Deviation between LnL-fit indices value and LnL-development

[janschreier]

Dear smartPLS-Team,

I just realized that the LnL-value in FIMIX differs between what is shown in the LnL-dev. Report and the fit indices report. Can you explain the reason fort hat?

Also I wonder how you would prove that a data source is sufficiently homogeneous. If I calculate a one segment solution in FIMIX I get an LnL of -32707, for the three segment solution LnL is -23524. While I guess this is a large enough difference, what deltas between LnL-values would you consider relevenat? (is it 1, 10, 100 or 1000?).

Thanks for your support, as always :)
Jan

[janschreier]

Hi smartPLS-team,

could you answer the first question of the previous post? ("why does the fit indices report show a worse value for LnL than the LnL-Report?")

Thanks!
Jan

[jmbecker]

Hi Jan,
there may be a bug in the last release version in how it handles the FIMIX repetitions. I can replicate the problem with the release, but in the current working version, I cannot replicate the problem. Thus, we may have fixed this already as we also have implemented saving the LnL of intermediary repetitions and providing a repetitions history for the LnL. This will all be part of the next release, but there the next release may take one or two more months.
Until then, I can only suggest using 1 repetition at a time to avoid that problem, as it also affects the other segment retention criteria.