Lectures Conjoint Analysis
Course objectives
At the end of this course, you will be able to help companies design products that work:
- Which new product to launch? à product design decisions are highly complex and a
product is characterized by many, many attributes. And some of these attributes are
very complex to understand.
- How to position existing products better?
- How to price existing and new products?
- How to manage product portfolios / product line designs?
Lecture 1 – Intro Conjoint Analysis
What is conjoint analysis?
Products are represented as bundles of attributes. And
levels of each attribute define the product.
What it is used for?
Conjoint analysis is a survey-based technique that
allows the analyst to understand people’s preferences for a product and especially the
trade-offs they make in making choices.
Why conjoint analysis?
- In direct surveys, respondents might say that they consider all attributes important
à that is not informative, because consumers make trade-offs. They can’t have all.
- Conjoint enforces tradeoffs between attributes as in real purchase occasion à all
attributes evaluated at once. Respondents evaluate “complete” products with both
strong and weak attributes.
- Conjoint reduces problem of socially desirable answers
- Conjoint adds realism à in real-life setting, consumers evaluate products and choose
what they like.
- Conjoint analysis is straightforward à there is a lot of software available perform the
analysis.
Conjoint in the age of big data?
Why should we run hypothetical choice experiments if firms can increasingly make use of
large amounts of transactional purchase data?
1. Lack of experimental price variations required to learn consumers’ preferences
2. Conjoint allows to measure consumer preferences for products or attribute levels not
yet introduced in the marketplace
3. Especially relevant for pricing of new product innovations
Field experiments are prominent alternatives for the goal of learning consumer preferences.
However…
1. Often, field experiments are difficult to conduct and not feasible in high-ticket
product categories (like cars, laptop)
2. Field experiments are limited to products already existing the marketplace
1
,You can use conjoint analysis in marketing, but also in medicines (what is the likelihood of
success) and law (select jury members) or human resources or economics.
Why is that relevant?
Every year there are many new products that are introduced. But very few succeed. With
conjoint analysis you can predict your market share. If your market share would be low, this
saves you a lot of money to invest.
Consumers complains a lot on social media. But how do you have to respond to customer
complaints? Therefore it is important that you’ve identified relevant attributes. Maybe you
have to be really fast in your response. Because a study (in airlines) shows that when a tweet
is answered in five minutes or less, the customer will pay almost $20 more for a ticket on
that airline in the future.
Steps in conjoint analysis
1. Conjoint design
2. Conjoint analysis
3. Choice simulator
Conjoint design
- Ranking-based conjoint à choose the most-preferred product, then the second
most-preferred product, until the least-preferred product
- Rating-based conjoint à Give a score to each product in turn. Problems of rating:
o Not realistic à In real-life we buy products rather than rating them.
o It is not clear whether spread in ratings is due to real preferences or due to
response style
o Implications for sales levels and market shares are not clear. Sales and market
shares result from consumer choices, not ratings.
- Choice-based conjoint à choice between
different variants (this course focus) and choose
the most-preferred product only.
Choice-based conjoint
- We record the choices made by every customer
during the n tasks (with n the number of
products, in practice between 10 and 15).
- Because, in every choice set, a different combination of attribute levels is used, we
can derive the effect of different combinations of attribute levels on choice. à
preferences = part-worths of attributes
- As we repeat the conjoint exercise across many customers, we can also detect
whether different customers have different preferences à customer-specific
preferences
Advantages of choice-based conjoint
- Trade-offs are enforced even more
- It is very realistic because of the trade-offs
2
, - You have the no-choice option included à none of the offered alternatives is
attractive. You must always include this in an experiment. This is needed to do a sales
proxy, so what is your sales of market share?
- Avoids the need of ad-hoc rules to predict market shares
- No subjective scaling
- Choice is cognitively less demanding than ratings
Disadvantages of choice-based conjoint
- Hypothetical bias à respondents’ product choices might be influenced by the
experimental setting. Respondents know that they are in an experimental setting,
which gives bias.
- Small individual level data à respondents become fatigue if exposed to a large
number of choice tasks that are actually required if the experiment includes large
numbers of attribute levels
- Bayesian methods and prior specifications à Bayesian statistical methods help as
they efficiently pool information across respondents. However, some analysts regard
the inclusion and specification of priors as subjective. This is really technical.
Conjoint analysis
Once you have your data set ready, you need a statistical model to estimate the customer
preferences. That is the logistic regression:
DV = logit(IV)
With DV the product chosen or not and IV the product attributes.
à But what if we have different tastes and what if you prefer the freedom of an Android
while I like the synchronization feature of the iPhone?
Than latent-class and hierarchical bayes are the answers.
