Ordere Logistic Model - observations completely determined

[ely mattos]

Hi!
I am running a Ordered Logistic Model (ologit) on a 350.000 cases database. Besides my 4-categories independent variable, I have around 4 continuous explanatory variables and 8 categorical variables (some of them just dummies).
I am sturgling to make the model converge. When it does (after some adjustments in the explanatory variables), I got the warning "observations completely determined. Standard errors questionable". Even when I run the model on only two continuous variables as explanatories I get that.

Any tips on that?

Thanks a lot!

[Clyde Schechter]

You are asking people to troubleshoot imaginary code applied to imaginary data based on a highly condensed, minimally informative report of the unexpected results. I think to get a helpful response, you have to show the exact code you used, the exact complete output Stata gave you, and probably a small example of your data as well.

Please read the FAQ fully for good advice on how to ask questions on this forum effectively. Also, focus on #12 so that when you post back with more information you do so in the most helpful ways possible.

[Richard Williams]

Like Clyde says we kind of have to make wild guesses here. My wild guess is that you might have something like, when x = 4, y always equals 3. I would try crosstabbing my 8 categorical variables with my dependent variable and see if there is anything like that happening.

If we see your code and output other problems might become obvious, e.g. you are saying i.x and x has hundreds of possible values.

[ely mattos]

Hello!
Thank you Clyde and Richard!
First post here... sorry about the misspecification of the problem.
Let me try to provide you with more detail that could help:

I am running the following:
ologit INDEPENDENT i.X1(DUMMY) i.X2(5 categories) i.X3(3 categories) X3(continuous) X4(continuous) i.X5(dummy) i.X6(dummy) i.X7(6 categories) X8(continuous) X9(continuous) i.X10(3 categories) [fweight = peoplew]

I have already performed crosstabs and checked for that issue Richard pointed out: it is ok! Nos missing cells.
I have no missings too. My continuous variables are quite well-behave with just a few zeros, for instance.

As for the output, everything is fine - except for that warning. The coefficients are significant, presenting the expected signals, and the confidence intervals are narrow. The test of parallelism is also ok.

Finally, one important observation: when I run the model without -fwheight- the problem is gone. Obviously my confidence intervals become larger, but no more warning.

Thank you for your help and patience!

[Richard Williams]

I wonder about fweight. In order to be using it, you must have multiple cases with the exact same values, which seems a little unusual given the number of categorical and continuous vars you have. I would run frequencies and descriptive stats on peoplew and make sure it is ok.