Apple Inc. v. Samsung Electronics Co. Ltd. et al
Filing
991
Administrative Motion to File Under Seal Documents Re Apples Opposition To Samsungs Motion To Exclude Opinions Of Certain Of Apple Experts filed by Apple Inc.. (Attachments: #1 Declaration Of Cyndi Wheeler In Support Of Apples Administrative Motion To File Under Seal Documents Re Apples Opposition to Exclude Apple Experts Opinions, #2 [Proposed] Order Granting Apples Administrative Motion To File Under Seal, #3 Apples Opposition To Samsungs Motion To Exclude Opinions Of Certain Of Apples Experts, #4 Declaration Of Mia Mazza In Support Of Apples Opposition To Samsungs Motion To Exclude Opinions Of Certain Of Apples Experts, #5 Exhibit Mazza Decl. Ex. D, #6 Exhibit Mazza Decl. Ex. F, #7 Exhibit Mazza Decl. Ex. G, #8 Exhibit Mazza Decl. Ex. J, #9 Exhibit Mazza Decl. Ex. K, #10 Exhibit Mazza Decl. Ex. L, #11 Exhibit Mazza Decl. Ex. R, #12 Exhibit Mazza Decl. Ex. S, #13 Exhibit Mazza Decl. Ex. T, #14 Exhibit Mazza Decl. Ex. U, #15 Exhibit Mazza Decl. Ex. V, #16 Exhibit Hauser Decl. Ex. B, #17 Exhibit Hauser Decl. Ex. C, #18 Exhibit Hauser Decl. Ex. D, #19 Exhibit Hauser Decl. Ex. E, #20 Exhibit Musika Decl. Ex. S, #21 Exhibit Musika Decl. Ex. T, #22 Exhibit Musika Decl. Ex. U, #23 [Proposed] Order Denying Samsungs Motion To Exclude Opinions Of Apples Experts)(Jacobs, Michael) (Filed on 5/31/2012) Modified on 6/3/2012 attachment #1 Sealed pursuant to General Order No. 62 (dhm, COURT STAFF).
Exhibit C
Commercial Use of Conjoint Analysis: A Survey
Author(s): Philippe Cattin and Dick R. Wittink
Reviewed work(s):
Source: Journal of Marketing, Vol. 46, No. 3 (Summer, 1982), pp. 44-53
Published by: American Marketing Association
Stable URL: http://www.jstor.org/stable/1251701 .
Accessed: 29/05/2012 17:56
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Marketing.
http://www.jstor.org
Philippe Cattin & Dick R. Wittink
Use
Conjoint analysis has been used extensively in
marketing research to estimate the impact of selected product (service) characteristics on customer preferences for products (services). In this
paper we discuss findings obtained from a survey
of commercial users of the methodology. We project that around 1,000 commercial applications have
been carried out during the last decade. We discuss the manner in which the methodology is
used commercially, remaining issues that deserve
further exploration, and recent advances or insights obtained by researchers working in this
area.
Commercial
of
Conjoint
Analysis: A
Survey
Introduction
ARKETING researchers have made considerable use of conjoint analysis to estimate the impact of selected product (or service) characteristics on
consumer preferences. Commercial use of the methodology appears to be widespread, although it is not
universally accepted. In designing a conjoint analysis
study, a researcher has many alternative approaches
to choose from. To gain insight into the extent of
usage and the manner in which the methodology is
applied, a mail questionnaire was developed and sent
to all research firms in the U.S. that were known to
the authors to offer conjoint analysis commercially.
Seventeen (all but three) of these firms cooperated
with the survey before the end of 1980. In this paper
we discuss the survey findings, and we elaborate on
Cattin anAssociate
is
Professor MarketingtheUniversity
of
at
Philippe
of Connecticut Dick Wittink an Associate
and
R.
is
Professor Marof
and
Methods Cornell
at
The
keting Quantitative
University. authors
thank E. Green,
Paul
Richard Johnson, Srinivasan Yoram
M.
V.
and
Wind their
for
with
to
instrument
helpful
suggestions regard thesurvey
andthe listof potential
The
Science
Institute
respondents. Marketing
for
The
provided
maintain refull
partial
support thisstudy. authors
for
version thesurvey
of
sponsibility thefinal
the
instrument, execution
of thestudy thecontent thispaper.
and
of
44 / Journalof Marketing,
Summer1982
some of the remaining issues associated with the application of conjoint analysis. The 17 research firms1
taken together have carried out approximately 700 research projects with conjoint analysis. Of these projects, 160 had been carried out during the most recent
12 months. The survey questions pertained to:
* the frequency of usage of the methodology by
product/service category;
0 the specific purpose of the research study;
0 the method used for generating (defining) product attributes or characteristics;
* the model (e.g., a main effects, part worth
model);
* the number of attributes;
* the method of data collection (i.e., full profiles,
trade-off matrices);
'Actually, two of the respondents represent advertising agencies,
and one response was obtained from a large corporation. We did not
contact other large corporations because the number of projects carried out for any one corporation is likely to be small. Furthermore,
most institutions employ outside agencies to conduct a study involving
conjoint analysis. However, we contacted all research and consulting
companies and advertising agencies known to us to offer conjoint
analysis commercially, as defined by Green and Srinivasan (1978, p.
104).
Journal of Marketing
Vol. 46 (Summer 1982), 44-53.
