Have an idea?

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

Batch logit runs with SMRT

Hi, I need to run numerous logit runs on my dataset for different segments. For example, I want to be able to compute utilities first for the entire dataset (total > 1500 respondents), then only for respondents in specific geographies (5 different possible locations), then by customer age, then by the combination of location and age, etc.

The outcome of all this will be used in a market model aiming at testing different offers on various customer segments in order to determine which one(s) we will keep.

The only way I found to be able to do this with SMRT is to create segments for first each segmentation criteria, and then for all the combination of those criteria (e.g. Location 1 and Age range 1, Location 1 and Age range 2, etc.). Overall, I will have to run more than 100 logit runs (even though all of them will not be exploitable).

Is there a way to process batch logit runs with SMRT, i.e. automatically performing logit runs on a predefined list of segments?

Thanks in advance for your help
asked Jun 4, 2014 by anonymous

1 Answer

0 votes
Most people avoid this type of a problem by computing utilities as the individual level via HB...then they don't have to run aggregate logit separately based on every subgroup they want and have to keep track of all these separate runs.  With HB, you get a single utility run (but each respondent has unique preference scores in the utility set) that you only have to compute once and then you can just move the segments into SMRT and run simulations via different banner points.  Or, if you are conducting the analysis in a cross-tab or statistical package separate from Sawtooth Software, you read the segment memberships and individual-level utilities into that package and crank through your analysis with the segments as filters or banner points.

Unless your data are especially sparse (something like 5 tasks or fewer per respondent), then running aggregate logit (and relying on it as a final model) tends to be a relic from the 1980s.   Not to say that HB works under all conditions and that aggregate logit has no place today...but making a general comment representing current practice.
answered Jun 4, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (131,390 points)
Thank you for your answer. I will look into that.

I've computed the individual level utilities thanks to CBC HB but am having trouble clustering the information for the groups I am interested in and could not find any information on the web.
Once I have all the individual utilities in a table (with each row corresponding to one respondent) , do I simply select the respondents I am interested in and average the utilities by attribute ?