There are many research methods I could review with you that are important to marketing
research and are especially important to research in sports marketing; for example, survey
research and questionnaire design. One especially valuable technique is conjoint analysis.
What is conjoint analysis? It’s the family of techniques that model choice by decomposing
overall preference and evaluation—in terms of the relative values or components of your
product’s attributes—of respondents. Conjoint analysis isn’t just one approach; it’s multiple
approaches for optimizing product offerings. Toothpastes have different components, such as
teeth whitening, breath freshening, and the like. Hotels offer different features, such as free
continental breakfast, free newspaper, free phone calls, or pay per view. To design the optimal
product for patrons at the maximum profit, it helps to know the monetary value that customers
assign to each product component.
In marketing, how might one typically use conjoint analysis? It could be used to design new
products. What features should be included? For a new brand in an existing product category,
this is more readily determined. Conjoint analysis can help design a new product, provided it
shares features with existing products. It can help to fine tune the marketing mix, especially
pricing. It can help identify segments of consumers who have similar product preferences.
Finally, it can simulate post-product-modified market shares. If a new brand is introduced, how
might it compete against current and likely competitor brands? Conjoint analysis allows market
simulations for estimating market shares for various configurations of a newly introduced brand.
Here are a few examples more specific to sports marketing. For equipment design, think in
terms of designing a set of golf clubs or large piece of exercise equipment. What features do
consumers find important? It’s profit sapping to include features that consumers don’t want, as
their additional willingness to pay won’t cover the cost of those features. To maximize features,
marketers try to identify features that people want and will purchase. Golf clubs—drivers and
putters, for example—are now adjustable or offer weighting systems that allow users to modify
the club. What features do consumers want in exercise equipment such as treadmills? Current
treadmills differ by top speed, programming (for exercise sessions), and maximum inclines.
Conjoint analysis can indicate what people are willing to pay for different levels of those
Sporting facilities include country clubs and sports arenas. What services should a country club
community offer? What kind of meal service? How should membership fees be structured?
What locker room features should be included? What type of golf course should be built? What
types of residences should be built around the golf course? Is a gated community preferred?
Will guards man the entrances? All these questions are necessary to determine what potential
residents want in that community.
Consider possible amenities for a sports arena. Is it an indoor or outdoor sporting facility? What
is the quality of the food service? Are the restrooms clean? What are the quality and size of
seats? Is the lighting adequate? Is there a jumbotron? These are important questions for
Page | 1
patrons who pay top dollar to attend such arenas and sporting venues. Ticket prices will reflect
preferences for those amenities. If you have a swimming pool inside an arena and no one uses
it, then that’s a needless expense. Here at NMSU, the availability of alcohol at sporting events is
an ongoing issue. That’s an amenities issue that conjoint analysis could address.
Finally, consider sports broadcasts. Team ownership must decide which media to use and in
what way. Now, most sports teams maintain an official website. During games, fans can use the
website to vote on the game MVP. Whether broadcasting games over network television,
satellite, radio, or Internet, teams face many decisions about the main and supplemental media
channels. The announcer talent is an important consideration, as an analysis like would include
fans’ willingness to watch. Pricing is one step removed, as ratings and viewership determine
sponsors’ willingness to pay for 30 and 60 second ad spots during broadcasts. Was it worth it
for NBC to pay for John Madden to announce football games? Is it worth it for FOX to pay Joe
Buck to announce baseball games? Apparently, the networks seem to think so. The networks
also pay reporters to walk around and interview the players and coaches; are viewers interested
enough in these interviews to justify the reporters’ salaries? Finally, broadcasts are supported
by many pieces of high-tech equipment. For example, new high-tech cameras offer dazzling
images that seem to attract viewers; hence, advertisers seemingly are willing to cover the cost
of this high-priced equipment.
