History of Attitude/Behavior Research
The "attitude" construct received its first serious attention from Darwin in 1872. Darwin defined attitude as a motor concept, or the physical expression of an emotion. For early psychologists, "attitude" was an emotion or thought with a motoric (behavioral) component. In some cases, the motoric component was subvocal speech; in other cases, gross behavior, such as postural change, was of interest. Beginning in the 1930's, psychologists began to argue actively about what components should comprise the attitude concept. Although there was agreement that all attitudes contain an evaluative component, theorists disagreed about whether beliefs (cognitions) and behaviors should be included as part of the attitude concept. The prevailing view among cognitive social psychologists was that "attitude" has both affective and belief components and that attitudes and behavior should be consistent; i.e., people with positive attitudes should behave positively toward the attitude object.
LaPiere (1934) reported that hotel managers' attitudes toward Chinese guests did not predict their responses to a Chinese couple who asked for a room. LaPiere's work was criticized on numerous grounds (e. g., the person who filled out the questionnaire may not have been the same person who later admitted the Chinese to the hotel), but many other researchers reported similar findings: attitudes did not predict behavior even when measured under optimal conditions, (Wicker, 1969).
In 1975, Fishbein and Ajzen published Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research, laying out the theory of reasoned action which they claimed would improve our ability to predict behavior. In published reports, the variables specified by the theory generally did account for more of the variance in behavior than had previous attitude/behavior measures. However, it soon became clear that some important limitations on the theory's domain were required, that additional variables would need to be included, and that the theory was perhaps better understood as a taxonomy, as opposed to an explanatory system. Ajzen (1987) has published an updated version of the theory of reasoned action called the theory of planned behavior. Although the theory of planned behavior has undergone relatively few empirical tests, it seems unlikely that it will fare significantly better than Fishbein and Ajzen's earlier work. Although Fishbein's model remains popular with some market researchers, the prevailing theory among psychologists is Fazio's (1986) attitude accessibility model.
In the remainder of this paper, each of these theories will be discussed along with the research undertaken to test them and the major problems each seems to have.
Fishbein and Ajzen's Theory of Reasoned Action
The theory of reasoned action actually applies to the prediction of intentions, as opposed to behavior itself. According to the theory, if behavior is under volitional control, then the intention to perform an action will correlate very highly with the action itself. By and large, this supposition has been found to be correct, with correlations between intention and behavior averaging 0.55. The full model is:
B I = wp Attitudebehavior + wp Subjective Norm
and Attitudebehavior = biei
and Subjective Norm = bi mi
and the w's are subjective weightings for a particular person.
Attitudes Toward the Behavior
Attitudes toward the behavior are made up of beliefs about engaging in the behavior and the associated evaluation of that belief. For example, consider the purchase of a car, X. In tests of the model, subjects are asked to list their beliefs associated with buying the car. These beliefs are consequences of the action. One belief might be: "Buying car X will cost me $300 a month." Another belief might be "Buying car X will make me more attractive to the opposite sex." Each belief is then rated for the likelihood that engaging in the behavior will produce that consequence. The likelihood ratings are an index of belief strength. After subjects rate the probability of each belief's being true, they evaluate how good or bad this aspect is. A car payment of $300 might be rated as quite bad, while being attractive to the opposite sex might be quite good. These ratings (both belief strength and evaluations) are quantified on -3 to +3 or 1 to 7 scales. The belief strength and evaluation ratings are multiplied together for each belief and summed across beliefs to give a measure of attitude toward the behavior.
Subjective Norm
The subjective norm term in the model is also multiplicative. The "b's" in this term are beliefs about what relevant others will think if the respondent engages in the behavior. For example, "People who are important to me would not want me to buy car X." Again, the certainty that this is true is rated by the respondent. Each belief receives a second rating: how strongly does the respondent wish to comply with the referent other's views. So, I might feel very certain that important others would not approve of my buying car X but I might have a very low desire to comply with their views.
Intention
Intention is usually measured by one to four questions asking the likelihood the respondent will engage in the behavior. Bagozzi, Baumgartner, and Yi (1989) have called attention to neglect of the reliability of the intention measure. A recent metanalysis (Sheppard, Harwick & Warshaw, 1988) found the mean correlation between intention and the attitudes + norm component to be 0.66.
