Conceptual and Operational Issues in the Extended Fishbein Model

Richard J. Lutz, University of California, Los Angeles
ABSTRACT - Fishbein's (1967a) model for the prediction and explanation of specific behaviors has attracted considerable attention among consumer researchers. The purpose of the present paper is to summarize some conceptual and operational issues surrounding the constructs in the model. In particular, attitude-toward-the-behavior and the relatively recent subjective norm construct are examined, and possible reconceptualizations are offered. Measurement problems with respect to the cognitive components in the model are discussed, and suggestions are made for the improved testing of model assumptions. Model testing and refinement are advocated, and appropriate research strategies, allowing causal analyses, are outlined.
[ to cite ]:
Richard J. Lutz (1976) ,"Conceptual and Operational Issues in the Extended Fishbein Model", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 469-476.

Advances in Consumer Research Volume 3, 1976      Pages 469-476

CONCEPTUAL AND OPERATIONAL ISSUES IN THE EXTENDED FISHBEIN MODEL

Richard J. Lutz, University of California, Los Angeles

ABSTRACT -

Fishbein's (1967a) model for the prediction and explanation of specific behaviors has attracted considerable attention among consumer researchers. The purpose of the present paper is to summarize some conceptual and operational issues surrounding the constructs in the model. In particular, attitude-toward-the-behavior and the relatively recent subjective norm construct are examined, and possible reconceptualizations are offered. Measurement problems with respect to the cognitive components in the model are discussed, and suggestions are made for the improved testing of model assumptions. Model testing and refinement are advocated, and appropriate research strategies, allowing causal analyses, are outlined.

Fishbein's (1967a, 1975) modification of the theory of propositional control (Dulany, 1968) has emerged over the past few years as one of the more popular theories in consumer research. This popularity is at least in part due to the fact that the theory is a promising one for consumer researchers interested primarily in the explanation of consumer behavior and, at the same time, provides a framework for devising behavior change strategies, a feature of interest to managers and policy makers.

In their recent review of research generated by this theory in the consumer domain, Ryan and Bonfield (1975) have provided an excellent summary of empirical findings, as well as raising some important research issues. The purpose of the present investigation is to discuss some additional conceptual and operational issues with respect to the Fishbein model. To accomplish this task, the most recent form of the model will be presented briefly, followed by detailed consideration of issues surrounding the various theoretical constructs in the model.

THE EXTENDED FISHBEIN MODEL

The Extended Fishbein Model (so designated because it is viewed as an extension of Fishbein's (1963) earlier model of attitude) incorporates attitudinal and social influences in attempting to explain the formation of behavioral intention, which is seen as the immediate predecessor of overt behavior. While nomenclature and operational specification of the constructs in the model have undergone several changes since the original formulation in 1967, Fishbein and Ajzen's (1975) recent book provides an update which will serve as the basis for the present discussion. Essentially, the model rests on three equations:

EQUATIONS  (1), (2), (3)

where B is the behavior under study; I is the individual's intention to perform the behavior; AB is his attitude toward performing the behavior; SN is the subjective norm with respect to the behavior; w1 and w2 are empirically derived weights; bi is the belief that performing the behavior leads to some consequence i; ei is the individual's evaluation of consequence i; n is the number of salient beliefs; bj is the normative belief regarding referent j's expectations as to whether the individual should or should not perform the behavior; mj is the individual's motivation to comply with referent j; and m is the number of salient referents (Fishbein and Ajzen, 1975, pp. 301-2).

Equation i is a theoretical statement of the determinants of intention, which is seen as more or less equivalent to behavior, depending upon several factors (e.g., the timing and specificity of measurement); Equation 2 specifies cognitive determinants of attitude. Equations 1 and 2 are familiar to consumer researchers; however, Equation 3, which deals with the determinants of the subjective norm, is a more recent addition to the model. Originally there was no SN construct in the model, and the Summative compound shown on the right side of Equation 3 was used directly in Equation i to represent normative influences. Thus SN has the same logical status in the model as does AB on the attitudinal side--both are summary measures of potential sources of influence (i.e., attitudinal and normative) on the formation of intentions.

Attention is now focused on issues surrounding each of the equations in the model, beginning with the attitudinal component (Eqn.2), followed by the normative component (Eqn. 3) and the intention equation. In each instance, issues have been separated into conceptual and operational categories to facilitate discussion.

THE ATTITUDINAL COMPONENT

Two major conceptual issues arise with respect to the attitudinal component, and both are concerned with the nature of the cognitive determination of AB. First, the status of the bi element as a probability dimension is considered; and second, the question of cognitive adding vs. cognitive averaging is discussed.

