getting problem with convergence in Truncated negative binomial count model

[yashodha]

Dear all

I am analyzing the recreation data, where trips taken takes non zero value. Trying with the truncated negative binomial model, am facing non-convergence when i use particular two variable in the model. When i use rest of the variable, the model convergence and fits well. I need to include one variable which is very important and leaving out that might yield biased results, but if i use it, it resulted in non-convergence.
Please could any one let me know what is the problem with this and is there any alternative?.

Note: Checking the dispersion the mean is far bellow that of variance (mean =4.2 and SD= 8.5)

please help me,
yashu

[Carlo Lazzaro]

Dear yashodha (as per FAQ, your full name would be welcome. Please re-register via the Contact us button at the bottom right of the screen):
- as per FAQ again, please report exactly what you typed and what Stata gave you back as a result.
As your query is currently posed, I do not to understand which mean and variance (probably a standard error) you are referring to.

Kind regards,
Carlo

[Clyde Schechter]

It is hard to know what is causing your problem without seeing your data and knowing exactly what commands you are trying.

One general point: some of these models have difficulty converging when there is a wide range of values among the independent variables. So, if one of the predictors ranges from, say, 1 to 40 and another ranges from 100 to 10,000,000 (or, in the other extreme, from 0.000001 to 0.001) this can create problems. You might try rescaling your variables so that they all have similar orders of magnitude. Sometimes that helps.

Also, in some models, convergence is easier if the variables are centered around some point in the middle of their range.

Those are a couple of general tips, that may or may not be useful in your rather unspecified situation.

[yashodha]

Schechter,
Thank you very much for you suggestion. Yes i realized it too, where income variable in 10000 to 1000000 order and multiple trip variable takes 0 or 1.
But the strange thing i get convergence when i use these both variable at a time. when i add purpose of visit (0= joint purpose, 1=recreation) along with monthly income and multi-destination trip variable, it gives me non-convergence problem.
Am using Truncated negative binomial model, since mean of number of trips in a year is far below that of variance.
the command i am using is

tnbreg trips tcost mincome tc*minc1 tc*minc2 tc*minc3 gender age edu mtrips -------------this one works ok
but if i use tnbreg trips tcost mincome tc*minc1 tc*minc2 tc*minc3 gender age edu mtrips pvisit--------------- gives non convergence.

note tc*minc2; here we divide monthly income in to four quantails and then interact with travel cost. purpose of doing this is we are want to know how htw high income people have when increase in the travel cost when compare to low income people.

Please anyone let me know why am facing this problem.

regards,
yashu

[Maarten Buis]

I would start with centering and rescaling your income variable, e.g.

gen mincome_c = (income - 20000)/1000
label var mincome_c "income centered at 20,000 (euros/dollars/cows?) measured in 1,000s of (euros/dollars/cows?)"

and enter mincome_c instead of mincome in your model.

[yashodha]

Hi,
Thank you. I will do that.