I am using the BIC to compare a mediated model with an interaction term to the same model without the interaction term. I am not sure if I should be comparing the BICs or the Sample Size Adjusted BICs. The BIC is lower without the interaction term, but the SSA BIC is lower with the interaction term.
Can you tell me when I should rely on the SSA BIC over the non-adjusted BIC? N = 342.
I can't recommend one BIC over another. I would look at the significance of the interaction.
Anna Austin posted on Monday, August 21, 2017 - 12:04 pm
I am conducting LCA with 11 indicator variables, N=537 with data from a complex survey design. I have been reading a few articles (Nylund KL, Asparouhov T, Muthén BO, 2007; Tofighi D, Enders CK, 2008) regarding the optimal model selection criteria to use. In this case, it seems to me as though BIC or the sample size adjusted BIC would be most appropriate. However, do you have any thoughts on which (BIC vs. sample size adjusted BIC) should be given preference when the two criteria indicate a different optimal number of classes? Thank you!