Status of the belief element. In developing his original A (attitude-toward-the-object) model, Fishbein (1967~) relied on the process of mediated generalization in behavioristic learning theory. Specifically, he proposed that the evaluative responses associated with certain salient characteristics of an object are transferred, or generalized, to the object itself, thus forming the individual's attitude toward the object. Further, the degree to which the evaluative response to any attribute was transferred to the object was assumed to be a function of the probability that the object possessed the attribute. Hence, a multiplicative summation identical to Equation 2 was the theoretical basis for Ao. The attitude-toward-the-behavior (AB) construct has never undergone independent theoretical development; rather, Fishbein (1967) posited that the same processes underlying the formation of A should also be the foundation of AB. However, the types of beliefs (bi) which tend to be salient for AB are no longer the descriptive beliefs salient for Ao, but rather are beliefs about the outcomes or consequences of engaging in the behavior. Thus, the "attributes" associated with the behavior are a set of outcomes, or valued states, which the individual wishes to either avoid or attain. The result of this rather subtle shift from descriptive beliefs to beliefs about behavioral consequences is that the attitude in question (AB) is one which has much more functional significance for the individual. That is, an individual's attitude toward engaging in a behavior will be favorable if the behavior leads to desirable outcomes or blocks undesirable outcomes for the individual. The key question then becomes whether the individual's beliefs about behavioral consequences take the form of subjective probability statements. There is some theoretical disagreement on this point.

Working within the functionalist tradition, Rosenberg (1956) developed his version of the "expectancy-value" model based on the assumption that the expectancy or belief component represented the degree to which the attitude object led to the blocking or attainment of certain values. Thus, rather than being concerned with the probability of association between an object and an outcome, Rosenberg dealt with the degree of association or dissociation between object and outcome.

When one analyzes the "attitude objects" studied by Rosenberg and the other functionalists, an interesting trend emerges: virtually every "object" studied was actually a behavior (see Table 1). While the behaviors under investigation were not generally those that the individual himself would have performed (e.g., most of the subjects in those studies would have had little power to remove Negro segregation), nevertheless those actions would have had the effect of blocking or facilitating the attainment of certain goals by the individual. Hence, the bipolar construct of "blocking/attainment'' or "association/dissociation" seems to be a reasonable one for representing beliefs about behavioral consequences.

TABLE 1

REPRESENTATIVE "ATTITUDE OBJECTS" USED IN RESEARCH BY FUNCTIONAL THEORISTS

Fishbein and Ajzen (1975) have pointed out that the association/dissociation construct can be represented as two probabilistic belief dimensions. This is certainly true logically and testable empirically. But which approach comes closer to representing the form of beliefs individuals actually hold in cognitive structure? Intuitively, it seems more plausible that association/dissociation would be operative in beliefs with respect to behaviors. Recent empirical evidence tends to support that notion (Bettman, Capon and Lutz, 1975a, c).

As conceptualized by Fishbein, the bi construct is a subjective probability dimension; it is typically measured on a 7-point scale ranging from "unlikely" to "likely". Fishbein and Ajzen (1975) advocate scale values for this measure of -3 to +3, thus treating it as a bipolar construct, which is hardly a probability statement. Interestingly, Bettman, et al. (1975a) found that subjects tend to respond to the b. measure in a bipolar fashion, thus providing justification for the use of the bipolar coding scheme in data analysis. Whether this response bipolarity is a function of respondents' "true" cognitive structures or is an artifact of the measurement procedure is uncertain. It is clear, however, that in its present form the bi construct does not behave as a probability dimension.

Ahtola (1975) has suggested a model for partitioning the bi dimension into belief content and belief strength, thereby taking into account both the blocking/attaining feature and the probability aspect of bi. This model, while having theoretical elegance, is quite cumbersome operationally. Further, it seems unlikely that individuals' cognitive structures are as differentiated as the Ahtola model implies. Instead, it seems more plausible that for the vast majority of beliefs, particularly those which are characterized by monotonic affect functions (Ahtola, 1975), belief content is the salient construct in cognitive structure. That is, the individual holds a belief regarding the degree to which the behavior in question leads to or blocks a particular goal. It is this belief, a bipolar construct, which should comprise the bi component in the Extended Fishbein Model. The probability dimension does not adequately capture the idea of association/dissociation, and consideration of both probability and association/dissociation overstates the information processing capabilities of most individuals.

