Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Is there ever a situation then when aggregated logit should be used instead of HB?

Thanks for the answer, Bryan. This leads me to wonder then what situation would one look to consider using aggregated logit over HB.
asked Aug 12, 2013 by anonymous

1 Answer

0 votes
When the data are especially sparse (few tasks relative to parameters to estimate, or partial-profile CBC), then aggregate logit or a low-dimensionality latent class solution could possibly be superior to the default CBC/HB run.

But, the HB enthusiasts (if facing very sparse conditions) would just counsel you to reduce the Prior Variance to something like 0.25 or 0.5 and to increase the Prior Degrees of Freedom (these are both Advanced settings in CBC/HB) to something like 1/4 or 1/2 the sample size.  This causes the individual HB results to smooth much more in the direction of the population means, and probably would erase the advantage of aggregate logit or Latent Class.
answered Aug 12, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (175,715 points)