Now that I’ve raved about the value of conjoint analysis for making optimal product and service
design decisions, let me illustrate the method with a few basic marketing examples. Don’t judge
conjoint analysis by the complexity of these examples, which are limited by this PowerPoint
medium. I’ve linked roughly 20 articles to the lecture that indicate conjoint design decisions.
This example entails determining the optimal design for packaged soup. For pedagogical ease,
it’s limited to four attributes, with a couple levels for each attribute. The soup can come in one of
three flavors: onion, chicken noodle, or country vegetable. The calories per serving can be 80,
100, or 140. The soup may or may not be salted. Salt provides an inexpensive flavor source; in
contrast, salt-free soup would require more expensive ingredients to produce a sufficiently
flavorful soup. Soups in this market are sold at two price points: $1.19 per can and $1.49. So,
what’s the optimal soup to market given customer preferences?
One way to collect conjoint data from consumers is to provide them what’s called full-profile
stimulus cards. In this example, these cards would show flavor, calorie level, salt content, and
price. Researchers could ask respondents to rate each combination of soup features on a 1 to
10 scale. Here, respondents would be asked to rate 36 different combinations.
This slide summarizes the ratings for the 36 combinations derived from three different flavors,
three different calorie levels, ‘yes’ or ‘no’ on salt freeness, and two different price levels (3 x 3 x
3 x 2 = 36). The slide show ratings as low as ‘1’ (least preferred) for ‘chicken noodle, 140
calories, no salt, and sells for $1.49’. Other combinations are rated ‘9’; for example, ‘onion, 80
calories, salt-free, and $1.19’. Conjoint analysis allows researchers to aggregate such ratings
from many respondents.
Page | 2
This slide shows utility functions. The beauty of conjoint analysis is it allows meaningful
comparisons among features measured on vastly different scales. In other words, it allows
comparing apples to oranges (or in this case, flavor versus calories versus salt freeness versus
price). Salt freeness is ‘yes’ or ‘no’, price is in dollar terms, and calories are something different,
so how can researchers identify how much consumers would value each level of each one
attribute? Conjoint analysis. In this example, the highest utility score is assigned to country
vegetable flavor at 3.66 units. Country vegetable and onion flavor are extremely important and
roughly equal in value to consumers. Chicken noodle is of no value. Calories are important:
limiting the soup to 80 calories is worth something, but only half as much as country vegetable
flavor. Dropping from 100 calories to 80 calories increases the utility relative to calories four-
fold, to 2.03 units. In contrast, dropping from 140 to 100 calories only increases utility by 0.46
units. Salt freeness also has some value, roughly in accord with 80 calories. Price isn’t as
important as salt freeness, 80 calories, or most preferred flavor.
Utility scores are ratio-scaled data, which almost no other research method creates. As a result,
it’s possible to make claims like ‘reducing the price $0.30 is only 1/5th as important as avoiding
chicken noodle flavor’. Hence, it possible to make informed decisions about the least expensive
soup that would provide the greatest utility to potential customers. Utility scores are additive, so
the soup with the highest value to customers would have these characteristics: country
vegetable flavor, 80 calories, salt free, and a $1.19 price. In contrast, the least desirable soup—
the one with no value to customers—would have these characteristics: chicken noodle flavor,
140 calories, salted, and a $1.49 price. The cost of production would now determine the most
profitable soup to produce. It would be absurd to sell a soup for $1.19 if production costs $1.29.
This is a graphical representation of the utility scores from the previous slide. The elbows or
breakpoints in the graphs indicate key attribute levels. The curve for calories provides an
excellent example; that elbow reinforces (1) the value of decreasing from 100 to 80 calories,
and (2) the value of decreasing from 140 to 100 calories is worth far less to consumers.
Here’s a graphical summary for PC attributes: different price points, memory capacities, and
different versatilities (in terms of number of possible tasks). Here, decreasing the price from
$2000 to $1500 is worth only 0.1 unit of utility; in contrast, decreasing the price from $1500 to
$1000 is worth 0.4 unit, or four times more utility gain to customers. Seemingly, people are
willing to pay a good price for a computer if they want it and it will perform needed tasks;
dropping the price several hundred dollars has little impact on the computer’s perceived value.