The above discussion argues against none of the empirical research on the Fishbein model, all of which has treated bi as a bipolar construct, though in theory it is a (unipolar) probability. The purpose of the discussion is simply to point out that the model has worked well, as operationalized, which happens to conflict with the model as conceptualized. What is argued, therefore, is that the weight of empirical evidence suggests a change in the conceptualization of the bi component to bring it more into line with empirical evidence. Once the bi dimension is accepted as being conceptually bipolar, then appropriate measurement procedures can be adopted to more accurately reflect that dimension. Bettman, et al. (1975c) were able to increase the strength of the relationship between AB and Sbiei by substituting a "high-low" scale for the commonly used "likely-unlikely" scale. The assumption was that subjects would be better able to respond to a scale which more closely approximated their own belief structures. Thus, "high-low" was seen as being more inherently bipolar than "likely-unlikely" and if subjects' beliefs tended to be bipolar, they should have been able to respond with less difficulty to the former scale. While the results were not conclusive, there was a good indication that the "high-low" scale performed better. Obviously, there may be other scale endpoints which would be superior to "high-low". Nevertheless, there is ample justification for questioning the status of the bi construct as a probability dimension.

Further research should be directed at uncovering the nature of the bi component. One approach would be the analysis of free response data generated by respondents to determine how they seemingly organize belief statements. The use of multitrait-multimethod procedures would also appear to be a valuable tool in this task.

Adding vs. averaging. Fishbein (1965), in discussing his Ao model, states: "...the assumption of 'summation' is not explicit in behavior theory. Thus, for example, a postulation of 'averaging' would also have been consistent with behavior theory" (p. 117). Much research has been devoted in the psychological literature to determining whether beliefs add or average to form overall attitude. Probably the major proponent of an averaging formulation is Anderson (1971), who has demonstrated averaging effects within his integration theory approach to cognitive processes. Fishbein and Ajzen (1975), however, dismiss most of the research on the adding/averaging issue on methodological grounds and prefer to stand by the summation assumption in Equation 2.

Bettman, et al. (1975b) found evidence that most of their subjects tended to average rather than add across beliefs, within the context of Fishbein's AB model. Thus, a serious question is raised regarding the appropriateness of the original summation assumption. The issue is unimportant when static tests of the model are undertaken, as the average is simply a linear transformation of the sum and is therefore perfectly correlated with it. However, in attitude change situations, the distinction is a crucial one, for the adding and averaging formulations often make opposite predictions. For instance, under a summation model, a piece of moderately favorable information added to a highly favorable initial attitude would be expected to increase the favor-ability of the attitude. However, an averaging assumption would predict a decrease in the favorability of the attitude. Obviously, this is an important concern for policy makers using the model for generating attitude change strategies.

Future research should use an attitude change paradigm to determine whether averaging or adding is the dominant cognitive process underlying AB. Undoubtedly, individual differences and situational influences may have impact on this process, and to the extent that they do, there can be no global generalization. In light of recent evidence, it seems inappropriate to continue with the assumption that Equation 2 is correct as it stands.

Operational Issue. The major operational issue confronting the Extended Fishbein Model, as well as all other expectancy-value formulations, is the problem with the scale properties of the expectancy (bi) and value (ei) terms. Schmidt (1973, 1975) has demonstrated that in the absence of ratio measure for both bi and el, the correlation between AB and Sbiei can be artificially manipulated to virtually any value by simple transformations of bi and ei. These transformations would not be possible on ratio data, but are perfectly legitimate on interval data. Therefore, from a mathematical perspective, Sbiei is a meaningless computation unless ratio data are obtained.

The problems inherent to obtaining ratio scale data are well known; it is safe to say that none of the hundreds of studies conducted on the various expectancy-value models in the literature have used such data. This is a serious concern, for it means that standard correlation approaches to model validation are essentially worthless in the absence of other kinds of data.

Table 2 shows the correlations calculated between AB and Sbiei for different combinations of coding assumptions. Data for these correlations came from 491 housewives responding to a questionnaire about a hypothetical laundry detergent. [For a more detailed description of these data, the reader is referred to Lutz (1975a).]

TABLE 2

PRODUCT-MOMENT CORRELATIONS BETWEEN AB AND Sbiei WITH VARYING CODING ASSUMPTIONS

The upper lefthand entry in the matrix is the correlation under the typical coding assumptions of -3/+3 for both the bi and ei components. The underlined entry in Table 2 is the maximum correlation attained under any of the coding schemes employed. This corresponds to a coding scheme wherein subjects' responses were coded from -2 to +4 for both bi and ei. The fact that the correlation is slightly higher under the latter approach suggests that subjects were displacing the neutral point on both scales toward the negative endpoint.

