Presentations

Upcoming Conference Presentations by Dr. Jay Magidson:
Sawtooth Software Conference 2016

Conference Presentations by Dr. Jay Magidson:
Advanced Research Techniques (ART) Forum 2016
Advanced Research Techniques (ART) Forum 2015
Modern Modeling Methods (M3) Conference 2015
Sawtooth Software Conference 2015


Conference:
Advanced Research Techniques (ART) Forum 2016
Date:
June, 2016
Title:
How to Develop a MaxDiff Typing Tool to Assign New Cases into Meaningful Segments
Presenters:               
Jay Magidson, Founder, Statistical Innovations Inc.
Gary Bennett, The Stats People Ltd
Abstract:
Use of latent class modeling to analyze MaxDiff response data has been found to be a good way
to obtain meaningful needs-based segments. In order to classify new cases into the most
appropriate segment, we describe and illustrate an innovative approach involving simulation to
develop a typing tool where a relatively small number of binary items is included in future
surveys, each item requesting a simple choice between a specified attribute pair. Both the
latent class analysis and simulation are performed using the Latent GOLD® program.
Download presentation: ART2016_MagidsonBennett (pdf)


Conference:
Advanced Research Techniques (ART) Forum 2015
Date:
June 15, 2015
Title:
A New Modeling Tool for Identifying Meaningful Segments and their Willingness to Pay: Improving Validity by Reducing the Confound between Scale and Preference Heterogeneity
Presenters:               
Jay Magidson, Founder, Statistical Innovations Inc.
Jeffrey Dumont, Senior Consultant, RSG
Jeroen K. Vermunt, Professor, Tilburg University
Abstract:
With discrete choice data, it is common practice to use HB or latent class modeling to capture heterogeneity across decision-makers. However, a significant part of the heterogeneity retrieved relates to differences in the amount of error variance, a phenomenon referred to as scale heterogeneity. As a result of the preference vs. scale heterogeneity confound, traditional approaches to segmentation may yield spurious segments that differ only in their scale heterogeneity, and may not differ at all in their willingness to pay. In this presentation we introduce a new scale adjusted latent class (SALC) choice model that accounts for both observed and unobserved scale heterogeneity, resulting in homogeneous segments that differ only in their preferences and willingness to pay, increasing the usefulness of segmentation analyses to marketers. We compare results from the SALC approach with other standard and nonstandard approaches.
Download presentation: ART2015_MagidsonDumontVermunt (pdf)


Conference:              
Modern Modeling Methods (M3) Conference 2015
Date:            
May 19-20, 2015
Title:    
Using a Scale-Adjusted Latent Class Model to Establish Measurement Equivalence in Cross-Cultural Surveys: An Application with the Myers-Briggs Personanlity Type Indicator (MBTI)
Presenter:        
Jay Magidson, Statistical Innovations Inc.
Abstract:
The Myers-Briggs personality indicator (MBTI) is one of the most commonly used and well known personality assessments in the world, with the items now being translated/adapted into 17 different languages across more than 20 countries. In this paper we describe how the challenge of  establishing measurement equivalence across cultures and different languages was achieved using a scale adjusted latent class (SALC) model.  In addition to describing the SALC model, we explain the assumptions underlying worldwide measurement equivalence, and describe the results.
Download presentation: M32015_Magidson (pdf)


Conference:    
Sawtooth Software Conference 2015
Date:
March 26, 2015
Title:
Models for MaxDiff and Related Data and their Performance with Latent Class and HB Methods
Presenters:   
Jay Magidson, Statistical Innovations Inc.
Gary Bennett, Logit Research
Abstract:
By eliciting the worst (least preferred) as well as best (most preferred) options, MaxDiff data are considered the “gold standard” for estimating preference for a group of stimuli (objects). In this presentation we use simulated and real MaxDiff data to compare various MaxDiff and extended Best-Worst (Multi-attribute CBC and “Dual Response None”) models and examine whether the utility estimates improve over those from the standard best-only design. We also compare Latent Class (LC) and Hierarchical Bayes (HB) methods that allow for individual differences. By focusing on the form and magnitude of bias in model estimates, and when it occurs, our results are somewhat surprising.
Download presentation: Sawtooth2015_MagidsonBennett (pdf)