topic Re: PROC CAUSALMED warning massages in Statistical Procedures

PROC CAUSALMED warning massages

llllllll1 — Tue, 10 May 2022 15:04:47 GMT

Hello!

When I'm trying to perform the PROC CAUSALMED recently, I receive these warnings:

WARNING: At least one element of the gradient for the outcome model is greater than 1e-3.
WARNING: The Hessian for the outcome model has been ridged with a maximum value of 524288. Standard errors might not be accurate.

Since the outcome variable is not rare, the log link was suggested in SAS help. And I've read the solutions on this website https://communities.sas.com/t5/Statistical-Procedures/PROC-CAUSALMED-warning-messages/m-p/811960#M40019 , but as you can see, I've no reason to reduce those covariates, and there is no interaction option initially, so what else can I do to treat this problem, or whether the results are still reliable even warnings are still existed? code is as follows:

proc casusalmed data=data all;
class X(ref=first) M(ref=last) Y(ref=first) C(ref=first);
model Y=X M /link=log;
mediator M=X;
covar c;
run;

the treatment, mediator, and outcome are all binary variables.

Re: PROC CAUSALMED warning massages

SAS_Rob — Tue, 10 May 2022 16:02:24 GMT

Most likely it is still an issue with either the predictors or the single covariate in the outcome model. It would be helpful to attach the following output from Proc FREQ to diagnose the issue.

proc freq data=all;

tables y*(m x c);

run;

Re: PROC CAUSALMED warning massages

llllllll1 — Wed, 11 May 2022 01:27:46 GMT

Thank you so much for your response! I checked the frequency chart but couldn’t find any abnormal from it. Then I tried to reduce the number of covariates and found when the main covariate was rejected, warnings disappeared. However, the covariate has been proved to be associated with the outcome in multivariate logistic analysis. And I don’t think it’s a good way to get rid of that covariate.