* the mode of presenting objects (stimuli) to respondents;
* the response mode for collecting preference
judgments (e.g., rank order preferences);
* the usage of alternative parameter estimation
procedures;
* reliability and validity of the results;
* the method employed to summarize the results
for presentation to management;
* issues associated with the implementation of
results; and
* remaining issues and unsolved problems.
We have also attempted to relate the survey findings
to the latest developments in conjoint analysis, and
we make recommendations regarding issues that warrant further study.
Survey Results
of Usage by Product Category
Based on the responses, the first commercial project
involving conjoint analysis was completed in 1971.
The number of projects carried out per respondent at
the time the survey was conducted (during the latter
half of 1980) varies from 2 to 200. Taken together,
the respondents have completed 698 projects during
this decade, or approximately 70 projects per year.2
During the most recent 12 months, however, a total
of 160 projects have reportedly been carried out by
the respondents.
The commercial applications of conjoint analysis
have been broken down by product category to determine for which types of products and services the
methodology has been employed by the survey respondents. Based on the results in Table 1, it appears
that the majority of commercial applications involved
consumer goods.3 Also, consumer and industrial
goods together account for more than 80% of the applications. There are no trends apparent in the category breakdown, as shown by the similarity between
the percentages for the period since a company started
using conjoint analysis and the figures reported for the
most recent 12 months.
Frequency
2This is a conservative estimate of the total number of projects.
Considering nonrespondents and companies not contacted (including
those in foreign countries), we project that around 1,000 projects have
been completed'as of 1981.
3Although the consumer goods category figures prominently in the
frequency of usage of the methodology, this does not imply that conjoint analysis is more appropriatefor such goods compared with other
goods.
TABLE 1
Frequency of Usage of Conjoint Analysis
by Product/Service Category
Category
Consumer goods
Industrialgoods
Transportation
Financial services
Government
Other services
Total
Since
Company
Started
429 (61%)
138 (20%)
25 (4%)
53
(8%)
18 (3%)
35
(5%)
698
During
Most
Recent
12 Months
96 (60%)
33 (21%)
5 (3%)
6 (4%)
7 (4%)
13 (8%)
160
Project Purpose
In a given commercial application of the conjoint
methodology, multiple objectives may be served.
Green and Srinivasan (1978) mention the evaluation
of new product or service concepts, consideration of
alternative communication campaigns, and market
segmentation among the objectives for applications in
the private sector. From the survey we find that new
product/concept identification was the purpose or one
of the purposes in 72% of the projects (see Table 2).
Perhaps somewhat surprisingly, pricing was mentioned second most frequently as an objective in 61%
of the projects. Based on the survey responses, market
segmentation was one of the objectives in 48% of the
projects, and advertising was identified 39% of the
time. The results were virtually identical if the responses for the most recent 12 months were used instead.
It should be noted that other procedures are available as well for the purposes discussed here. Specifically, several survey respondents mentioned that conjoint analysis is not used when the attributes used to
define objects tend to be "soft." Thus, if the preference for an item is determined by perceptual dimensions and the perceptions are difficult or impossible to relate to physical attributes, conjoint analysis
TABLE 2
Percentage of Applications Involving
Specified Purpose
Purpose
New product/concept identification
Pricing
Marketsegmentation
Advertising
Distribution
Percentage of
Applications
72%
61%
48%
39%
7%
Commercial of Conjoint
Use
A
Analysis: Survey 45
/
may not be a suitable methodology. Some respondents indicated that they do not use conjoint analysis
when the numberof attributesis large or when attributes tend to be highly correlated.4
I
TABLE3
Frequency of Usage of Alternative Attribute
Generation Methods
Attribute Generation
One of the key assumptions underlying the methodology is that an individual's preferencefor an object
can be decomposed into preference scores for components or characteristicsof the object. If a main effects model is used, the preferencescore for a given
attributelevel does not depend on any other attribute.
In order to gain insight into how preferencesare determinedand how the preferences for actual objects
in the marketplacecan be influenced, it is essential
that considerable effort be expended on generating
attributes.In a typical application (excluding applications involving computerinteractivetechniques), a
set of attributesis defined prior to the collection of
preferencejudgments.
In some applications, management may exclusively decide the set of relevant attributes. For example, in one study not covered by this survey, management was primarilyinterestedin the influence of
certain price variables relative to other dimensions,
on the preferencefor subscriptionseries of arts programs(Currim,Weinbergand Wittink 1981). In other
cases, the set of attributes
may be based more heavily
on direct consumer input. In general, input from the
targetmarketas well as -frommanagementshould be
used. Thus the attributesshould include those most
relevant to potential customers and those which satisfy the managerialconstraint(variablesto be manipulated either in productdesign, pricing, communication campaignsor distributionefforts).
Every survey respondentmentionedthat the client
(i.e., management)was involved in the generationof
attributes,as shown by the numberof respondentsindicating usage of "Expert judgment of client's personnel" in Table 3. Input from the target market is
obtained through a variety of procedures, including
group interviews and direct questioning of individual
subjects. Only five respondents indicated that they
used "protocols." However, these respondentstend
to favor this procedurequite strongly, as indicatedby
the median rank value for this procedure.
