Optional Workshops and Tutorials
Optional 1.5-Day Workshop
Monday 8 am - 5 pm & Tuesday 8 am - Noon	CBC MODELING WORKSHOP Bryan Orme, Aaron Hill, and John Howell, Sawtooth Software
	***As of October 3rd, this workshop is full. No more registrations allowed.*** The CBC Modeling Workshop is designed to introduce participants to discrete choice (CBC) analysis through an interactive, hands-on workshop. The course will include an overview of conjoint analysis techniques and methodology, with specific attention to CBC/Web. Attendees will also receive practical experience creating surveys in Sawtooth Software’s SSI Web system. Topics include: Conjoint methodology overview Formulating attributes and levels Designing conjoint experiments Analyzing CBC data using Counts, Logit, Latent Class, and hierarchical Bayes (HB) Using market simulators to estimate preference for competitive products in market scenarios, including price sensitivity Best practices / common mistakes related to CBC. Participants will create several conjoint surveys and analyze sample data using a team-oriented case study approach. Cost is $700 (you must register separately from the conference to attend this workshop).
Optional Half-Day Tutorials
	Tutorial workshops are being offered to provide opportunities for a more in-depth learning experience. Each tutorial will be led by an outstanding professional with pertinent research and teaching experience. These classes are offered only on Tuesday. Tutorials are optional and are an additional cost. Please note that you must register separately for the tutorials.
Tuesday 8 am - Noon	RESEARCH FOR SOLID PRICING DECISIONS David Lyon, Aurora Market Modeling
	Pricing decisions go right to a firm’s bottom line, so pricing research is one of the most important areas of marketing research. A number of very different approaches are used to measure price sensitivity, and sorting out their competing claims of accuracy, realism and simplicity can be very confusing. While none are a panacea, all have their place, and this session will focus on understanding the basics, pros and cons of each to help establish what to use when. Price as an object of survey questioning exaggerates psychological response problems that may not be noticed with other product attributes. It also elicits entirely new problems: bargaining behavior, price as proxy for quality, formation of reference prices, etc. Some research approaches, including many forms of conjoint analysis, can amplify the effects of these problems. In short, price is not “just another attribute” – it has both special importance and special problems. We will survey a number of approaches to the pricing problem, all ones based on surveys, and focus on when and where each works relatively well or relatively poorly. These will include: simple willingness-to-pay questions, monadic designs, the van Westendorp approach, including the Newton-Miller-Smith variant of it, ratings-based conjoint discrete choice modeling (or choice-based conjoint) and others!
Tuesday 8 am - Noon	ADVANCED CBC/WEB APPLICATIONS FOR POWER USERS Bryan Orme, Sawtooth Software
	This course is for individuals who have already developed a solid understanding of CBC and who have substantial practical experience. Knowledge of HTML, the use of SSI Scripting (SSI Web’s instruction language), SSI Web’s Unverified Perl, and the layout of CBC/Web’s .CHO/.CHS files is highly recommended. This tutorial will first introduce the standard advanced techniques formally supported by CBC/Web (with its Advanced Design Module): conditional pricing, dual-response None, constant sum, partial-profile and alternative-specific designs. After that, we’ll indulge in a tour of a variety of “power tricks” and case studies that take us well beyond the standard applications of CBC/Web. Attendees may be surprised at the versatility of CBC/Web, once one learns to customize CBC questions using Free Format or to modify the .CHO/.CHS files to accomplish specific advanced outcomes (such as volumetric CBC). This course is designed as a fast-paced, advanced-level lecture. Students will not be using the software and will not require a laptop PC.
Tuesday 8 am - Noon	DESIGN CONSIDERATIONS FOR CBC STUDIES Jon Pinnell, MarketVision Research
	This tutorial will provide the practicing choice researcher an overview of design considerations when conducting CBC studies. The session will provide a brief background on the history and evolution of discrete choice as well as the terms and definitions common with choice studies. The session will include a discussion on study specifications, including developing attributes and levels, the strengths and weaknesses of different design strategies, the estimation and interpretation of utilities, and conducting simulations. Examples will be provided of the various applications of discrete choice. The session will include a review of ‘best practices’ based on empirical findings presented at related conferences in the past.
