Beliefs affect behavior through a sequence of effects. Specific beliefs about a behavior inform attitude, perceived norm, and perceived behavioral control regarding the behavior, which in turn determine intention to perform the behavior. If one has the necessary abilities to perform the behavior and if there are no situational obstacles that impede behavioral performance, then intention should lead to behavior. The conceptualization of behavior formation as a process makes clear that a persuasive message cannot directly change behavior. Although the ultimate objective of persuasive messages is to reinforce or change a particular behavior, persuasive messages at best create or change beliefs. When beliefs are appropriately selected, changes in those beliefs should affect attitude, perceived norm, or perceived behavioral control, which in turn should affect intention and behavior. Those beliefs that most strongly discriminate between people who do and do not (intend to) perform a particular behavior, are the choice candidates to address in persuasive messages (Fishbein & Ajzen, 2010; Fishbein & Yzer, 2003).
In terms of reasoned action theory, persuasion thus concerns the effects of exposure to a persuasive message on beliefs about performing a behavior, and through effects on those beliefs on behavior. Clearly, then, the precision with which one can predict behavior is directly relevant for persuasion scholarship. The remainder of this chapter will therefore be used to review the ability of reasoned action theory to predict behavior. For this purpose it is useful to first discuss the historical context in which reasoned action theory was developed.
Historical Context
* * *
In the early 20th century there was widespread consensus that attitude should matter as a basis for human behavior. For example, most contemporary definitions emphasized attitude as a tendency to act (for an overview see Allport, 1935). By the 1960s, however, accumulated empirical support for the hypothesis that people act on their attitude was inconsistent at best, with many studies reporting no effect of attitude on behavior at all. As a result, many scholars questioned the usefulness of attitude for behavioral prediction. Most widely cited in this regard is Wicker (1969), who, on a review of studies that correlated self-reported attitude with lagged observations of behavior, concluded that it is unlikely that people act on their attitude. In counterpoint, others argued that measurement issues were at least in part responsible for weak correlations between attitude and behavioral data. Particularly pertinent is Triandis’s (1964) finding that the prediction of behavior from attitude improved when measures of attitude and behavior represented the same dimensions.
The debate on the question whether attitude predicts behavior helps understand the origins of reasoned action propositions. In effect, what was under discussion was whether contemporary attitude theory offered valid hypotheses about how thoughts, feelings, and behavior regarding an object are associated. Fishbein observed that the confusion surrounding the attitude-behavior relation had to do with the wide range of different variables that were included under the umbrella label of “attitude.” Similar to Thurstone (1928), Fishbein (1967) viewed attitude as “a relatively simple unidimensional concept, referring to the amount of affect for or against a psychosocial object” (p. 478). Building on Dulany’s (1968) theory of propositional control over verbal responses, he argued that attitude should be separated from its antecedents and consequences. Moreover, in order to improve prediction of behavior, he urged scholars to focus on the relations between these variables, that is, beliefs, attitude, behavioral intention, and behavior (Fishbein, 1963, 1967).
A number of principles have been developed to aid such inquiry (e.g., Ajzen & Fishbein, 1973). A first holds that prediction of behavior (e.g., running) is more precise than prediction of behavioral categories (e.g., exercise) or goals (e.g., losing weight). Exercise includes many different behaviors, and each of these behaviors may be associated with quite different beliefs. From the author’s perspective, for example, running is fun but swimming is not. Whether or not I will report to like and engage in exercise therefore depends on whether I think about running, swimming, or both when asked about my exercise. Similarly, losing weight is a goal that can be achieved by many different behaviors, and one may hold positive beliefs about losing weight yet in fact not achieve that goal because necessary dieting and exercise behaviors are not performed due to negative beliefs about those behaviors.
Second, prediction of specific behaviors is more precise than prediction of general behaviors. Levels of specificity vary by the extent to which a behavioral definition includes each of four components, that is, action (e.g., running), target (e.g., at a 9-minute per mile pace), context (e.g., on a treadmill at the YMCA), and time (e.g., twice a week). Clearly, “running” can be interpreted more broadly than “running twice a week at a 9-minute pace on a treadmill at the YMCA.” When two people think about “running,” they may therefore think about quite different behaviors, each associated with different, behavior-specific beliefs. It is for this reason that persuasive messages are more effective when they promote a specific behavior and its underlying beliefs than a general, more broadly interpretable behavior (Fishbein, 2000).
