Dear stata community,
I am new to the latent class analysis and I have some questions regarding my model and the results in Stata, so I hope someone here can help me out.
The current model includes 11 indicator variables (3-level) that assessed the (no, unmet or met) needs of old-aged people. With LCA, we would like to see how these needs possibly cluster into certain classes of (no, unmet or met) needs.
The results were interpretable, but when I tried to run the 4-class model, endless iterations came up. I am very unsure about how to make use of startvalues (number of draws and seeds) and if they make sense in this case at all. Anyhow, I tried to limit the number of iterations to 350 and got a result, but it has missing standard errors at some of the variables, so it is not really interpretable (or model not identified?!). Now I tested for the fit statistics like the BIC and got a lower value for the 4-class model than for the 3-class model. However, as I said, I don't know how to make use of the 4-class model due to the described problems.
So my question is if anybody knows how to deal with this or how to constrain certain variables with missing standard errors?
After all, I am not sure whether LCA works at all with our sample because I read that each category/level of the variables should cover around 10% of the sample. However, the unmet needs in our sample are comparably low for each variable (1-8% unmet need) which also means that the variance is small. I also tried to dichotomize the variables but that wasn't successful in the analysis.
I would be very thankful for any help!
Best regards,
Sophia
I am new to the latent class analysis and I have some questions regarding my model and the results in Stata, so I hope someone here can help me out.
The current model includes 11 indicator variables (3-level) that assessed the (no, unmet or met) needs of old-aged people. With LCA, we would like to see how these needs possibly cluster into certain classes of (no, unmet or met) needs.
Code:
gsem (cseh1_p ckgesund1_p cpsych1_p chaus1_p cmobil1_p ckont1_p caktiv1_p cgeszu1_p csoz1_p cbezieh1_p cfinunt1_p <- ), mlogit lclass (A 3)
So my question is if anybody knows how to deal with this or how to constrain certain variables with missing standard errors?
After all, I am not sure whether LCA works at all with our sample because I read that each category/level of the variables should cover around 10% of the sample. However, the unmet needs in our sample are comparably low for each variable (1-8% unmet need) which also means that the variance is small. I also tried to dichotomize the variables but that wasn't successful in the analysis.
I would be very thankful for any help!
Best regards,
Sophia
