Conjoint analysis - Pennsylvania State University analysis Compositional vs. decompositional...
Transcript of Conjoint analysis - Pennsylvania State University analysis Compositional vs. decompositional...
Conjoint analysis
Conjoint analysis
Outline
▪ Conjoint analysis as a decompositional
preference model
▪ Steps in conjoint analysis
▪ Uses of conjoint analysis
Conjoint analysis
Compositional vs. decompositional
preference models
▪ Compositional: respondents evaluate all the
features (levels of particular attributes) characterizing
a product; combining these feature evaluations
(possibly weighted by their importance) yields a
product’s overall evaluation;
▪ Decompositional: respondents provide overall
evaluations of a series of products composed of
various combinations of attribute levels; the overall
evaluations are then decomposed into the utilities
associated with different levels of the various
attributes;
Conjoint analysis
Example of a compositional model
▪ Consider the following laptop computer:
□ Dell
□ 320 GB hard drive
□ 4 GB of RAM
□ 12.1 inch screen
□ Price of $1,200
▪ On a scale from 0 (lowest) to 10 (highest), how
would you rate this computer on each attribute?
▪ Assign a total of 100 points to each of the 5
attributes so that the points reflect the relative
importance of each attribute.
Conjoint analysis
Example of a decompositional model
▪ Rank the following four descriptions of laptop
computers in terms of overall preference:
Profile 1 Profile 2 Profile 3 Profile 3
Brand Dell Apple Dell Apple
Size of Hard
Drive320 320 320 160
Amount of
RAM2 4 4 4
Screen size 15.4 15.4 12.1 12.1
Price $1,200 $1,200 $1,500 $1,200
Rank
Conjoint analysis
Basic idea of conjoint analysis
▪ Overall utility for a product can be decomposed
into the utilities (called part-worths) associated with
the levels of the individual attributes of the product;
▪ The relative importance of a given attribute is
given by the ratio of the part-worth range for that
attribute divided by the sum of all part-worth
ranges;
Conjoint analysis
Steps in conjoint analysis
▪ Determine attributes and attribute levels
▪ Select product profiles to be measured
▪ Choose a method of stimulus presentation
▪ Decide on the response method
▪ Collect and analyze the data
▪ Interpret the results
Conjoint analysis
Attributes and attribute levels
▪ Identify the relevant product attributes that are
considered during choice
▪ Select attribute levels that represent the options
actually available in the market
▪ Trade-off between the completeness of the
representation and the complexity of the design
Conjoint analysis
Example: Laptop Profiles
Brand Hard Drive RAM Screen Price
Dell 320 GB 2 GB 15.4 in $1,200
Apple 320 GB 4 GB 15.4 in $1,200
Dell 160 GB 4 GB 15.4 in $900
Apple 320 GB 2 GB 15.4 in $900
Dell 320 GB 4 GB 12.1 in $1,500
Apple 320 GB 2 GB 12.1 in $1,500
Apple 160 GB 4 GB 15.4 in $1,500
Apple 160 GB 2 GB 12.1 in $900
Apple 160 GB 4 GB 12.1 in $1,200
Dell 160 GB 2 GB 12.1 in $1,200
Dell 320 GB 4 GB 12.1 in $900
Dell 160 GB 2 GB 15.4 in $1,500
Conjoint analysis
Product profiles
▪ Full factorial designs:
all possible combinations of the levels of the various
attributes are used
▪ Fractional factorial designs:
□ a subset of all possible combinations is used
□ orthogonal designs in which each level of one
attribute is paired equally with all the levels of other
attributes are beneficial
Conjoint analysis
Example: Laptop Profiles
Brand Hard Drive RAM Screen Price
Dell 320 GB 2 GB 15.4 in $1,200
Apple 320 GB 4 GB 15.4 in $1,200
Dell 160 GB 4 GB 15.4 in $900
Apple 320 GB 2 GB 15.4 in $900
Dell 320 GB 4 GB 12.1 in $1,500
Apple 320 GB 2 GB 12.1 in $1,500
Apple 160 GB 4 GB 15.4 in $1,500
Apple 160 GB 2 GB 12.1 in $900
Apple 160 GB 4 GB 12.1 in $1,200
Dell 160 GB 2 GB 12.1 in $1,200
Dell 320 GB 4 GB 12.1 in $900
Dell 160 GB 2 GB 15.4 in $1,500
Conjoint analysis
Methods of stimulus presentation
▪ Verbal descriptions
▪ Pictures
▪ Actual products or prototypes
Apple Laptop
with 320 GB of Hard Disk Space,
4 GB of RAM, and a
Screen Size of 15.4 inches –
at a Price of $1,200.