Other procedures mentioned include the examinationof alternative
products/servicesavailablein the
and in-depth interviews of individual
marketplace
subjects. The examination of existing items in the
marketplaceshould not be relied upon exclusively,
4Some respondents have used conjoint analysis with a large number
of attributes. The methodology can also accommodate correlated attributes. Such problems have been handled successfully in commercial and other applications.
46 / Journal Marketing,
of
Summer
1982
Method
Expert judgment of
client's personnel
Group interviews
Direct questioning of
individual subjects
Kelly's repertory grid
Protocols
Other
Number of
Respondents
Indicating
Usage
Median
Rank*
17
15
2
2
10
7
5
3
3
3
1
3
*Based on "used most often" (1) to "used least often" and
tabulated only for respondents who indicated that they used
the method in question.
however. An advantage of conjoint analysis is the
possibility of obtaining informationabout the influence of an attributeon preference, even when the existing items available in the marketplacedo not vary
on the attribute. For example, the available brands
may be offered at identical prices. Yet this does not
mean that price should be excluded as an attributein
the conjoint analysis project. It should be noted, however, that the other proceduresmay also be limiting
in the sense that the attributesgeneratedmay reflect
the variationin existing items ratherthanthe potential
impact an attributehas on consumer choice. In general, the generation of attributesand the amount of
variationon a given attributeallowed for in the definition of hypothetical stimuli, should allow for the
not
discovery of opportunitiesin the marketplace otherwise evident. In addition, conjoint analysis can be
used to obtain quantitativeassessments of the market
potentialfor new or modified products.
Model Specification
The most common model used by the survey respondents is the part worth model. In this model preference
for an object is assumed to be an additive function of
the values (worths) of its components (attributelevels). Companiesthat carry out only a limited number
of conjoint analysis studies per year tend to rely exclusively on this model. In the vector model a continuousfunction is used to representthe influences of
attributeson preference. For continuous functions an
attributeshould be measured on at least an interval
scale. For example, price was mentionedby respondents as an attributefor which a continuous and often
nonlinearfunction would be used to estimate the effect on preference. A quadraticfunction (Green and
Srinivasan 1978, pp. 105-6; Pekelman and Sen,
1979a, 1979b) is an example of a continuous nonlinear function. Such a function can be used to approximate cases where preference is assumed to be monotone decreasing (increasing), such as for automobile
preference as a function of miles per gallon. This
specification can also be used for attributesthat are
of the ideal point type, such as the amount of sugar
in a dessert.
Three of the survey respondents, one being a
heavy user of the methodology, indicated that they
use noncompensatory well as compensatory
as
models.
A noncompensatory
model may, for example, include
a cutoff rule indicatingthat an object will not be considered at all if its specification on a given attribute
is below a minimumlevel. Specifically, an individual
may consider only medium sized automobilesno matter how attractivea compact might be on other characteristics. One survey respondentindicated that the
specification of the (noncompensatoryor compensatory) model would depend on the decision making
structure,as identified in qualitative marketresearch
conducted prior to the collection of preferencejudgments. Of course, such a decision making structure
may change as the natureof the available alternatives
changes.
Another survey respondent uses a hierarchy of
noncompensatory(threshold, disjunctive and lexicographic) and compensatory rules to approximatean
individual's approachto determiningpreferences for
alternativeobjects. One respondentindicatedthat the
goodness of fit is usually higher for noncompensatory
models compared with compensatory models. However, in a recent study Olshavsky and Acito (1980)
found no significant difference in either internal or
external validity between selected compensatoryand
noncompensatorymodels.
Some respondents are using recently developed
modeling procedures that may be especially useful
when the numberof attributesis relatively large, such
as componential segmentation (Green and DeSarbo
1979) and hybridmodels (Green, Goldbergand Montemayor 1981). The hybrid modeling procedure includes direct ratingsof attributelevels and of attribute
importances.
Number of Attributes
Clients of the survey respondents appear to identify
many attributesas being potentiallyrelevant (up to as
many as 50 attributes).However, there may be substantial overlap between the attributesinitially identified. Thus a smaller set of attributescan be defined
to capturemuch, if not all, of the initial set. Across
all respondentsthe median numberof attributesvaries
from 3 to 15, but for most respondents the median
numberof attributesactually used in conjoint analysis
is 6 or 7. The number is kept relatively small, espe-
cially if the preference judgments are collected by
means of the full profile approach. Individualshave
difficulty evaluating objects defined on more than six
attributesat a time because of informationoverload
(Green and Srinivasan 1978).
Data Collection Procedures
The two main alternativeproceduresare the full profile or concept evaluation approachand the trade-off
matrix or two factors at a time approach.Extensions
or variations of these procedures(and issues related
to data collection) are discussed by Green and Srinivasan (1978, pp. 107-9). We have summarized in
Table 4 the relative frequencywith which commercial
applications completed by the survey respondents
have involved alternativeapproaches.From this table
it can be seen that a majorityof the conjoint analysis
applications has involved the full profile approach.
The relative popularityof the full profile approachis
especially pronounced if only the most recent 12
months are considered (69% of all applications)compared with the time since a company started using
conjoint analysis (56% of the applications).
Eight survey respondentsstatedthatthey favor the
full profile approachbecause it is more realistic, as
exemplified by the statement, "It is the most realistic
reflection of the choice environment." Otherbut less
frequent reasons given in favor of this approachinclude speed, ease of administration,validity, interviewee convenience, flexibility in analysis anidless
respondentfatigue. The reasons given for the use of
the two factors at a time approachinclude the ability
to use many attributes,the speed with which the interview is completed, and the clarity of understanding
of the task by respondentsas well as by management.
