After I ran latent class analysis with five classes, I used CPROBABILITIES to save all class probabilities and a class membership for each individual. However, final class counts based on their most likely latent class membership indicated in the output file are different to the class counts based on the file saved by CPROBABILITIES. The analysis used sample weights.
Thank you for responding to my question, Dr.Muthen. However, after I weighted the whole analysis in Stata by using svyset, the means between the one generated in TECH 7 (sample statistics based on posterior probabilities) and the one analyzed in Stata that weighted by using person weights are different. Should I use class probabilities such as CPROB to weight the save values in Stata?
You have to weight the observations by CPROB and by the sampling weight. Multiply the two weights. This will match the tech7 output. The Cprob and the most likely latent class membership are different: Cprob allows for the fractional class membership (according to the posterior class probabilities) while the most likely latent class membership does not allow fractional class membership and just categorizes the individuals into classes.
Raphael Kim posted on Tuesday, April 10, 2018 - 10:23 am