ApolloChoiceModelling forum

Hi all,
I have estimated an MMNL model with one latent variable using Apollo 0.2.9 on R 4.3.1
I have 723 respondents answering 4 menus each with four alternatives (so the Number of modelled outcomes should be 11568)
When I ran the model again using Apollo 0.3.1 on R 4.3.2, for some reason, the Number of modelled outcomes: 5061, the Model diagnosis was : Relative function convergence and obviously, the estimated parameters were different
The code files are exactly the same, the only difference is that in the old versioned file, I did not specify the "rows=(task==1)" for the LV estimation.
I could not attach the output files as this format is not accepted.
Any idea what could have led to these different estimation results (same code, same data) ?
Anat

I think that understand the problem, the correct number of models is indeed 5061 = 723*7, I should have specified the "rows=(task==1)" for the LV estimation. I just don't understand how it converged in the previous version (0.2.9).

Anat

modelled outcomes is the number of separate dependent variable observations. Relative function convergence is the BGW terminology for convergence.

Using rows=(task==1) is correct, and will indeed give you different estimates as you don't overstate the weight for the measurement model part

Stephane