Hi!

I'm conducting an study that aims to obtain the correlation between life satisfaction (dep. var.) and several socioeconomic variables. I have done this already with an OLS and, worried about heteroskedasticy, I ran the Brusch-Pagan test, founding that, indeed, heteroskedasticity is a problem. Therefore, I used robust errors to deal with this. However, as there is a discussion in the literature over whether people interpret life satisfaction in an ordinal or cardinal way, I would like to double check the robustness of my results using an ordered probit model. I would like to test for heteroskedasticy here too and I have read that the new Stata's command "hetoprob" can both detect if heteroskedasticity is a problem and run the model solving it. My questions are:

1) Is there something I should have in mind when making the decision over whether "hetoprobit" command is the best way to check and work with heteroskedasticy (like if there are better alternatives)?

2) I don't know which variables are causing the heteroskedasticity. The command "hetoprobit" requires the specification of every variable for which there might be differential variance for different values at the end of the syntax line. Then, should I include all my dependent variables there?

3) Finally, I would like to know how does Heteroskedasticy works. Does it operate by giving different weights to the errors as "robust" in the OLS?


Thank you in advance!