Tuesday 1 - 5 pm	POWER TRICKS FOR EXPERIENCED SSI WEB USERS Justin Luster, Dave Squire, and Aaron Hill, Sawtooth Software
	This tutorial is for those who are proficient users of SSI Web and are ready to advance to the next level. We will cover many advanced concepts that will enable you to create more sophisticated surveys and we’ll show you a few things that should make your current job easier. Some of the items that we will cover include: Advanced Constructed List Building Unverified Perl JavaScript Time saving tips (such as how to search and replace text across the entire survey) Custom Question Verification Free Format including dynamic HTML Using Flash from within SSI Web Custom CSS Passwords and Quota Control The programmers of SSI Web will be presenting this tutorial so make sure to bring your questions and suggestions. This is a presentation only (not a hands-on programming workshop) so no laptop PC or software is required.
Tuesday 1 - 5 pm	RESEARCH FOR SOLID PRICING DECISIONS (Repeat Session) David Lyon, Aurora Market Modeling
Tuesday 1 - 5 pm	DESIGN CONSIDERATIONS FOR CBC STUDIES (Repeat Session) Jon Pinnell, MarketVision Research
Tuesday 1 - 5 pm	MARKET SIMULATIONS WORKSHOP Bryan Orme, Sawtooth Software
	In this hands-on workshop, attendees will use Sawtooth Software’s tool for market simulation (SMRT). We’ll assume attendees already have some experience with conjoint methods. However, we’ll review basic part-worth utility and importance interpretation. The focus of the course will be on the motivation and mechanics for conducting market simulations. While the topics will mostly be conceptual, some math will be involved. We’ll introduce and explain the mechanics of First Choice, Logit Rule (Share of Preference), Randomized First Choice, and Purchase Likelihood simulation models. We’ll also cover base case formulation, sensitivity analysis, line-extension, segmentation analysis, respondent weighting, scale factor (exponent), external effects, and elasticity estimation. We’ll demonstrate the IIA issue and show how capturing heterogeneity significantly reduces the problem. Class members must bring a laptop PC, with Windows 2000/Windows XP (or later) installed. A temporary student lab version of the software needed for this workshop (for non-commercial, training use only), will be given to attendees. Those with a professional version of SMRT may use their own installation.
Main Conference Sessions
Wednesday October 17, 2007
David G. Bakken Harris Interactive	The Weakest Link: A Cognitive Approach to Improving Survey Data Quality
	Questionnaire construction is more art than science. Recent advances in the understanding of cognitive aspects of survey response offer a new approach to improving the quality of survey data. Observations from think-aloud pretests of online surveys support hypotheses about survey response processes. We share our experience with think aloud pre-tests and best practices for using think aloud cognitive interviews to improve the quality of survey data.
Don Peterson, Matthew Siegel, and Paul Venditti, General Electric	Evaluating Financial Deals Using a Holistic Decision Modeling Approach
	We demonstrate that complex investment decisions can be modeled with an ACA-based decision support tool. The tool validates strongly to past expert decisions and saves time for screening new deals. We discuss modifications to ACA's self-explicated importance question along with real-world challenges when translating executive preferences into actionable decision support tools.
Break
Edwin Love, University of Washington School of Business and Christopher N. Chapman, Microsoft Corporation	Issues and Cases in User Research for Technology Firms
	Understanding customer preferences and making business decisions for technology products presents several unique challenges, so technology marketing research must differ from traditional marketing research. Using several real-world examples, we show that a combination of iterative product definition and choice-based research yields improved actionable results early in the product development cycle.
L. Allen Slade, Covenant College	Minimizing Promises and Fears: Defining the Decision Space for Conjoint Research for Employees versus Customers
	Conjoint research with employees requires careful definition of the decision space-–the attributes and levels of rewards--to avoid creating fear or false promises. A case study of employment brand research at Microsoft shows how to define the decision space in a way that manages employee expectations while maximizing research value.
Ely Dahan, UCLA Anderson School	Conjoint Adaptive Ranking Database System (CARDS)
	We combine real-time computation & database retrieval with a novel form of adaptive conjoint analysis that imposes internal consistency as the criterion for selecting stimuli to be displayed to each respondent. The theory underlying this work is presented, along with empirical tests using iPods and SmartPhones as examples.
Lunch
Steven Gaskin, Applied Marketing Science, Inc. and John R. Hauser, MIT	Two-Stage Models: Identifying Non-compensatory Heuristics for the Consideration Set then Adaptive Polyhedral Methods within the Consideration Set
	Scientific evidence suggests a decision process of forming a consideration set and then choosing from the consideration set. We explore a practical procedure to first identify the product features that each consumer uses heuristically to form consideration sets. An adaptive CBC method automatically explores features critical for choice within these consideration sets.
