I am developing a SEM model for factors associated with dental caries. It has 48 observed variables and 13 latent variables. I have both independent and dependent latent variables in the model. Independent latent variable is knowledge. Some of the dependent variables are family oral health behaviors, parenting style, oral health related attitudes, Perceived oral health risk, Sweet consumption, toothbrushing, dental service utilization and so on. All the variables were measured by categorical data including few binary variables. My final outcome variable is binary(dental caries present/dental caries absent). My final sample size is 1023. Can I use LISREL for this model and what should be the estimation method?


1 Answer 1


In theory, yes. Since you have categorical measured variables (i.e., items), I would suggest Diagonally Weighted Least Squares (DWLS) or Full Information Maximum Likelihood (FIML). If you only had a handful of factors (e.g., 3 or 4), you would be in better shape. However, I would be skeptical of fitting such a complex model (i.e., 13 latent variables) with such a small sample. It is still worth a shot, however!

Finally, given your small sample, I would also consider the Bayesian framework, if none of the above suggestions work. Though I do not believe LISREL accommodates Bayesian methods, so you would need to look elsewhere (e.g., R and Mplus).

  • $\begingroup$ Thank you very much for the informative response. I used Robust Maximum Likelihood estimation using asymtotic covariance matrix to analyze the model. I got a model with meaningful values as the outcome using the above estimation with the final sample size of 1023. Is it technically correct to use Robust ML for this model as the estimation method? $\endgroup$
    – Nisha
    Apr 10, 2021 at 16:14
  • $\begingroup$ Yes I would say so. $\endgroup$ Apr 10, 2021 at 16:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.