Pricing Executive Summary
Week 06 Conjoint Analysis
https://www.smh.com.au/business/companies/david‐jones‐and‐bp‐ink‐deal‐to‐bring‐fancy‐food‐to‐petrol‐ stations‐20190827‐p52l2z.html
Customer Value
• Customer Value is the total amount of money that the customer is willing to pay for the benefits received from the product.
• For pricing, each customer benefit should be equated to dollars and cents that customers are willing to pay (WTP) for it.
• Benefit 1 + Benefit 2 + ….. = WTP 1 + WTP 2 + … = Total WTP • Customer value sets the ceiling or the highest possible price that can be charged for the product.
• Understanding customer value requires an understanding of the types and number of benefits customers receive from the product and the product/ service features that contribute.
Source: Dholakia, How to price effectively, 2017
Attributes define a product
What are attributes that define a mobile phone?
What is Conjoint Analysis?
Which car should I get?
Conjoint Analysis: The Underlying Model
• A Product is a “bundle” of attributes • Consumers evaluate the alternatives in the marketplace by examining how much they offer on the various attributes and how critical each attribute is to them
• Total Value of product = sum of sub‐values (partworths) of its attribute levels to the individual
• A consumer prefers the product that delivers the greatest Total Value to him/her • Decompose the product into the value of each sub‐part in order to determine preference for the composed product/service
Example: A Consumer’s Value System for a car
v(brand) + v(engine type) + v(body type) + v(price)) = V(
Conjoint analysis model
Consumer’s overall judgment about a set of complex alternatives
Rank a set of alternatives; State their preferences
Decompose overall judgment into separate utilities for individual attributes
Statistical analysis to recover individual attribute weights, w
Preference = ∑ (w x µ) = w1 µ1 + w2µ2 + w3 µ3 + …
Given attribute levels for the item (0 or 1)
• If you choose left, you prefer Power. If you choose right, you prefer Fuel Economy.
• Rather than ask directly whether you prefer Power over Fuel Economy, we present realistic tradeoff scenarios and infer preference from your product choices.
Simple example of Conjoint Analysis
Would you prefer…
210 Horsepower or 140 Horsepower 17 MPG 28 MPG
Another simple choice‐based conjoint
More elegant ranking‐based conjoint
Far more complicated examples Discrete Choice Experiment
1. Identify a set of relevant product attributes (based on discussions with a car company)
2. Define reasonable levels for these attributes (based on carsales.com)
Stages in Conjoint Analysis A real example: Buying a car (ratings task)
Source: Havard Business School
3. Create product profiles
4. Obtain consumer preferences for profiles via survey
Concrete Conjoint Example
Source: Havard Business School
Q: With 4 attributes and 3 levels each, how many possible profiles available?
• Complete (Full) Factorial Design (CFD) • All possible combinations of attribute levels (L) of k attributes. • CFD design size = L1 x L2 x L3 x … x Lk. • E.g., 3 attributes with 2 levels, 4 levels and 5 levels each will result in 40 profiles (= 2 x 4 x 5)
• Fractional Factorial Design (FFD) • Minimum design size = 1 + (L1‐1) + (L2‐1) + (L3‐1) + … + (Lk‐1) • E.g., above case requires at least size of 9 (=1 + 1 + 3 + 4) product profiles.
• Desirable properties • Balanced – each level in an attribute appears equal number of times. • Orthogonality – each attribute is designed to be independent of one another. It requires special software such as SPSS, SAS, or R.
Create Conjoint Analysis Design
• A respondent rates every profile on a rating scale (e.g., 7‐point or 10‐ point)
• A respondent ranks each profile in terms of preference, e.g., 1st, 2nd, 3rd, … for preference rank.
• A respondent chooses the most preferred option out of a set of multiple options, e.g., one out of 4 alternatives. It requires multiple sets of choice tasks.
• A respondent chooses the most preferred option and the least preferred option out of a set of small number of options, e.g., 3 or 4 options. It requires multiple sets of choice tasks.
Data Collection
5. Analyze the Data For ratings data, simple regression can be used to compute the part‐worths for the attribute levels. Choice task requires logit or probit.
‐ Dummy coding or Effect Coding is required (see workshop)
Create a “baseline” profile • E.g., “Japanese,” “Sedan,” “Gasoline,” “$20,000” • Partworths for these levels set to 0 • Partworths of other levels = deviations from this baseline profile • Total Value of baseline profile captured by the intercept
Concrete Conjoint Example
Interpreting the Output
Intercept = Total Value for the Baseline Option
v(Japanese) + v(Sedan) + v(Gasoline) + v($20000) = 4.2 “value units”) = V(
Coefficients
Intercept 4.20
American 0.33
European ‐0.84
SUV 0.96
Sports Car ‐0.10
Hybrid 1.78
Electric 0.86
$30,000 ‐0.58
$40,000 ‐1.20
Part‐worths
Part‐Worth Plots
-1
0
1
2
3
Japanese American European
Brand Origin
-1.00
0.00
1.00
2.00
3.00
Sedan SUV Sports Car
Body Type
0.00
1.00
2.00
3.00
Gasoline Hybrid Electric
Engine Type
-2.00
-1.00
0.00
1.00
2.00
3.00
$20000 $30000 $40000
Price
What do you make of this?
