In preparing for a conference paper, we realised that the choice community as a whole has adopted an unrealistic method for dealing with the “No Buy” option in Choice models when asked as a second question after a “forced” choice between the alternatives. This is known as the “Dual Response None” case. The net result was that if simulations contained more or fewer alternatives than the original number of alternatives in the choice exercises, the simulations were less likely to be good, resulting in too high or too low a share for the “no buy” options and biased shares for the products. The advice was always to not simulate products that were “too similar” but we were not satisfied with that guidance and wanted to fix this issue.
After reviewing the original papers on the “dual response none” question, we realised there was a better way of modelling the “No buy” option in these cases, which more closely mimics peoples real world decision making processes. This splits the decision into two parts: they first choose the “best” alternative and then decide whether they would buy that alternative, rather than evaluating all the alternatives again with the none option. All of our simulators have been modified to take this into account and give much better simulations when the number of available products are varied.
We presented this approach at a breakout session at the Sawtooth 2015 conference in Florida.