Can Z values be used to determine relative importance of variables in a logistic regression model? So in this example:
http://www.ats.ucla.edu/stat/stata/output/stata_logistic.htm
Is "read" the most important variable to the model?
 A: No. "z values" are the ratio of the coefficient estimate divided by the standard error of the estimator (easily verifiable in the link).
As such, they provide an indication as to how much uncertainty "surrounds" the point estimate of the coefficient. The larger the "z value" gets, the less uncertainty there is, since it means that the standard error gets smaller and smaller relative to the coefficient value. So a confidence interval around the point estimate will be tighter.  
Moreover, in logistic regression, the point estimates do not reflect the "partial marginal effect" of each regressor on the dependent variable, as in linear regression. Here regressors affect the probabilities around the dependent variable, and also, due to the non-linear structure, the effect of the one is not immediately separable from the effect of the others. So it is a tad more elaborate (compared to linear regression) to accurately assess the contribution of each regressor to the probabilities related to the dependent variable.
