xtlogit with fe or re option take tens of hours to run

[Wei Lyu]

Hi All,

I am running a xtlogit with re and fe option to see the difference between the random effect and fixed effect. My dataset has nearly 2 million observations and my xtset variable has 532 groups. I have another 14 control variables. the RE model costs me about over 16 hours to get the outcome, and the FE have been run for almost 20 hours but no outcome shows until now.

Anyone has any thoughts about how to speed it? My stata version is Stata/SE 12.0. I used a server to run my program, it has 2 Xeon E5620 processors with 64 GB RAM.

Thanks,
Wei Lyu

[Steve Samuels]

Unfortunately, Stata SE can't take advantage of your server's multiple processors and cores. (Stata MP can.)

You have more than enough observations to fit a separate intercept to each group with xtlogit, pa, which is superior, I think , to conditioning out the intercepts with xtlogit, fe.

[Wei Lyu]

Thank you very much for your information about Stata/MP, I think it's time to write an e-mail to ask for an update.

I just did some changes to my model, and then the RE is much faster (around 2 hours); the FE is still running (almost 4 hours currently).