Third, and known as the compatibility principle, prediction of behavior improves when behavior is measured at the same level of specificity as beliefs, attitude, and intention (cf. Triandis, 1964). For example, intention to recycle hazardous materials may not correlate with frequency of recycling batteries, because people may intend to perform the more general behavior of recycling hazardous materials but not intend to perform the specific behavior of recycling batteries.
Adherence to these principles should improve the precision of behavioral prediction, and consequently, the effectiveness of persuasive efforts. Remarkably, however, although these principles are as relevant for the prediction of behavior today as when they were first introduced, they continue to be violated in research that applies reasoned action theory (Hale, Householder, & Greene, 2002; Trafimow, 2004). This has important implications. For example, it has been shown that measurement in accordance with the compatibility principle strengthens relations among reasoned action variables, which suggests that studies that do not adhere to this principle underestimate the ability of reasoned action variables to explain intention and behavior (Cooke & Sheeran, 2004; van den Putte, 1993).
Key Components and Their Relations
* * *
Reasoned action theory has three structural parts that together explain behavior formation: (a) the prediction of behavior from behavioral intention; (b) the explanation of intention as a function of attitude, perceived norm, perceived behavioral control, and their underlying beliefs; and (c) the exposition of beliefs as originating from a multitude of potential sources. I will use this partition to structure a discussion of issues related to each reasoned action component and the proposed relations between components.
Behavior
The precision with which behavior can be predicted improves when specific behaviors rather than behavioral categories or goals are measured, and when the behavior that one wants to predict is measured at the same level of specificity as the variables that are used to predict it. Another noteworthy measurement issue has to do with the question whether behavior should be observed or assessed with self-report measures.
Whereas for pragmatic reasons most reasoned action research uses self-reports of behavior, observed behavior has an intuitive appeal because it does not, or at least to a lesser extent, suffer from validity issues known to affect self-reports of behavior (Albarracín et al., 2001). Key among those is that self-reports of behavior can be exaggerated (e.g., male’s reports of sexual activity; Brown & Sinclair, 1999) or understated (e.g., reports of at-risk health behavior; Newell, Girgis, Sanson-Fisher, & Savolainen, 1999). Regardless of whether these biases are deliberate or reflect fallible cognitive estimation processes (Brown & Sinclair, 1999), they render behavioral self-reports less than perfectly accurate. This does not mean that prediction of observed behavior is always more precise than predictio
n of self-reported behavior.
Consider, for example, Armitage’s (2005) study of physical activity among members of a gym. Armitage measured attitude, perceived norm, perceived control, and intention at baseline with items framed in terms of “participating in regular physical activity.” At a three-month follow-up he assessed behavior by both asking gym members enrolled in his study “How often have you participated in regular physical activity in the last 3 months?” and by electronically logging gym entrance. Clearly, baseline measures were more compatible with the self-report behavior measure than with the observed behavior measure. As just one example, when people think about regular physical exercise, they may think about activities outside the gym that are not reflected in records of gym attendance, but that likely are reflected in self-reports of physical exercise. In support of this contention Armitage found a stronger correlation of intention to participate in regular physical exercise with self-reported regular physical exercise, r = .51, than with records of gym attendance, r = .42. This finding has been corroborated in meta-analytic research (Armitage & Conner, 2001; but see Webb & Sheeran, 2006).
A moment’s reflection shows that the attitude, perceived norm, perceived control, and intention measures that Armitage used would have been more compatible with, and thus more predictive of, the self-report behavior measure used three months after baseline if the former would have asked about “participating in regular physical activity in the next three months.” This is an issue that affects many prospective studies. Interestingly, however, discussions about improving behavioral prediction predominantly focus on variables that possibly moderate effects of reasoned action variables on self-reported behavior, and remain largely silent on measurement of behavior itself (for a notable exception, see Falk, Berkman, Whalen, & Lieberman, 2011). To be sure, moderator analysis has important potential for determining when the theory’s propositions are particularly likely to apply, which not only directs investigators to appropriate application but also suggests areas for further theory development (Weinstein & Rothman, 2005). Even so, the scarcity of work that tests the validity of self-report behavior measures, for example, by assessing compatibility between behavioral determinant and behavior measures, is striking (Albarracín et al., 2001).
