the ratio
of the log-likelihood from the intercept only model to the specified
model. The Cox-Snell R-squared is a measure of improvement over
an intercept-only model. These Pseudo R-Squared measures are