Case-series Poisson Regression - "not concave"

[Siobhan Dooley]

Hi,

I am carrying out a case-series analysis looking at the risk of drug types on falls.
I have a final dataset that looks as follows:

ID	fall1	fall2	fall3 etc.	drug	exposure status	date	duration	logduration	outcome_indicator
1	12/3/98	6/7/02	5/2/08	7687687	1	4/6/02	50		1

an outcome indicator of 1 indicates that a fall occurred during a particular exposure period (example above).

There are no missing values, except when duration=0 therefore logduration=.

Each time I run the regression [xtpoisson outcome_indicator i.exposurestatus, i(id) offset(logduration) irr] - Stata says (not concave) after the 11th iteration.

Does anyone know why this is the case?

Thanks

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

If it quits saying not concave and converges, then you can ignore it. If it hangs with not concave, then you have a problem. Diagnosing this is tricky. I'd try simplifying the model first then adding bits until it hangs. I'd also try running it with regression - sometimes you'll see something odd in the regression results.

[Joao Santos Silva]

It is likely that you have perfect predictors of the zeros.

Best wishes,

Joao