Selection of the Beliefs
In some studies, subjects list their own beliefs about the consequences of engaging in a behavior. In other studies, pilot testing of a large sample is used to discover the most common relevant beliefs for a particular group; these common beliefs (usually 7 to 15 different beliefs) are then given to the experimental sample for rating.
This model has been used extensively with health and social issues, particularly intentions to use birth control, stop smoking, smoke marijuana, recycle, use alcohol, etc.
Problems in this Area of Research
Limitations on the Domain of Application
Conscious control. As implied by its name, the theory of reasoned action does not apply to habitual actions that are presumably not under continual conscious processing. In other words, the theory applies to behavior the individual consciously elects to do. Many tests of the model have been conducted with habitual behaviors, however. Kahle & Beatty (1987) applied the model to coffee drinking and found good statistical support for the theory. However, a model that included only habit and situation did an even better job of prediction.
Correspondence Ajzen and Fishbein (1977) subsequently published an article further limiting the theory to those situations in which the attitude and behavior demonstrate correspondence. According to Ajzen and Fishbein, attitudes and behavior each have four elements: action, target, context, and time. Correspondence occurs to the extent that attitudes and behaviors are identical on all four elements. To predict intention, attitudes and intention must measure exactly the same four elements. To return to the car purchase example: my intention to buy car X has an action (buy), a target (car X), and to expand the example, a context (at a particular dealer's with a particular loan), and a time (next month). If I want to predict a specific intention, I must measure a specific attitude. The resulting experimental procedure seems extremely trivial; instead of measuring these extremely specific components, one can simply ask "Do you intend to purchase car X from this dealership next month?" Fishbein and Ajzen's primary goal in developing the notion of correspondence was to show why one can't predict specific intentions from general measures of attitude, such as, "What do you think of car X?" To predict specific intentions (behaviors), equally specific attitudes must be measured.
Behavior scaling. Fishbein and Ajzen (1976) also addressed a measurement problem that makes the prediction of intentions problematic. They noted that researchers spend great effort to develop attitude scales that are reliable, valid, and satisfy certain measurement criteria, such as Guttman or Thurstone scales. Behaviors are often chosen haphazardly. Other than the researcher's intuition, there is no way to scale how positive or negative a particular behavior might be. Consider someone who has a "very positive attitude" toward abortion rights. I may find that the person does not sport a bumper sticker advocating abortion rights. Fishbein would not find this puzzling because we don't know anything about how the behavior, displaying the bumper sticker, scales. Is it an extremely positive behavior? Is it slightly positive? Fishbein and Ajzen found that scaling behaviors and attitudes on the same scale (e. g., Likert, Guttman, etc.) resulted in dramatic improvements in the attitude/intention (behavior) correlation. In addition, the predictive power of general attitude measures, in particular Likert scales, improved when a number of behaviors were presented and subjects were asked how many behaviors they had performed or intended to perform. Prediction of individual behaviors from a general attitude measure was extremely poor. Prediction of the score on this so-called behavioral composite scale was much better. People with more positive attitudes performed a greater number of positive behaviors but which specific behavior was performed was not predictable from the general attitude measure.
Additional Variables
Moderators. Although Fishbein and Ajzen believed that any other variable affecting the attitude-intention (behavior) link exerted its effect on one of the terms in the model, Ajzen's own research proved this was erroneous. Self-monitoring refers to a stable individual difference (Snyder, 1974) in the tendency to vary one's behavior in different situations. High self-monitors are sensitive to situation cues and tailor their behavior, dress, and speech to the situation. Low self-monitors are indifferent to situational cues and act on the basis of their principles. Ajzen, Timko and White (1982) found that the attitude/intention model was more predictive of the behavior of low self-monitors than high self-monitors. High self-monitors' intentions did not correlate with their behavior. Low self-monitors apparently tend to act on their attitudes no matter what the situation. High self-monitors may not express an attitude in behavior if they feel the behavior is inappropriate for the situation. To summarize, the Fishbein-Ajzen model works better for low self-monitors because these people are more likely to translate their attitudes into behavior across a variety of situations.
Private self-consciousness. This is a second individual difference moderator currently receiving attention. Private self-consciousness is the dispositional tendency to be aware of one's own internal thoughts and feelings. Miller & Grush (1986) found higher attitude/behavior consistency for people high in private self-consciousness, presumably because they were more aware of their own attitudes.
