For my dissertation analyses I'm running a mediation model in SEM. I have one dichotomous dependent variable as well as continuous dependent variables. When I run my various models with my categorical outcome variable, I have to use WLS or WLSMV estimation. However, when I run my models with only my continuous variables, the default estimation of ML is used. The problem is that across my models, my sample size differs. Can you suggest what type of estimation I can use to make my models have the same sample size.
Can you recommend any papers that use mediation modeling in SEM with dichotomous and continuous outcomes.
You can use ML for categorical variables as well as continuous variables. You should have the same sample size unless observations are being eliminated due to missingness on covariates and the covariates in the model differ.
Thank you for your quick response. I had one more question about specifying certain variables as missing.
1. When I run my analyses using a missing are all (-9); statement for each of my models that I run, I get different number of observations for each of the models. However, when I specify specific variables to be missing For examples missing are H1MO9Rr H1MO10Rr H1MO11Rr H1MO12Rr H1MO13Rr (-9); I am able to get the same sample size across groups, however, my coefficients are different. So my question is, is it more preferred to have the same number of observations across the models being run for my various mediation analyses, or use the missing are all command for all of my variables.