non-nested RE in xtmixed returning error

[Megan Lawrence]

Hi,

I am using xtmixed to try to calculate non-nested RE for multiple levels and receiving the error “likelihood evaluates to missing” after >15 minutes of evaluation. The code I am using is as follows:

xtmixed depvar ||_all: R.sic ||_all: R.country ||_all: firm_id

There are approximately 15,000 rows in the data and about 11,000 unique firm_ids. I noticed that there was one other post on Statalist that didn’t seem to be answered about the same issue (by David Chan). For reference, I’m using Windows 7 Enterprise and Stata 13.1.

Thanks,
Megan

[Steve Samuels]

You've apparently tried a three-way crossed-effects model.

Page 394 of the Manual entry for mixed, the updated name for xtmixed, has this to say about a two-way crossed model:

Although we were able to fit the crossed-effects model (8), it came at the expense of increasing the column dimension of our random-effects design from 2 in model (5) to 57 in model (8). Computation time and memory requirements grow (roughly) quadratically with the dimension of the random effects. As a result, fitting such crossed-effects models is feasible only when the total column dimension is small to moderate.

It looks like you'll be unable to cross firm with anything else

You have about 15/11 = 1.36 observations per firm, so the data are highly imbalanced. I suggest that you group SIC codes into broader classes, possibly cross with country as fixed or random effect, then nest firms at the lowest level.

Good luck!

[Steve Samuels]

It's been a long time since I worked with SIC (Standard Industrial Classification) codes. If I recall, many US companies had only one SIC (four digit code) . If this is the case in your data, then a failure to cross firm with SIC is understandable apart from the dimension problem. would itself