© Copy Right: Rai University212 11.556

RE

SE

AR

CH

ME

TH

OD

OLO

GY

We will now learn of an important technique, which helps usdetermine the best or optimal combination of product featuresor attributes. Conjoint measure-ment is the statistical techniquetypically used to identify the most desirable com-bination ofattributes or features for the particular product or service under in-vestigation. This is a highly advanced technique and we shall onlybe able to touch on it intuitively. A more detailed explanation isbeyond the scope of our course. However it is an importanttechnique, which helps a marketing person, decides on what is theoptimal combination of product features to be offered.By the end of this lesson you should have an intuitiveunderstanding of what is conjoint analysis and its applications inmarketing and other types of research problems.

What is Conjoint Analysis?Conjoint analysis is a technique used to identify the most desirablecombination of features to be offered in a new product. Itaddresses the problem of how the customer will value the varioustangible and intangible features offered by a particular firm’sproduct.Conjoint measurement also tells us the extent to whichrespondents are willing to give up (trade off) some features andattributes to retain others.Thus conjoint analysis is done to determine what utility a consumerattaches to attributes such as:• Price (high, low,)• After sales service (frequent, monthly, yearly, guarantee)• Product features

Conjoint analysis - How it worksA consumer is asked to compare different products attributecombinations and rank them. Respondents are to indicate thecombination they most prefer, the second most preferred, etc.Conjoint analysis is applied to categorical variables, which reflectdifferent features or characteristics of products. For example for anew product the features may be:• Colour (different shades)• Size (largest vs. medium vs. small)• Shape (square vs. cylindrical)• Price (different price levels)It differs from factor analysis because it is only applied to categoricalvariables. It is similar to factor analysis in that it tries to identifyinterdependencies between a number of variables where thevariables are the different features.We can best understand Conjoint analysis with the help of anexample:

Example 1Suppose we have to design a public transport system. We wish totest the relative desirability of three attributes:

The company aims to provide a service. They wish to test threelevels of frequency, and three levels of prices. Further they want totest the weightage given by consumer to add on features such asAC and music. The conjoint problem can be presented as follows:Fare (three levels Rs10, Rs15, Rs 20)Frequency of service (10 minutes, 15 minutes, 20 minutes)AC vs non AC vs. music (Ac & music, AC, music, nothing)A sample of 500 respondents are selected and asked to rank theirpreferences for all possible combinations and for each level. Theseare shown below along with one respondent’s sample rankings.We can present our trade off information in the form of a table:Table 1

Frequency Ac AC&music Music Nothing

10

15

20

1

5

9

2

6

10

3

7

11

4

8

12

Basically the respondent’s preference ranking help reveal howdesirable a particular feature is to a respondent. Featuresrespondents are unwilling to give up from one preference rankingto the next are given a higher utility. Thus in the above examplethe respondent gives a high weightage to service followed by AC.the offer of music is clearly not very important as he ranks it belowAC. However he is not willing to trade off frequency of servicewith either AC or music.Conjoint analysis uses preference rankings to calculate a set ofutilities for each respondent where one utility is calculated for eachrespondent for each attribute or feature. The calculation of utilitiesis such that the sum of utilities for a particular combination showsa good correspondence with that combination’s position in theindividual’s original preference rankings. The utilities basically showthe importance of each level of each importance to respondents.We can also identify the more important attributes by looking atthe range of utilities for each of the different levels.For Example• Frequency of service has a range from 1.6 to .04. The

range is therefore equal to =1.2.A high range implies thatthe respondent is more sensitive to changes in the levelof this attribute.

• These utilities are calculated across all respondents for allattributes and for different levels of each attribute.

At the end of the analysis we would identify 3-4 of the mostpopular combinations would be identified for which the relativecosts and benefits can be worked out.

LESSON 36:CONJOINT ANALYSIS

© Copy Right: Rai University11.556 213

RE

SE

AR

CH

ME

TH

OD

OLO

GY

Uses of conjoint analysis

• It is used in industrial marketing where a product can havemany combinations and features and not all features would beimportant to all consumers. In industrial marketing the analysiscan be done at the individual level, as each individual isimportant.