It should be noted that the task of evaluatingconcepts becomes more complex as the numberof attributes used to define the concepts increases. For this
reason many researchersdo not vary more than ap-
TABLE 4
Relative Frequency of Usage of Alternative Data
Collection Methods
Method
Full profile (concept
evaluation)
Two factors at a time
(trade-off matrices)
Combination of full profile
and two factors at a time
Other
Since
Company
Started
During
Most
Recent 12
Months
56%
69%
27%
13%
14%
3%
15%
3%
Commercial of Conjoint
Use
A
Analysis: Survey 47
/
proximately six attributesin the set of profiles presented to an individual. Nevertheless, some survey
respondentshave used the full profile approach for
applicationsinvolving substantiallymore than six attributes. However, usually special designs are used
in such cases. For instance, a variation of the full
profile approachcan be used on two or more subsets
of profiles defined on, say, six attributes,where some
attributesare used in more than one subset (Green
1974). Moreover, as indicated earlier, some survey
respondents have begun to use recently developed
methods that involve alternative data collection approaches (e.g., Green, Goldberg and Montemayor
1981).
Interactive data collection procedures (Johnson
1980) are used by three of the survey respondents.
Reasons given for the use of interactive approaches
include speed of data collection, managementinterest, respondentinterest, data quality and breadthof
coverage. With interactivetechniques the researcher
has flexibility in varying the numberand natureof the
attributesas well as the specification of attributelevels across interviewees. Additional flexibility is obtained when the preference judgments requested at
any particularstage in the data collection process depend on the preferencejudgments already provided
by the individual. The consistency of an individual's
judgments can be tested, and the interviewee can receive immediate feedback on his/her preferences.
Furthermore,the arduous task of processing information from questionnairesis avoided.
Such interactiveproceduresare likely to become
more popular as the technology required becomes
more widely available, although the choice of interviewees is constrainedunless the equipment needed
to collect the informationis portable and can be set
up quickly in alternativelocations. Respondentreaction to the use of computerinteractivetechniquesappears to be favorable as indicated by the results of a
study conducted for Xerox (MacBride and Johnson
1980). In this study, responses obtained throughthe
electronic approach had higher predictive validity
than a paper and pencil interview.
Methods of Presenting Stimuli to Interviewees
Verbaland paragraph
descriptionsof hypotheticalobjects are the most commonly used methods, as shown
in Table 5. The survey respondents mentioned that
these procedures are convenient, inexpensive and
Pictorialrepresentations also used,
straightforward.
are
but typically in combinationwith verbal descriptions.
Only on rareoccasions does the set of objects involve
actualproducts,presumablybecause the development
of prototypesis frequentlynot feasible.
It should not be surprisingthat the methodof presentation may affect the responses. Verbal and para48 / Journal Marketing,
of
Summer
1982
TABLE5
Relative Frequency of Usage of Alternative
Methods of Presenting Stimuli to Respondents
Since
Company
Method
Started
Verbal descriptions
50%
20%
Paragraph descriptions
Pictorial descriptions
19%
Actual products
7%
Other*
4%
*Other
includesmodels or pseudoproducts.
During
Most
Recent 12
Months
46%
23%
17%
9%
5%
graph descriptions are subject to response biases reare
sultingfrom the orderin which attributes presented
(Johnson1981). There is also some evidence thatpictorial representations more likely to produceconare
figural processing of the informationpresented(Holbrook and Moore 1981).
Response Mode for Preference Judgments
Based on the answers provided for the period since
a company started, a preferencerank order of hypothetical objects was elicited more frequently comparedwith the use of rating scales. However, the difference in relative frequency of usage between these
two response modes is minimal for the most recent
12-monthperiod (see Table 6). Reasons provided by
the surveyrespondentsfor using rankorderjudgments
include ease of use, ease of administration,
and a desire to keep the judgment task as close as possible to
TABLE6
Relative Frequency of Usage of Alternative
Response Modes for Preference Judgments
Response Mode
Rank Order
Paired comparison
Rating scale
Other*
Since
Company
Started
45%
11%
34%
10%
During
Most
Recent 12
Months
41%
5%
39%
15%
Variable Definition
Preference
33%
44%
Liking
10%
8%
Intention to buy
54%
46%
Other**
3%
2%
*Other
consists primarily gradedpairedcomparisons.
of
**Otherincludesactual purchaseor order placementby respondents.
a consumer's usual shopping behavior. Rating scales
are favored by some survey respondentsbecause, it
is claimed, the rating scales are less time-consuming
for an interviewee and also because of interviewee
convenience and ease of analysis.
Rating scales may not be practicalwhen trade-off
matrices are used to collect preference judgments.
However, graded paired comparisonjudgments indicating the strengthof preference for one object over
another can be used instead, as shown by Johnson
(1980). If rank order preferencejudgments are used,
researchers should consider the possibility that the
numberof levels used for the attributesmay have a
systematic influence on the substantive results. Adjustments may be necessary before the results can be
compared across attributeswith varying numbers of
levels (Wittink, Krishnamurthi Nutter 1982).
and
The remainderof the results in Table 6 shows the
relativeusage of alternativedefinitions for the dependent (criterion)variable. "Intention to buy" is most
frequently used, with "Preference" a close second
choice, particularlyif measured during the most recent 12-monthperiod.
