Have an idea?

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

MBC Combinations Observed vs. Predicted

Dear All,

we conducted a MBC Project and estimated separate (HB) Logit models for each choice alternative  in the MBC excercise. Based on Count Analysis we are able to obtain the probability/frequency of two way choice combinations.

Is there a way to estimate the two way combinations from the logit models in order to approximate the frequencies of the Counts Analysis as good as possible?

Many Thanks Christiane
asked Aug 17, 2015 by Christiane

1 Answer

0 votes
As described in the documentation, you can indeed predict combinatorial outcomes using the market simulator within MBC (as long as you have used HB estimation).  You need to request combinatorial predictions within the simulator interface (click "Simulate Choice Combinations" then also click the checkbox to "Display Individual Results When Available".  

One of the output tabs will show the combinatorial predictions by respondent.  Then, analyze the output representing the individual-level combinatorial predictions using a tabulation software (or you could even construct this counting analysis within Excel) so that you can tabulate 2-way frequencies between the target dependent variables of interest.

As a side note: combinatorial prediction accuracy will be better if you initially formulate the dependent variables as combinatorial combinations (combinatorial dependent variables), where you collapse certain groupings of binary variables that seem to make sense to combine.  Those collapsed combinatorial coded dependent variables shouldn't exceed a few dozen potential outcomes, however.  For example, you might code up 4 dependent binaries as a single 16-level dependent variable.
answered Aug 18, 2015 by Bryan Orme Platinum Sawtooth Software, Inc. (128,265 points)
...