# Attitude Research and Behavioral Intentions: a Critical Review

^{[ to cite ]:}

Jeffrey E. Danes (1981) ,"Attitude Research and Behavioral Intentions: a Critical Review", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 57-60.

^{[ direct url ]:}

http://acrwebsite.org/volumes/9785/volumes/v08/NA-08

This paper reviews the work by Holbrook, Velez, and Tabouret, "Attitude Structure and Search: An Integrative Model of Importance-Directed Information Processing," Brinberg, "A Comparison of Two Behavioral Intention Models," and Miniard, "Examining the Diagnostic Utility of the Fishbein Behavioral Intentions Model."

ATTITUDE STRUCTURE AND SEARCH

Holbrook, Velez, and Tabouret begin their paper with the following statement, "In spite of extensive research on multiattribute attitude models and information-acquisition paradigms, the linkage between these two important facets of buyer behavior is not well understood." Later on they assert, "YMultiattribute Attitude Research has tended to focus primarily on the __static__ structure of attitude while neglecting* *the __process__ by which new information inputs are incorporated into the determination of affect" (emphasis added).

At the conceptual level, Holbrook, __et al__ move away from a static description and provide a process, dynamic view of the linkages between information and attitude. One would expect, therefore, that dynamic information-attitude -models would be presented to compliment their conceptual work. Unfortunately, this was not the case. The four algebraic multiattribute attitude models appear as static representations of attitude. The modeling work provided by Holbrook, __et al__ could be improved if dynamics were explicitly incorporated into their four models. We now address this issue.

A Dynamic Representation

Holbrook, __et al__ presented four multiattribute models to predict attitude __after__ individuals have acquired information (messages) about a product: a full and partial additive mode and a full and partial averaging model. What is the attitude value before information (messages) is received? Holbrook, __et al__ do not address this issue, and neither do they identify or develop multiattribute models that gives us the pre-information search (pre-message) attitude. It is argued below, however, that they implicitly assume that pre-message attitude is given by Fishbein's model (or an "averaging" variant of his model). Without loss of generality, let us restrict our discussion to Holbrook's, __et al__ partial additive model.

After individuals acquire information (receive messages) about the object; Holbrook __et al__, assumed that if the message, m, associates an attribute with an object, the __new__ belief value equals 1. Or, if the message disassociates an attribute with an object, the __new__ belief value equals -l. Evaluation, e, is not affected by the messages. The post-message attitude as given by Holbrook's, __et al__, partial additive model may be expressed as:

where, a is the attitude toward the object, b is belief, is evaluation, and acq is the information (message) variable which is coded 1 or 0 depending upon whether the message was received.

As stated above the attitude, a, reported in Equation 1 is a post-message or "new" attitude. Also, the belief values are post-message or "new" beliefs. With new notation, let us rewrite Equation 1:

where the message, m, is used denote acquired information, acq. Both new a and new b may be expressed as their pre-message values plus the change, i.e.,

new a = a + Da (4)

new b = b + Db (5)

where a and b is the pre-message attitude and belief, and Da and Db is the change in attitude and belief. If we make the appropriate substitutions, Equation 3 may be expressed as:

One perplexing problem with the Holbrook __et al__ model is that the message, m, is entered twice in Equation 7. Note here that the pre-message attitude,

includes the message which the individual has not yet received: It makes more sense to define the pre-message attitude as:

As implied by Holbrook, et al by Equation 7, change in attitude may be expressed as:

where, Db stands for change in belief.

How Do Beliefs Change?

Holbrook, __et al__, explicitly assume that a message changes belief in the direction of the object-attribute association made in the message. They also assume maximum change; hence, the post-message belief is equal to the belief-value communicated in the message. Thus, Holbrook, __et al__, have simply re-stated Anderson and Hovland's (1957) "Distance-Proportional'' Model:

New b = b + a(m - b), (11)

or

b + Db = b + a(m - b), (12)

or

Db = a(m - b). (13)

Additionally, since the authors assume maximum change, the proportionally constant, a, must equal 1.

Db = m - b (14)

Replacing (m - b) for Db in Equation 10 gives:

Once again there are two message terms. The first message term was derived directly from Holbrook __et al's __assumptions; the last message term appears to be unnecessary and may be deleted:

The model presented in Equation 16 is more consistent, than is Equation 15, with Holbrook's __et al__ logic and assumptions. More importantly, if we began with Equation 9 and worked to build a dynamic model of the information-attitude process, we would arrive directly at Equation 16--given the belief change assumptions made by Holbrook, et al.

Holbrook, et al was interested in attitude after messages were received. They "new" or post-message attitude may be obtained by adding Equations 9 and 16; this gives:

Equation 17 produces a numerical result __identical __to Holbrook's __et al__, partial additive model. However, the dynamic assumptions made by Holbrook, __et al __are explicitly stated in the new model.