An identical conclusion was reached by Bettman, et al. (1975a) using an ANOVA framework for investigating the model; In their data, an expected symmetric interaction pattern was shifted toward the negative ends of the scales, indicating that a -3/+3 coding scheme tends to underestimate the favorability of the biei contribution to overall attitude.

In another context, Bentler (1969) has shown semantic space to be approximately bipolar. Since the good-bad and likely-unlikely scales used by Fishbein to measure bi and ei are derived from the work of Osgood, et al. on the semantic differential technique, it seems plausible that the bi and ei scales should also be approximately bipolar. This is in agreement with the empirical results discussed above.

Thus, while there is no direct evidence pertaining to the scale properties of the bi and ei measures, it appears that they are at least fairly adequately represented by the -3/+3 coding scheme advocated by Fishbein and Ajzen (1975). Nevertheless, there is an indication that a slightly asymmetric coding procedure would be more appropriate.

The only strictly correct operational procedure for testing the relationship between AB and Sbiei, as currently conceptualized by Fishbein and Ajzen (1975), would be to obtain ratio measures of bi and ei. For the probability component, this would be a relatively easy task, as subjects could be asked to provide subjective provide subjective probability estimates on a "chances out of 100" scale. (Incidentally, data of this type would be necessary to refute the arguments against the use of subjective probabilities which were presented in the preceding section of this paper.)

Measuring ei on a ratio scale is a more imposing task. As yet, no completely satisfactory procedure exists. Thurstone and Jones (1957) have presented a model for determining the "rational origin" for subjective values, but the method is quite cumbersome and offers no guarantee that an origin can be located in every instance.

Adequate operationalization of the bi and ei constructs in Equation 2 as subjective values with a natural origin is the only procedure by which the Fishbein model can be tested correlationally. ANOVA procedures allow tests without making scale assumptions regarding the bi and ei components, but are fraught with problems of external validity (Bettman, Capon and Lutz, 1975d). It appears that further tests of the validity of Equation 2 in the model must rely on multimethod approaches wherein correlational and ANOVA procedures are combined, wherever possible within attitude change contexts.

THE NORMATIVE COMPONENT

Conceptual Issues. The normative component of the Extended Fishbein Model has always been its weak link. Only recently has the subjective norm (SN) construct been formulated, and empirical evidence in support of it is sparse. Similarly, the theoretical antecedents of SN are poorly developed conceptually. Therefore, both sides of Equation 3 raise major conceptual issues: both the criterion variable SN and its antecedents, bj and mj, need fuller development.

Subjective norm. The normative component has suffered from the lack of an overall criterion measure (similar to AB for the attitudinal component) since the genesis of the model. Absence of this construct has prevented serious validation work on the determinants of normative pressure, i.e., normative beliefs (bj) and motivation to comply (mj).

In their recent book, Fishbein and Ajzen (1975) report the use of the following scale to measure SN:

Most people who are important to me think

I should ___:___:___:___:___:___:___ I should not

perform behavior X.

Correlations ranging from .625 to .910 with Sbjmj were found; yet, the above statement does not seem to capture the essence of motivation to comply. Rather, the above scale appears to measure a generalized normative belief, i.e., what others think the person should do. If the phrase, "who are important to me," is meant to somehow pick up mj, then it seems inadequate. There is little chance that "negative" referents, for whom mj would be small or even in the opposite direction, would be evoked by this phrase.

Ryan (1975) has recently proposed a social compliance (SC) construct similar in purpose to SN and has used Kelman's (1958) notions as a conceptual basis. Specifically, a scale was constructed to measure only the compliance process in Kelman's typology, the assumption being that the processes of identification and internalization should be captured by the attitudinal component. Initial empirical work showed a statistically significant correlation SC and Sbjmj; however, SC also correlated highly with Sbiei and AB, suggesting the need for further work on this construct.

The essential conceptual content which must be captured by the SN construct is the general social pressure on the person to perform the behavior. Thus, both normative beliefs and the motivation to comply with these beliefs should be represented in this single unidimensional construct. One way to conceptualize SN is to treat it in the same manner as AB. AB is an overall affective dimension, where the intensity of affect is seen as deriving from a number of affective (ei) dimensions (Fishbein, 1963). The degree to which each of the ei dimensions contributes to AB is a function of belief strength (bi).

To treat SN analogously, the SN construct would be viewed as a global motivation to comply with other people in deciding whether or not to perform the behavior in question. The source of this motivation to comply would be the individual mj elements which represent motivation to comply with certain referents, in general. To the extent that these referents are perceived as having beliefs about what the individual should do with respect to the behavior in question, then the various mj elements will combine to form a general motivation to comply with others in that situation. Under this approach, SN might be measured on a scale such as:

With respect to performing behavior X,

I very much want to do ___:____:___:___:___:___:___ I very much want not to do

what other people think I should do.

