how to control for heteroskedasticity and use fixed effects at the same time?

[Andreas Kazamias]

Hello all!
I have a panel data and for my regression I use a Linear Probability Model (LPM). I also employ individual Fixed Effects however I am aware that I can't use robust and fe at the same time just like I would do with random effects. So, how to control for the heteroskedastic nature of the LPM model? should I use xtreg or areg and should I cluster at the id level?

Thanks in advance!

[Jeff Wooldridge]

Assuming N is reasonably large and T not too large, clustering by id is easily the best option. It accounts for serial correlation as well as the heteroskedasticity.

[Andreas Kazamias]

Hello Professor and thank you for your answer!
Indeed I have a large number of N(740000 observations and 94000 individuals). Also my T is not too large (8 waves).

So, would xtreg y x1 x2 x3 x4 x5 x7, fe vce(cluster id) account for the heteroskedasticity and fixed effects at once or should I use areg y x1 x2 x3 x4 x5 x6 x7, absorb(id)?

Kind Regards

[Jeff Wooldridge]

Use xtreg just as you've specified. As it turns out, areg does not always produce the correct clustered standard errors.

[Andreas Kazamias]

Great, thank you very much for your time and advise, Professor Wooldridge!

Have a nice day,
Andreas Kazamias