Summary

Holbrook, __et al__ presented a promising conceptualization of the dynamic relation between information acquisition (or message reception) and attribute formation/change. The four algebraic models presented, however, failed to capture the dynamic assumptions asserted in their conceptual work. Our discussion centered upon translating Holbrook's __et al__ assumptions about the dynamics of the information (message) - attitude process into change equations isomorphic with their assumptions.

TWO BEHAVIORAL INTENTION MODELS

Brinberg's paper compares two models for the prediction of behavioral intentions: the extended Fishbein model and Triandis' model. The major differences between the two models are (1) Fishbein assumes "affect" and "evaluation'' are indicators of different constructs; (2) Fishbein asserts "subjective norms" influence behavioral intentions and Triandis argues for "social determinants."

To compare the two models, Brinberg compares 'the factor structure of the __predictor__ variables. He argues against a standard multiple regression analysis "...since the regression technique does not allow for the effects of measurement error, a researcher may conclude that two (or more) constructs exist, when, in fact the constructs are imperfect measures of the same underlying dimension." This, indeed, is an interesting criticism, for Brinberg puts the blame on regression when the blame ought to be placed elsewhere--on __measurement__. If Brinberg had obtained reliability coefficients for his scales, he could have used the corrected correlations and standard deviations in his regression analyses. Correlations and standard deviations that are corrected for attenuation (unreliability) are "free" of measurement error. If only standardized regression coefficients are of interest, only the correlation matrix need be corrected for attenuation. The more obvious problems associated with multicollinearity were not addressed.

"To circumvent some of the limitations of thc regression technique ..."Brinberg relies on a principle components orthogonal factor analysis (with varimax rotation) of the predictor variables. Two factor analyses were performed, one for evaluation, affect, and consequences. And, one for Triandis' social components and Fishbein's normative components. In each case, two factors were obtained. The first two-factor solution was interpreted as support for Triandis' Hypothesis that the affective and evaluative components are not the same. The second two-factor solution was interpreted as "what relevant others think" and as "what is appropriate to do." Hence, Brinberg argues that both Fishbein and Triandis have misspecified the "social influence" component of behavioral intentions. Warshaw (1981) has criticized the scales used for the measurement of "affect" and "evaluation;" the same criticism may be made for the "social influence" constructs. Brinberg docs not cite any evidence for the __validity__ of his measures.

At this point a few questions become apparent:

(1) Why was a

principal Components Factoranalysis used?(2) Why was the Factor Analysis constrained to be orthogonal?

(3) Would an oblique solution yield highly correlated factors?

(4) What are the relationships between "affect," evaluation,'' "social determinants," "subjective norm," and behavioral intentions?

As to the first question ("Why was a principal components factor analysis used") a principal components analysis makes assumptions about the data which may not be true--it assumes perfect measurement. The principal components analysis operates on a correlation matrix with ones in the diagonal; i.e., it assumes the variables are measured without error. Other factor techniques do not make this assumption; perhaps, they should have been explored.

For the second and third questions ("Why was the Factor Analysis constrained to be orthogonal," and "Would an oblique solution produce highly correlated factors") there appears to be no Justification for constraining the factors to be uncorrelated. We suspect that "affect" and "evaluation'' are highly correlated; likewise that "social determinants'' and "subjective norms" are also highly correlated. Brinberg could have performed an __Oblique Multiple Groups Factor Analysis__ (Harmon 1967) and the "natural" factor correlations could have been found. Assuming uncorrelated factors is arbitrary and unjustified.

The last question ("What are the relationships between 'affect,' 'evaluation,' 'social determinants,' 'subjective norms,' and 'behavioral intentions") could be found using a variety of highly related methodologies: (1) an oblique multiple groups factor analysis followed by a path analysis performed upon the factor correlation matrix of the five constructs, or (2) a multiple indicator path analysis, or to use the newer terminology, a structural equation (causal model) analysis with latent variables.

In summary Brinberg attempted to compare two models for the prediction of behavioral intentions: the extended Fishbein model and Triandis' model. For the comparison of the two models, Brinberg used an orthogonal principle components factor analysis on the predictor variables. Their results were interpreted as support for Triandis' notion that "affect" and "evaluation" are separate constructs; that the "social influence" construct consists of two dimensions: "what relevant others think," and "what is appropriate to do."

Several methodological problems were discussed, including:

(1) Reliability and validity of measurement;

(2) Choice of factor analysis technique

(3) Use of orthogonal solution;

(4) Structural equations analysis as an alternate analytic tool.