Conjoint analysis
Response method
▪ Rankings or ratings of the product profiles in terms
of preference, purchase probability, etc.
▪ Pairwise comparisons of product profiles in terms of
preference, purchase probability, etc.
▪ Choice of a product from a set of product profiles
Conjoint analysis
Brand Hard Drive RAM Screen Price A B
Dell 320 GB 2 GB 15.4 in $1,200 9 6
Apple 320 GB 4 GB 15.4 in $1,200 6 12
Dell 160 GB 4 GB 15.4 in $900 12 5
Apple 320 GB 2 GB 15.4 in $900 11 11
Dell 320 GB 4 GB 12.1 in $1,500 4 3
Apple 320 GB 2 GB 12.1 in $1,500 1 9
Apple 160 GB 4 GB 15.4 in $1,500 3 10
Apple 160 GB 2 GB 12.1 in $900 8 7
Apple 160 GB 4 GB 12.1 in $1,200 5 8
Dell 160 GB 2 GB 12.1 in $1,200 7 1
Dell 320 GB 4 GB 12.1 in $900 10 4
Dell 160 GB 2 GB 15.4 in $1,500 2 2
Example: Laptop Profiles
Conjoint analysis
In-class exerciseUsing the data in the table, answer the following questions:
a. How much utility does each of the two consumers attach to the
different levels of the five attributes? (Hint: Compute each
consumer’s average rating of all the options with a given
feature. For example, to figure out how much consumer A
values the Apple brand name, compute the average rating of
the six Apple laptops.)
b. What’s the relative importance of the five attributes for the two
consumers?
c. Consider consumer A’s ratings. For this consumer, what’s the
predicted utility of a Dell computer with 160 GB of hard drive
space and 2 GB of RAM, a 12.1 inch screen, and a price of
$1,200?
-4
-3
-2
-1
0
1
2
3
4
Apple Dell
A
B
-4
-3
-2
-1
0
1
2
3
4
HD 160 GB HD 320 GB
A
B
-4
-3
-2
-1
0
1
2
3
4
2 GB RAM 4 GB RAM
A
B
-4
-3
-2
-1
0
1
2
3
4
$900 $1,200 $1,500
A
B
-4
-3
-2
-1
0
1
2
3
4
12.1 in Screen 15.4 in Screen
A
B
Conjoint analysis
In-class exercise
Using the data in the table, answer the following questions:
a. How much utility does each of the two consumers attach to the
different levels of the five attributes? (Hint: Compute each
consumer’s average rating of all the options with a given
feature. For example, to figure out how much consumer A
values the Apple brand name, compute the average rating of
the six Apple laptops.)
b. What’s the relative importance of the five attributes for the two
consumers?
c. Consider consumer A’s ratings. For this consumer, what’s the
predicted utility of a Dell computer with 160 GB of hard drive
space and 2 GB of RAM, a 12.1 inch screen, and a price of
$1,200?
Conjoint analysis
In-class exercise
Using the data in the table, answer the following questions:
a. How much utility does each of the two consumers attach to the
different levels of the five attributes? (Hint: Compute each
consumer’s average rating of all the options with a given
feature. For example, to figure out how much consumer A
values the Apple brand name, compute the average rating of
the six Apple laptops.)
b. What’s the relative importance of the five attributes for the two
consumers?
c. Consider consumer A’s ratings. For this consumer, what’s the
predicted utility of a Dell computer with 160 GB of hard drive
space and 2 GB of RAM, a 12.1 inch screen, and a price of
$1,200?