To the extent that intention (I) is under normative control in a given situation, then SN should be highly correlated with I. Note that under this approach it would not be necessary to represent directly in SN the direction of normative beliefs; all that is important is the degree to which the individual is motivated to comply with these beliefs. The regression weight (w2) attached to the normative component would reflect the direction of normative beliefs, in general. This is exactly the same approach as is used in measuring AB: affect is the only important feature, not the beliefs from which that affect derives. In order to use the normative component for diagnosis, individual bj elements would have to be analyzed, just as is presently done with bi elements underlying AB-

Thus the determinants of intention are viewed as two response tendencies of the individual. One response tendency is attitudinal and reflects the individual's own personal desire to engage in the behavior. The other response tendency is normative and summarizes the person's motivation to go along with others in deciding whether or not to perform the behavior. These two response tendencies may be competing or complementary and should in general behave in accordance with the propositions outlined by Fishbein (1967a, 1975).

It is not recommended that a single scale such as the one shown above be used to measure SN. Rather, a series of items, in either semantic differential or Likert format, should be employed to achieve a reasonable degree of reliability. Again, Ryan's (1975) measure of SC is a useful first step in this direction.

Determinants of subjective norm. Perhaps due to the absence of the SN construct until recently, there has been considerable ambiguity surrounding the exact nature of the normative belief (bj) and motivation to comply (mj) elements underlying the normative component. In its most recent form (Fishbein and Ajzen, 1975), bj is treated in a fashion similar to bi for the attitudinal component. Thus a normative belief (bi) is seen as a probabilistic statement that a particular referent thinks the individual should or should not perform the behavior in question. Motivation to comply (mj) with the normative belief is defined as the general tendency of the individual to go along with the referent's expectations, although there is some sentiment for adopting a more situation specific conceptualization of mj (Fishbein and Ajzen, 1975, p. 306).

A useful approach to more clearly delineating the conceptual properties of bj and mj might be to consider, separate from AB, the social consequences of performing the behavior in question. Thus the individual may be asked to respond to bj statements like:

If I purchase Brand X, my family will hate me

likely_____:_____:_____:_____:_____:_____:_____:_____:unlikely

If I purchase Brand X, my family will cut off my allowance

likely_____:_____:_____:_____:_____:_____:_____:_____:unlikely

[While it will not be discussed again here, arguments against the use of probability dimensions in conceptualizing the bi component of AB apply equally to the use of probability to represent bj. The likely-unlikely scales are used here to show the similarity between the construct being proposed here and the bi component as currently operationalized.]

Note that this procedure would permit a multidimensional representation of any particular referent's relationship to the individual, thus allowing possible incorporation of Kelman's (1958) typology of social influence and French and Raven's (1959) bases of social power.

Corresponding to the above bj dimensions would be mj dimensions more closely akin to the ei elements underlying AB.$ Thus,

For my family to hate me is

good_____:_____:_____:_____:_____:_____:_____:_____:bad

For my family to cut off my allowance is

good_____:_____:_____:_____:_____:_____:_____:_____:bad

represent evaluations of the social consequences of engaging in the behavior. The impact of these changes in bj and mj on the SN construct would be to convert it in-to an attitude toward engaging in the behavior due to social demands (rather than personal evaluations). This may be thought of as similar to Rokeach's (1968) notion of attitude toward the situation. Thus, intention would be determined by the interaction of two attitudes--AB, which is based on personal beliefs about the consequences of engaging in the behavior, and SAB (social attitude toward the behavior), which is an attitude toward engaging in the behavior based exclusively on social outcomes.

Whether these two attitudes can be operationalized such that they are independent of each other is an empirical question. The important point of the above discussion is that the normative component can be conceptualized using the same basic elements (i.e., beliefs and evaluations) as are used in deriving AB. This approach is at least worthy of some consideration, given the relative success of the attitudinal component in past research and the general lack of success with respect to the normative component.

Operational Issue. The chief operational issue with respect to the normative component of the Extended Fishbein Model is similar to the problem encountered in scaling bi and ei elements. Both bj and mj should be ratio scales to allow multiplication of the terms as specified in Equation 3. In practice, both elements underlying the normative component have typically been measured on 7 point scales, coded -3/+3. Schmidt's (1973, 1975) analysis applies to this practice, as well as the coding procedures for the attitudinal component which were discussed earlier. Correlations can be manipulated drastically by simply changing the coding assumptions. Table 3 shows a wide range in correlation between Sbjmj and intention, based on the same data set as the attitudinal correlations reported earlier in this paper. Intention was used as the dependent measure, as this research was conducted prior to the formulation of the SN construct.

