mixed command on STATA -- how many iterations??

[Jeffrey Leow]

hi i ran this code on my STATA dataset
mixed pat_cost_inf age i.white i.married i.charlsoncat i.payor i.hosp_type i.location i.bedsize i.region surg_total hosp_total i.proc_type [fweight = proj_wgt] || prov_id:

dataset itself is not that big - 141 variables, 84972 observations
no of distinct prov_id = 500

it has been running for the past 5 days and is onto

Iteration 7136: log likelihood = -6229132 (backed up)
Iteration 7137: log likelihood = -6229132 (backed up)
Iteration 7138: log likelihood = -6229132 (backed up)

and still going!

how long should i expect for it to run?
or does it not converge at all for eternity?

[Jeffrey Leow]

also, does the log likelihood remaining the same, mean that it is not converging and will not converge?

[Carlo Lazzaro]

Jeffrey:
what you're experiencing is, in general, a symptom that the model should be better specified.
For instance, what hit my eyes is the relevant number of categorical variables : do you really need all of them? Can't you go for a more parsimonious model? What does the literature in your reserch field suggesting when reports the same research topic?.
In those instances, the recipe is always the same: start your modelization all over again, add one predictor at a time and see when Stata starts behaving problematically.

[Oded Mcdossi]

In addition to Carlo's good advices I noticed that you specify frequency weights for your weight type. Are you really sure that you need fweight and not pweight.

[Jeffrey Leow]

Carlo - thanks a lot. ill try to choose a few covariates first and try it out. yeah actually they all have been implicate in one way or other. in any case, ill do them one by one and see where this leads.

Oded -- yeap you're right! i'm using sampling weights so it should be PWEIGHT instead. thanks for picking that up!