Dear all,
I have estimated a fractional logit model and a heteroskedastic fractional probit model with the same specification using Stata 15. In fractional logit model, the marginal effects of variables of interest are all significant (at 1%, 5% and 10% levels) and sizes and signs are all as expected according to the hypothesis. On the other hand, in heteroskedastic fractional probit model, some of the marginal effects are not statistically significant, although their sizes and signs are as expected. My understanding is, in heteroskedastic model, standard errors are modified so as to give statistically insignificant effects.
My dilemma is which results I should report? According to the hypothesis, it would be easier for me to claim the results of fractional logit model. However, the lnsigma tests associated with heteroskedastic model indicate that data is suitable for heteroskedastic model. Should I report both the models? In that case, how do I align statistically insignificant results of heteroskedastic models with the hypothesis? Or should I only report fractional logit model and drop the results of heteroskedastic model? Please give me some advice. Thanks.
Ujjwal Kumar Das