Other variables. In many tests of Fishbein's model, additional variables have been included and found to increase the attitude/behavior correlation. Some of these include: economic variables (Lynne & Rola, 1988) in predicting farmers' behavior regarding soil conservation; moral values (Boyd & Wandersman, 1991) in predicting condom use; and academic achievement and friends' intentions (Carpenter & Fleishman, 1987) to predict college entry.
Measurement Problems
Methodology. Most tests of Fishbein's model ask subjects to indicate how strongly they believe that a series of belief statements are true. These ratings are followed by a request to indicate how good or bad each belief consequence is. Research by Budd and Spencer (1986) suggests that this format creates a serious confound. First, when the Fishbein items are scattered within a larger questionnaire, the correlations between intention and the [bi ei + bi mi] term are much lower than when all the ratings are presented in a cluster. Second, Budd and Spencer asked students to rate how honestly they felt a questionnaire measuring attitudes, beliefs, norms, and intentions had been answered. When the theory of reasoned action was violated (attitudes were inconsistent with intentions), the hypothetical respondent was seen as more dishonest. The authors argue that Fishbein's theory is part of an intuitive psychology of intention and that this intuitive psychology acts as a source of response bias, promoting consistency between responses.
Self-reports. A second methodology issue is that tests of the Fishbein-Ajzen model rely almost exclusively on self-reports. Behavior itself is rarely directly observed. Self-reports of behavior are notoriously unreliable and have been found to vary with attitude (Ross, McFarland, Conway & Zanna, 1986); people with more positive attitudes report more positive actions than they actually performed; people with negative attitudes report more negative actions than actually performed. Manfredo and Shelby (1988) studied wildlife tax-fund donations. Fishbein and Ajzen's model was applied to both actual behavior gathered from the fund's records and self-reports from donors. The correlations for predicting actual and self-reported behaviors were both significant but were significantly different from each other.
Statistical issues. Evans (1991) has criticized the Fishbein and Ajzen model on statistical grounds. Evans notes that when one is using a multiplicative component to predict a simple variable (e.g., attitudes toward the behavior to predict intention), one must include the main effects (the belief strengths and the evaluations of each belief) in the model prior to the entry of the multiplicative component. Evans notes that multiplicative components have a peculiar property: a change in the zero point or a change in the interval size of either component scale can have marked effects on the size of the correlation coefficient. Evans singles out the Fishbein-Ajzen model for scrutiny. He notes that of 40 studies covered in a recent meta analysis (Sheppard et al., 1988), none have tested a full additive model that included main effects. Evans could find only a single study (Hewstone & Young, 1988) in which attitude researchers studied main effects. These authors examined the relationship between beliefs and evaluations of outcomes to overall attitudes toward the European Economic Community. Hewstone and Young actually compared an additive model that included main effects and a multiplicative model without the main effects. They also compared two different ways of scaling beliefs and evaluations, a -3 to +3 scale and a +1 to +7 scale. For the full additive model, scaling had no effect on the multiple R which was 0.46. For the multiplicative model, the -3 to +3 scale version correlated 0.30 with attitudes. Given the variety of scales used to study the attitude/behavior link and the failure of investigators to enter the main effects first, Evans concludes that a whole body of literature is rendered suspect.
Ajzen's Theory of Planned Behavior
The theory of planned behavior (Schifter & Ajzen, 1985) is an extension of the theory of reasoned action. The theory of planned behavior includes one additional variable: perceived behavioral control. Perceived behavioral control is assessed by asking people how much control they have over performing a particular behavior. In Ajzen's tests of the theory (Ajzen & Madden, 1986; Madden, Ellen & Ajzen, 1992) the measurement of the attitudinal component has also been simplified. Attitudes toward the behavior are measured on a five item semantic differential scale. Including the perceived behavior control variable does lead to significant improvements in R2 for behaviors perceived to be low in control. "Getting a good night's sleep" is an example of a low control behavior; "taking vitamins" is a high control behavior. The behavioral control variable did not improve prediction for the latter behavior, presumably because the behavior itself is already perceived as high control.
Problems. All the problems associated with the theory of reasoned action are also problems for the theory of planned behavior.