Alternative Estimation Procedures
The results in Table 7 suggest that regression analysis
(and analysis of variance) is now the most commonly
used estimation procedure, based on the survey responses for the most recent 12 months. This contrasts
with the result that MONANOVA is the single most
popularmethodduringthe period since the companies
startedusing conjoint analysis. Note, however, that
the "Other" category has the largest relative frequency. This category includes the PERMUTE al-
TABLE7
Relative Frequency of Usage of Alternative
Estimation Procedures*
Since
Company
Started
MONANOVA
PREFMAP
LINMAP
Monotone regression
Regression/ANOVA
LOGITanalysis
Other**
During
Most
Recent 12
Months
24%
3%
22%
3%
4%
16%
10%
48%
5%
28%
15%
36%
*The percentages do not add up to 100 because some respondents often use more than one method for "convergence."
**Other includes the PERMUTE
algorithm (an algorithm similar to MONONOVA)
and a repertoire of methods that includes
most of the above.
gorithm and proceduresconsisting of a repertoireof
methods (involving the identificationof noncompensatory rules and the specification of decision-tree
structures).None of the respondentsindicated usage
of LINMAP (Srinivasan and Shocker 1973) for parameterestimation, even though this technique was
listed in the questionnaireas one of the possible procedures. We do, however, know that LINMAP has
been applied commercially. Approximately70 orders
(half to the commercial sector) have been filled for
this program (Shocker 1981). Based on the survey
responses, there is a definite trend toward increasing
use of techniquessuch as LOGITand regressionanalysis. This trendmay be relatedto the increasingusage
of rating scales for the collection of preferencejudgments.
The substantive results obtained from conjoint
analysis do not seem to depend very much on the specific estimationprocedure.MONANOVA and regression analysis tend to provide similar results in terms
of parameter
estimates for both rankordersand rating
scales (Carmone,Green and Jain 1978). This is probably due to the fact that MONANOVA suffers from
local optimumsolutions (Wittinkand Cattin 1981, pp.
104-5). LINMAP is guaranteedto provide a global
optimum. With simulated data, LINMAP outperformedMONANOVA,regression
analysisandLOGIT,
in terms of external validity, when rank order preference data were obtained from a dominantattribute
model. On the other hand, for compensatorymodels
with approximatelynormally distributedpart worths
for the attributelevels, the externalvalidity was highest for regression analysis (Wittinkand Cattin 1981).
The comparisons are based on analytical procedures
which in some cases have been updated.For example,
Srinivasanhas extended the linear programmingprocedure to include a "strict paired comparison" approach. He obtained higher predictive validity, in
terms of the percentageof first choices predictedcorrectly, with this extension (Srinivasan1981) than was
obtainedwith the earlier version of LINMAP used in
the study by Jain et al. (1979).
Other factors that may dictate or influence the
choice of estimationprocedureinclude the availability
of software and the ability to constrainparameterestimates. For example, with functionsthatare expected
to be monotone increasing (decreasing), simulation
results indicate that predictive validity is improvedif
a quadraticor part worth function is constrainedto be
monotone over the range of attributelevels (Cattin
1981). With LINMAP such constraintscan be introduced quite readily. Least squaresprocedurescan also
be adaptedto accommodatesuch constraints.
Hybrid models have been introduced (Green,
Goldberg and Montemayor 1981) to reduce the data
requiredfrom an individual respondent.Interviewees
Commercial of Conjoint
Use
A
Analysis: Survey 49
/
are first clustered based on self-explicated data. The
preferencejudgments are used to estimate main effects and selected interactionsat the segment (or aggregate) level. Ultimately, each interviewee's model
is a weighted sum of the person's (individual) selfexplicatedmodel and a segment or aggregateconjoint
model. Cattin, Gelfand and Danes (1981) recently
proposed a simple Bayesian regression procedureto
combine self-explicated data with an individualized
conjoint model. Their analytical results indicate that
the Bayesian procedure should outperforma procedure based on the individual preference judgments
only. In eithercase, the idea behindthe hybridmodels
is to augment, not to replace, the conjoint data. Selfexplicatedmodels by themselves appearto have lower
external validity compared with the conjoint models
(Cattin, Gelfand and Danes 1981; Green, Goldberg
and Wiley 1981).
Reliability and Validity
The survey respondentsindicatedthat they frequently
obtain measures of reliability. The measuresused include replications (i.e., asking interviewees to evaluate one or more stimuli twice, at different times),
consistency checks and split half reliabilitymeasures.
With respect to validity, most of the survey respondents(11) indicatedthat they do obtain measures
of validity, at least sometimes. Holdout stimuli or
validation samples are used to compare actual and
predictedpreferencejudgmentsor to compute a crossvalidated correlation (external validity). A few respondentscheck only the internalvalidity to eliminate
data from interviewees ". . . whose data are not ex-
plainableby a model." The predictive validity is assessed occasionally through a comparison of an interviewee's preferencerankingof brandsavailable in
the marketplacewith the predicted ranking obtained
from the individual's preferencemodel.
The reliabilityof the results in an aggregatesense
is influenced by the numberof interviewees included
in a study. This sample size is determinedpartly by
the purposeof the study and the allocatedbudget. The
summaryresults in Table 8 show that the sample size
varies considerably across the survey respondents.
