Conjoint Analysis

Conjoint Analysis

1. Introduction

Conjoint analysis, one of the widely applied marketing research and psychology research method, was used for understanding and predicting the consumer's preference, decision, and choice. The conjoint study forces the consumer to make more "real world" decision by looking at complete description of all attributes and related sub-attributes. The assumption of this research method is based on that complex decision is made by several factors considered jointly. The conjoint analysis helps to understand the consumer's preference on specific products.

The conjoint analysis which starts with the participant's overall judgements about a set of complex alternatives decomposes the participant's original evaluation into separate and compatible utility scales. In the conjoint analysis, the designed products or services are described by "profiles". Each "profile" is constructed by the arbitrary combination of one arbitrary selected level of each attribute. Through decomposing the overall judgements into components, the conjoint analysis reveals the consumer's preference and the "true" value that underlies the consumer's decision. In essence, the conjoint analysis provides more accurate information that enable the researcher to gain deep insight into the consumer's preference and decision making process. The primary advantage of this method is that the participant is forced to make tradeoffs between decisions as the real decision making process and the researcher can model human's decision making process in a more realistic way. Indeed, the decision maker's preferences for multiattribute options can be predicted.

2. Basic Issues Of Conjoint Analysis

The major issues in designing conjoint study include (1) selection attributes and levels, (2) Selection of a model of preference, (3) data collection method, (4) stimulus presentation, (5) measurement scale for the dependent variable, and (6) estimation method.

Selection attributes and levels

The Attribute is the key feature of a product or service, and the level is the specific variation or points within an attribute. In the conjoint analysis, the attribute should be able to cover major features of a product or service. In addition, an attribute should be measurable and able to be decomposed into several levels, which should successfully reflect the variation of each attribute during the process of decision making.

Preference models

Alternative models for preference include vector model, ideal-point model, and part-worth function model. The vector model assumes that the preference for the jth stimulus is given by

where the subscript p (p=1,2,...,t) denotes the set of t attributes or factors that have been chosen, is the individual's weight for the t attributes, and is the continuous variable such as travel time or price. In this vector model, the individual's weight function is different for each individual. When a negotiation is conducted within specific conditions, this model has limitation in the real application.

The ideal-point model assumes that the preference is negatively related to the squared distance of the location of the jth stimulus from the individual's ideal point . The is defined as

The preferred stimulation is the one that is close to the ideal- point. The major limitation of this ideal-point model is the difficulty in measuring the ideal point.

The part-worth function model, which provides the flexibility for each preference function along each of the attribute, can be defined as

where is the function denoting the part- worth of different levels of for the pth attribute. This preference model is widely accepted because of the flexibility and the ability to be compatible with any preference function of arbitrary shape. In general, the part-worth function model is used to measure the consumer's preference and attitude.

Data collection alternatives

.The hybrid conjoint model requires the researcher to collect three types of raw data from each respondent (Green 1984), however the traditional conjoint analysis requires only three types of data described as follows:

(1) Level desirability data - Likert scaling is usually applied to obtain the respondent's preference through ranking the levels within each attribute. The likert scaling allows each level of each attribute to be ranked independently. Also, the researcher can obtain similar data from ordinal scaling measurement.

(2) Attribute importance data - The respondent is asked to allocate 100 points over all attributes. Combining with level desirability, the respondent self-expresses his/her preference and utility level.

(3) Full-profile evaluation data - Each respondent is asked to provide an overall rating of a profile, which indicates the respondent's "intention to accept" towards a specific profile. Each profile contains one specific level of each attribute that is designed by the researcher.

Stimulus presentation

In most of experimental designs, presentation of stimulation usually combines two basic approaches - verbal description and pictorial representation.

The verbal description is widely used in the experimental design. The two-factor-at-a-time approach is primarily conducted by the verbal description. The pictorial representation, advantaged from the visual pros, is used to convey the information that is insufficient to use verbal description. The major advantage of this approach is to deliver more realistic information to the respondent. Some research reported that the combination of the verbal and pictorial description has similar effect as the verbal description, but the respondent can finish the evaluation more quickly yet with less fatigue.

Measurement scale for the dependent variable

Various alternatives for the measurement scale can be broadly categorized into metric and nonmetric scale. The metric scale includes rating scale and ratio scale. The nonmetric scale includes paired comparison and rank order. The major advantage of the rating scale is that it is easy to be conducted in any form of survey.

The rank order scale is especially useful in personal interview that enables the researcher to explain the difficult concept. The pair-comparison approach usually collects more data than needed. Compared to the other measurement scale, the pair-comparison approach is less efficient. The rank order scale is more reliable since it allows the respondent to compare his/her preference in the rank order.

Estimation methods

Estimation methods of the conjoint analysis can be roughly classified into the following three categories:

(1) Methods which assume the dependent variable is ordinally scaled. This category includes MONANOVA, PREFMAP, and LINMAP. The MONANOVA is restricted to the use in the part-worth function model only. The LINMAP is especially suited in the ideal-point model. Other models can be used either in the part-worth functional model or the vector model.

