Conjoint analysis attempts to determine the relative importance, consumers attach to salient attributes and the utilities they attach to the levels of attributes. This information is derived from consumers’ evaluations of brands, or brand profiles composed of these attributes and their levels.
The respondents are presented with stimuli that consist of combinations of attribute levels. They are asked to evaluate these stimuli in terms of their desirability. Conjoint procedures attempt to assign values to the levels of each attribute, so that the resulting values or utilities attached to the stimuli match, as closely as possible, the input evaluations provided by the respondents. The underlying assumption is that any set of stimuli, such as products, brands, or stores, is evaluated as a bundle of attributes.20
Conjoint analysis has been used in marketing for a variety of purposes, including:
Determining the relative importance of attributes in the consumer choice process. A standard output from conjoint analysis consists of derived relative importance weights for all the attributes used to construct the stimuli used in the evaluation task. The relative importance weights indicate which attributes are important in influencing consumer choice.
• Estimating market share of brands that differ in attribute levels. The utilities derived from conjoint analysis can be used as input into a choice simulator to determine the share of choices, and hence the market share, of different brands.
• Determining the composition of the most preferred brand. The brand features can be varied in terms of attribute levels and the corresponding utilities determined. The brand features that yield the highest 0 utility indicate the composition of the most preferred brand.
• Segmenting the market based on similarity of preferences for attribute levels. The part worth functions derived for the attributes may be used as a basis for clustering respondents to arrive at homogeneous preference segments
Applications of conjoint analysis have been made in consumer goods, goods, financial, and other services. Moreover, these applications have spanned all areas of marketing.
A survey of conjoint analysis reported applications in the areas of new product/concept identification, competitive analysis, pricing, market segmentation, advertising, and distribution
Statistics and Sierras Associated with Conjoin! Analysis
The important statistics and terms associated with analysis include
Part-worth functions. The part-worth functions or utility functional describe the utility consumers attach to.the levers bf each attribute,
Attribute levels. The attribute levels denote the values assumed by the attributes. Full profiles. Full profiles or complete profiles of brands are constructed in terms of all the attributes by using the attribute levels specified by the design.
Fractional factorial designs. Fractional factorial designs are designs employed to reduce the number of stimulus profiles to be evaluated in the full profile approach
Orthogonal arrays. Orthogonal arrays are a special class of fractional designs that e~able the efficient estimation of all main effects.’
internal validity. This involves correlations of the predicted evaluations for the holdout or validation stimuli with those obtained from the respondents
Conducting Conjoint Analysis
Figure 21.5 lists the steps in ‘con Joint analysis’. Formulating the ‘problem involves identifying the sentimentalizes and their levels. These attributes-and levels are used for constructing the stimuli to be used in a conjoint evaluation task. The respondents rate or rank the stimuli using a suitable scale and the data obtained are analyzed. The results are interpreted and their reliability and validity assessed.
Formulate the Problem
In formulating the conjoint analysis problem, the researcher must Identify he attributes and attribute levels to be used in constructing the stimuli. Attribute levels denote the values assumed by the attributes. From a theoretical standpoint, the attributes selected should be salient in in consumer preference and choice. For example, in the choice of an automobile brand, price, gas mileage, interior space: and so forth should be included. From a managerial perspective, the attributes and their levels should be actionable. To tell a manager that consumers prefer
a sporty car to one that ,is conservative looking is not helpful, unless sportiness and conservative ness are defined in terms of attributes over’ which ‘a manager has control. The attributes cat), be identified through discussions.with management and industry experts, anal~,sis of secondary data, qualitative research, and pilot surveys, A typical conjoint, analysis study involves six or seven attributes
We illustrate the conjoint methodology by considering the problem of how students evaluate sneakers, Qualitative research identified three attributes as, salient sole, the upper, and the price,23 Each-was defied in :terms of three levels, as shown in These attributes and their levels were used constructing the analysis stimuli. Note that to keep the illustration simple, we are using limited number of attributes, that is, only three, It has been argued that pictorial stimuli should be used when consumers’ marketplace choices are strongly
_guided by the product’s styling, such that the choices are heavily based on an inspection of actual products or pictures of products 5.24
Construct the Stimuli
Two broad approaches are available ‘for Constructing conjoint analysis stimuli: the pairwise approach and the full-profile procedure in the pairwise approach, also called’two-factor evaluations, the respondents evaluate two attributes at a time unfit all the possible pairs of attributes have been evaluated> This approach fs illustrated in the context of the sneakier example in Figure 21,16, For each pair, respondents evaluate all the combinations of levels of both the attributes, which are presented in a matrix, In the full-profile approach, also called multiple-factor evaluations, full or
The sneaker example follows the full-profile approach. Given three attributes, defined at three levels each, a total of 3 X 3 X 3 = 27 profiles can be constructed. To reduce the respondent evaluation task, a fractional factorial design was employed and a set of nine profiles was constructed to constitute the estimation stimuli set (see Table 21.6). Another set of nine stimuli was constructed for
validation purposes. Input data were obtained for both the estimation and validation stimuli. However, before the data could be obtained, it was necessary to decide on the form of the input data.
