This content has been updated with new exercises, new readings, and new PowerPoint slides. Please visit: Curriculum
5 Lectures / 5 Exercises, Plus Primary Research Project
This is a suggested curriculum for technically capable graduate students, appropriate for a graduate-level market research or multivariate methods course. The purpose is to introduce students to the three main forms of conjoint analysis (traditional full-profile, ACA and discrete choice (CBC)) and give them insight into the mechanics of experimental design and part worth estimation. Each student may conduct a primary research project and produce a research report, using commercially available conjoint programs from Sawtooth Software. This curriculum may be taught as five lectures. Each lecture is accompanied by a set of readings and subsequent exercises that should take about 2 hours each to complete. The primary research project (including software training, data collection, data analysis, and report) should take about 12-20 hours to complete.
Day 1, Conjoint Analysis: Motivation and Theory
- Understanding the Value of Conjoint Analysis (2009)
- Formulating Attributes and Levels in Conjoint Analysis
- Pick a product or service that interests you, and formulate attributes and levels to describe the essential characteristics of that product. Make sure that the levels are mutually exclusive within each attribute, and that the attributes do not have overlap in meaning.
- How many attributes and levels should a researcher include in a conjoint study? Is there any need to maintain the number of levels consistent across attributes?
- Why should levels be mutually exclusive within each attribute?
- Describe the differences between nominal (categorical), ordinal, and quantitative attributes. What implications do these distinctions have on analysis?
- What danger is there in prohibiting certain attribute levels from ever appearing with other attribute levels in a conjoint study?
- Power Point Slides (curriculum3_1.ppt)
- Conjoint Analysis: How We Got Here and Where We Are--An Update (2004)
Day 2, Traditional Conjoint (CVA)
- Understanding Conjoint Analysis in 15 Minutes (1996)
- Analysis of Traditional Conjoint Using Excel: An Introductory Example (2009)
- Interpreting Conjoint Analysis Data (2009)
- Formulate a conjoint analysis problem, with three attributes of your choice. The first attribute should have 4 levels, the second 3 levels, and the last one 2 levels. Generate a "full factorial" experimental design, by generating all possible combinations of the three attributes (twenty-four total cards). Have a few people complete the survey, using a ratings scale. Dummy-code and estimate part worth utilities separately for each respondent. Chart the results (part worth utilities and importance scores), and interpret.
- For each respondent, how many degrees of freedom are there? (Hint: degrees of freedom are equal to the number of observations minus the number of parameters to be estimated.) How might the degrees of freedom relate to the precision of the parameters estimated in the model?
- How many possible product profiles can be created with a conjoint analysis problem including 5 brands, 4 designs, 4 colors, and 3 prices? Would each respondent need to evaluate each product combination to obtain good results?
- What is the difference between estimating part worth utilities for each individual and aggregating the results versus pooling all of the ratings together for all respondents and estimating a single regression model? What implications are there for market segmentation, and market simulations?
- Power Point Slides (curriculum3_2.ppt)
- Card Sort Spreadsheet Example
- CVA Technical Paper (2002)
- SSI Web Help Manual (CVA Section)
Day 3, Adaptive Conjoint Analysis (ACA)
- Access the Adaptive Conjoint Analysis (ACA) Sample Study and complete the questionnaire. Answer the questions realistically, to reflect your opinions and preferences. At the end of the interview, the computer estimates your preference scores (part worth utilities) and importance scores for the attributes. Cut-and-Paste the part worth utility and importance results into your work.
- What was the survey experience like? Do these part worth utility and importance scores seem to reflect your preferences?
According to your utilities, which of these product alternatives would you be expected to prefer?
(Sum the scores across the features to determine the total value of each alternative.)
- Is this an accurate prediction of your preference?
- How might this survey had been different had traditional full-profile conjoint been used? How many cards would have been needed for each respondent to answer with traditional conjoint to enable stable individual-level part worth utility estimation?
