(Page 9 of 26) - Gender and Vote Choice in the 2006 Canadian Election authored by Gidengil, Elisabeth., Everitt, Joanna., Blais, Andr??., Fournier, Patrick. and Nevitte, Neil.

Since the dependent variable has three unordered categories (Conservative, Liberal and

NDP), all of the vote estimations are based on multinomial logistic regression. This enables us to
model the vote as a choice among all three parties and thus capture the “multifaceted process of
choosing among multiple parties at once” (Whitten and Palmer 1996). The advantage of this
approach is that it highlights the inter-party dynamics of support and it allows for different
variables to play into different sets of choices. Take a variable like public sector employment.
Being a public sector worker may be very relevant to the choice between the Conservatives and
the NDP, but count for little when the choice is between the Liberals and the NDP.

The coefficients estimated by multinomial logistic regression represent the predicted

marginal impact of a given independent variable on the log-odds of choosing a given party
relative to a baseline party. Their meaning depends on the values of the other variables included
in the model. As such, they lack a straightforward, intuitively obvious interpretation. To facilitate
interpretation, we use the coefficients to estimate the independent impact of each variable on the
probability of voting for each of the parties. These estimations take the form of a series of “what
if?” simulations. Say we want to estimate the impact of public sector employment on a woman’s
probability of voting NDP. On the basis of the estimations, we can compute the mean probability
of voting NDP, first if every woman was a public sector worker and, second if no woman
worked in the public sector, keeping the effects of the other social background characteristics
unchanged. The difference in the mean probabilities gives us an estimate of the average impact
of public sector employment on the probability of a woman voting NDP, everything else being
equal. We can do the same calculation for men and then compare the probabilities.

Findings

Table 1 presents the most important findings, based on the models for women and men

combined. The column entries indicate the estimated effect on vote choice of being female,
beginning with a model that contains only the gender variable, then adding the structural and
situational variables (and the socio-demographic controls), and ending with a model that includes
value orientations as well.

It is quite clear that situational and structural factors cannot explain why women were

more likely than men to choose the NDP over the Conservatives. When a wide array of
potentially relevant structural and situational variables (and controls for other social background
characteristics) are added to a model that includes only gender, the coefficient for being female
barely changes. On the other hand, when values and beliefs are added to the model, the gender
gap effectively disappears and the coefficient for being female approaches zero. As Inglehart and
Norris (2003) argue, the roots of the modern gender gap in vote choice are clearly cultural. The
modern gender gap in Canada reflects differences in women’s and men’s value orientations
rather than structural and situational differences in their lives and experiences.

[Table 1 about here]

Structural and Situational Explanations

It would be a mistake, though, to conclude that structural and situational factors are

therefore irrelevant to understanding the impact of gender on vote choice. This becomes clearer