Clogit with fixed effects

[Masha Zakharova]

Dear statalist,

I am trying to replicate a model that has a vote choice as a dependent variable, and a set of regressors that measure various factors for each party for each individual (for example, distance between a party and an individual on the left-right scale), as well as fixed effects for parties (the model I am trying to replicate is in Bolsted and Dinas, 2016, Table A9 in Appendix, it is run for many elections/countries with parties' fixed effects). I tried emailing the authors but never got a response (it has been a month now). Anyway, I ran following script on my data:

clogit VOTE distance i.partynumber, group (idd), where partynumber is a different number for each party in all elections, and idd is a voter id variable.

So this is what an example of few rows of data (simplified) looks like:

idd	VOTE	distance	partynumber	electionid	number_within_election
1	0	2	21	1	1
1	1	0	22	1	2
1	0	3	23	1	3
2	0	4	21	1	1
2	0	3	22	1	2
2	1	1	23	1	3
3	1	2	40	2	1
3	0	3	44	2	2

The last 2 variables in the table are not used in analyses, they are just shown here to show how the data is structured.

If I run this model on one election, it runs just fine. However, when I try to run it on more than one election it just never converges. My question is, is my model specification wrong, or is there something wrong with the way I structured the data?

I tried interacting party number within election with electionid, as well as creating a long dataset that has as many rows for each individuals as there are parties in all elections (128) - took 2 days to create it, but the model never converged either.

Thank you for your time,
Masha

[Phil Bromiley]

Please check the FAQ on asking questions - providing Stata code (using code delimiters), Stata output, and sample data (using dataex) will increase the chances you get a helpful answer.

I'm not sure what went wrong. I'd try running the model with regress and i.idd as a regressor, or xtreg with idd as the panel. Sometimes oddities will turn up in such estimators that inform you on other estimators.

I wonder if you don't need some voter or party characteristics in your model. If you're including idd as a panel, then the only explanatory variables are party number dummies.

[Masha Zakharova]

Paul,

Thank you very much for your answer. I will look into FAQ and try running the commands you suggested, will see if that helps!

Thank you,
Masha