[Kwaku Agyei Adu]

Carlo,
just a quick one. In an attempt to compare the males to that of females (i run a separate model for both males and females).One professor mentioned you cannot make direct comparison between males and females in an adoption study because of the differences in the slope of the two categories and as such I should rather run the a joint model with gender dummied as well as creating a dummy for the slope of gender. Any suggestion on this please??

Regards!
Kwaku

[Carlo Lazzaro]

Kwaku:
you actually ran a joint model including -gender- (or, equivalently, -i.gender-) among predictors.
Conversely, I do not follow you about the need to run separate regression models for male and female.

[Kwaku Agyei Adu]

Alright Carlo,

But then how do you dummy the slope of gender in a model (aside just the normal dummy for gender)??

Regards,

Kwaku

[Carlo Lazzaro]

Kwaku:
you probaly need to interact gender with some other predictor (which one I do not know, though).

[Kwaku Agyei Adu]

Carlo,

I am using this approach. I interacted gender with all the explanatory variables and then run the model together with the explanatory and interacted variables. I then used joint F-test to test the significance of the interacted variables including gender itself. If I found the test to be significant, I will then drop the interactions that were not significant and only use those that were significant in the model. Is my approach statistically right?


Regards,

Kwaku

[Carlo Lazzaro]

Kwaku:
no, it is not.
Your model specification should be driven by what Others did in the past when presented with your research topic, regardless of what is statistically significant or not.
Besides, you cannot use F-test and then drop what is not significant, as the test included the non significant coefficìents, too.
As an aside, replies would be possibly more helpful if you post what you typed and what Stata gave you back (kind reminder). Thanks.

[Kwaku Agyei Adu]

Carlo,
I have attached a document for your perusal. Could you please take a look at it (more precisely page 15 of the document)

Regards,
Kwaku

[Carlo Lazzaro]

Kwaku.
thanks for the article (however, please be sure not to breach any copyright agreement with posting published article in electronic format).
You may follow the article approach, even though I find difficult to believe that interacting -gender- with all the other predictors makes sense in informative terms: think of how you would explain to an audience the marginal contributions of each one of them when adjuster for other predictors.