Conjoint analysis
Review questions
▪ What’s the difference between compositional and
decompositional preference models?
▪ What’s a fractional factorial design in conjoint
analysis and why is it useful?
▪ What are part-worths in a conjoint study?
Conjoint analysis
Review questions (cont’d)
• A conjoint study was conducted for LCD TV’s,
using three brands (LG, Samsung, and Sony),
three screen sizes (46, 54, and 63 in.) and three
price levels ($2,300; $2,800; and $3,600).
• The utility differences between the lowest and
highest levels of each attribute were 3 for brand
name, 2 for screen size, and 5 for price.
• Based on these findings, price is how many times
more important than screen size?
Conjoint analysis
ME output for laptop computer examples
Respondents' Preference Partworths
Respondents' preference partworths. The most preferred profiles sum up to 100, the least preferred to 0.
Respondents / Attributes and Levels
Apple Dell 160 320 2 4
Respondent 1 0 14 0 6 0 3
Respondent 2 50 0 0 17 0 8
12.1 15.4 900 1200 1500
0 11 66 36 0
0 19 6 6 0
Respondents / Attributes and Levels
Respondent 1
Respondent 2
Conjoint analysis
Office Star data
Conjoint Study Design
Attributes and attribute levels of the Conjoint study.
Attributes / Levels Level 1 Level 2 Level 3 Ordering
LocationLess than 2
milesWithin 2-5
milesWithin 5-10
milesDecreasing
Office suppliesVery large
assortmentLarge
assortmentLimited
AssortmentUnordered
FurnitureOffice
FurnitureNo Furniture Unordered
Computers No computers Software onlySoftware and
computersUnordered
Conjoint analysis
Office Star bundles
BundlesAttribute levels for a full-profile, fractional design Conjoint study
Attributes / Bundles Bundle 1 Bundle 2 Bundle 3 Bundle 4 Bundle 5 Bundle 6 Bundle 7 Bundle 8
LocationLess than 2
milesLess than 2
milesLess than 2
milesLess than 2
milesWithin 2-5
milesWithin 2-5
milesWithin 2-5
milesWithin 2-5
miles
Office suppliesVery large
assortmentLarge
assortmentLimited
AssortmentLarge
assortmentVery large
assortmentLarge
assortmentLimited
AssortmentLarge
assortment
FurnitureOffice
FurnitureNo Furniture No Furniture
Office Furniture
Office Furniture
No Furniture No FurnitureOffice
Furniture
Computers No computers Software onlySoftware and
computersSoftware only Software only No computers Software only
Software and computers
Bundle 9 Bundle 10 Bundle 11 Bundle 12 Bundle 13 Bundle 14 Bundle 15 Bundle 16
Within 5-10 miles
Within 5-10 miles
Within 5-10 miles
Within 5-10 miles
Within 2-5 miles
Within 2-5 miles
Within 2-5 miles
Within 2-5 miles
Very large assortment
Large assortment
Limited Assortment
Large assortment
Very large assortment
Large assortment
Limited Assortment
Large assortment
No FurnitureOffice
FurnitureOffice
FurnitureNo Furniture No Furniture
Office Furniture
Office Furniture
No Furniture
Software and computers
Software only No computers Software only Software onlySoftware and
computersSoftware only No computers
Attributes / Bundles
Location
Office supplies
Furniture
Computers
Conjoint analysis
Respondents’ ratings of Office Star bundles
Respondents' Ratings
Respondents' ratings for each bundle (use consistent scale, e.g., between 0 and 100)
Respondents / Ratings Bundle 1 Bundle 2 Bundle 3 Bundle 4 Bundle 5 Bundle 6 Bundle 7 Bundle 8
Respondent 1 90 50 50 80 85 40 40 90
Respondent 2 50 55 95 50 50 40 40 85
Respondent 3 40 60 90 60 45 35 45 85
