Using Fixed effect regression on cross sectional data with binary outcome.

[chintu red]

I'm investigating nutritional outcomes (Yes and No) across districts, which are the smallest administrative regions, using cross-sectional data. I'm interested in understanding the impact of various socio-economic variables on nutritional outcomes. Since all the children in the district are exposed to a similar nutritional program. However, the data for the districts is not included in the dataset. To address this, I've employed fixed-effect regression to control for district-level variation and compared how various socio-economic factors influence nutritional outcomes for individuals living in the same district.

I used the "reghdfe" syntax with the following command:

reghdfe nutritional_out residence caste religion mpce_quintile age mother_edu father_edu cooking_fuel improved_water improved_sanitation diarh fever any_illness distance_school, absorb(distid) clus(psu)

In this command, "nutritional_out" is a binary variable indicating nutritional outcomes.

My question is: What if I report regression coefficients as beta-coefficients? Would this make any difference in my estimates? Alternatively, should I run a logistic regression while controlling for "distid"? What statistical differences do these approaches make, if any? I would appreciate your suggestions and references on this matter. Please correct me if I have any inaccuracies in my approach.

Thank you in advance for your assistance.

[Carlo Lazzaro]

Chintu:
the community-contributed module reghdfe- is for continuous regressand.
Unless you're planning to go Linear Probability Model, you should consider -xtlogit,fe- that assumes conditional fixed effect due to incidental parameter bias.