What is this person’s ideal car?
How Important is Each Attribute?
• For each attribute: • Range of an attribute = max part‐worth – min part‐worth • Importance of an attribute = Range / (sum of ranges across all attributes)
ATTRIBUTE IMPORTANCES Attribute Range Importance
Brand Origin 1.17 0.22 Body Type 1.06 0.20 Engine Type 1.78 0.34 Price 1.20 0.23
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Brand Origin Body Type Engine Type Price
Sum 4.6 1.0
Choice Prediction
If presented with these three options, which one would this individual choose?
A Japanese, Sedan, Hybrid, $20,000 B European, Sports Car, Gasoline, $40,000 C American, SUV, Gasoline, $30,000
V(A)= 4.21+ v(Japanese)=0 + v(Sedan)=0 + v(Hybrid)=1.78 + v(20,000)=0 5.99 V(B)= 4.21+ v(European)=‐0.84 + v(Sports Car)=‐0.1 + v(Gasoline)=0 + v($40,000)=‐1.2 2.07 V(C)= 4.21+ v(American)=0.37 + v(SUV)=0.96 + v(Gasoline)=0 + v($30,000)=‐0.58 4.96
-1
0
1
2
3
Japanese American European
Brand Origin
-1.00 0.00 1.00 2.00 3.00
Sedan SUV Sports Car
Body Type
0.00
1.00
2.00
3.00
Gasoline Hybrid Electric
Engine Type
-2.00 -1.00 0.00 1.00 2.00 3.00
$20000 $30000 $40000
Price
So it looks like we can raise the price of option A…by how much?
Coefficients
Intercept 4.20
American 0.33
European ‐0.84
SUV 0.96
Sports Car ‐0.10
Hybrid 1.78
Electric 0.86
$30,000 ‐0.58
$40,000 ‐1.20
• Convert utilities of each level of attribute to price level that a person is willing to pay for the desired level of attribute
• From the output, we know that V($20,000) = 0 and V($40,000) = ‐1.20. Thus, $20,000 increase implies 1.2 utilities decrease ($1 = 1.2/20,000 util).
• SUV has higher utility (0.96) than Sports car (‐0.10). So, the person is willing to pay more for SUV than for Sports car of same brand.
• Utility difference between two types = 0.96 – (‐0.10) = 1.06 $17.667 (= 20,000 x (1.06/1.2)).
Willingness‐To‐Pay Trade‐Off Analysis
Coefficients
Intercept 4.20
American 0.33
European ‐0.84
SUV 0.96
Sports Car ‐0.10
Hybrid 1.78
Electric 0.86
$30,000 ‐0.58
$40,000 ‐1.20
• Try to select a representative sample of the market of interest • Below are the average importance weights across such a sample
From Individual to Market Level Analysis
0 .1
.2 .3
Brand Origin Body Type Engine Type Price
• Often more insightful to analyze the market by looking at all individuals to properly account for the heterogeneity in preferences
From Individual to Market Level Analysis
0 .2
.4 .6
P ric
e
0 .2 .4 .6 .8 Brand
Brand v. Price
0 .2
.4 .6
P ric
e
0 .2 .4 .6 .8 Body Type
Body Type v. Price
0 .2
.4 .6
P ric
e
0 .2 .4 .6 Engine Type
Engine Type v. Price
• Based on the preferences of each individual in the sample, what would the market share be of the following three options be?
A Japanese, Sedan, Hybrid, $20,000 B European, Sports Car, Gasoline, $40,000 C American, SUV, Gasoline, $30,000
From Individual to Market Level Analysis
47.7%
32.5%
19.8%
A B
C 2. Identify most preferred option by each individual
1. Compute utilities of each option by each individual
3. Sum of all individuals who would choose each option
Decision Support System Example Flight configuration
Decision Support System Example Wine configuration
Decision Support System Example Wine configuration
• Ideal product development based on preferred attributes • Product is a bundle of attributes. So, combining most preferred attributes will result in the most preferred product offering.
• Segmentation based on individual part‐worths • Some emphasize brands, and others emphasize prices, etc. • Some are price sensitive and some are not.
• Market share forecast leads to Decision Support System • Willingness‐To‐Pay trade‐off analyses
• Compare utilities of price with utilities of any other attributes. • Pursuing higher level of an attribute, e.g., higher horsepower, increases price. WTP for higher horsepower can be computed.
Usage of Conjoint Study
• Service or experiential goods are hard to evaluate without actual experience
• Defining attributes and levels are hard • Preferences are not well formed
• Some customer segments may find the conjoint ratings task to be very difficult
• Ways of better implementation • Prototype or Testable products • Artificial environment mimicking real experience • Advanced conjoint such as Adaptive Conjoint (ACA) or Information Acceleration (IA)
Difficulty of Conjoint Analysis
• Parle‐G case (Workshops) • Case report is due at 3pm on Friday, 6 September. • Submit an electronic copy via Turnitin on UTSOnline
• NO Lecture unless demanded
Next week