What if we could do “HB on the fly,” estimating and updating each respondent’s utilities after each individual choice task is answered? Doing so would open all sorts of opportunities for adaptation during an interview. Can we? Well, no, not quite. But with a few minor compromises, we can come darn close, while staying true to the standard HB/MNL model and assumptions. This talk will explore just how close we can come, explain the key computational idea involved, and consider how it could be practically implemented.