Fazio's Attitude Accessibility Theory
Fazio's (1986) model of the process by which attitudes guide behavior is currently receiving a fair amount of attention in the social psychological literature. Fazio defines attitude as a learned association between a concept and an evaluation. Like any construct based on associative learning, attitude strength varies. Fazio indexes strength using a reaction time paradigm. The more rapidly an attitude can be expressed, the greater its strength. The stronger the attitude the more accessible it is.
To guide behavior, attitudes must be accessible. Attitudes that are highly accessible from memory are much more likely to guide behavior than less accessible attitudes. Fazio, Sanbonmatsu, Powell and Kardes (1986) have demonstrated that accessible attitudes are activated spontaneously upon presentation of the attitude issue. Their emphasis on the automatic activation of attitudes differs markedly from Fishbein's view that attitudes result from a controlled effortful process of attribute consideration and evaluation.
Fazio and his colleagues have shown that correlations between attitudes and behavior are much higher among people with highly accessible attitudes. In one study (Fazio and Williams, 1986), accessibility was assessed by how quickly respondents rated the 1984 candidates for U.S. President. Four months later on the day after the elections, the respondents were asked if they had voted and for whom. Among voters with highly accessible attitudes, 80 percent of the variance in voting behavior was explained by attitudes; among voters with less accessible attitudes, only 44 percent of the voting behavior was accounted for by attitudes. Fazio and Williams believe the greater consistency of the highly accessible group is a function of greater attitudinal stability. Highly accessible attitudes are linked to selective processing of information and even selective attention (Fazio, 1989; Roskos-Ewoldson & Fazio, 1992). To the extent that accessible attitudes are accessed each time an individual encounters the relevant concept, the attitude protects its holder against counter-attitudinal information and potential attitude/behavior inconsistency.
Accessibility is weakly related (0.30) to attitudinal polarity. Extreme attitudes do have a tendency to be more accessible. Accessibility, measured by reaction time to an attitudinal query, is a function of: number of previous expressions of the attitude; opportunities for review or rehearsal of the beliefs and behaviors associated with the attitude; direct experience with the attitude object; and anticipation of future interaction with the attitude object. Highly accessible attitudes are more difficult to change (Wu and Shaffer, 1987).
Problems
There have been few published criticisms of the attitude accessibility model. Bargh and Chaiken (1992) have recently claimed that variations in associative strength are more a function of word frequency and cultural norms than individual differences in experience with the attitude object. Bargh and Chaiken were able to replicate Fazio and colleagues' reaction time findings.
The lack of criticism may result from a number of factors: the relative recency of the theory is one obvious consideration. Because the computers needed to measure reaction times are more expensive than a paper questionnaire, attempts to replicate and extend the theory may be more difficult. Finally, social psychologists essentially abandoned interest in learning theories of attitudes in the 1950's; Fazio's work on the power of learned associations to guide behavior may raise issues with which cognitive theorists are less comfortable. One interesting development is that Ajzen discusses Fazio's model without criticism (but possibly without enthusiasm) in a recent chapter.
Summary
As Evans (1991) noted, interest in the Fishbein/Ajzen models has waned in recent years. Attitude accessibility would appear to be the promising newcomer. There are, however, two strands of thought that suggest accurate prediction of specific volitional actions may be a difficult, if not impossible, task.
First, consistency between nonhabitual behaviors themselves tends to be quite low. Epstein (1979) tracked the behaviors of college students over a 2-week period. Behaviors were actions such as number of telephone calls made, number of letters written, number of social contacts initiated, etc. Consistency between actions on any two days was extremely low. With 7-day means, correlations improved dramatically. Thus, even the old adage that the best predictor of behavior is past behavior held true only for aggregated behaviors. To the extent that behaviors themselves exhibit temporal instability, a stable construct, such as an attitude, cannot predict a particular behavior successfully.
Second, Mischel (1983) has noted the power of situations, relative to attitudes or traits, in the control of behavior. He has argued that consistency between an internal state and a behavior is an epiphenomenon constructed by the observer to simplify the task of making sense of the world. Mischel believes behavior is situationally specific and cannot be understood by aggregation to remove temporal instability.
One should also note that the attitude/behavior relationship is currently receiving relatively little research attention. Whether this declining interest is a consequence of the difficulty of demonstrating the relationship or simply a manifestation of the "fads and fashions" of social psychology is a question yet unanswered.