The median sample size varies across the respondents
from 100 to 1,000, althoughthe median is in the 300
to 550 range for most respondents.
Analysis of Conjoint Analysis Results
To summarizeresults to managementand to identify
promising marketingactions, the survey respondents
carry out at least one of the following procedures:
* a marketsimulation,based on preference
models
estimatedat the individuallevel, to predictmar50 / Journal Marketing,
of
Summer
1982
TABLE8
Sample Size Used by Survey Respondents
Mean
Median
(across respondents)
2107
1200
High
Low
138
100-150
Median
466
500
II
ket sharesundervarious scenariosinvolving the
introductionof a new productor modifications
of an existing product (including changes in
marketingmix variables);
* aggregationof preferencejudgments, and estimation of preference model parameters,at the
segment level, or aggregation of individual
level parameterestimates at the segment level;
* full aggregation of preference data across all
interviewees.
To make marketshare predictionsit seems advisable to use individualizedpreferencemodels. Wittink
and Montgomery (1979) obtained higher predictive
validity for preference models estimated at the individuallevel thanfor segmentbased preference
models.
The segment based models, in turn, were more predictive of actual choice behavior, subsequent to the
collection of data for conjoint analysis, than models
estimatedacross all individuals. With respect to market simulation, special optimization procedures are
available, such as POSSE (Green, Carrolland Goldberg 1981). Some survey respondentsindicated they
use such procedures.
to
However, it is not straightforward make market
share predictionsbased on conjoint analysis, for several reasons. (1) The conjoint models are based on
preferenceor intent-to-purchase
behavior, not on actual behavior. (2) There are likely to be attributesexcluded from the models that may affect behavior in
the marketplace.(3) The models should be extended
to include the effect of mass communication, distribution effort and competitive reactions. (4) Perceptions of a product and its attributesmay have to be
modeled to incorporate the differences (links) between perceptionsand objective or physical features
included in the conjoint study. (5) New productsmay
take several years to be developed and marketed,after
which the natureof the competitionmay be different,
and systematic changes in customer preferencesmay
have taken place. Therefore, it is importantto update
the models or to determinehow economic and other
conditions influence the models' parameters.
It should be noted that even if market share predictions are made based on individualizedmodels, the
results may also be summarized at some aggregate
(i.e., segment) level to provide insight to management
about the part worths of attribute levels for selected
subgroups of potential customers. In this regard, Currim and Wittink (1980) have shown the equivalence
of several alternative aggregation schemes when least
squares analysis is used to estimate parameters. In a
strict sense, however, the parameter estimates may
not be comparable across individuals. At a more practical level, this difficulty is perhaps not severe. Researchers may, however, want to standardize the parameter estimates for each interviewee before making
interpersonal comparisons.
A common way of summarizing results is to compute attribute importance weights, based on the conjoint analysis results, at the individual level and average these weights across individuals within a market
segment. Usually, such derived importance weights
are based on taking the difference between the parameter estimates of the most preferred and least preferred
levels of a given attribute. In this manner, the importances reflect the amount of variation used for a
given attribute in creating hypothetical objects. Furthermore, the importance of one attribute is measured
relative to the importance of the other attributes used
in the study.
Recent research results suggest, however, that the
derived attribute importances may also reflect the
number of attribute levels used in a study if rank order
preference judgments are collected, even when the
most and least preferred levels are held constant
(Wittink, Krishnamurthi and Nutter 1982). Thus, if
price is an attribute and a decision has been made to
use as maximum and minimum levels $20 and $15,
respectively, then the derived importance of price
may be influenced by the number of intermediate
price levels used to define hypothetical objects. This
implies that the parameter estimates (e.g., part worths)
may not be comparable across attributes. Hence, market share estimates may also be influenced systematically by the number of attribute levels (Wittink and
Krishnamurthi 1981). We emphasize that this potential problem has been observed only for rank order
preference judgments and not for cases when rating
or other scales are used.
Implementation
of Results
The survey respondents reported that the conjoint
analysis results have had impact on concept (product)
design, the selection of features for a new product,
modification of existing products and pricing. A number of survey respondents expressed reservations
about the extent of impact because ". . . it is difficult to assess the role of research in the final execution
and implementation."
Unsolved Problems
The survey respondentsmentioned the following aspects as problems that may require attention:
* the validity of the results (especially with respect to estimated marketsharesbased on preference judgments for hypotheticalobjects);
* the numberof observationsneeded for reliable
estimation of parameters;
* how to handle a large number(i.e., more than
eight) of attributes;
* how to deal with problems involving multiple
decision makers, for example, industrialbuying
centers and family decision making;
* how to detect and incorporateinteractionsand
nonlinearitiesin preferencemodels; and
* the lack of "good software."
With respect to these points we note that the hybrid utility estimation approachadvocated by Green,
Goldberg and Montemayor(1981) is designed to incorporateselected interactions, estimated at the segment level. This approach can also accommodate a
large numberof attributes.A recent study by Krishnamurthi(1980) addresses the modeling of joint decision making situations.
Concluding Comments
Finally, we want to address a few other issues associated with conjoint analysis. Althoughthe survey respondents did not specifically mention these issues,
we feel that these points deserve additionalthought.
In informal discussions with other researchers,some
of these points have been raised.