(2) Methods which assume the dependent variable is intervally scaled. This category includes ordinary least squares (OLS) regress, and minimizing sum of absolute errors (MSAE) regression. The primary advantage of the OLS methods is the providing standard error for the estimated parameters. The MSAE method that allows researchers to set priori constraints on the estimated parameters is usually considered more robust than OLS.

(3) Methods which relate pair-comparison data to a choice probability model. This category includes LOGIT and PROBIT. These probability choice model assumes that the pair-comparison is probability independent. If the dependent variable is collected directly from the pair-comparison, then these models are realistic.

3. Behavior Foundations of Conjoint AnalysisThe conceptual issues of the consumer behavior and decision making can be illustrated as the series relationship shown in Table 3.1.

Table 3.1: Conceptual framework for decision-making and choice behavior (Richard P. Bagozzi)

Table 3.1 implies the information used by decision makers to evaluate and compare options. During the comparing process, the decision maker learns not only about the options but also about the relative advantage of the option attributes. In order to compare these attributes, the decision maker develops evaluation rules that can analyze the relative importance of each attribute by decomposing an attribute into components. The evaluation rules are the utility functions of the decision maker. Based on the relative importance of the attribute, the decision maker's preferences for each option are formed. These preferences are the basis for decision making. Essentially, the purpose of the conjoint analysis is to study and model the conditions that the decision maker uses in evaluation .

The random utility view

Random utility theory, first proposed by Thurstone (1927), suggests that consumers try to choose those alternatives that they like the best under certain constraints such as income and time. (Richard P. Bagozzi). However, consumers do not always choose the product that they like best. The fluctuation over the behavior can be illustrated by the random component of the utility function.

where is the unobservable but true utility of alternative i, is the observable or systematic component of utility; and is the random component that can not be explained. Therefore, the probability for the consumer to choose product i is

where is the probability of choosing product or brand i from a set of competing product offerings C.

The systematic component of utility is predicted by the information or resource that influence choice behavior. In theory, the systematic component consists of several sub-components. For example, the preference for product i is determined by price, service, quality, and functionality. The relationship between the explainable variables and choice behavior can be illustrated as a linear function as follows:

where is a k 1 vector of utility coefficients associated with a vector of k explanatory variables, x'. All k vectors are independent with one another. Thus, equation (3.2) can be rewritten as

Equation (3.4) implies that the probability of choosing product i by a consumer is equal to the probability that the systematic component and its associated error for alternative i is higher than the systematic component and associated error components for all other competing offerings. In practice, the consumer evaluates attributes under some constraints such as time, money, and emotion. The differences in the individual consumer can be explained by two variables - and z. z is the vector of the measures of the individual differences. is 1 1 vector of coefficients associated with z. Therefore, the equation (3.4) can be generalized as:

where all terms are previously defined. Equation (3.5) represents different individual's preference for different attribute such as price, service, and quality.

The conjoint analysis is especially focused on the explainable component of utility. In general, the consumer reveals his/her utility by the judgements or choices that s/he makes in response to a preference. Therefore, utility maximization is one of the basic assumptions in the conjoint analysis.

4. Procedures of Conducting A Conjoint Analysis

The procedures of conducting a conjoint analysis requires several steps. Each step involves important decision. The guidelines for conjoint study is indicated as follows:

(1) Identify the research objective - Conjoint analysis is not globally applicable. By clearly identifying the research problems and objectives, the researcher can evaluate if the conjoint analysis is appropriate for his research. Typical conjoint research problem, which can be solved by identifying tradeoffs among the levels of each attribute, should be as focused as possible .

(2) Determine the sample size and characteristics - It is appropriate to limit the conjoint analysis to those populations who need to make tradeoffs decision such as product buying or designing. The appropriate sample size depends upon the variation of the populations' characteristic and the desired precision of the study. In general, the larger the number of attributes, the larger the sample size should be.

(3) Choose the data collection method - Some options of data collection include mail or telephone survey and personal or computer-aid interview.

(4) Identify the attribute and the level of each attribute - The attribute and the level of each attribute should be realistic and appropriate for a research problem. The level of each attribute should be able to reflect the variation that is critical in decision making. It is wise to keep the number of attributes in minimum size to minimize the estimation effort. The success of the conjoint analysis heavily depends upon the attribute selection.

(5) Configure attributes and levels into individual profiles - Using a factorial or fractional factorial design to generate the orthogonal array.

(6) Select an appropriate method for data collection - The appropriate method depends on the characteristic of the samples and how they can be reached. In addition, one must give attention of choosing the appropriate representing method for the levels such as text or pictorial description.

(7) Conduct the survey and collect the needed data for analysis.

(8) Analyze the data - The primary stage of data analysis is to develop the utility and relative importance of each attribute for each individual.

(9) Validate the findings - The data which is collected through experimental design is internal ly valid (Green).

(10) Report the results in terms of utility, importance rating and preference measurement.

5. Example I This example is taken from the report "Conjoint Analysis" (INTELLIQUEST). In a automobile purchasing case, it is assumed that the relevant attributes of an automobile are body type, manufacturing, fuel economy, and price. The attributes and levels of attributes are represented in Table 5.1.