• In case of consumer goods the analysis should be done segmentwise. To avoid unnecessarily long questionnaires a preliminaryfactor analysis should be run to select only testable attributes.Also the number of attributes should be restricted.

ProblemsIt is important that the attributes be selected carefully. The analysisassumes the attributes are important to consumers.We Now Present some Applications of Conjoint Analysis fromthe Internet.Conjoint helps understand why consumers prefer certain products:

U [product] = U [attribute1 (i)] + U [attribute2(j)] + ... + U[attribute(k)]

Where:U [product]=overall utility, or “worth” of the productU [attribute (y)] = “part worth” of the yth level of the Xth Attribute J = number of attributes

Conjoint analysis is a sophisticated technique for measuringconsumer attitudes and preferences. Like the multi-attributemodel, it helps understand why consumers prefer certain products.Also like the multi-attribute model, it decomposes overallpreference into a series of additive terms.However, there is an important difference between conjoint analysisand the multiattribute model:• The multiattribute model is compositional - it builds up an

inferred overall attitude as the sum of measured sub-components.

• The conjoint model is deco positional - it measures overallpreference and decomposes this into inferred sub-components.

2. Example of Conjoint Analysis TechniquePackaged soups.Four attributes with the following levels:

a. Flavor onion, chicken noodle, country vegetableb. Calories 80, 100, 140c. Salt-free yes, nod. Price $1.19, $1,49Altogether there are 3x3x2x2 = 36 possible combinations.A consumer could, in theory, rate each of the 36 combinationson a 9-point preference scale.

3. Conjoint ExamplePackaged Soups Results for One Subject Salt

Flavor Cal. Free Price RatingOnion 80 yes $1.19 9Onion 80 yes $1.49 8Onion 80 no $1.19 6

Onion 80 no $1.49 6

Onion 100 yes $1.19 7Onion 100 yes $1.49 6Onion 100 no $1.19 5Onion 100 no $1.49 5

Onion 140 yes $1.19 7Onion 140 yes $1.49 6Onion 140 no $1.19 5Onion 140 no $1.49 5

Chicken 80 yes $1.19 7Chicken 80 yes $1.49 6Chicken 80 no $1.19 2Chicken 80 no $1.49 2

Chicken 100 yes $1.19 3Chicken 100 yes $1.49 3Chicken 100 no $1.19 2Chicken 100 no $1.49 1

Chicken 140 yes $1.19 2Chicken 140 yes $1.49 2Chicken 140 no $1.19 2Chicken 140 no $1.49 1

Vegetable 80 yes $1.19 9Vegetable 80 yes $1.49 8Vegetable 80 no $1.19 7Vegetable 80 no $1.49 6

Vegetable 100 yes $1.19 8Vegetable 100 yes $1.49 7Vegetable 100 no $1.19 6Vegetable 100 no $1.49 5

Vegetable 140 yes $1.19 6Vegetable 140 yes $1.49 5Vegetable 140 no $1.19 5Vegetable 140 no $1.49 4

4. Part-worth calculated for one subject: normalized relative

Attribute level mean mean importance Flavor vegetable 6.33 1.00 43%

onion 6.25 .98chicken 2.75 .00

© Copy Right: Rai University214 11.556

RE

SE

AR

CH

ME

TH

OD

OLO

GY

Calories 80 6.33 1.00 26%100 4.83 .58140 4.17 .40

Salt-Free yes 6.06 .92 23%no 4.17 .40

Price $1.19 5.44 .75 8%$1.49 4.78 .56

5. Orthogonal ArraysIn actual applications, it becomes impossible to present all possiblecombinations of attributes to a consumer. Consider carpetcleaners:

3 package designs3 brand names3 price pointsGood Housekeeping Seal yes/noMoney Back guarantee yes/no

b. There are 108 possible combinations?c. Q. what is the fewest number of combinations we can get by

with?a. Sum of the number of degrees-of-freedom for the main

effects of each attribute:(3-1) + (3-1) + (3-1) + (2-1) + (2-1) = 8

d. We need to find an “orthogonal array” of 8 profiles, whichallows us to estimate all additive main effects in the conjointmodel. Typically, we at least double the minimum number forgreater stability.

