|
||||||
Technical Papers| Home |
Solutions |
Products |
Services |
Education |
Support |
| | ||||||||||||||||||||||||||
|   SS Winter 2000Qualitative Techniques for Enhancing Conjoint AnalysisMost of the articles we publish in this newsletter focus on the quantitative aspects of conjoint analysis. This article highlights some of the more qualitative aspects that are often inadequately treated or entirely ignored.Determinant Attributes To model the decision process, we develop a list of determinant attributes. (The term determinance was first coined by James Myers and Mark Alpert nearly three decades ago.) A determinant attribute is a product feature that is both important to the buyer and on which competing brands are perceived to differ. It is common for the researcher and the client to co-develop a list of attributes based on their knowledge of the product and the marketplace. Sometimes relying on "expert opinion" is not adequate to faithfully represent the key decision factors. Qualitative up-front work with respondents is often used to generate a list of determinant attributes. Jordan Louviere has suggested a simple series of questions to facilitate that process:
Another technique for identifying determinant attributes is referred to as the repertory grid. Both Louviere's and Ray Poynter's (referenced below) papers mention this approach. Briefly, respondents are asked to compare products or brands three-at-a-time and then list in what ways the products are similar and different. Assessing Perceptions During the conjoint analysis interview, other questions can be used to better understand the decision process. One can probe how respondents perceive the brands/products to be positioned on the various attributes and why respondents have certain preferences. Conjoint simulations assume that respondents both know and believe that the brands offer the specified features. In the actual marketplace, this assumption may not hold. Brands hindered by ineffective marketing communications campaigns may never achieve the degree of preference suggested by the simulator. In some markets, brands not perceived to offer essential features may never be considered. Therefore, it may be useful to include a series of questions outside of the conjoint exercise wherein respondents state how they perceive each brand is positioned on each attribute. Such information might be used to direct effective marketing communications strategy. Some researchers have advocated using the perceptions and the conjoint utilities together in conjoint simulations, by assigning each brand the utilities of the levels associated with it for each respondent. We have tried this ourselves, but generally have not been pleased with the results. Asking "Why?" In his 1999 Sawtooth Software Conference paper, Ray Poynter points out that researchers should not only learn what respondents prefer, but why they have those preferences: "Failure to answer this seemingly straightforward question," Ray argues, "will frequently result in the client doubting the wisdom of both the researcher and of the executive who commissioned the project" (Poynter, Ray. "But Why? Putting the Understanding into Conjoint," 1999 Sawtooth Software Conference Proceedings). It is possible when using techniques such as Adaptive Conjoint Analysis (ACA) with the Ci3 system to reference the current respondent's conjoint utilities in real time and ask follow-up questions about specific preferences. For example, if your client is interested in why some customers strongly prefer packages of 12 rather than 24, you could program Ci3 to ask a qualitative follow-up question if the 12 package count was preferred to the 24 and if that attribute was one of the top three attributes in terms of conjoint importance. Sign Up for Sawtooth Software E-NewsAt Sawtooth Software, our website (www.sawtoothsoftware.com) has been invaluable both for marketing and technical support. We are taking steps to leverage the Internet and particularly email even more. Last November we launched an email newsletter to supplement this quarterly publication.One of our greatest concerns about an email newsletter was that we not irritate users and potential customers. We despise unsolicited "spam" messages. We decided that we must send only to existing customers and to individuals who have contacted us for information and volunteered their email addresses. We decided that E-News should arrive only periodically (not more than once per month), carry a short message with hyperlinks, and provide a clear way for people to un-subscribe. It was not without trepidation that we fired off our first batch of Sawtooth Software E-News, but it took just a few hours to see that it had been a success. We received many unsolicited return email messages with positive feedback and requests to additionally subscribe colleagues. Please join our E-News list. Send an email to enews@sawtoothsoftware.com with "Subscribe" in either the subject or message field. New Software: CBC Advanced Design ModuleSince 1993, our CBC software for Choice-Based Conjoint has been used successfully in thousands of studies. After years of listening to our customer's research needs and by paying attention to outside experts, we recently created the Advanced Design Module for CBC. Industry experts have commented that our CBC software worked well for the kinds of studies where all attributes freely combine with one another. But they also pointed out that some research requires "alternative-specific" designs, for which our CBC package was not appropriate.Alternative-Specific Designs Alternative-specific designs let you customize the list of attributes or attribute levels that apply to each alternative (product concept). Consider the following choice task:
