Hi,

I have a puzzling result that I cannot reconcile, so I am writing to seek some help. I have tried running #model 1 with interaction variable between two variables while the model #2 runs the independent variable across two subgroups (which was used to interact in the model #1).

For the interaction variable, I get a positive coefficient, but for sub-group analysis, I get a positive coefficient for the gender == 0 while I get a negative coefficient for the gender == 1. What is troubling is that the model#1 and model #2 are not showing a consistent results in terms of sign and magnitude. (Coefficients are bold-fased).

I have following regression models that attemp to measure the effect of period dummy (2013-2017) on labor market outcome.

labor_outcome: equals 1 if labor movement occurs and 0 otherwise
period_dummy: equals 1 if years 2013-2017 and 0 otherwise
gender: equals 1 if female and 0 if male

Model #1: xtreg labor_outcome i.period_dummy##i.gender, fe

Simplified results:
1.period_dummy | -.0010372 .0016781 -0.62 0.537 -.0043267 .0022522
1.gender | 0 (omitted)
period_dummy#gender |1 1 | .0031244 .0018389 1.70 0.089 -.0004802 .006729

I get 0.0031244 for the coefficient in front of the interaction variable.

Model #2:

xtreg labor_outcome period_dummy if gender == 0, fe

Simplified results:
1.period_dummy | 0.0026596 0.0020307 1.31 0.190 -0.0013214 0.0066406

xtreg labor_outcome period_dummy if gender == 1, fe

Simplified results:
1.peirod_dummy | -0.0028907 0.002115 -1.37 0.172 -0.0070374 0.0012559


Can anyone help me reconcile these?