latent class analysis (LCA) multinomial logistic regression - no coverage issue

[Haley Ou]

Hello everyone,

I've encountered an issue while working with ordinal variables (with three levels) as independent variables and binary/ordinal/continuous variables as dependent variables. I'm seeking some clarification on the appropriate link function to use in different scenarios:

1. For a binary dependent variable (Y), should I use the logit link function?
2. For an ordinal dependent variable (Y), should I use the ologit link function? And use "i." for that ordinal variable
3. For a continuous dependent variable (Y), should I use the Gaussian link function?

Here's a snippet of the code I'm using:

All challenges variables are ordinal variables as independent variables.

For a binary dependent variable:

gsem (cov_pan_chal_hlth_base covid_pandemic_challenges_abode_ covid_pandemic_challenges_food_b covid_pandemic_challenges_water_ cov_pan_chal_med_base cov_pan_chlng_trans_base <-, logit) (C <- english_pro), lclass(C 4, base(2))
However, I encountered a "convergence not achieved" error. Could this be problematic? And if so, is the coefficient table still valid?

For a continuous dependent variable:

gsem (cov_pan_chal_hlth_base covid_pandemic_challenges_abode_ covid_pandemic_challenges_food_b covid_pandemic_challenges_water_ cov_pan_chal_med_base cov_pan_chlng_trans_base <-, gaussian) (C <- yrs_us), lclass(C 4, base(2))
But I received an error stating "option gaussian not allowed".

When I used ologit as the linkage for all, there were no errors.

What could be the issue here, and what suggestions do you have for resolving it? Any insights would be greatly appreciated.