Conversely, some PC users are price sensitive, so price point is critical to them; for the,
dropping the price several hundred dollars may bring them into the market. For versatility,
increasing from 4 to 8 functions is worth a lot, but after 8 functions there isn’t much gain in
additional versatility. If increasing from 8 to 12 tasks requires increasing the price from $1000 to
$1500 (to cover higher production costs), then it’s not worth it to consumers; this PC would lose
0.4 units of price utility (from .8 to .4 units) but gain only 0.1 units of versatility utility. This
conjoint (also called ‘tradeoff’) analysis allows marketers to understand the tradeoffs that people
make when deciding to buy one brand versus another brand.
Page | 3
This example shows coffee makers with three attributes: brewing time, capacity, and price.
Brewing time is 3, 6, 9, or 12 minutes, the capacity is 4, 8, or 10 cups, and the price is $18, $22,
or $28. This matrix summarizes many people’s preferences for these attributes. Here, the
lowest number denotes the least preferred combination of attributes and the highest number
denotes the most preferred combination. This example is a bit odd because the most preferred
combination has a longer brewing time than another model with the same price point and
capacity. Regardless, the $18, 10 cup, and 6 minute brewing time combination is most
preferred, and the $28, 4 cup, 12 minute brewing time combination is least preferred. So, how
does conjoint analysis convert this rank-order preference data into utility scores?
This slide suggests how conjoint analysis software creates utility scores from rank-order
preference data. Assume initially that all utilities are equal for capacity, price, and brewing time.
As conjoint analysis assumes an additive model, total utility is the utility for chosen capacity plus
price plus brewing time. This additive model is simple to understand. The software then adds
the utility scores for each set of attribute levels to create a matrix of total utilities. Next, the
software compares that summed utilities matrix to the initial rank-order preference matrix (the
one with numbers 1 through 35). If the relative summed utility matrix doesn’t correspond well to
the initial rank-order preference matrix, then the software modifies the utility scores for one of
the attribute levels, recomputes the summed utility matrix, and recompares it to the initial rank-
order preference matrix. The software continues this modification-comparison routine until the
two matrices correspond at some acceptable level. In essence, the software is generating
different utility scores and then comparing the preference and utility matrices for acceptance
correspondence. For the coffee maker, the most preferred attribute combination has a utility
score based on summing 0.6 units of utility (for an $18 price), 0.5 units of utility (for a 10 cup
capacity), and 0.4 units of utility (for a 6 minute brewing time), which equals 1.4 units of utility.
The highest utility goes to the largest capacity, the cheapest, and the fastest machine, at 1.6
units of utility. What the software is doing is to create a matrix with utility scores that correspond
as closely as possible to the initial rank-order preference data.
I hope the previous examples have given you a basic idea about the data generated from a
conjoint analysis, how you might use that data to make an informed decision about the
combination of features you might design into your product, and how the computer takes
respondents’ rank-order preference data and converts it into part-worths or utility scores. What
other things would you need to worry about? I assume you’re a consumer of conjoint analysis
and not a do-er, so you’re not running the software or collecting the data. What things should
you be doing to help your researcher to design a conjoint study that will help you make the best
possible decision about the design of your product or service? The first thing to think about is
The product attributes should to be determinant, which means those attributes
completely determine what people buy or don’t buy. If you include an irrelevant product
feature in a conjoint analysis, then you’re discovering nothing about the ideal product
The attributes should be easily measured and easily communicated. The attributes in the
examples—such as soup flavor, number of calories, price point, cost of PC, number of
Page | 4
cups capacity, and brewing times—are all objective, easily communicated, and easily
measured. To be avoided are attributes like ‘plush styling’ or luxurious or any vague
descriptor that’s difficult to communicate in a way that all respondents understand. Such
attributes can’t help marketers design better products.
