Fit complementary log-log models to DOS data, using dichotomized depression as response and gender as the predictor. Comparing the results between modeling the probability of No depression and modeling the probability of Depression. This confirms that the complementary log-log link function is not symmetric.