Aggregate level (same part- Segment level (different Individual level (different
worths for all respondents) part-worths for segments) part-worths per respondent)
- assuming same + realistic, as segments take + realistic, with different
preferences may give into account different preferences
misleading results preferences + high precision thanks to
+ high precision, as all + high precision if all the joint estimation
respondents are combined respondents are used in one - how may strategies should
big analysis the firm implement given
- there is no heterogenity the diversity of preferences
Method: LOGIT Method: LATENT CLASS Method: HIERARCHIAL
ANALYSIS (LCA) BAYES (HB)
Market simulations
What are the market shares after introducing a new product or change the consumer price?
We assume competitive market shares.
What can the choice simulator do?
3
, - It lets predict you which SKU respondents or segments of the population will choose
(estimate demand and market share)
- Lets play “what-if” games to investigate the value of modifications to an existing
product or alternative
- Lets you investigate product line extensions
Lecture 2 - Conjoint Design
Conjoint is to decompose a product into attributes. The car
can be decomposed in various attributes. The price, brand,
trunk size, etc.
These are all the attributes that I can use to describe the car
that I show to consumers. Each attribute has different levels.
To find out what the optimal price is, you as a manager can
use conjoint analysis. I can design a car with more fuel
sufficient or less. I cannot so easily change my brand, this is more fixed. But I can learn
whether consumers like my brand, Fiat, in comparison to my
competitor, Mercedes or vs Skoda.
Conjoint helps me to understand what customers value.
The fiat is one option, but the Skoda is another option. It has a
different brand, different price, different fuel consumption, etc.
Showing consumers these two options, and if I do this
repeatedly, I can learn what consumers value in terms of
attribute levels. How can I do this? I need to decompose the
product into its attribute levels.
Decompositional view
To make it more general, you first have the
product (STIMULUS), then I have the different
attributes and then I have the different levels. I
can decompose all the products in the world. I
can make it as complex as I want, when I add
more attributes or more levels. That is what you
need to decide about if you want to do a
conjoint.
How do I bring this to utility? I decompose a
product into attribute levels, but in the end of the
day I want to measure consumer utility for the
different attribute levels. The way we do it is that
we estimate the utility for the different attribute levels.
We decompose a product into the sum of its attribute levels and want to estimate
consumer’s preferences (utilities) for the different levels. Then I sum it up and I end with a
total utility for a product. Then I can compare total utilities across products and then say
that the consumer chooses the product that has the highest utility.
The utility of a stimulus à the sum of the utilities of the various attributes’ levels.
Utility of Fiat 500 = brand (Fiat) + Trunk size (<300L) + …
4
Course objectives
At the end of this course, you will be able to help companies design products that work:
- Which new product to launch? à product design decisions are highly complex and a
product is characterized by many, many attributes. And some of these attributes are
very complex to understand.
- How to position existing products better?
- How to price existing and new products?
- How to manage product portfolios / product line designs?
Lecture 1 – Intro Conjoint Analysis
What is conjoint analysis?
Products are represented as bundles of attributes. And
levels of each attribute define the product.
What it is used for?
Conjoint analysis is a survey-based technique that
allows the analyst to understand people’s preferences for a product and especially the
trade-offs they make in making choices.
Why conjoint analysis?
- In direct surveys, respondents might say that they consider all attributes important
à that is not informative, because consumers make trade-offs. They can’t have all.
- Conjoint enforces tradeoffs between attributes as in real purchase occasion à all
attributes evaluated at once. Respondents evaluate “complete” products with both
strong and weak attributes.
- Conjoint reduces problem of socially desirable answers
- Conjoint adds realism à in real-life setting, consumers evaluate products and choose
what they like.
- Conjoint analysis is straightforward à there is a lot of software available perform the
analysis.
Conjoint in the age of big data?
Why should we run hypothetical choice experiments if firms can increasingly make use of
large amounts of transactional purchase data?
1. Lack of experimental price variations required to learn consumers’ preferences
2. Conjoint allows to measure consumer preferences for products or attribute levels not
yet introduced in the marketplace
3. Especially relevant for pricing of new product innovations
Field experiments are prominent alternatives for the goal of learning consumer preferences.
However…
1. Often, field experiments are difficult to conduct and not feasible in high-ticket
product categories (like cars, laptop)
2. Field experiments are limited to products already existing the marketplace
1
,You can use conjoint analysis in marketing, but also in medicines (what is the likelihood of
success) and law (select jury members) or human resources or economics.
Why is that relevant?
Every year there are many new products that are introduced. But very few succeed. With
conjoint analysis you can predict your market share. If your market share would be low, this
saves you a lot of money to invest.
Consumers complains a lot on social media. But how do you have to respond to customer
complaints? Therefore it is important that you’ve identified relevant attributes. Maybe you
have to be really fast in your response. Because a study (in airlines) shows that when a tweet
is answered in five minutes or less, the customer will pay almost $20 more for a ticket on
that airline in the future.
