Absorbing interaction terms and including them in the regression shows different results

[toh2972]

Dear Sergio

I wanted to run this regression of individuals' grade on sat and race with a list of interaction terms which is just an interaction of individuals' colleges with the running year variable:

reghdfe grade sat race, absorb( i.college#c.year)

However the coefficient on sat and race is different if I include the interaction terms individually in the regression using "reg"

reg grade sat race i.college#c.year

Is there an intuitive reason behind this discrepancy and if so is there a way to have reghdfe show the same results as just using reg when including these interaction terms?
Thank you in advance.

[sergiocorreia]

Three things:

1. What version of reghdfe are you using? the latest from Github?
2. Are you sure you don't want to also add a different intercept for each college? i.e. use absorb(i.college##c.year)
3. If you just want to use a simple # (i.e. different slope but not intercept), the help file mentions that the computation is not as numerically accurate, so you might want to use a stricter tolerance.

[toh2972]

Thank you Sergio for the helpful reply.