Latent Class

[col54]

I am designing a choice experiment which I will use latent class modelling to analyse. I have seen it discussed elsewhere in the Forum that Ngene doesn't enable efficient latent class designs. Therefore, am I better using a fractional factorial design or an efficient designed optimised for mnl? I am assuming the former?

[Michiel Bliemer]

I am not sure what you mean, an efficient design is a type of fractional factorial design, just like an orthogonal design is a fractional factorial design.

If you want to optimise a design for estimating a latent class model then you could consider doing this by specifying a model for each class with corresponding class-specific parameter priors. For example:

;eff = fish(mnl,d)
;fisher(fish) = des1(class1[0.5],class2[0.5])
;model(class1):
U(alt1) = b1[0.1]*x1 + ... /
...
:model(class2):
U(alt1) = b1[0.2]*x1 + ... /
...

If you do not have information on the priors for each class then there is no advantage in optimising for a latent class model and you may simply optimise for a multinomial logit model with a single set of parameter priors.

Michiel