TABLE 3

PRODUCT-MOMENT CORRELATIONS BETWEEN INTENTION AND Sbjmj WITH VARYING CODING ASSUMPTIONS

In this case, bj was measured on scales like the following:

My family would_____:_____:_____:_____:_____:_____:_____:_____:would not

expect me to try Brand M detergent.

Motivation to comply (mj) was a situation specific measure, as follows:

With respect to my choice of laundry detergent

I want to_____:_____:_____:_____:_____:_____:_____:_____:do not want to

do what my family expects me to do.

As can be seen in Table 3, the assumption of -3/+3 coding for both elements is incorrect (if correlation with intention can be accepted as a criterion). While bj appeared to operate in a bipolar manner, mj was clearly treated as a unipolar construct by the respondents. Hence, a 1 to 7 coding scheme for mj increases the correlation between Sbjmj and I to its maximum (lower left-hand entry in Table 3), while the standard coding procedure results in the minimum correlation between the two variables (upper left-hand entry in the table).

Obviously, much more work must be devoted to obtaining satisfactory measures of bj and mj. The likelihood of obtaining ratio scales seems quite small; however, the use of ANOVA procedures similar to those used for the attitudinal component (e.g., Bettman, et al., 1975a) should provide useful insights.

 

DETERMINANTS OF INTENTION

Conceptual Issues. Conceptual issues with respect to the overall intention equation (Eqn. 1) concern 1) the status of intention (I) as a variable mediating the effects of all other cognitive constructs on behavior (B), and 2) the psychological meaning of the weights (w1 and w2).

Status of intention. Intention is a familiar construct to consumer researchers. It has been used at an aggregate level since the early 1950's when Katona and his associates at the University of Michigan's Survey Research Center began incorporating an intention variable in their economic forecasts. Recent consumer theories (e.g., Howard and Sheth, 1969) have postulated an intention variable as the immediate precursor of purchase behavior and have treated it as an explanatory construct for individual purchase behavior.

In a similar vein, Fishbein (1967) has elevated intention from the status of being one of the components of attitude (i.e., conation) to a construct in its own right. Equation 1 above clearly specifies the role of intention as a mediating, summary cognitive construct in which attitudinal and normative factors are combined to reach a decision to behave in a certain manner. Thus, a causal flow of influence is postulated wherein I is the resultant of AB and SN and, in turn, is the antecedent of B (Ryan, 1975)- Substantiation of these proposed relationships is essential if the Extended Fishbein Model is to be used in the formulation of behavior change strategies.

Recent research on causal patterns within the Fishbein model (Lutz, 1975b; Ryan, 1975) has shown promising results for the hypothesized relationships. Neither of these studies measured overt behavior, however. Further research using change experiments and longitudinal designs should be undertaken, and, wherever possible, measures of overt behavior should be obtained.

Psychological meaning of w1 and w2. The weights attached to the attitudinal and normative components in Equation 1 have traditionally been estimated through regression procedures. Fishbein and Ajzen (1975) state: "Ideally, the weights for the attitudinal and normative components would be available for each individual with respect to each behavior in a given situation. Since adequate estimates of this kind are not presently available, the practice has been to use multiple regression techniques..." (p. 303).

But what do the weights represent conceptually? If they represent the relative importance that the individual places on attitudinal vs. normative considerations, then surely such estimates could be obtained. Of course, such judgments would not be very reliable, as recent clinical judgment research has shown. Nevertheless, a conceptual explanation of the weights seems necessary before it can be concluded that independent estimates of their magnitudes are not available. The present regression procedure has several drawbacks, as shall be seen in the next section.

Operational Issues. Three areas of concern arise with respect to the operationalization of Equation 1: first, measurement of intention can be improved; second, there is a persistent problem of multicollinearity between the attitudinal and normative components; and third, more work needs to be done on the testing of the model at the individual level.

Measurement of intention. Juster (1966) reports that a purchase probability scale (an 11-point scale ranging from "0 chances out of 10" to "10 chances out of 10") was "markedly superior" (in the prediction of behavior) to traditional measures of purchase intention. Little attention has been paid to the measurement of I within the Fishbein model. Typically, multiple item semantic differential inventories are used to provide a single score; however, there has been no attempt made to relate this score to the concept of purchase probability. While Juster's (1966) results were based on aggregate level analyses, his conclusion about the usefulness of the probability scale suggests that a similar form of measurement may be useful for the explanation of individual purchase behavior.