Table 5.1: Attributes, levels of each attributes, and utility of each level of each attribute

In this example, the four major attributes are body type, manufacturer, fuel economy and price. The levels of each attribute are: three levels for attribute "body type", three levels for attribute "manufacturer", four levels for attribute "fuel economy", and three levels for attribute "price". In the conjoint study, respondents express their preferences for products that combine specific level of product attributes. Under this condition, the respondents need to make trade-offs. The typical example is to make a choice between two products that combine different levels of attributes. An example is given in Table 5.2.

Table 5.2: The comparison of two products

The data obtained from a series of such product comparison reveals the utility of each level of each attribute and the importance of each attribute in the decision process. The utility is a numerical expression of the relative importance of each attribute to the decision maker. The lower utility means less importance for an individual decision maker. From Table 5.1, we can conclude that the decision maker

1. prefers station wagon that van or sedan (utility level: 66 > 36 > 0)

2. does not differ much for different manufacturer

3. prefers fuel economy of 35 MPG than the lower fuel economy car

4. places a highest value for lowest price

Observing the different utility levels, we can figure out the relative importance of each attribute of the decision maker by calculating the range of utility. The range of utility can be calculated by the way that the maximum value of utility of each attribute subtracts the minimum value of utility of each attribute. This range represents the maximum impact of each attribute in the final decision making. In practice, the importance of each attributes is calculated in terms of its range as a percentage of the sum of the ranges across attributes. Using the above data, we can calculate the relative importance of each attribute as shown in Table 5.3.

Table 5.3. Attribute importance

The results of the above analysis can be expressed as the pie chart (Figure 5.1). The pie chart clearly displays how influential each attribute is in the decision maker's evaluation. As noted, the price and body type are the most important factors, and together they occupy more than two third of the total range of utility. Therefore, the consumer's utility level towards product A and B can be calculated as shown in Table 5.4.Table 5.4: Total utility of product A and product B

From Table 5.4, the consumer has higher probability to choose product A. By calculating the total utility level for each product, the researcher can find the interest level of each product and the new interest level of feature exchange. Through preference stimulation such as the change of product B's price from $12,000 to $11,000 and the manufacturer from Japanese to U.S., the total utility of each product A and B will change as shown in Table 5.5.

Table 5.5: Price sensitivity for product B

The consumer changes his/her preference from product A to product B (Figure 5.2).

In the conjoint analysis, the most important steps are determining the attributes and the levels of each attributes. The features of the attribute includes comprehensive, measurable, determinable, and meaningful. Comprehensive means that the attribute is adequate in indicating the degree to which the overall objective is met. Measurable means the ability to obtain a probability distribution for each alternative over the possible levels of the attribute. In addition, the decision maker's preference for different possible levels of the attribute is accessible. Determinable means that each attribute in the list should influence choice while none that are irrelevant to choice should be included. Meaningful means the attribute accurately delivered the meaningful information about the product. Possible attributes could be price, service, promotion, delivery, physical property..., etc. The attribute can be represented either by character or by graphic.

The levels of the attribute should be constructed under the criteria of meaningful, informative, and realistic. The levels should be captured in the way that the decision maker thinks. The levels should also reveal the real situation of making decision. For example, it is unrealistic to put the price range from $ 8,000 to $ 12,000 for BMW or Cadillac. Recent research found that the number of intermediate attribute levels have impacts on the relative importance of each attribute.

6. ConclusionThe conjoint analysis which successfully predicts the measurable features of the attributes has some potential usabilities and limitations. This technique is widely used in marketing research to predict the consumer behavior and decision making process. In essence, this analysis is applied to predict the change of product's features and characteristics. One of the typical example is to predict the consumer's response for quality improvement. The researcher can receive valuable information to further calculate the benefit and cost of the change. In addition, it is also able to provide valuable information for the new product and service design. Through the analysis, relative importance of each attribute is revealed and the product designer can take into account each attribute during the product design. Moreover, the conjoint analysis is applied to study the issue of price sensitivity, market segmentation, and some other psychological images.

Although the conjoint analysis has the potential usability, this technique also has some potential limitations. The technique of the conjoint analysis involves the utility function, attribute and the component which is decomposed for each attribute. In practice, some products or services may involve the utility functions that are not measurable or inadequate to be captured by the conjoint analysis. In the current situation, all the applied models are simple models. However, in the real life, most of the attributes are interacted with one another. How well can the simple models represent the complex interactions is still unknown. In addition, some attributes may not be able to be decomposed into components. Typical examples are movie, weather, emotion and personality.

In order to successfully apply the conjoint analysis, some guidelines are provided for the experimental design. First of all, the researcher needs to identify the research objectives in order to collected valuable information in an experiment. Secondly, the researcher has to select the attribute with care. The selected attributes should be able to reflect the major factors that influence the decision making. In the conjoint analysis, the attribute's characteristic is more important than the quantity of the attribute. In other words, during the attribute design process, the research team also need to face the trade-off process to pick up the most important ones.

Last Updated on November 22, 1996 by Ying-Hueih Chen