The attributes should be controllable by the company. If government has mandated that
certain features be incorporated in a product, it’s superfluous to ask consumers if they
want that feature. Why ask people if they are willing to pay $100 for legally mandatory
The attributes should be realistic. It’s silly to ask people if they would buy a car for $5.00
that gets 200 miles to the gallon, as such a vehicle is technologically impossible at
present. Only possible combinations of features should be presented to respondents.
People should differ in their preferences for different levels of each attribute. If people
are indifferent among soup flavors, then it’s superfluous to ask them their preferences on
The attributes should be compensatory, which is consistent with conjoint analysis as an
additive model. Recall that total utility is the sum of utilities for each included attribute
level. Compensatory means that a product may be acceptable if its low utility score on
one feature is offset by high utility scores on other features. Also, attributes should not
be complimentary, like DVDs and DVD players. If attributes are complimentary, then
both are required, which precludes a compensatory model.
The attributes should not be redundant, which is an important issue because price often
is included in a traditional conjoint analysis. The problem with price, to some extent, is
that it incorporates the other features. For example, respondents will make assumptions
about the quality and features of a low-price PC. If people are asked to rate the same
attribute more than once, then their utility scores for that attribute will be split between
the redundant attributes; as a result, the attribute will seem less important. Capacity is
not redundant of brewing time and flavor is not redundant of calories.
Here are some other kinds of considerations that are important for two reasons.
Don’t overload the software. Asking about too many attributes at too many levels makes
it difficult for the conjoint analysis software to parse the utility scores for each level of
Respondents will burn out if asked about too many attributes at too many levels. There
are ways to sidestep this problem—even with full profile methods—with balanced
designs that limit respondents to a subset of possible product configurations.
Attribute levels and their ranges must be considered to ensure they are meaningful,
informative, and realistic to both respondents and marketers. Bizarre configurations
should be avoided because they cause respondents to discount the research as silly.
Remember, adding one or two attributes levels, depending on the conjoint method used,
can increase the complexity of the respondent’s task markedly.
Page | 5
If at this point you’re enthused about using conjoint analysis to identify the optimal design of
goods and services, then I hope you’ll read (or at least skim) some of the articles I’ve provided
from Sawtooth Software, as they are free and should be accessible to most readers. Although
Sawtooth Software produces user-friendly conjoint analysis software, I’m not advocating you
buy it from them. However, some Sawtooth website content can help you to understand what
using this software entails.
The full profile method discussed previously is limited. The examples in this lecture imply a full
profile approach. Full profile conjoint analyses should be limited to six or seven attributes on
only a few levels. A full profile method with four attributes, two at three levels and two at two
levels, requires respondents to consider 36 combinations. Seven attributes at four levels would
require 4 to the 7th power (4 x 4 x 4 x 4 x 4 x 4 x 4) combinations, which clearly is unworkable. In
essence, the full profile approach yields good utility data, but respondents only can process a
few attributes on a few levels.
There are other possibilities, such as asking respondents to rate pairs of attributes in a matrix,
similar to the coffee maker example. The problem with that task is it lacks realism, which can
produce inconsistent responses and suspect data. Instead, I use Sawtooth Software’s Adaptive
Conjoint method, which allows people to initially rate the desirability of each attribute at each
level. That rating data is used to generate preliminary utility scores, which are fine tuned by
subsequently asking respondents to indicate their relative preferences for different attribute
combinations. This hybrid approach works fairly well and provides good quality data. There’s
also choice-based conjoint, which I’ve yet to use. Regardless, the Sawtooth Software articles
discuss all these approaches and when to use them.
Finally, if you’d like to walk through some conjoint exercises, Dr. Tom Novak of Vanderbuilt
University has created these two exercises. I found these on the Internet and they are offered to
you free, from Tom and me. Either the airline travel preferences or movie theatre preferences
exercise are similar to, but slightly different from, the three examples I gave in this lecture.
Page | 6