Conjoint Analysis

Conjoint Analysis-Learning with Pradeep Chintagunta

1. Conjoint Analysis

2. What is Conjoint Analysis?
•CA is a multivariate technique used specifically to understand
how respondents develop preferences for products or services.
It is based on the simple premise that consumers evaluate the
value or utility of a product / service / concept / idea (real or
hypothetical) by combining the utility provided by each
attribute characterizing the product / service / concept / idea
•CA is a decompositional method. Respondents provide overall
evaluations of products that are presented to them as combos of
attributes. These evaluations are then used to infer the utilities
of the individual attributes comprising the products. In many
situations, this is preferable to asking respondents how
important certain attributes are, or to rate how well a product
performs on each of a number of attributes

3. Managerial uses of Conjoint Analysis
After determining the contribution of each attribute to the
consumer’s overall evaluation, one could
1. Define the object with the optimal combo of features
2. Predict market shares of different objects with different
sets of features
3. Isolate groups of customers who place differing
importances on different features
4. Identify marketing opportunities by exploring the market
potential for feature combos not currently available
5. Show the relative contributions of each attribute and each
level to the overall evaluation of the object

4. Commercial Applications
•Technique is widely used by consumer and industrial product
companies, service companies, marketing research, advertising and
consulting firms
•Over 400 commercial applications per year even in the mid 80s
•Types of applications include
–Consumer durables: automobiles, refrigerators, car stereos, condos, food
processors, HDTV
–Industrial products: copy machines, forklift trucks, computer software, aircraft
–Consumer nondurables: bar soaps, hair shampoos, disposable diapers
–Services: car rentals, credit cards, hotels, performance art series, rural health
care systems, BART
–Other: MBA job choice

5. A Survey
•Familiarity & usage of value assessment methods
•58 industrial firms in the top 125 of the Fortune 500 list
•16 market research firms from the top 40

6. Survey Results
Method Industrial Market Research
Familiarity %Usage % Familiarity %Usage %
Internal Engg.
Assessment
61.3 42.5 - -
Field value-in-use63.8 36.3 25 5
Focus group 92.5 60 90 60
Direct survey 91.3 48.8 85 55
Benchmarks 83.8 27.5 80 25
Conjoint 75 28.8 90 60
Compositional
methods
45 10 40 5

7. P&G and Disposable Diapers
•P&G makes extensive use of CA to guide product modification
•Question: What value do consumers associate with two improved
features in disposable diapers:
–Improved absorbency
–Elastic waistband
•Context: P&G had a patent on the elastic waistband, but a competitor
imitated the modification. If the imitation was illegal, what damage
should P&G claim?
•Potential answers:
1. Use market data to estimate the effect of the elastic waistband on
market share. Problem: Elastic waistband + Increased absorbency
were introduced simultaneously
2. Use CA to separately estimate the effects

8. Steps in CA
•Identification of respondents
•Identification and definition of attributes in customer
language
•Specification of attribute variation and levels
•Creation of objects (experimental design)
•Creation of instrument, including socioeconomic,
demographic and usage questions
•Sampling plan
•Data collection
•Data analysis: Typically, regression analysis separately by
respondent
•Market simulation: exploration of “what-if” questions

9. Preferences for Sports Cars
You are provided 18 hypothetical sports cars each described
on 5 features:
Point of origin: US, Japan, Europe
Convertibility: Sunroof, Removable top (Manual),
Removable top (Automatic)
Styling: Coupe (2-door), Sedan (4-door)
ABS: No, Yes
Acceleration: 0 to 60 in 5.5 secs, 0 to 60 in 8.5 secs
Assume all 18 cars are roughly equivalent on attributes not
mentioned above such as gas mileage, safety, price, etc.

10. Selecting the stimulus set of profiles
•In the above example, there are 72 possible profile combos or
“cars”. Typically, not all combos of attribute-levels are required to
estimate the conjoint model, i.e., fractional factorial designs may be
adequate
•How many profiles to include in design?
–Degrees of freedom to estimate individual level parameters
–Data collection costs and respondent load
•Criteria for profile selection
–Look out for dominated profiles and unrealistic profiles
–Most software do the appropriate selection

11. Steps in the analysis
•Each of the 18 selected profiles is presented to respondent
•Respondent indicates her/his preference for each of the
profiles by:
–Rank ordering the profiles, or
–Rating them on a 1-100 scale, or
–Choosing the most preferred alternative
•Depending on the above, an ordinal regression (LINMAP),
a regular regression or a logit model is fitted to the data
•Dependent variable is the preference measure. Independent
variables are dummy variables, i.e., presence / absence of
each of the attribute-levels
•Estimated coefficient are called part worths

12. ProfileOriginConvertibleStyleABSAccelEuroJapanAutoManualCoupeABS YFast
1 US Sun SedanNo 8.5
2 JapanSun CoupeYes5.5
3 Euro Manual SedanYes8.5
4 Euro Auto CoupeNo 8.5
5 JapanSun SedanNo 8.5
6 Euro Sun CoupeNo 8.5
7 US Manual CoupeNo 5.5
8 JapanManual SedanNo 8.5
9 Euro Sun SedanNo 5.5
10 US Maual SedanNo 5.5
11 JapanManual CoupeYes8.5
12 Euro Manual SedanNo 8.5
13 US Auto SedanYes8.5
14 JapanAuto SedanNo 8.5
15 US Sun SedanYes8.5
16 US Auto CoupeNo 8.5
17 JapanAuto SedanNo 5.5
18 Euro Auto SedanYes5.5

13. Interpreting the Coefficients or PART WORTHS
18K 17K 16K Sun Manual Auto No ABS ABS
PRICE
CONVERTIBLE BRAKING
UTILITIES UTILITIES UTILITIES
30 40
10
40
20

14. Simulating aggregate choices
Objective is to forecast likely market shares of attribute combos which
represent potential management actions, in a defined competitive
scenario
Translating Utilities into Choice Predictions
First Choice Rule
Highest utility profile chosen
by each respondent
Share of Preference Rule
Predict choice probabilities
using a model such as Logit
Both methods ignore marketing variables such as advertising weight
and distribution which are typically not in the conjoint design. Fix:
“Adjust” the market shares using this additional information

15. Using CA for segmentation
Two-Stage Approaches
A priori
Researcher selects
specific attributes
Post hoc
Full set of
attributes used
Clustering (K-means)
Relate clusters to background variables
such as demographics using techniques
like discriminant analysis
One-Stage Approach
Concomitant variable
Latent Class Conjoint
Simultaneous clustering
and profiling using
background characteristics

16. CA with large numbers of attributes
•Full profile models are unrealistic with a large number of attributes
•Two alternatives
–Self-explicated models: Respondent provides
•a) Rating of desirability of each level of each attribute
•b) Relative importance of each attribute
•Part-worths are given by (a) * (b)
•Compositional, not decompositional approach
–Hybrid models: Combine self explicated with part worth
conjoint approaches. Self explicated info is used to pare down
the number of attributes / profiles. Then a fractional factorial
design is used on the remaining. Hence, needs to be customized
for each respondent
•Sawtooth software’s ACA

17. Choice Based Conjoint
Motivation: Using conjoint judgment studies to forecast choices is
theoretically unappealing because of the ad hoc assumptions
required
In choice based conjoint, the respondent chooses one profile from the
set of alternative profiles known as the choice set. The stated
choices are used to estimate the parameters of the choice model such
as the logit model.
Advantage: Greater realism of respondent’s task
Disadvantage: Given limited information on each respondent,
individual level estimation is precluded. Hence, individual
differences (heterogeneity) needs to be accounted for in other ways