Day 4, Choice-Based Conjoint (CBC)
- Which Conjoint Method Should I Use? (2013)
- Using Conjoint Analysis in Pricing Studies: Is One Price Variable Enough? (1992)
Create a Choice-Based Conjoint (CBC) survey design by selecting two attributes, each with three levels. The first attribute should be three brands of a product or service (e.g. Coke, Pepsi, Sprite). The second attribute should have varying price levels (e.g. $1.00, $1.35, $1.70). Choose a product category that interests you with realistic prices that cover a fairly wide range. Create an experimental design plan. The design is to have nine choice tasks (questions). Each task offers the respondent two concepts (products) to choose from.
Create a written CBC questionnaire (either electronically, or using index cards). An example of a choice task is as follows:
First, create nine product concepts (all possible combinations of the two attributes). These nine concepts form the left-side concept within each choice task. Create the second (right-hand) concept for each of these nine tasks as a "shift" of the product concept on the left. For example, if the product on the left is represented as (2,1) (Level 2 of the first attribute and Level 1 of the second attribute), we shift (increment) the levels by 1, resulting in the new concept (3,2) (Level 3 of the first attribute and Level 2 of the second attribute). To shift level 3, we revert back to level 1.
Record the design using the table below, filling in the remaining cells (we have provided the first and last concept specifications).
- What are the characteristics (in terms of experimental design) of the original nine-concept design and its "shift." Hint: tally how many times each level appears with another level (within the same concept), and opposing other levels (in the competitive concept).
- Administer the questionnaire to a few individuals. Each person should choose one of the concepts in each of the nine choice tasks.
Use the table below to record the summary information for all interviews and estimate the choice probability for the brands and prices (main effects):
Use the table below to record the summary information for all interviews and estimate the choice probabilities for the joint effects of each brand at each price:
- Chart the choice probabilities for each of the brands at each of the prices, where the Y axis is choice probability, the X axis is price. Interpret the results.
- What would happen with the results with many more respondents? What would happen if you included a "None of these" selection in each choice task? How would the questionnaire and design be better or worse if you included all three brands in each choice question?
- Power Point Slides (curriculum3_4.ppt)
- CBC Technical Paper (2013)
- What We Have Learned from 20 Years of Conjoint Research: When to Use Self-Explicated, Graded Pairs, Full Profiles or Choice Experiments (1997)
- An Overview and Comparison of Design Strategies for Choice-Based Conjoint Analysis (2000)
- How Many Questions Should You Ask in Choice-Based Conjoint Studies? (1996)
- The Benefits of Accounting for Respondent Heterogeneity in Choice Modeling (1998)
- SSI Web Help Manual (CBC Section)
Day 5, Market Simulations
- Download the data set in the Excel file utilities.xls. This is an Excel spreadsheet containing hypothetical part worth utilities for a conjoint analysis problem. The conjoint design has three attributes: Brand, Color, and Price. Chart the average part worth utilities and importance scores.
- Create a "First Choice" (Maximum Utility Rule) simulator using Excel or another program of your choice. The simulator should simulate shares of choice for product scenarios that include three products. The simulator should sum the total utility for each product, project which product alternative each respondent would prefer, and summarize those product preferences as shares of preference summing to 100%.
Assume that you work for the company making Brand A. Further assume that there are two competitors in the market:
Product 1: Brand B, Red, $100
Product 2: Brand C, Blue, $150
What product could your company offer to maximize units sold? How many units are sold of that product? (Assume each respondent "purchases" one unit, and these respondents represent 1/1000 of the total market.)
What product could your company offer to maximize total revenue (#units x price). What is the total revenue for that product?
- Should these findings bear out in the real world? What assumptions does the market simulator make?
- Power Point Slides (curriculum3_5.ppt)
- Dealing with Product Similarity in Conjoint Simulations (1999)
- tvdata.xls -- real conjoint data set for television preferences
- SSI Web Help Manual (ACA Section)
Student Primary Research Project
The student uses ACA, CVA or CBC to plan a conjoint study using a computerized interview, field the study among a small convenience sample, and analyze the results. Prior to designing and programming the study, the student should complete the tutorial available in the online help.
An academic license covering the school can be obtained from Sawtooth Software at a price of $1,000. The software includes on-line documentation and useful tutorials.