Popular for their practical, practitioner-oriented focus and depth in the fields of choice/conjoint analysis, segmentation, and data collection/analysis, the Sawtooth Software conference was back this year with us presenting for the first 3 days in this 5-day event.

Whether you are a methodologist or a general survey programmer, there were many sessions that would interest you. Among other sessions presented by experts in their respective fields, our in-house method heavyweights presented a couple of sessions at this conference:

Two-day Turbo Choice Modeling (Mon 5 – Tue, 6 March 2018)

While many topics mentioned Sawtooth Software’s CBC- and MBC-related programs, the principles generalize to any other software for choice modeling. The sessions emphasized practical issues and practical solutions more than theoretical academic research.

Topics include:

  • what we’ve learned from eye-tracking in CBC,
  • system 1 vs. 2 thinking, convergence in HB,
  • price optimization with Nash Equilibrium,
  • optimization algorithms,
  • upper-level model and context effects,
  • bandit MaxDiff,
  • searching for interaction terms with HB analysis,
  • dual-response price,
  • sparse/express MaxDiff,
  • making CBC more engaging.

Practical tips and tricks for Conjoint and MaxDiff (Tue, 6 March 2018)

In this practical session, you would partly be in charge of what we taught. We provided you with a large set of practical topics, of which you decide which would be most relevant. If you struggle with questions like:

  • Which conjoint method is the most appropriate for the business question?
  • Which kind of None should I use?
  • First choice or share of preference?
  • What is the difference between preference share, market share and volume share?
  • How many concepts do I show?
  • Should I line-price my products?
  • How many products can I include in my study?
  • I have too many items to do a regular MaxDiff, now what?

Preference Based Conjoint: Can it be used to model markets with many dozens of products? (Wed, 7 March 2018)

Conjoint analysis is often used for complex markets, with dozens of products in the market. Ideally we would replicate the existing complexity of the market as well as we can in the design of the conjoint survey but that is not always feasible. The key question in this presentation was to check if a different way of constructing the statistical design can improve the prediction for simulators with many dozens of products.

Tools for dealing with Correlated Alternatives (Wed, 7 March 2018)

Correlated alternatives violate our standard conjoint modeling assumptions (IIA). While respondent level utilities help, sometimes that is not enough. We describe and compare several tools for dealing with correlated alternatives. These include full blown nested logit, error components logit, and post-hoc simulator adjustments.