Issues in designing a conjoint studya. Attribute selectionb. Collecting preference datac. Typical sample sizesd. Profile vs. two-factor evaluatione. Computerized (adaptive) approaches7. Desirable problem situations for conjoint analysisa. Product is realistically decomposableb. Product is reasoned high-stake decisionc. All combinations that are presented to respondent are

reasonabled. Product/service alternatives can be realistically describede. New product alternatives can be synthesized from basic

attributes8. Airline Example: Stagesa.  Develop relevant set of attributes and select appropriate

levelsb. Use a fractional factorial design to create an orthogonal

array of stimuli.c. Rate or rank the stimulid. Estimate part-worths for attribute levels.e. Estimate relative importance of attributesf. Interpretation

9. Airline Example: Discussiona. Interpretation of resultsb. Selecting attributesc. Selecting attribute levelsd. Applications to market segmentatione. Applications to product development10. ExampleConjoint analysis for faculty chair candidate. Conjoint as aid todecision making.

Attributes Used were

a. Area of specialization (quantitative, consumer behavior,strategy, management, international)

b. Research orientation (star, active, inactive, minor)c. Teaching orientation (star, good, average, below average)d. Current position (chair, full, associate)e. Role in department (work with junior faculty, work with

faculty at other schools, work with business community)• Discuss Part-Worths and relative importances for eight

faculty members.• Discuss use of non-metric data and monotone

transformation of faculty rank orders, which optimizedthe fit of the conjoint model

Sawtooth Software

Research Paper SeriesUnderstanding Conjoint Analysis in15 MinutesJoseph Curry,Sawtooth Technologies, Inc.1996© Copyright 1996 - 2001, Sawtooth Software, Inc.530 W. Fir St.Sequim, WA 98382(360) 681-2300www.sawtoothsoftware.comUnderstanding Conjoint Analysis in 15 MinutesJoseph Curry(Originally published in Quirk’s Marketing Research Review)Copyright 1996, Sawtooth SoftwareConjoint analysis is a popular marketing research technique thatmarketers use to determine what features a new product shouldhave and how it should be priced. Conjoint analysis became popularbecause it was a far less expensive and more flexible way to addressthese issues thanconcept testing.The basics of conjoint analysis are not hard to understand. I’llattempt to acquaint you withthese basics in the next 15 minutesso that you can appreciate what conjoint analysis has to offer.A simple example is all that’s required.

© Copy Right: Rai University11.556 215

RE

SE

AR

CH

ME

TH

OD

OLO

GY

Suppose we want to market a new golf ball. We know fromexperience and from talking withgolfers that there are threeimportant product features:! Average Driving Distance! Average Ball Life! PriceWe further know that there is a range of feasible alternatives foreach of these features, for instance:Average Driving Distance Average Ball Life Price275 yards 54 holes $1.25250 yards 36 holes $1.50225 yards 18 holes $1.75Obviously, the market’s “ideal” ball would be:Average Driving Distance Average Ball Life Price275 yards 54 holes $1.25 and the “ideal” ball from a cost ofmanufacturing perspective would be:Average Driving Distance Average Ball Life Price225 yards 18 holes $1.75assuming that it costs less to produce aball that travels a shorter distance and has a shorter life.Here’s the basic marketing issue: We’d lose our shirts selling thefirst ball and the marketwouldn’t buy the second. The most viable product is somewherein between, but where?Conjoint analysis lets us find out where.A traditional research project might start by considering the rankingsfor distance and ball life inFigure 1.