Of general concern is the question about the realism of the task required from interviewees. The
trade-offs involved in comparing alternative hypothetical objects may seem quite unreal to individuals
cooperatingwith a conjoint analysis study. If the preference judgments involve hypotheticalobjects representative of what is available in the marketplace,the
interviewee may not have any difficulty providingrealistic evaluations. If, on the other hand, the hypothetical objects differ dramatically from the actual
objects (products) available, the task will be more
demanding and the judgments may not be as representative of what an individual would actually do in
a marketplacesetting. If the quality of the preference
judgmentsdeclines as the task for the interviewee becomes more artificial (i.e., less representative of
choices available in the marketplace),then the applicability of the methodology may be severely limited.
Conceivably then the results from conjoint analysis
Commercial of Conjoint
Use
A
Analysis: Survey 51
/
if performedat some aggregate level may not differ
dramatically from results that could be obtained
throughquantalchoice models.5 The main advantage
of conjoint analysis would then consist of the opportunityto estimate models at the individuallevel based
on an experimentallycontrolled set of objects.
Related to this issue is the ability and/or willingness of interviewees to provide accurate preference
judgments. This problem is unique to conjoint analysis (and other survey research), as opposed to researchsuch as quantalchoice models which are based
on revealedpreferences.Intervieweeswho have agreed
to cooperatewith a study will differ on the degree to
which they are devoted to providingaccurateanswers.
If the preferencejudgment task is perceived to be an
interestingone by an interviewee, we would expect
greater accuracy compared with a task that is perceived to be uninteresting.For example, if an automobile manufacturerwants to select individuals to
react to hypothetical automobiles, the manufacturer
should take a sample from the targetmarket. Individuals currentlyconsidering the purchase of a new car
should be motivatedand equippedto provide accurate
preference judgments. However, such individuals
may be reluctantto cooperate with a research study
if they cannot distinguish the research from a sales
pitch. Recent automobile purchasersdo not have to
worryabout a possible sales pitch. Nevertheless, their
motivation to provide accuratepreferencejudgments
is likely to be considerablylower. Researchersshould
pay careful attentionto the problem of motivatinginterviewees.
5Inquantal
choice of an objectis explained
choice, the marketplace
as a functionof the object'scharacteristics
acrossa set of customers,
with limitedopportunity allow for individualheterogeneity the
to
in
estimates.See Madansky(1980), Plott (1980) and Sriniparameter
vasan(1980a)for comparisons conjointanalysisandquantal
of
choice
models.
Consumer choice in the marketplaceusually involves variousperceptualdimensions as well as physical characteristicsof alternativebrands. The insight
providedby the researchresults may be improvedby
explicitly incorporating
perceptualdimensions. For an
example of recent research involving the estimation
of relationshipsbetween preferences,perceptionsand
physical features, see Hauser and Simmie (1981) and
Holbrook (1981).
With respect to the selection of attributes,the use
of brandname and price are somewhat controversial.
Brand names may capture a number of aspects that
may be covered separately by other attributes.As a
consequence, conjoint analysis results for brandmay
be very difficult to interpret.Nevertheless, actual or
perceived advantagesassociated with the brandname
are relevant to the questions addressedin a conjoint
analysis study. Similarly, price is controversial because interviewees may view price as an indication
of productquality. For a discussion of issues associated with the use of price as an attribute,see Rao and
Gautschi (1980) and Srinivasan(1980b).
Apart from these issues, we speculate that computer interactive techniques will receive increasing
attention. This and other developments have been
motivated, at least in part, by a desire to increase respondent motivation and to minimize respondentfatigue. There is also potential to integratethe effects
of marketingmix and other (e.g., economic) variables
with the results obtained from conjoint analysis. We
also projectincreasingapplications industrial
in
goods,
such applicationsrequire acceptable procealthough
dures for developing group preference models. We
encouragecommercialusers to sharetheirexperiences
with the marketingcommunity so that the procedures
can be adapted furtherand the effectiveness and efficiency of marketingresearchcan be increased.
REFERENCES
Carmone, Frank, J., Paul E. Green and Arun K. Jain (1978),
"Robustness of Conjoint Analysis: Some Monte Carlo Results," Journal of Marketing Research, 15 (May), 300-3.
Cattin, Philippe (1981), "On the Estimation of Continuous
Utility Functions in Conjoint Analysis," working paper
No. 60-81, Center for Research and Management Development, University of Connecticut.
, Alan E. Gelfand and Jeffrey Danes (1981), "A
Simple Bayesian Procedure for Estimation in a Conjoint
Model," working paper No. 10-81, Center for Research
and Management Development, University of Connecticut.
Currim, Imran S. and Dick R. Wittink (1980), "Issues in the
Development of a Marketing Support System Using Segment-Based Consumer Preference Models," in Market
of
52 / Journal Marketing,
Summer
1982
Measurement and Analysis, David B. Montgomery and
Dick R. Wittink, eds. Cambridge, MA: Marketing Science
Institute, 386-96.
, Charles B. Weinberg and Dick R. Wittink (1981),
"Design of Subscription Programs for a Performing Arts
Series," Journal of Consumer Research, 8 (June), 67-75.
Green, Paul E. (1974), "On the Design of Choice Experiments Involving Multifactor Alternatives," Journal of
Consumer Research, 1 (September), 61-8.
