No announcement yet.
  • Filter
  • Time
  • Show
Clear All
new posts

  • Help to figure out GSEM with ordinal variable


    I am using GSEM to estimate and test direct and indirect effects in a system of regression models. Specifically,

    gsem(x2 <- X1)(Y <- X1 X2), vce(robust)

    where y and x2 are the endogenous and continuous variable. X1 is the exogenous and ordinal variable. My goal is to evaluate the direct and indirect effect of X1 on Y.

    My question is this, can I simply treat X1 as normal continuous variable? If not, is there any function in gsem package that deal with my current setup?



  • #2
    Originally posted by KC Lin View Post
    X1 is the exogenous and ordinal variable. . . . can I simply treat X1 as normal continuous variable?
    I've seen that done in conventional regression modeling, although it seems to be just as common to treat an ordered-categorical explanatory variable as a factor variable. It depends, I think, on what's traditional in the field of study, how many categories there are to the variable and whether there is an expectation of a prominent nonlinearity in the relationship. With gsem, you can use factor variable notation if you choose to go that route.

    I don't know whether or how all of this plays into the direct-versus-indirect question, and so I cannot help you there if that's a consideration to you.


    • #3
      What hypothesis are you trying to test? That should help you identify the best path forward. There isn’t really a good way to handle this case so it really is a matter of which assumptions you believe are best suited to your question and model more generally. You should also consider the measurement scale of your mediator since it can be challenging to interpret decomposed effects if the measurement scales vary between the variables in the model.