Dear Stata users,

I have a question regarding fixed-effects individual-slopes models, also known as heterogeneous slopes models. I would like to estimate the following model: Income = age + married. This could be done using -reghdfe- as follows:
Code:
reghdfe income married, absorb(id##c.age)
If I'm not mistaken, the coefficient of married has two interpretations when this model is estimated on a balanced panel:
(1) The difference between the average income as predicted by individual age-trajectories and the average observed income.
(2) The average difference between individual incomes as predicted by age-trajectories and individual observed incomes.

Now, I have noticed that things change when using unbalanced panel data. In that case, (1) and (2) do not match each other nor do they match the coefficient, probably because some observations are missing. This made me question my understanding of these models. Hence: how should the coefficients of fixed-effects individual-slopes models be interpreted?


Many thanks,
Bram