Let's suppose I have an outcome variable that is not continuous. For example, it could be an ordered categorical variable or a nominal (unordered) categorical variable. This implies that I would need to model the variable as an ordered logit model or as a multinomial model.

I know lavaan and other softwares does not include the option to fit a whole structural equation model SEM with outcome variables that are not continuous (there is only a workaround for binary variables). As I would like to still use latent variables in the categorical model, I was wondering if it is correct to calculate the factor scores using the lavPredict() function and just include these latent variables as regressors in an external categorical model. I have seen many examples where people follow this approach but using Exploratory Factor Analysis EFA or Principal Component Analysis PCA. I would preper to rely on CFA since I do have a clear and solid theory about the latent variables.

Moreover, the lavPredict() function has the argument method with the options regression and Bartlett. Should I use the regression or the Bartlett factor scores? what are the differences among the two alternatives?

In other words:

HS.model <- ' visual  =~ x1 + x2 + x3 
              textual =~ x4 + x5 + x6
              speed   =~ x7 + x8 + x9 '

fit <- cfa(HS.model, data=HolzingerSwineford1939)

Latents <- lavPredict(fit)
Latents <- as.data.frame(Latents)

Latents_Bartlett <- lavPredict(fit, method = "Bartlett")
Latents_Bartlett <- as.data.frame(Latents_Bartlett) 

It is correct to use the latent variables visual, textual, and speed in the dataframes Latents or Latents_Bartlett as regressors in a subsequent ordered logit model or multinomial model?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.