SF Wang posted on Tuesday, November 11, 2014 - 8:08 am
In order to select # of latent classes, do we generally start from 2 and increase until k class is no longer significantly better than k-1 class in LO-MENDELL-RUBIN LRT test?
What if c=3 has p-value 0.4, but c=4 has p-value .03, and BIC of c=4 is lower than BIC of c=2? should we consider to use BIC to select from c=4 and c=2? or should we stop at c=3 due to non-significantce observed at c=3?
When looking at the Lo_Mendell-Rubin LRT test, stop interpreting once you obtain a non-significant p-value. See how many classes are suggested by the statistics you look at and use those number along with theory to make a final decision. See the Topic 5 course video and handout on the website where this type of decision making is discussed.