Rich Johnson and Bryan Orme, Sawtooth Software	A New Approach to Adaptive CBC
	Despite its popularity, CBC has weaknesses. These issues might be solved with an improved method of adaptive design and data collection. Past research in adaptive CBC has assumed respondents answer in an additive, compensatory manner. Our approach recognizes both non-compensatory and compensatory decision making, and our results show significant improvement vs. standard CBC.
Break
Thomas Otter, Ohio State University	Hierarchical Bayesian Analysis for Multi-Format Adaptive CBC
	The new approach to adaptive CBC introduced by Johnson and Orme combines a configurator, a profile screening task and CBC questions. I discuss HB analysis for ACBC that adaptively pools information across the different tasks, accounting for potential scale differences. I also investigate different likelihood specifications for the individual tasks.
Kevin Lattery, Maritz Research	EM CBC: A New Framework for Deriving Individual Conjoint Utilities by Estimating Responses to Unobserved Tasks via Expectation-Maximization
	Rather than estimating individual utilities via sampling of coefficient distributions (HB and others), this presentation outlines an application of the Expectation-Maximization (EM) algorithm to estimate each individual’s responses to conjoint scenarios in the experimental design, but not shown to them. This allows modeling at the individual level with individual constraints.
Jay Magidson, Statistical Innovations and Jeroen K. Vermunt, Tilburg University	Removing the Scale Factor Confound in Multinomial Logit Choice Models to Obtain Better Estimates of Preference
	A theoretical weakness of CBC as currently practiced is that individual utility estimates are confounded by differential measures of uncertainty (error variances). By separating the scale factor from the utilities we obtain clearer estimates of preference. Results from extended latent class models indicate that quick respondents have the highest variances.
Thursday October 18, 2007
Keith Chrzan and Doug Malcom, Maritz Research	Empirical Tests of Seven Brand Image Scaling Techniques
	Using data from three commercial studies collected in the past two years, we compare seven different ways of brand image scaling in terms of (a) their face validity (similarity of perceptual maps they produce), (b) their ability to discriminate among brands, and (c) their ability to support predictions via MNL choice models.
Phil Hendrix, immr and Stuart Drucker, Drucker Analytics	Alternative Approaches to MaxDiff with Large Sets of Disparate Items
	Increasingly, MaxDiff is being used to scale large sets (30+) of disparate items, despite the obvious challenges for respondents. We report the results of a study comparing conventional MaxDiff with three adaptive approaches that use information gleaned from respondents to shorten the task, reduce respondent burden, and improve estimation.
Break
Chris Diener, Lieberman Research Worldwide	Segmentation Using Choice Model Optimization
	Often we segment using CBC coefficients or latent class. This presentation illustrates an approach using optimization in simulations to create emergent segments. We optimize for multiple products and assign respondents to segments based on their most preferred product. The presentation will compare approaches and illustrate benefits, drawbacks and implementation issues.
Luiz Sá Lucas, IDS Market Analysis	Joint Segmenting Consumers Using Both Behavioral and Attitudinal Data
	The paper presents a novel way to segment consumers from behavioral and attitudinal data. A fusion of Machine Learning and Marketing Research techniques, the idea is to develop a couple of segmentation and classification devices, using a distance matrix that is a weighted combination of attitudinal and behavioral distance matrices. The resulting classifier must be as easy to implement as possible, and will be the resulting segmenter in the process.
Lunch
Patrick Moriarty, Scott Porter, OTX Research and John Fiedler, 12 Americans	America’s Social Network: Mapping Consumers’ Connection with Cultural Icons
	In this era of media and brand fragmentation, marketers must demonstrate connection with consumers to drive successful brands. In this paper, the authors build from theories of social identity development and utilize MaxDiff methodology via an Internet survey to deliver an assessment protocol that enables marketers to identify key consumer groups and understand the social network of brands, personalities and media.
Joseph Retzer and Ming Shan, Maritz Research	Cluster Ensemble Analysis and Graphical Depiction of Cluster Partitions
	This presentation introduces an approach for improving on base cluster analysis algorithms, many of which are likely familiar to most attendees. “Cluster ensemble analysis”, a relatively new tool for unsupervised learning, along with valuable graphical tools for evaluating and communicating cluster analysis results, will be described and illustrated.