Figure 1Rank Average Driving Distance Rank Average Ball Life1. 275 yards 1 54 holes2. 250 yards 2 36 holes3. 225 yards 3 18 holesThis type of information doesn’t tell us anything that we didn’talready know about which ball toproduce.Now consider the same two features taken conjointly. Figures 2aand 2b show the rankings of the 9 possible products for twobuyers assuming price is the same for all combinations.Figure 2aBuyer 1 Average Ball Life54 holes 36 holes 18 holes275 yards 1 2 4250 yards 3 5 7AverageDrivingDistance 225 yards 6 8 9Figure 2bBuyer 2 Average Ball Life54 holes 36 holes 18 holes275 yards 1 3 6

250 yards 2 5 8AverageDrivingDistance 225 yards 4 7 9Both buyers agree on the most and least preferred ball. But as wecan see from their otherchoices, Buyer 1 tends to trade-off ball life for distance, whereasBuyer 2 makes the oppositetrade-off.The knowledge we gain in going from Figure 1 to Figures 2a and2b is the essence of conjointanalysis. If you understand this, youunderstand the power behind this technique.Next, let’s figure out a set of values for driving distance and asecond set for ball life for Buyer 1so that when we add these valuestogether for each ball they reproduce Buyer 1’s rank orders.Figure 3 shows one possible scheme.

Figure 3

Buyer 1 Average Ball Life54 holes5036 holes2518 holes0275 yards100(1)150(2)125(4)100250 yards60(3)110(5)85(6)60AverageDrivingDistance225 yards0(7)50(8)25

© Copy Right: Rai University216 11.556

RE

SE

AR

CH

ME

TH

OD

OLO

GY

(9)0Notice that we could have picked many other sets of numbersthat would have worked, so thereis some arbitrariness in themagnitudes of these numbers even though their relationships toeachother are fixed.Next suppose that Figure 4a represents the trade-offs Buyer 1 iswilling to make between ball lifeand price. Starting with the values we just derived for ball life,Figure 4b shows a set of valuesfor price that when added to thoseball life reproduce the rankings for Buyer 1 in Figure 4a.

Figure 4aBuyer 1 Average Ball Life54 holes 36 holes 18 holes$1.25 1 4 7$1.50 2 5 8 Price$1.75 3 6 9Figure 4bBuyer 1 Average Ball Life54 holes5036 holes2518 holes0$1.2520(1)70(4)45(7)20$1.505(2)55(5)30(8)5Price$1.750(3)50(6)25

(9)0We now have in Figure 5 a complete set of values (referred to as“utilities” or “part-worths”) thatcapture Buyer 1’s trade-offs.

Figure 5

275 yards 100 54 holes 50 $1.25 20250 yards 60 36 holes 25 $1.50 5225 yards 0 18 holes 0 $1.75 0Average Driving Distance Average Ball Life PriceLet’s see how we would use this information to determine whichball to produce. Suppose wewere considering one of two golf balls shown in Figure 6.

Figure 6Distance Ball Long-Life BallDistance 275 250Life 18 54Price $1.50 $1.75The values for Buyer 1 in Figure 5 when added together give us anestimate of his preferences.Applying these to the two golf balls we’re considering, we get theresults in Figure 7.

Figure 7Buyer 1Distance 275 100 250 60Life 18 0 54 50Price $1.50 5 $1.75 0Total Utility 105 110Distance Ball Long-Life BallWe’d expect buyer 1 to prefer the long-life ball over the distanceball since it has the larger total value.It’s easy to see how this can be generalized to several different ballsand to a representative sample ofbuyers.These three steps—collecting trade-offs, estimating buyer valuesystems, and making choice predictions— form the basics ofconjoint analysis. Although trade-off matrices are useful forexplaining conjoint analysis as in this example, not many researchersuse them nowadays. It’s easier to collect conjoint data by havingrespondents rank or rate concept statements or by using PC-basedinterviewing software that decides what questions to ask eachrespondent, based on his previous answers. As you may expectthere is more to applying conjoint analysis than is presented here.But if you understand this example, you understand what conjointanalysis is and what it can do for you as a marketer.

Point to Ponder• Conjoint analysis is a technique that typically handles non

metric independent variables.• Conjoint analysis allows the researcher to determine the

importance of product or service attributes and the levels offeatures that are most desirable.

© Copy Right: Rai University11.556 217

RE

SE

AR

CH

ME

TH

OD

OLO

GY

• Respondents provide preferences data by ranking or rating cardsthat describe products

• These data become utility weight of product characteristics bymeans of optimal scaling and loglinear algorithms.

Notes