Significance

[Magda Cala]

I am analyzing the effect of gender on homeownership in the US and I keep finding that my female dummy is insignificant after I add age and age squared to the regression. I am using probit model with dependent variable of value 1 for homeowner and 0 for not. I also tried a model with age and female interaction term and then female dummy turned out to be significant. I was trying to see if there is some correlation between female and age but it doesnt look like it. I only notice that there is more older women then men. I am not sure which tests to run to see what is the issue and also if it would be justified to include the interaction term or should i report that the female dummy is not significant?

[Maarten Buis]

In the model with interaction terms, the main effect for gender is the effect when age zero is (so for a baby). That is probably a bit of a extrapolation...

[Joro Kolev]

In the lines of what Maarten said, a trick that my labour economists mentors taught me is to demean age when you do age interactions. Or to subtract from age the median. Or certain percentile. Or certain age that is of interest to you.

Then the main effect of the female dummy would be the effect when age is at the mean, or at the median, or at the certain percentile, or at he certain age you subtracted.

[Magda Cala]

that is very helpful! because i am actually not even including people younger then 18 so the effect at age 0 is not really of my interest. I was wondering though actually how to justify adding the interaction effect, because it changes the significance of other factors that i am interested in