Charles E. Cunningham, Heather Rimas, and Ken Deal, McMaster University	Do Depressive Disorders Influence Performance on Discrete Choice Conjoint Experiments?
	We examined how depression, which often accompanies health problems, influences performance on discrete choice experiments designed to understand patient preferences. Although depression did not increase inconsistent responding to identical hold out tasks, it did influence health service preferences and segment membership. We consider issues in the design and interpretation of hold out tasks.
Break
Michael G. Mulhern, Mulhern Consulting, Douglas L. MacLachlan, and Stephen Samaha, University of Washington Business School	Determining Product Line Pricing by Combining Choice Based Conjoint and Automated Optimization Algorithms: A Case Example
	Pricing decisions that incorporate consumer preferences, competitive entrants, and well-defined segments are critical to the success of any product or product line. This case study illustrates an approach to developing a pricing strategy that maximizes product line revenue by implementing choice based conjoint and automated search algorithms.
Greg Rogers, P&G	Using Choice Models to Forecast Sales of Fast Moving Consumer Goods – A Comparison of CBC and Chip Allocation Based Forecasts
	CBC studies can be used to forecast new product sales, but implementation can be limited by their cost and complexity. Chip allocation (also known as 'constant sum') questions can provide similar data and are much easier to implement—but are the forecasts as good? This research explores using chip allocation data in a Dirichlet model to forecast sales for a new product.
Larry Goldberger, Adelphi Research by Design	A Simple Solution to the Challenge of Measuring Adding and Switching in a Polytherapy Choice Allocation Model
	Physicians commonly prescribe multiple treatments for a given patient, resulting in choice allocations that exceed 100%. Standard choice models do not easily capture cannibalization in these situations. This paper presents a simple solution to this problem: modeling the likelihood that a new product will substitute for an existing product.
Friday October 19, 2007
Frank Berkers, Gerard Loosschilder, SKIM Group and Mary Anne Cronk, Philips Lifeline Systems	Data Fusion to Support Product Development in the Subscriber Service Business
	This paper focuses on a practical implementation of data fusion at a large health insurance company in the subscriber service business. It shows how the impact of new product and marketing ideas on subscriber lifetime value is assessed by appending research results to the subscriber base and forecasting changes in lifetime value.
Jorge Alejandro and Kurt A. Pflughoeft, Market Probe	Multiple Imputation as a Benchmark for Comparison within Models of Customer Satisfaction
	Missing values are a reality that many market researchers have to face on a regular basis. Several popular techniques to handle missing values are reviewed and compared against multiple imputation. A simulation study is used to compare the accuracy of the statistical estimates of a regression model under each technique.
Lynd Bacon, LBA, Peter Lenk, University of Michigan, Katya Seryakova & Ellen Veccia, Knowledge Networks	Making MaxDiff More Informative: Statistical Data Fusion by way of Latent Variable Modeling
	MaxDiff scaling has many advantages, but one of its weaknesses is that it doesn’t retain overall level differences between respondents. We describe a latent variable approach to combining ratings data with MaxDiff choices that preserves the benefits of MaxDiff while also referencing respondents’ scores to a common origin.
Break
Greg Allenby, Thomas Otter, and Qing Liu Ohio State University	Endogeneity Bias – Fact or Fiction?
	Adaptive designs can lead to an endogeneity bias in sampling experiments. We re-examine this issue in light of the likelihood principle, and show that once the data are collected, this bias is ignorable. The likelihood principle is implicit to Bayesian analysis, and discussion is offered for detecting and dealing with endogeneity in marketing.
Well Howell, Harris Interactive	CBC/HB, bayesm and Other Alternatives for Bayesian Analysis of Trade-off Data
	Bayesm and other packages in R provide for Bayesian trade-off analysis. While CBC/HB offers more ease of use, the R packages prove useful during choice and maxdiff model development in the areas of convergence testing and the use of upper-level covariates as will be demonstrated on synthetic and real data.
John Howell, Sawtooth Software	Respondent Weighting in HB
	It has been shown that CBC/HB shrinks the parameter estimates for each individual toward the sample mean. Three methods are proposed to adjust the amount of shrinkage when it would be desirable to weight the sample in order to improve individual level utility prediction.
Conference Adjourned