Multicollinearity of AR and SN. [Strictly speaking, most instances of multicollinearity have been between AB and Sbjmj , since SN has only recently been included in the model. However, Ryan (1975) observed multicollinearity between AB and SC.] It is well known that high correlations between predictor variables in a multiple regression cause the regression weights attached to those variables to become unstable. This is an unsatisfactory situation for the manager interested in basing strategy decisions on the relative magnitude of those weights. Yet in many cases, due to a halo effect present in the measurement procedure or perhaps a true relationship between the two components, AB and SN have been highly correlated. Under such conditions, the Extended Fishbein Model is weakened for explanatory purposes. This points to the need, discussed in the previous section, for independent assessment of w1 and w2.

Individual level analysis. Most tests of Equation I have been conducted via cross-sectional regressions. This procedure is not completely satisfactory, as there is no information provided regarding individuals' w1 and w2 parameters. In order to understand consumer behavior at the individual level, a separate set of weights for each person in the sample would be desirable. Recently, Wilson, et al. (1975) and Kakkar (1975) have reported the results of individual level analyses which support the Fishbein model. Wilson, et al. were able to find significant clusters of respondents with respect to the patterns of their regression weights, which suggests the possibility of segmentation analysis. However, individual level analysis may be a two-edged sword; Kakkar (1975) found heightened multicollinearity between AB and Sbjmj in his study. Wilson, et al. do not report the correlation between AB and Sbjmj.

In order to test the model at the individual level without the threat of multicollinearity or the problems with scale properties, Anderson's (1971) information integration framework can be employed. Similar to the studies reported by Bettman, Capon and Lutz (1975a, b, c, d), a factorial design incorporating bi, ei, bj and mj as four completely crossed factors and intention-as the dependent variable would allow tests of several assumptions. First, the multiplicative relationship of bj and mj could be examined (the multiplicative relationship of bi and ei has already been established). Second, the assumption that the attitudinal and normative components are additive, as shown in Equation 1, could be tested. This would provide evidence as to whether the observed multicollinearity between AB and SN is due to measurement effects or a true configural relationship between the two components.

CONCLUSION

Although a relatively recent addition to the consumer literature, the Extended Fishbein Model has shown considerable promise as both a managerial and research tool. Recent developments with respect to the normative component point to the need for further testing and refinement of the model.

As was true with early research on the multiattribute attitude model, much of the research on the Extended Fishbein Model has been within a static, cross-sectional paradigm. In order to test a theory as rich as the Fishbein theory, other approaches are essential. In particular, dynamic tests of the model are useful for investigating causal relationships. In a similar vein, longitudinal designs would help to identify the intricate feedback mechanisms between behavior and beliefs, behavior and attitude, etc. Continued emphasis should be placed on individual level analysis, and ANOVA procedures should be used to complement the more commonly used correlational designs.

REFERENCES

Ahtola, Olli T. "The Vector Model of Preferences: An Alternative to the Fishbein Model," Journal of Marketing Research, 12 (February 1975), 52-59.

Anderson, Norman H. "Integration Theory and Attitude Change," Psychological Review, 78 (1971), 171-206.

Axelrod, Joel N. "Induced Moods and Attitudes Toward Products," Journal of Advertising Research, 3 (June 1963), 19-24.

Bentler, P.M. "Semantic Space is (Approximately) Bipolar," Journal of Psychology, 71 (1969), 33-40.

Bettman, James R., Noel Capon, and Richard J. Lutz. "Cognitive Algebra in Multi-Attribute Attitude Models," Journal of Marketing Research, 12 (May 1975a), 151-64.

Bettman, James R. "Information Processing in Attitude Formation and Change," Communication Research, 2 (3) (July 1975b), 267-78.

Bettman, James R. "Multiattribute Measurement Models and Multiattribute Attitude Theory: A Test of Construct Validity," Journal of Consumer Research, 1 (March, 1975c), 1-15.

Bettman, James R. "A Multimethod Approach to Validating Multi-Attribute Models." In M. J. Schlinger (Ed.), Advances in Consumer Research, Volume 2. Chicago: The Association for Consumer Research, 1975d, 357-74.

Carlson, Earl R. "Attitude Change Through Modification of Cognitive Structure," Journal of Abnormal and Social Psychology, 52 (1956), 256-61.

Divesta, Francis J., and Jack C. Merwin. "The Effects of Need-Oriented Communications on Attitude Change," Journal of Abnormal and Social Psychology, 60 (1960), 80-5.