Respondent 4 75 80 60 70 90 65 60 85
Respondent 5 90 80 70 70 80 75 50 75
Bundle 9 Bundle 10 Bundle 11 Bundle 12 Bundle 13 Bundle 14 Bundle 15 Bundle 16
30 60 60 30 30 90 80 40
75 35 20 25 25 85 35 45
80 55 40 45 50 85 50 35
85 70 55 60 90 85 70 65
80 75 50 60 80 75 50 75
Respondents / Ratings
Respondent 1
Respondent 2
Respondent 3
Respondent 4
Respondent 5
Conjoint analysis
Estimating part-worths
Conjoint analysis
Conjoint analysis
Respondents' Preference PartworthsRespondents' preference partworths. The most preferred profiles sum up to 100, the least preferred to 0.Respondents / Attributes and Levels
Less than 2 miles
Within 2-5 miles
Within 5-10 miles
Very large assortment
Large assortment
Limited Assortment
Office Furniture
No FurnitureNo
computersSoftware
onlySoftware and
computers
Respondent 1 31 23 0 2 3 0 55 0 1 0 11
Respondent 2 31 16 0 3 7 0 2 0 0 1 61
Respondent 3 14 0 0 0 6 4 4 0 0 22 76
Respondent 4 17 17 0 47 22 0 9 0 0 17 27
Respondent 5 24 8 0 59 39 0 0 1 9 0 15
Respondent 6 49 17 0 22 12 0 26 0 2 0 0
Respondent 7 31 15 0 0 5 0 52 0 8 0 12
Respondent 8 22 6 0 8 9 0 0 1 3 0 68
Respondent 9 24 7 0 0 1 0 0 1 0 21 74
Respondent 10 18 18 0 43 20 0 12 0 0 15 27
Respondent 11 15 10 0 50 32 0 0 13 22 0 4
Respondent 12 50 3 0 21 17 0 22 0 7 0 1
Respondent 13 27 19 0 4 5 0 55 0 0 1 13
Respondent 14 29 12 0 9 13 0 1 0 4 0 58
Respondent 15 17 3 0 0 1 13 0 9 0 11 61
Respondent 16 16 16 0 51 24 0 2 0 0 21 32
Respondent 17 23 0 0 48 31 0 0 12 0 4 17
Respondent 18 45 12 0 33 22 0 18 0 4 4 0
Respondent 19 35 13 0 22 25 0 17 0 0 6 22
Respondent 20 29 15 0 29 32 0 18 0 3 0 21
Respondents’ part-worths for Office Star data
Conjoint analysis
Detailed preference partworths (Enginius)
Less than 2 miles
Within 2 to 5 miles
Within 5 to 10 miles
Very large assortment
Large assortment
Limited assortment
Office furniture
No furniture
No computers
Software only
Software and
computers
1 30.99 22.18 0.00 2.30 2.99 0.00 54.96 0.00 0.71 0.00 11.07
2 31.13 16.35 0.00 2.43 6.92 0.00 1.72 0.00 0.00 1.23 60.23
3 13.58 0.20 0.00 0.00 7.07 4.82 4.32 0.00 0.00 22.50 75.04
4 18.03 18.00 0.00 45.82 21.39 0.00 8.65 0.00 0.00 16.41 27.50
5 23.98 8.00 0.00 60.25 39.81 0.00 0.00 1.38 8.11 0.00 14.39
6 49.67 17.26 0.00 22.43 12.03 0.00 26.00 0.00 1.90 0.00 0.07
7 30.81 14.78 0.00 0.68 5.42 0.00 52.09 0.00 7.48 0.00 11.68
8 22.01 5.31 0.00 8.18 9.02 0.00 0.00 1.31 2.81 0.00 67.65
9 23.83 7.45 0.00 0.00 0.86 0.38 0.00 0.43 0.00 22.35 74.88
10 17.83 17.53 0.00 43.17 20.87 0.00 12.21 0.00 0.00 14.38 26.79
11 15.55 10.11 0.00 49.78 31.83 0.00 0.00 12.77 21.90 0.00 4.35
12 50.78 3.36 0.00 20.76 15.99 0.00 21.77 0.00 6.69 0.00 0.32
13 27.21 18.29 0.00 3.97 5.24 0.00 54.17 0.00 0.00 1.16 13.38
14 28.79 11.88 0.00 8.45 12.68 0.00 0.92 0.00 4.12 0.00 57.61
15 17.50 2.90 0.00 0.00 1.22 12.99 0.00 8.46 0.00 11.38 61.04
16 15.35 15.32 0.00 50.54 22.76 0.00 1.44 0.00 0.00 22.05 32.67
17 23.32 0.00 0.38 47.61 31.00 0.00 0.00 12.25 0.00 3.42 16.82
18 44.15 11.92 0.00 32.99 22.04 0.00 19.03 0.00 3.83 3.22 0.00
19 35.91 13.16 0.00 22.36 24.63 0.00 17.05 0.00 0.00 5.31 22.41
20 28.78 14.61 0.00 29.02 31.74 0.00 18.46 0.00 3.04 0.00 21.01
Conjoint analysis
Summary statistics for preference partworths
Conjoint analysis
Conjoint analysis
Conjoint analysis
Uses of conjoint analysis▪ Market segmentation
Q: How would you segment the market using individual-level
conjoint analysis output?