Decide on the Form of Input Data
As in the case of MDS, conjoint analysis input data can be either nonmetric or metric. For nonmetric data, the respondents are typically required to provide rank order evaluations. For the pairwise approach, respondents rank all the cells of each matrix in terms of their desirability. For the full-profile approach, they rank all the stimulus profiles. Rankings involve relative evaluations of the attribute levels. Proponents of ranking data believe that such data accurately reflect the behavior of consumers in the marketplace.
In evaluating sneaker profiles, respondents were required to provide preference ratings for the sneakers described by the nine profiles in the estimation set. These ratings were obtained using a 9-point Likert scale (I = not preferred, 9 = greatly preferred). Ratings obtained from one respondent are shown in Table 21.11
Select a Conjoint Analysis Procedure
The basic conjoint analysis model may be represented by the following formula
Several different procedures are available for estimating the basic model. The simplest, and one that is gaining in popularity, is dummy variable regression (see Chapter 17). In this _ case, the predictor variables consist of dummy variables for the attribute levels. If an attribute has k, levels, it is coded in terms of k, – I dummy variables (see Chapter 14). If metric data are obtained, the ratings, assumed to be interval scaled, form the dependent variable. If the data are nonrnetric, the rankings may be converted to 0 or I by making paired comparisons between brands. In this case, the predictor variables represent the differences in the attribute levels of the brands being compared. Other procedures that are appropriate for non metric data include LINMAP, MONANOVA, and the LOGIT model (see)
The researcher must also decide whether the data will be analyzed at the individual-responlIent or the aggregate level. At the individual level, the data of each respondent are analyzed separately. If an aggregate-level analysis is to be conducted, some procedure for grouping the respondents must be devised. One ‘common approach is first to estimate individual-level part worth or utility functions. The respondents are then clustered on the basis of the similarity of their part-worths. Aggregate analysis is then conducted for each cluster. An appropriate model for estimating the parameters should be specified
The data reported in Table 21.11 were analyzed using ordinary least-squares (OLS) regression with dummy vf!riables. The dependent variable was the preference ratings. The independent variables or predictors were six dummy variables, two for each variable. The transformed data are shown in Table 21.12. Because the data pertain to a single respondent, an individual-level
analysis was conducted. The part-worth or utility functions estimated for each attribute, as well the relative importance of the attributes, are given in Table 21.13.29
The model estimated may be represented as
For Sole. the attribute levels were coded as follows:
The levels of the other attributes were coded similarly. The parameters were estimated as follows
To solve for the part-worths, an additional constraint is necessary. The part-worths are estimated on an interval scale. so the origin is arbitrary. Therefore. the additional constraint that is imposed is of the form:
These equations for the first attribute, Sole, are
Solving these equations, we get
The relative importance weights were calculated based on ranges of part-worths, as follows:
The estimation of the part worths and the relative importance weights provides the basis for interpreting the results.
For interpreting the results, it is helpful to plot the part-worth functions. The part-worth function values for each attribute given in Table 21.13 are graphed in Figure 21.17. As can be seen from Table 21.13 and Figure 21.17, this respondent has the greatest preference for a rubber sole when evaluating sneakers. Second preference is for a plastic sole, and a polyurethane sole is least preferred. A leather upper is most preferred, followed by canvas and nylon. As expected. a price of $30.00 has the highest utility and a price of $90.00 the lowest. The utility values reported in
Table 21.13 have only interval-scale properties, and their origin is arbitrary. In terms of relative importance of the attributes, we see that Price is number one. Second most important is Sole, followed closely by Upper. Because price is by far the most important attribute for this respondent, this person could be labeled as price sensitive.
Assessing Reliability and Validity
Several procedures are available for assessing the reliability and validity of conjoint analysis results.
1. The goodness of fit of the estimated model should be evaluated. For example, if dummy variable regression is used, the value of R2 will indicate the extent to which the model fits the data. Models with poor fit are suspect.
2. Test-retest reliability can be assessed by obtaining a few replicated judgments later in data collection. In other words. at a later stage in the interview, the respondents are asked to
evaluate certain selected stimuli again. The two values of these stimuli are then correlated to assess test-retest reliability.
3. The evaluations for the holdout or validation stimuli can be predicted by the estimated pan-worth functions. The predicted evaluations can then be correlated with those obtained from the respondents to determine internal validity.
4. If an aggregate-level analysis has been conducted, the estimation sample can be split in several ways and conjoint analysis conducted on each subsample. The results can be compared across subsamples to assess the stability of conjoint analysis solutions.
In running a regression analysis on the data of Table 21.12, an R2 of 0.934 was obtained, indicating a good fit. The preference ratings for the nine validation profiles were predicted from the utilities reported in Table 21.13. These were correlated with the input ratings for these profiles obtained from the respondent. The correlation coefficient was 0.95, indicating good predictive ability. This correlation coefficient is significant at a = 0.05.