Steps in conjoint analysis
1. Conjoint design
2. Conjoint analysis
3. Choice simulator
Conjoint design
- Ranking-based conjoint à choose the most-preferred product, then the second
most-preferred product, until the least-preferred product
- Rating-based conjoint à Give a score to each product in turn. Problems of rating:
o Not realistic à In real-life we buy products rather than rating them.
o It is not clear whether spread in ratings is due to real preferences or due to
response style
o Implications for sales levels and market shares are not clear. Sales and market
shares result from consumer choices, not ratings.
- Choice-based conjoint à choice between
different variants (this course focus) and choose
the most-preferred product only.
Choice-based conjoint
- We record the choices made by every customer
during the n tasks (with n the number of
products, in practice between 10 and 15).
- Because, in every choice set, a different combination of attribute levels is used, we
can derive the effect of different combinations of attribute levels on choice. à
preferences = part-worths of attributes
- As we repeat the conjoint exercise across many customers, we can also detect
whether different customers have different preferences à customer-specific
preferences
Advantages of choice-based conjoint
- Trade-offs are enforced even more
- It is very realistic because of the trade-offs
2
, - You have the no-choice option included à none of the offered alternatives is
attractive. You must always include this in an experiment. This is needed to do a sales
proxy, so what is your sales of market share?
- Avoids the need of ad-hoc rules to predict market shares
- No subjective scaling
- Choice is cognitively less demanding than ratings
Disadvantages of choice-based conjoint
- Hypothetical bias à respondents’ product choices might be influenced by the
experimental setting. Respondents know that they are in an experimental setting,
which gives bias.
- Small individual level data à respondents become fatigue if exposed to a large
number of choice tasks that are actually required if the experiment includes large
numbers of attribute levels
- Bayesian methods and prior specifications à Bayesian statistical methods help as
they efficiently pool information across respondents. However, some analysts regard
the inclusion and specification of priors as subjective. This is really technical.
Conjoint analysis
Once you have your data set ready, you need a statistical model to estimate the customer
preferences. That is the logistic regression:
DV = logit(IV)
With DV the product chosen or not and IV the product attributes.
à But what if we have different tastes and what if you prefer the freedom of an Android
while I like the synchronization feature of the iPhone?
Than latent-class and hierarchical bayes are the answers.
Aggregate level (same part- Segment level (different Individual level (different
worths for all respondents) part-worths for segments) part-worths per respondent)
- assuming same + realistic, as segments take + realistic, with different
preferences may give into account different preferences
misleading results preferences + high precision thanks to
+ high precision, as all + high precision if all the joint estimation
respondents are combined respondents are used in one - how may strategies should
big analysis the firm implement given
- there is no heterogenity the diversity of preferences
Method: LOGIT Method: LATENT CLASS Method: HIERARCHIAL
ANALYSIS (LCA) BAYES (HB)
Market simulations
What are the market shares after introducing a new product or change the consumer price?
We assume competitive market shares.
What can the choice simulator do?
3
, - It lets predict you which SKU respondents or segments of the population will choose
(estimate demand and market share)
- Lets play “what-if” games to investigate the value of modifications to an existing
product or alternative
- Lets you investigate product line extensions
Lecture 2 - Conjoint Design
Conjoint is to decompose a product into attributes. The car
can be decomposed in various attributes. The price, brand,
trunk size, etc.
These are all the attributes that I can use to describe the car
that I show to consumers. Each attribute has different levels.
To find out what the optimal price is, you as a manager can
use conjoint analysis. I can design a car with more fuel
sufficient or less. I cannot so easily change my brand, this is more fixed. But I can learn
whether consumers like my brand, Fiat, in comparison to my
competitor, Mercedes or vs Skoda.
Conjoint helps me to understand what customers value.
The fiat is one option, but the Skoda is another option. It has a
different brand, different price, different fuel consumption, etc.
Showing consumers these two options, and if I do this
repeatedly, I can learn what consumers value in terms of
attribute levels. How can I do this? I need to decompose the
product into its attribute levels.
Decompositional view
To make it more general, you first have the
product (STIMULUS), then I have the different
attributes and then I have the different levels. I
can decompose all the products in the world. I
can make it as complex as I want, when I add
more attributes or more levels. That is what you
need to decide about if you want to do a
conjoint.
How do I bring this to utility? I decompose a
product into attribute levels, but in the end of the
day I want to measure consumer utility for the
different attribute levels. The way we do it is that
we estimate the utility for the different attribute levels.
We decompose a product into the sum of its attribute levels and want to estimate
consumer’s preferences (utilities) for the different levels. Then I sum it up and I end with a
total utility for a product. Then I can compare total utilities across products and then say
that the consumer chooses the product that has the highest utility.
The utility of a stimulus à the sum of the utilities of the various attributes’ levels.
Utility of Fiat 500 = brand (Fiat) + Trunk size (<300L) + …
4
