ologit command error

[sara jh]

Dear All,

I am trying to run a logistic regrssion using the ologit as follows:

ologit dummay_vari vari1 vari2 vari3 vari4 vari5 etc i.sic2 i.fyear, cluster(gvkey)



However, stata continue running showing the following without any result.

Iteration 0: log pseudolikelihood = -10177.617
Iteration 1: log pseudolikelihood = -35.320993 (not concave)
Iteration 2: log pseudolikelihood = -16.382337 (not concave)
Iteration 3: log pseudolikelihood = -1.9022478 (not concave)
Iteration 4: log pseudolikelihood = -.61051939 (not concave)
Iteration 5: log pseudolikelihood = -.06930767 (not concave)
Iteration 6: log pseudolikelihood = -.03565096 (not concave)
etc....


Thank you very much,

[Carlo Lazzaro]

Sara:
your model has a convergence issue.
Try it all over again, adding one predictor at a time and see when Stata starts to complain.

[sara jh]

thank you very much.

the independent variable on my data is financial restatement, which 1 when their is financial restatement and 0 otherwise, and data with value of 1 is only about %4 of the data, while the convergence issue is because one of the independent variable is whether the financial restatement is negative or not, and the majority of the financial restatement is negative.


Another question I need your help with, which is that one of the independent variables is the length of time the financial restatement last, the length could be for only a short period of time and sometime for longer period. This variable all time zero except when the dependent variable is one, it causes the outcome not normal. For example, the Pseudo R2 increases after I add this variable from 0.10 to 0.60.

How can I solve this issue please?




Thank you very much,

[Carlo Lazzaro]

Sara:
if your dependent variable ranges between 0 and 1, why using -ologit- instead og -logit- (or -logistic-)?
In addition:
1)

data with value of 1 is only about %4 of the data

is clearly an issue as far as convergence is concerned;
2) why the outcome should be normal?
3) why not sharing what you typed and what Stata gave you back via CODE delimiters (as per FAQ)? Thanks.

[sara jh]

Thank you,

Actually, after I added one control variable, the sign of the main interest variable change the sign from significant negative to significant positive.

Is that normal?

[Carlo Lazzaro]

Sarah:
as per FAQ, on order to get (more) helpful replies from interested listers, you are kindly requested to share what you typed and what Stata gave you back via CODE delimiters, please. Thanks.

[Richard Williams]

As Carlo keeps saying, we really could use a bit more information here.

A possible cause of sign flips is suppressor effects. Whether that is plausible here, I don’t know, since we know so little about what you are doing. See

https://www3.nd.edu/~rwilliam/stats2/l35.pdf