Note that the attributes that apply to buses (how often it arrives and cost per one-way trip) do not apply to cars or bicycles. The parking lot fee for cars doesn't apply to buses or bicycles. Also note that a bicycle doesn't have other attributes to further describe it. Bicycle is a "constant alternative" in addition to the constant "None" alternative. In general, the base CBC system could not handle alternative-specific designs (although you could use alternative-specific prices with conditional pricing). Alternative-specific designs can solve situations in which it seems that specifying multiple prohibitions is the only way to make valid product combinations. The Advanced Design Module helps you avoid the detrimental conditional dependencies caused by specifying too many prohibitions between pairs of attributes. In addition to classic transportation studies, there are many situations in which alternative-specific designs may be useful. In pharmaceutical research, the attributes that apply to one drug treatment may not necessarily apply to other treatments. In technology markets, some attributes that describe palmtops may not necessarily apply to laptops, but we may want respondents to make decisions between palmtops and laptops. Alternative-specific designs can lead to long attribute lists (even though each alternative may involve just a few relevant attributes). For that reason, the Advanced Design Module permits CBC studies with up to 30 attributes. We still caution that conjoint studies should probably display around six or fewer attributes for any one alternative. Partial-Profile Designs The Advanced Design Module also lets researchers conduct "partial profile" choice studies. Those of you who have used our ACA (Adaptive Conjoint Analysis) system are familiar with the idea of partial profiles: showing just a subset of the attributes in any one conjoint question. With our implementation of partial profile choice, you can study up to 30 total attributes, but only display a handful at a time in any one choice task. Unlike ACA, our partial profile choice method doesn't involve an adaptive algorithm; it randomly cycles across all attributes to create a balanced and efficient design. The user can also specify that some attributes should be displayed in every task. For example, brand and price might be included in each task while other attributes rotate in and out of the tasks. Graphics and multimedia elements can also be included in partial-profile questionnaires. Partial-profile choice is a relatively new technique that has shown promise. Users may naturally ask whether ACA or partial-profile choice is better for studies involving many attributes. We don't know the answer, but expect that it depends upon the situation. We invite you to design carefully controlled studies to investigate those issues for yourself, and of course to share the findings if you can. You can compute individual-level utilities from partial-profile choice studies using our CBC/HB module. However, we expect that those utilities will contain more noise at the individual-level than ACA utilities which have benefitted from a customized design, ratings-based data and stabilizing priors. On the other hand, it is argued that CBC studies present a more realistic task than other conjoint methods like ACA, so the debate is certainly alive and interesting. Please call us if you would like to order the Advanced Design Module. For those who already own CBC v2, the price is $2,000 for a Category I license. As with our other software, it comes with a 60-day money-back guarantee. ACA, CVA and CBC Receive New Windows Market SimulatorImagine delivering a polished Windows-based market simulator to clients. Imagine that it lets you select banner points for a "cross-tab" like display, showing attribute utilities, importances, and shares of preference in separate columns for each sub-group. Imagine that it includes the new Randomized First Choice model to adjust for product similarity. Imagine that you can cut-and-paste simulation results directly to a spreadsheet. Imagine that no matter the utility estimation procedure (ACA, CVA, logit, Lclass, ICE or HB), you can use the same market simulator to analyze the results.All of these capabilities are provided by the new upgrade to ACA v4.5, CVA v2.5 and CBC v2.5. Upgrade: ACA v4.5, CVA v2.5 and CBC v2.5 We are pleased to announce upgrades to our conjoint analysis software. The CBC v2.5 upgrade is free to those owning CBC v2. For those owning CVA v2, the CVA v2.5 upgrade is $500. The ACA upgrade costs $500 for those owning ACA/10 v4; and $1,000 for those owning ACA/30 v4. Please note that you only need to upgrade the software system that permits the largest number of attributes. For example, if you own both ACA/30 and CVA, you only need to purchase the ACA/30 upgrade. The market simulator can work with any Sawtooth Software data files, with the number of attributes restricted by your license. Therefore, upgrading your 30-attribute ACA software is sufficient for running simulations from CVA data, which at maximum have 10 attributes. Please call for more information or to order. Academics Cite Sawtooth Software Conference ProceedingsWe have hosted the Sawtooth Software Conferences for over a decade now, with the goal of bringing academic researchers and practitioners together to discuss practical research issues. We have been pleased with the quality of the resulting papers, which have been published in a series of Proceedings. A recent book on conjoint analysis suggests that our conferences and the resulting Proceedings have had a favorable impact on the academic research community.Conjoint Measurement - Methods and Applications (Eds. Gustafsson, Herrmann, Huber. Springer-Verlag, Berlin, 2000) is a collection of sixteen recent essays, mostly by academics but with a few practitioners. As we examined the essays, we were delighted to notice how often papers presented at the Sawtooth Software conferences were referenced. The table below shows the top seven sources cited in that volume according to our informal tally:
If we take this book as a proxy for the modern state of conjoint thought, the venerable Journal of Marketing Research is by far the most influential source of research about conjoint analysis. But, by the same standard, the Sawtooth Software Conferences are the second most influential source. Even though the focus of our conferences has been practitioner-oriented, this group of mostly academic authors has referred to them more often than several academic journals. We expect Sawtooth Software conferences to remain an important resource for the research community. Sawtooth Software Conference 2000 Hilton Head, South Carolina, March 21 - 24The Sawtooth Software conferences are renowned for their practitioner-oriented focus and depth in the fields of conjoint analysis, segmentation, and data collection/analysis. It is not a sales-oriented program, but a forum to exchange ideas and learn about quantitative methods in marketing research.The 2000 conference will focus on conjoint/choice analysis and general issues regarding data collection (both PC and Internet). The first day includes optional tutorials. The general session starts on March 22. We have provided summaries of the tutorials and presentations here. For more information, please see our web site www.sawtoothsoftware.com. March 21 - 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. Class size is limited to foster interaction and individual instructor attention. These classes are offered on Tuesday only. Tutorials are optional and are an additional cost.
