Conference Presentations & Research Contributions
For nearly two decades, my work in marketing science has explored a recurring question: How can we ensure that sophisticated analytical methods produce evidence worthy of consequential business decisions?
The answer has sometimes required developing new statistical approaches, sometimes testing the limitations of established methods, and sometimes creating more flexible tools for practitioners. The work has been presented at the American Marketing Association’s Advanced Research Techniques (ART) Forum, Sawtooth Software conferences, and the Joint Statistical Meetings.
These contributions reflect a continuing research interest in the relationship between experimental design, statistical modeling, valid inference, and business decisions.
Developing Methods Around Business Problems
Some methodological advances begin with a statistical question. Others begin with a business problem that existing statistical methods cannot adequately solve.
An early example was a unified framework for global and local market segmentation, presented at the 2009 ART Forum. Multinational organizations faced a practical conflict: global segmentation could obscure meaningful differences among markets, while separate segmentations for each geography could overwhelm decision-makers with incompatible solutions. The framework was developed in practice to balance global consistency with local market differences.
Later work explored Bayesian estimation and the use of informed priors to improve choice models in difficult research settings. Rather than accepting conventional estimation defaults, the research investigated whether additional methodological judgment could produce better models.
The same principle motivated work on willingness-to-pay profiles. Instead of describing preference heterogeneity only through conventional model parameters, this research explored how to identify groups with meaningfully different economic valuations and use those differences in product portfolio decisions.
The common theme is that model architecture should respond to the requirements of the problem, not simply follow the conventions of an established analytical workflow.
Testing the Limits of Analytical Inference
A second stream of work asks a different question: When do apparently reasonable statistical results support conclusions that are actually misleading?
In 2018, I investigated emerging direct-utility approaches to volumetric conjoint analysis. The academic ideas were promising, but their usefulness for practical inference had not been established. Empirical evaluation revealed serious limitations in the properties needed to support managerial conclusions.
In 2019, a split-sample comparison led to a more unexpected finding. MaxDiff performed reasonably well in recovering aggregate preference orderings, but its estimated respondent-level utilities produced problematic correlation structures. Additional empirical tests and simulations showed that increasing the number of tasks, altering coding, and changing covariance assumptions did not resolve the underlying issue.
The practical warning was specific: a method suitable for measuring relative preferences is not necessarily suitable for every downstream multivariate analysis.
The 2021 Price Fixing presentation examined a related but distinct problem. Poorly constructed choice experiments could make price-sensitive consumers appear insensitive to price or even price-seeking. The research demonstrated the problem through an actual experimental design, deliberately constructed alternatives, and simulation.
These investigations illustrate an important principle: statistical estimation cannot automatically repair inadequate information or justify interpretations that the research design does not support.
Extending the Tools Available to Practitioners
Methodological expertise also involves implementation.
At the 2016 Sawtooth Software Conference and 2017 ART Forum, I presented tutorials on building interactive conjoint simulators using R and Shiny. These sessions demonstrated how researchers could construct customized simulation applications and provide capabilities beyond those offered by standard software interfaces.
The broader purpose was to expand the options available to practitioners. Analytical tools should be adaptable to the decision problem rather than becoming a constraint on the research itself.
This interest in flexible implementation continues through independent statistical programming, Bayesian estimation, and customized decision-support applications.
Conference Presentations and Tutorials
The following presentations document methodological research, applied developments, and practitioner education. Where available, links lead to conference records, research summaries, or presentation materials.
2021
Price Fixing: Can You Construct an Experimental Design to Make Price-Sensitive Consumers Appear Price-Seeking?
Sawtooth Software Conference
Demonstrated how inadequate experimental designs can distort the apparent price sensitivity of respondents, using empirical evidence, constructed designs, and simulation.
Willingness-to-Pay Profiles: A Model-Based Approach to Identify Heterogeneous Demand and Craft a Portfolio Strategy
Joint Statistical Meetings, invited session on Advances in Choice Modeling
Proposed modeling preference heterogeneity in willingness-to-pay space to identify economically meaningful consumer groups and inform product portfolio decisions.
2019
Multivariate Analysis with MaxDiff: A Curious Look at the Properties of MaxDiff Utilities
American Marketing Association, Advanced Research Techniques Forum
With Jeff D. Brazell. Investigated whether respondent-level MaxDiff utilities preserve the covariance structure required for downstream multivariate analysis. Empirical comparisons and simulation exposed significant limitations despite reasonable aggregate preference measurement.
Read the presentation’s methodological record
2018
Managerial Inference from a Direct Utility Model for Volumetric Conjoint
American Marketing Association, Advanced Research Techniques Forum
Evaluated whether emerging volumetric conjoint methods produced evidence suitable for managerial inference, identifying important limitations in their practical application.
Properties of Direct Utility Models for Volumetric Conjoint
Sawtooth Software Conference
Examined the statistical properties of direct-utility approaches to volumetric conjoint using empirical research, with particular attention to the reliability of conclusions drawn from the models.
2017
Web-based Conjoint Simulators via Shiny
American Marketing Association, Advanced Research Techniques Forum
A practitioner tutorial demonstrating the construction of interactive web-based conjoint simulators using R and Shiny.
2016
Conjoint Analysis Upper-level Models
American Marketing Association, Advanced Research Techniques Forum
Presented work on modeling and estimation beyond conventional default specifications, examining opportunities for more appropriate model structures.
Make Your Conjoint Simulator Shiny
Sawtooth Software Conference
A tutorial showing researchers how to develop customized interactive conjoint simulation applications using R and Shiny.
2015
Better MNL Conjoint through Informed VB
American Marketing Association, Advanced Research Techniques Forum
Investigated the use of informed prior distributions to improve Bayesian multinomial logit estimation in challenging choice modeling applications.
2013
Evaluating Anchored MaxDiff via Simulation
Sawtooth Software Conference
With Jeffrey P. Dotson. Used simulation to examine alternative anchored MaxDiff approaches, their statistical properties, and implications for research practice.
2009
A Gibbs-like Framework for Global Segmentation
American Marketing Association, Advanced Research Techniques Forum
Introduced a unified segmentation framework developed to preserve meaningful local differences within a coherent global segmentation system, reducing the complexity faced by multinational decision-makers.
Continuing Research: The Minimum Evidence Standard
The questions investigated in these presentations have continued to develop.
A recurring finding is that conventional indicators of analytical success do not necessarily establish whether the resulting model can support the decision for which it was constructed.
The Minimum Evidence Standard (MES) is an ongoing research program intended to make the requirements for credible choice modeling evidence more explicit. A central principle is evaluating predictive performance on respondents excluded from estimation, rather than relying solely on in-sample fit or predictions for respondents already represented in the model.
Recent work examines respondent covariates, price specifications, covariance structures, and other modeling decisions through this evidence standard.
This research is not a departure from the earlier work. It extends a longstanding interest in developing useful analytical methods while investigating the conditions under which their conclusions deserve to be trusted.
The objective remains consistent: better methods, stronger evidence, and more defensible business decisions.
