SKIM names Maura Coracini as Managing Director, Latin America
SKIM has named Maura Coracini as Managing Director, Latin America. Based in the São Paulo area.
On the 27th of June and the 28th, we presented two different topics on Preference Based Conjoint and Iterative Multilevel Empirical Bayes (IMEB) at the 4-day 2017 Advanced Research Techniques (ART) Forum held by American Marketing Association (AMA).
Ranging from data scientists, marketing professional and to academics, the ART Forum is an event where you find a network of colleagues who are serious about developing tools that can solve the next generation of problems in the field of marketing.
Our Director Methodology and Innovation EU, Jeroen Hardon and Research Director, Marco Hoogerbrugge, presented a topic on “Preference Based Conjoint: Can it predict dozens of product?”. As for the 2nd presentation, our VP Methodology and Innovation, Kevin Lattery presented “Iterative Multilevel Empirical Bayes (IMEB): An efficient flexible and robust solution for large scale conjoint.”
Insights from “Preference Based Conjoint: Can it predict dozens of products?”
Presenters: Jeroen Hardon and Marco Hoogerbrugge
https://www.slideshare.net/SKIMgroup/big-simulators-with-dozens-of-products-which-conjoint-method-is-most-suitable-at-art-forum-2017
Conjoint analysis is often used for very complex markets, with a large number of attributes and products in the market. Ideally we would replicate the complexity of the market that exists in reality as good as we can in the design of the conjoint survey.
Although we already proved in the past that having more reality makes the predictions better, this is not always feasible. Most particularly, when there are quite a few attributes, we cannot show too many concepts in a single task as this would lead to information overload for the respondent.
The key question in this presentation was to check what extent the discrepancy between the number of concepts in a task (3, or 4) versus the number of products in the conjoint simulation model (20, or 50) may influence the results of the study, by comparing CBC, ACBC and PBC (preference based conjoint).
Insights from “Iterative Multilevel Empirical Bayes (IMEB): An efficient flexible and robust solution for large scale conjoint”
Presenter: Kevin Lattery
https://www.slideshare.net/SKIMgroup/iterative-multilevel-empirical-bayes-imeb-an-efficient-flexible-and-robust-solution-for-large-scale-conjoint
The typical approach to conjoint analysis uses Hierarchical Bayes with Gibbs Sampling to integrate over an upper level multivariate normal. This works well with small to medium size data, but with many parameters may be inadequate or impractical.
We described an approach that fixes the upper level prior based on the data (hence Empirical Bayes) and cross-validation of a scale parameter. This fixed prior eliminates MCMC iterations, making it much faster than standard HB. The Empirical Bayes approach also avoids issues of inadequate convergence and improper scaling that may occur with MCMC.
We also described some of the additional flexibility of this approach, including respondent level customization and the ability to easily estimate utilities for additional respondents.