Interviewing over the Internet (8 am - 12 pm) There is a lot of discussion about the web and marketing research. Some believe that the technology is too new and has limited use for general purpose marketing research. We will try to separate the facts from the fiction to help you decide whether it is useful for you, and how to get started. The first half of the tutorial will cover mostly theory. The second half will step through a sample internet study from construction to data collection to analysis using Sawtooth Software and other tools. Attendees should have some familiarity with the web from a user's perspective. Some exposure to HTML is helpful but not necessary.
An Introduction to Hierarchical Bayes Methods (8 am - 12 pm) Hierarchical Bayes methods have generated a lot of excitement in the marketing research community in recent years because of their ability to estimate models we could not before estimate. Now we have off-the-shelf computer hardware and software to make these techniques available to a wide audience. Tim and Mark have been using hierarchical Bayes methods for the past five years to study a wide variety of research problems, and in this tutorial they describe the theory underlying hierarchical Bayes methods as well as some interesting case studies.
Advanced Research Methods Using SPSS and SAS Software (1 - 5 pm ) This session will feature representatives from two well-known statistical software companies. They will provide instruction on advanced methods along with a demonstration of how their respective software can solve specific analytical problems. Neural Networks and Decision Trees--Tony Babinec, SPSS Neural networks and decision trees are two popular modern techniques that can be used in classification and prediction problems. Both techniques are very flexible, and both can be implemented in a variety of ways. Roughly half the presentation time will be spent on exposition, and the other half will be spent on demonstration using marketing examples and current software such as Clementine and AnswerTree. Design Efficiency for Conjoint and Choice Experiments-- Warren Kuhfeld, SAS Institute Marketing researchers have increasingly turned to conjoint and discrete choice experiments to address questions about product and price optimization and strategy. Fundamental to these techniques is the plan for running the study, which is called the experimental design. The purpose of this tutorial is to provide a practical introduction to basic experimental designs for conjoint and discrete choice analysis. Topics will include orthogonality, balance, efficiency and coding.
Introduction to CBC (1 - 5 pm) Choice-Based Conjoint has become very popular in the last five years. It is well-suited for both product-design issues and pricing research. One of the main strengths of CBC interviews has been their ability to mimic what buyers do in the marketplace: choose among available products. Until recently, this realism came at a cost: CBC data could only be analyzed in the aggregate. Recent advances have made it possible to analyze CBC data at the segment- or individual-level, making CBC even more useful and accurate than before. e will provide a brief history of conjoint methods including traditional full-profile conjoint, ACA and choice based conjoint. Following that, we'll talk about how to design, conduct and analyze a CBC study. March 22-24, General Session The main session will feature 24 presentations. Each presentation is followed by approximately 10 minutes of audience discussion. Following is a list of presentations and authors.
The Sawtooth Software 2000 Conference will be held on March 21-24, 2000 at the Hyatt Regency Hilton Head Resort hotel, Hilton Head Island, SC. We have negotiated special room rates for the conference ($129 per night). We recommend that you reserve your rooms as early as possible to not miss this low rate (843/785-1234). When reserving rooms, make sure to say that you are attending the Sawtooth Software Conference.
|
|||||||||||||||||||||||||
Technical Papers
|
About Us | Contact Us | Privacy | FAQ's | Forum |
|
© 2008 Sawtooth Software, Inc. All rights reserved. |