I am studying the effect of two risk determinant in isolation and in combination. Using Stata software, I can generate a variable of the different possible stratas of the combination of the two risk determinant variables and then adjust a classical stratified logistic regression model. I have learnt that could also use the lincom command available in Stata. It appear to me that the results of both analyses are similar. Does anyone have experience with the lincom command and could explain the difference from the stratified logistic regression?
Hi Daniela,
I think it's hard to answer your question without knowing a little more.
Did you _svyset_ the data with your strata and run the logistic model using the svy: prefix?
Also, how did you set up the alternative logistic model? Did you run each model separately and combine them with _suest_ to make the linear combination?
- joe
Why don't you do a multivariate logistic regression with an interaction term?
The syntax would be: "logistic [outcome] [predictorA] [predictorB] [interaction term between predictors A and B]"
the lincom command allows you to add and subtract coefficients of the regression that you should have done beforehand. If you can provide more details about your question it would be easier to explain it.
I agree with Mr. Dany Matar's command. This command will help to find the Linear combinations of estimators, ie to find the effect of the exposure variable on outcome with the interaction effect and also its 95% CI. Without this command we need to calculate it manually by asking for the Variance Covariance matrices..
Hi Joseph, I haven't tried the options you suggest. To be honest, I don't have experience with the complex survey data analysis you suggested. I will study those and try to apply to my data. I believe the _suest command will need the lincom to estimate linear combinations anyway. Thanks for letting me know some alternative analysis.
Hi Dany,
I had tried this option and then used the lincom command find point estimates for linear combinations, as Devika pointed out. I was wondering whether the lincom command would consider the matched design the study has, which we control by using the clogit command in Stata. But I now guess it will.
If you used clogit for your regression, then the lincom command would account for the design since it uses the regression results to estimate combinations. So if the regression results account for matching (which they do when you use clogit) then the lincom would do that as well.
Many thanks to everyone for sharing your knowledge. I appreciated it very much!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics