sunyforever posted on Monday, October 28, 2013 - 10:44 am
Dear Linda or Bengt,
Do variables not included in the USEV list make a difference?
I have been running a LCA model with covariates. After fitting a model with nice estimated probabilities plot, I imported more variables as possible covariates (with missing values).
However, when I run the same LCA model (the Usevariables line includes exactly the same variables) with the new dataset, the outcomes are slightly different. The BIC, AIC, LMR, and BLR remain unchanged, but "PROPORTION OF DATA PRESENT FOR U," "UNIVARIATE PROPORTIONS AND COUNTS FOR CATEGORICAL VARIABLES," "FINAL CLASS COUNTS," Thresholds, and so forth are somewhat different. More importantly, the plot of estimated probabilities turns out not as nice as before (given the theory).
Do the newly included variables (with missing values) somehow have an effect on the model? Is there any way to fix this problem?