▪ New product design
Q: How can conjoint analysis be used for new product design?
▪ Trade-off analysis (esp. in pricing decisions)
Q: How much could the price of a Dell computer with 160 GB of
hard drive space and 2 GB of RAM, which currently sells for
$1,200, be raised if the screen size were increased from
12.1 in to 15.4 in?
▪ Competitive analysis
Q: How can conjoint analysis be used to simulate market
shares?
Conjoint analysis
Market share simulations
Conjoint analysis
Choice rules for simulations
▪ First-choice rule: the option with the highest utility
is chosen;
▪ Share of preference rule: the choice probability is
equal to the utility of an option relative to the total
utility of all options in the choice set;
▪ Alpha rule: a weighted combination of the first two
rules which maximizes the correspondence with
market shares;
▪ Logit choice rule: similar to the share of preference
rule, but the choice probabilities are computed as in
the logit model;
Conjoint analysis
Market share simulations:
Existing product profiles
Existing Product Profiles
Labels and attribute levels for each existing product profile that already exists in the market.
Attributes / Existing Product Profiles
Office Equipment Department Store
Location Within 2-5 miles Within 2-5 miles
Office supplies Large assortment Limited assortment
Furniture Office furniture No furniture
Computers Software and computers Software only
Conjoint analysis
Market share simulations with
optimal product profiles
Market Share Simulations
Market share predictions for different scenarios, using the Logit Rule.
Scenario / Product profiles
Office Equipment Department StoreMarket Share of Optimal Product
Profile
Predicted market shares 73% 27% n/a
...with Optimal Product 1 34% 13% 53%
...with Optimal Product 2 36% 14% 50%
...with Optimal Product 3 39% 14% 47%
...with Optimal Product 4 40% 15% 45%
...with Optimal Product 5 41% 15% 43%
Conjoint analysis
Optimal product profiles
Optimal Product ProfilesLabels and attribute levels for each optimal product profile that the software recommends you introduce in this market
Attributes / Optimal Product Profiles
Optimal Product 1
Optimal Product 2
Optimal Product 3
Optimal Product 4
Optimal Product 5
LocationLess than 2
milesLess than 2
milesLess than 2
milesWithin 2-5 miles
Less than 2 miles
Office suppliesVery large
assortmentLarge
assortmentVery large
assortmentVery large
assortmentLarge
assortment
Furniture Office furniture Office furniture No furniture Office furniture No furniture
ComputersSoftware and
computersSoftware and
computersSoftware and
computersSoftware and
computersSoftware and
computers
Conjoint analysis
Review: Conjoint analysis
Based on the conjoint study, LG management
knows that a price increase from $2,300 to $2,800
leads to a decrease in utility of 3. If utility goes up
by 1 when the screen size is increased from 46 to
54 inches, how much can LG charge for the TV set
with the larger screen?
Conjoint analysis
Next two classes
▪ Thursday: Text analysis and Google analytics
□ In-class exercise: Ottos’ reviews from TripAdvisor
▪ Tuesday: Dürr Environmental, Inc
□ No segmentation, introduction of 1 product only