Hi,

the interpretation of a categorical predictor in SEM is just the same as it is in any convenient regression model. However, as in simple regression, it strictly depends on how you have coded/included the categorical variable in your model. For a three-categorical variable, one would typically use dummy-coding, i.e. you form two indicator variables (1-0-0 and 0-1-0) representing two of your three categories, include them in your model, and interpret the results in reference to your third category.

If you have obtained only one estimate for your categorical variable in your SEM I would assume that you do not have included two dummies but just have included the whole three-categorical variable. Thus, the model assumes a linear relationship with your outcome - which does not make sense. So please have a look at your model and consider respecification as described above.

Regards,

Alexander Pabst

Dear Jaspreet,

I am not sure you need to use SEM in your case. Are your dependent variable latent constructs or just observed variables? You have a simple model, and maybe MANOVA will give an answer to your research question.

Dear Jaspreet,

I agree with Marina. It seems you don't need to use SEM. You have several continuous dependents variable and categorical IV variable. I simpler model, that accounts for the multiple DVs will test the mean differences accross the levels of the IV.

Thanks everyone. Here I have asked question only with respect to my categorical variable but in reality I have continuous predictors also. That is why I was using SEM to put everything in context.

Please find attached my datasheet. I am specifying the following model for SEM in R.

model

I recommend not using categorical variables in your SEM  and to interpret the direct and indirect effects as if your are using Standardized Beta Coefficients if your variables are continuous variables.

Dear Jaspreet,

If you have several metric predictors and one categorical, you should run multiple group analysis. The categorical will be the grouping variables, and will work as a moderator.