SEM with control variables

[Michael Chen]

Hi everyone,

In a few links of my structural equation model (SEM), I seek to adjust or control for individual age (continuous), education attainment (ordinal with 5 categories), and race (nominal with 3 categories). I've read elsewhere online that the categorical variables (i.e., education and race) will need to be recoded and entered as dummy (binary) variables. Is that really the case? Or could they be entered directly into the SEM command?

For example:

sem (DV <- IV age i.educ i.race) ///

...{I'm omitting the measurement component of the syntax}

where
DV = dependent variable (outcome of interest, continuous)
IV = independent variable (primary predictor, also continuous)

Thanks,

Michael

[Phil Bromiley]

You can include a categorical variable as either continuous (imposing the assumption that 3 is three times 1) or as binary. Normally, binary is better. The continuous might occur if you had income 0=10000, 10000-20000, 20000-30000.

Yes, i.educ etc is the right way to do it. You may want to check GSEM - SEM makes normality assumptions, but I'm not use whether the issue applies here.

[Richard Williams]

sem, alas, does not allow for factor variable notation. I am not totally sure why. gsem does. So either compute the dummies yourself or see if you can use gsem for the model.