Dulany, Don E. "Awareness, Rules, and Propositional Control: A Confrontation with S-R Behavior Theory," In Theodore R. Dixon and David L. Horton (Eds.), Verbal Behavior and General Behavior Theory. Englewood Cliffs, N.J.: Prentice-Hall, 1968, 340-87.

Fishbein, Martin A. "An Investigation of the Relationships Between Beliefs About an Object and the Attitude Toward That Object," Human Relations, 16 (August 1963), 233-9.

Fishbein, Martin A. "A Consideration of Beliefs, Attitudes, and Their Relationship." In Ivan D. Steiner and Martin Fishbein (Eds.), Current Studies in Social Psychology. New York: Holt, Rinehart, and Winston, 1965, 107-20.

Fishbein, Martin A. "Attitude and the Prediction of Behavior." In Martin Fishbein (Ed.), Readings in Attitude Theory and Measurement. New York: Wiley, 1967a, 477-92.

Fishbein, Martin A. "A Behavior Theory Approach to the Relations Between Beliefs About an Object and the Attitude Toward the Object," in Martin Fishbein (Ed.), Readings in Attitude Theory and Measurement. New York: Wiley, 1967b, 389-400.

Fishbein, Martin A., and Icek Ajzen. Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research. Reading, Mass.: Addison-Wesley, 1975.

French, J. R. P., Jr., and Bertram Raven. "The Bases of Social Power," in D. Cartwright (Ed.), Studies in Social Power. Ann Arbor: University of Michigan Press, 1959.

Howard, John A., and Jagdish N. Sheth. The Theory of Buyer Behavior. New York: Wiley, 1969.

Juster, F. Thomas. "Consumer Buying Intentions and Purchase Probability: An Experiment in Survey Design," Journal of the American Statistical Association, 61 (September 1966), 658-96.

Kakkar, Pradeep. Situational Influence on Emotional Response, Cognitive Structure and Cognitive Process. Unpublished doctoral dissertation, University of California, Los Angeles, 1975.

Kelman, Herbert C. "Processes of Opinion Change," Public Opinion Quarterly, 25 (Spring 1961), 57-78.

Lutz, Richard J. "Changing Brand Attitudes Through Modification of Cognitive Structure," Journal of Consumer Research, 1 (March 1975a), 49-50.

Lutz, Richard J. "An Experimental Investigation of Causal Relations Among Cognitions, Affect and Behavioral Intention,'' Working Paper No. 34, Center for Marketing Studies, University of California, Los Angeles, 1975b.

Osgood, Charles E., George J. Suci, and Percy H. Tannenbaum. The Measurement of Meaning. Urbana: University of Illinois Press, 1957.

Peak, Helen. "The Effect of Aroused Motivation on Attitudes," Journal of Abnormal and Social Psychology, 61 (1960), 463-8.

Peak, Helen., and H. William Morrison. "The Acceptance of Information into Attitude Structure," Journal of Abnor-and Social Psychology, 57 (1958), 127-35.

Rokeach, Milton. Beliefs, Attitudes and Values: A Theory of Organization and Change. San Francisco: Jossey-Bass, 1968.

Rosenberg, Milton J. "Cognitive Structure and Attitudinal Effect," Journal of Abnormal and Social Psychology, 53 (1956), 367-72.

Rosenberg, Milton J. "Cognitive Reorganization in Response to the Hypnotic Reversal of Attitudinal Affect," Journal of Personality, 28 (1960), 39-63.

Ryan, Michael J. "A Test of Associations in a Causal Chain Derived from Fishbein's Behavioral Intention Model," unpublished working paper, University of Alabama, 1975.

Ryan, Michael J., and E. H. Bonfield. "The Fishbein Extended Model and Consumer Behavior," Journal of Consumer Research, 2 (September 1975), 118-37.

Schmidt, Frank L. "Implications of a Measurement Problem for Expectancy Theory Research," Organizational Behavior and Human Performance, 10 (October 1973), 243-51.

Schmidt, Frank L., and Terry C. Wilson. "Expectancy Value Models of Attitude Measurement: A Measurement Problem," Journal of Marketing Research, 12 (August 1975), 366-8.

Thurstone, L. L., and Lyle V. Jones. "The Rational Origin for Measuring Subjective Values," Journal of the American Statistical Association, 52 (1957), 458-71.

Wilson, David T., H. Lee Mathews, and James W. Harvey, "An Empirical Test of the Fishbein Behavioral Intention Model," Journal of Consumer Research, 1 (March 1975), 39-48.

---------------------------------------