Power analysis

[d.asioli]

Hello,

I need to do sample size calculations and power analysis for a choice experiment (compare different treatments groups) and I have never done it. I will be analyzing data by using Mixed Logit Model after I complete the experiment. Is anyone can help where and how to start?

Thanks a lot,

Daniele

[Michiel Bliemer]

These publications describe sample size calculations for choice experiments, and the first discusses specifically for mixed logit.

https://www.sciencedirect.com/science/article/abs/pii/S0191261509001398
https://link.springer.com/article/10.1007/s11116-013-9451-z
https://link.springer.com/article/10.1007/s40271-015-0118-z

Ngene can do sample size calculations for mixed logit if you specify the model with random coefficients and appropriate parameter priors (preferably from a pilot study).

Michiel

[d.asioli]

Dear Michiel,

Thanks a lot for this reply. It was very useful.

Now in another experiment I use the Best Worst design (using incomplete block design), and I need to compare different treatments groups. How I can run power analysis to calculate the minimum sample sizes of the treatments?

Thanks
Daniele

[Michiel Bliemer]

I do not have much expertise in best-worst scaling and I am not sure if formulas for sample size calculations have been derived for this type of method. To do something similar as in discrete choice experiments, you would need to derive its analytical variance-covariance matrix, noting that it needs to account for two choices in each choice task (namely the best and the worst choice), and have priors for each parameter. If analytical derivations do not exist in the literature, you could resort to simulation in which you simulate choice observations for different sample sizes, estimate the model for each simulated data set, and observe the t-ratios or p-values of the estimated parameters.

Michiel