THE DIAGNOSTIC UTILITY OF THE FISHBEIN MODEL

The purpose of Miniard's paper is to assess whether the subjective norm component is detectable when it "is present" and when "it is not present." Miniard reports two studies. The first experiment was designed to be void of normative influences; the primary task of this experiment was to assess whether normative influences could be detected. The results of his experiment supported the extended Fishbein model. There are, however, several methodological problems associated with the first study:

(1) No manipulation checks on the experimental treatments were made; hence, we have no guarantee that normative influences were actually absent;

(2) The reliability of measurement (apparently) was not assessed. Unreliability may have distorted the obtained omega squared's. (Below we demonstrate such distortion using regression.)

Miniard's second study reports a reanalysis of his previous published work. In this study, he asserts that normative influences were present. Two out of four regression analyses, however, failed to detect them. There are three apparent problems with the second study:

(1) The manipulation may not have been relevant to males.

(2) The neighbor, husband, and parent significant others were arbitrarily selected. No evidence is given justifying: this choice.

(3) Reliability of measurement is not assessed; the regression analyses were not corrected for unreliability.

The last criticism appears to be widespread in consumer research; thus, let us focus our attention on the impact measurement error may have upon multiple regression. As stated above, two of four regression models failed to detect the expected subjective norm influence. The four models differed in their operationalization of subjective norm. One model contained a direct measure of subjective norm, SN; the others varied in terms of specificity general moderate, and situation specific. The direct and general models failed to receive support. For the direct measure of subjective norm, SN, the b coefficient equaled .08; and the bcoefficient for attitude toward behavior, A_{b} equaled .55. Unfortunately, Miniard __does__ not provide us with his correlation matrix; however, the matrix supplied in Table 1 will be used for discussion.

EXAMPLE CORRELATION MATRIX FOR BEHAVIORAL INTENTION, ATTITUDE TOWARD BEHAVIOR, AND SUBJECTIVE NORM

From Table 1, the following regression weights may be derived: b_{Ab} = .555 and B_{SN} = .084. It is assumed, however, that all three measures are perfectly reliable. Could measurement unreliability account for the nonsignificant weight for subjective norm? Indeed it could, and we show such an example below. Miniard __did not__ provide reliability estimates for his measures; so let us, for discussion purposes, assume that the reliabilities are: .64, .98, and .68 for behavioral intentions, BI, attitude toward behavior, A_{b} and subjective norm, SN. If we replace the diagonal of ones with the reliability coefficients and correct (for attenuation) the off-diagonal correlations, we arrive at the corrected correlation matrix reported in Table 2.

Does unreliability influence the regression weights? It certainly does! The corrected 8 coefficient for subjective norm now equals .19 and the corrected b coefficient for attitude toward behavior now equals .62; both coefficients may be significant using Miniard's degrees of freedom. Since Miniard did not estimate the reliability of measurement or correct his analyses for unreliability, we are unable to assess how much his regression weights were distorted.

In summary, Miniard proposed an interesting test of the extended Fishbein model. He presented two studies. The first study was designed to be void of subjective norm information. In this experiment, the extended Fishbein model accurately detected the lack of a normative influence. The second study was designed to produce normative influences; and only two of four tests were able to detect the expected influence. Our discussion centered upon several methodological problems:

(1) Failure to use manipulation checks or possible irrelevant manipulations;

(2) Failure to assess reliability of measurement and correct analyses for unreliability.

REFERENCES

Anderson, Norman H. and Hovland, Carl I. (1957), "The Presentation of Order Effects in Communication Research," in __The Order of Presentation Persuasion, __ed., Carl I. Hovland, New Haven, CT: Yale University Press, pp. 158-69.

Brinberg, David (1981), "A Comparison of Two Behavioral Intention Models," in __Advances in Consumer Research__ ed., Kent B. Monroe, Washington, D.C.: Association for Consumer Research.

Harmon, Harry J. (1967), __Modern Factor Analysis__, Chicago: The University of Chicago Press.

Holbrook, Morris B., Velez, David A., and Tabouret, Gerard R. (1981), "Attitude Structure and Search: An Integrative Model of Importance-Directed Information Processing," in __Advances in Consumer Research__, ed., Kent B. Monroe, Washington, D.C.: Association for Consumer Research.

Warshaw, Paul R. (1981), "A Discussion of Attitude Research and Behavioral Intentions," in __Advances in Consumer Research__, ed., Kent B. Monroe, Washington, D.C.: Association for Consumer Research.

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

Tweet
window.twttr = (function (d, s, id) { var js, fjs = d.getElementsByTagName(s)[0], t = window.twttr || {}; if (d.getElementById(id)) return; js = d.createElement(s); js.id = id; js.src = "https://platform.twitter.com/widgets.js"; fjs.parentNode.insertBefore(js, fjs); t._e = []; t.ready = function (f) { t._e.push(f); }; return t; } (document, "script", "twitter-wjs"));