, J. Douglas Carroll and Stephen M. Goldberg
(1981), "A General Approach to Product Design Optimization Via Conjoint Analysis," Journal of Marketing, 45
(Summer), 17-37.
and Wayne S. DeSarbo (1979), "Componential
Segmentation in the Analysis of Consumer Trade-Offs,"
Journal of Marketing, 43 (Fall), 83-91.
, Stephen M. Goldberg and Mila Montemayor(1981),
"A Hybrid Utility Estimation Model for Conjoint Analysis," Journal of Marketing, 45 (Winter), 33-41.
and James B. Wiley (1981), "A CrossValidation Test of Hybrid Conjoint Models," working paper, University of Pennsylvania.
and V. Srinivasan (1978), "Conjoint Analysis in
Consumer Research: Issues and Outlook," Journal of Consumer Research, 5 (September), 103-23.
Hauser, John R. and Patricia Simmie (1981), "Profit Maximizing Perceptual Positions," Management Science, 27
(January), 33-56.
Holbrook, Morris B. (1981), "Integrating Compositional and
Decompositional Analyses to Represent the Intervening
Role of Perceptions in Evaluative Judgments," Journal of
Marketing Research, 18 (February), 12-28.
and William L. Moore (1981), "Feature Interactions in Consumer Judgments of Verbal Versus Pictorial
Presentations," Journal of Consumer Research, 8 (June),
103-113.
Jain, Arun K., Franklin Acito, Naresh K. Malhotra and Vijay
Mahajan (1979), "A Comparison of the Internal Validity
of Alternative Parameter Estimation Methods in Decompositional Multiattribute Preference Models," Journal of
Marketing Research, 16 (August), 313-22.
Johnson, Richard M. (1980), "Measurement of Consumer
Values Using Computer Interactive Techniques," in Market Measurement and Analysis, David B. Montgomery and
Dick R. Wittink, eds., Cambridge, MA: Marketing Science
Institute, 271-7.
(1981), "Problems in Applying Conjoint Analysis," paper presented at the Conference on Analytical Approaches to Product and Marketing Planning, Vanderbilt
University, October.
Krishnamurthi,Lakshman (1980), "Modeling Joint Decision
Making Through Relative Influence," Ph.D. dissertation,
Stanford University.
MacBride, James N. and Richard M. Johnson (1980), "Respondent Reaction to Computer-Interactive Interviewing
Techniques," paper presented at ESOMAR Conference,
Monte Carlo, September.
Madansky, Albert (1980), "On Conjoint Analysis and Quantal
Choice Models," The Journal of Business, 53 (July),
537-44.
Olshavsky, Richard W. and Franklin Acito (1980), "An In-
formation Processing Probe into Conjoint Analysis," Decision Sciences, 11 (July), 451-70.
Pekelman, Dov and Subrata K. Sen (1979a), "Measurement
and Estimation of Conjoint Utility Functions," Journal of
Consumer Research, 5 (March), 263-71.
and
(1979b), "Improving Prediction in
Conjoint Measurement," Journal of Marketing Research,
16 (May), 211-20.
Plott, Charles R. (1980), "Comments on 'On Conjoint Analysis and Quantal Choice Models,'" The Journal of Business, 53 (July), 545-6.
Rao, Vithala R. and David A. Gautschi (1980), "The Role
of Price in Individual Utility Judgments: Development and
Empirical Validation of Alternative Models," working paper, Cornell University.
Shocker, Alan D. (1981), personal communication.
Srinivasan, V. (1980a), "Comments on 'On Conjoint Analysis
and Quantal Choice Models,'" The Journal of Business,
53 (July), 547-50.
(1980b), "Comments on the Role of Price in Individual Utility Judgments," Research Paper No. 569,
Stanford University.
(1981), "A Strict Paired Comparison Linear Programming Approach to Nonmetric Conjoint Analysis,"
Research Paper No. 620, Graduate School of Business,
Stanford University.
and Alan D. Shocker (1973), "Linear Programming
Techniques for Multi-dimensional Analysis of Preferences," Psychometrika, 38 (September), 337-69.
Wittink, Dick R. and David B. Montgomery (1979), "Predictive Validity of Trade-off Analysis for Alternative Segmentation Schemes," in 1979 AMA Educators' Conference
Proceedings, Neil Beckwith et al., eds., Chicago: American Marketing Association, 68-73.
and Philippe Cattin (1981), "Alternative Estimation
Methods for Conjoint Analysis: A Monte Carlo Study,"
Journal of Marketing Research, 18 (February), 101-6.
and Lakshman Krishnamurthi(1981), "Rank Order
Preferences and the Part-Worth Model: Implications for
Derived Attribute Importances and Choice Predictions," in
Proceedings Third ORSA/TIMS Special Interest Conference on Market Measurement and Analysis, John W. Keon,
ed., Providence, RI: The Institute of Management Sciences, 8-20.
and Julia B. Nutter (1982), "Comparing Derived Importance Weights Across Attributes," Journal of Consumer Research, 8 (March), 471-4.
Commercial of Conjoint
Use
A
Analysis: Survey 53
/
Disclaimer: Justia Dockets & Filings provides public litigation records from the federal appellate and district courts. These filings and docket sheets should not be considered findings of fact or liability, nor do they necessarily reflect the view of Justia.
Why Is My Information Online?