Looking for Stata package for mediation analysis allowing for factorial variables

[Mengmeng Li]

Hi everyone,

I am writing to know if Stata packages for mediation analysis are available.
I found some information about "-paramed-" and "-medeff-" commands, but they did not quite fit into my data structure.

I have a panel data, for some observations they have repeated measurements (two repeats) for the primary predictor, for others only one measurement per observation. Outcome variable is continuous. I am applying multiple linear regression model with -cluster- option for standard error computation. Based on these, -medeff- seems more appropriate because it allows for -cluster- option, but the issue here is that it does not support factorial variables. Both of my key predictor and other confounders are categorical variables, so have to be included as i.varname in the model.

I greatly appreciate if anyone could give me some insights on the alternative methods to overcome the issue or some other ways to perform mediation analysis.

Thanks so much.

Best regards,
Mengmeng

[Roman Mostazir]

You can use Stata's -gsem- (Generalized Structural Equation) command to fit the model. It will all factor variable notation and can adjust the model for repeated measures. See help for -gsem-

[Phil Bromiley]

It is also possible to run dummy variables without using factor variable notation - look at the xi command. Most Stata-provided routines don't require xi since they work with factor variables, but there is nothing to stop you from generating your dummy variables yourself. This is not generally recommended since it wastes time and also makes using margins trickier.

[Mike Lacy]

The -khb- user-written routine for mediation analysis will accept factor variables. See -ssc describe khb-.