Undiscovered mistake in regression with interaction term

[Lena Buettner]

Hey everyone.

I'm having troubles with the outcome of a linear regression model, since I expected something completely different:

This ist how I coded the interaction:
g number_younger_siblings_female=number_younger_sibl ings*(sex-1)

The variable sex is coded 1 for males and 2 for females. (sex-1) makes it 0 for males and 1 for females.

I expected the outcome to be a larger negative number for females, but what I found is a lower negative effect for females, what is unlikely according to the literature. It would make sense to have an negative coefficient of the interaction So I think there must be a mistake and I just can't find it. I would be super grateful if someone would find the mistake!

This is my output:

[Carlo Lazzaro]

Lena:
1) the best practice to number the levels of a categorical variable is starting from 0 onwards (and if I'm not mistaken, Stata recode levels this way, i.e., starting from 0);
2) the best practice to create interactions and categorcal variable is using -fvvarlist- notation;

c.numbero_of_younger_siblings##i.female

I would give a shot to what above and see if the issue get fixed.

[Lena Buettner]

Thank you so much for your ideas! So the code you provided is the interaction itself? So is this the correct whole regression?:

xtreg enrollment female number_younger_siblings c.number_younger_siblings##i.female i.age, fe cluster(hh_idn)

I don't really get what to do

[Maarten Buis]

No ## indicates that Stata should include the interaction and the main effects, so your command should be:

xtreg enrollment c.number_younger_siblings##i.female i.age, fe cluster(hh_idn)

You can read more about this if you type in Stata help fvvarlist

[Lena Buettner]

Ahh now I understand and it worked!! Thank you so much!