How to suppress categorical outputs in proportional hazard model

[Meng JI]

Hi everyone,

I have a question about how to output part of the regression results for main IVs in the cox proportional hazard model.

For instance, my code is something like : stcox age ndrugtx treat site

In the regression result, I only want to keep the result for age while suppressing coefficients for other variables.

I wonder if anyone knows how to achieve this in Stata?

Thanks

[Clyde Schechter]

You cannot do that to the -stcox- output directly. However, if you pass the results through -esttab- or -estout-, you can use -keep()- and -drop()- options to show only those coefficients you want to show.

[Meng JI]

You cannot do that to the -stcox- output directly. However, if you pass the results through -esttab- or -estout-, you can use -keep()- and -drop()- options to show only those coefficients you want to show.

Hi Clyde,

Thanks a lot for your reply. I've used code below to get an output of the coefficients and significance of the variables:

However, I wonder if you know how to add the significance level of model (e.g. Prob>chi2=0.7957) into the output?

Thank you and look forward to your reply.

esttab, label keep(age)

[Clyde Schechter]

I think you can do that by specifying the -scalars(chi2)- option in -esttab-.

Added: No, sorry that's the chi square statistic, not the p-value. It seems -stcox- does not return the overall p-value in e(), so I don't believe it is possible to get it into the -esttab- output. At least I don't know how.

I suspect the reason that the model p-value is not returned in e() is because it is seldom, if ever, of real interest. I see, in your case, there are only 2 model df, so with that few variables, perhaps that is of more interest than it would usually be. I would also note, however, that your N is only 11, and maximum likelihood estimation is asymptotically correct in large samples. And with N = 11 and 2 predictors, you barely have 5 observations per predictor, which means your model is probably overfit to the noise in the data. So I'm not sure that that p-value would mean much even if you could somehow extract it.

[Meng JI]

I think you can do that by specifying the -scalars(chi2)- option in -esttab-.

Added: No, sorry that's the chi square statistic, not the p-value. It seems -stcox- does not return the overall p-value in e(), so I don't believe it is possible to get it into the -esttab- output. At least I don't know how.

I suspect the reason that the model p-value is not returned in e() is because it is seldom, if ever, of real interest. I see, in your case, there are only 2 model df, so with that few variables, perhaps that is of more interest than it would usually be. I would also note, however, that your N is only 11, and maximum likelihood estimation is asymptotically correct in large samples. And with N = 11 and 2 predictors, you barely have 5 observations per predictor, which means your model is probably overfit to the noise in the data. So I'm not sure that that p-value would mean much even if you could somehow extract it.

Hi Clyde,

I see. I will keep the chi-square statistic for now. Thank you so much for your detailed explanation of what happened in my current model. It really helps a lot!

Best wishes
Meng