Behavioral Intention
Behavioral intention is the most immediate determinant of behavior. It is defined as people’s readiness to perform a behavior: “Intentions are assumed to capture the motivational factors that influence a behavior; they are indications of how hard people are willing to try, of how much of an effort they are planning to exert, in order to perform the behavior” (Ajzen, 1985, p. 181). Intention is indicated by the subjective probability of behavioral performance, that is, by people’s estimate of how likely it is that they will or will not perform a particular behavior. Examples of widely used intention items are How likely is it that you … (followed by the definition of the behavior under investigation; scale anchors I definitely will not—I definitely will) and I intend to … (scale anchors I completely disagree—I completely agree).
The intention concept and its operationalization have not been universally accepted, however. Concerned about the sufficiency of intention as the only variable that directly determines behavior, investigators have proposed several alternative intention concepts and measures. This section reviews three such measures.
Warshaw and Davis (1985) proposed that behavioral expectations, or people’s self- predictions regarding their behavior, are superior to behavioral intention in predicting behavior, because behavioral expectations take possible barriers to behavioral performance into account more so than intention. Items such as I expect to … and I will … (scale anchors highly unlikely to highly likely) are commonly used to measure behavioral expectation. Empirical findings suggest that behavioral expectation measures do not outperform intention measures (Armitage & Conner, 2001; Fishbein & Stasson, 1990; Sheeran & Orbell, 1998; but see Sheppard, Hartwick, & Warshaw, 1988), and it is not uncommon to combine the two types of measures into a single intention scale (e.g., Fielding, McDonald, & Louis, 2008).
Gibbons, Gerrard, Blanton, and Russell (1998) proposed behavioral willingness as another alternative for intention. Gibbons and colleagues argued that an intention to act implies rational deliberation, whereas behavior often is irrational and triggered by situational factors. Developed in the context of health-risky behavior, the behavioral willingness hypothesis holds that people may intend to engage in safe behavior, but be willing to engage in risky behavior if the situation would offer opportunities for doing so. For example, someone may intend to have no more than three drinks at a party, but drink more when at the party an attractive person offers a fourth drink. Similar to this example, behavioral willingness measures ask whether people would be willing to engage in a particular behavior given a particular scenario, that is, under specified circumstances. It is therefore unclear whether behavioral willingness is truly different from intention or simply a more specific intention (Fishbein & Ajzen, 2010).
Gollwitzer’s (1999) concept of implementation intentions offers a greater contribution to behavioral prediction. Implementation intentions are highly specific plans people make about when, where, and how to act on a motivation to act, that is, on their intention to act. There is evidence that implementation intentions improve the prediction of behavior (e.g., Ziegelmann, Luszczynska, Lippke, & Schwarzer, 2007), but not always (e.g., Budden & Sagarin, 2007; for a review, see Gollwitzer & Sheeran, 2006). Instead of a viable alternative to the intention variable, implementation intentions are perhaps better interpreted as a useful moderator, such that people who formed positive intentions are more likely to act on their intentions if they have also thought about how to implement their plans.
Predicting Behavior From Intention
Reasoned action theory has been able to account for behavior with a good measure of success. For example, meta-analyses of studies that prospectively examined behavior found intention-behavior correlations to average around r = .45 (e.g., Albarracín et al., 2001; Armitage & Conner, 2001; Cooke & Sheeran, 2004; Hagger, Chatzisarantis, & Biddle, 2002; Sheeran & Orbell, 1998; Sheppard et al., 1998). Whereas these average correlations usefully indicate the theory’s general ability to account for behavior, it is important to understand which factors increase or decrease the strength of association between intention and behavior. Before discussing two such factors, I first address an important methodological implication of the hypothesis that intention predicts behavior.
Testing Prediction
To test the hypothesis that intention predicts behavior, behavior should be measured some time after the variables that theoretically predict it were measured. Because behavior assessed at a certain time point indicates what people did at that same time (for observed behavior) or have done prior to that time (for self-reported behavior), correlating cross-sectional intention and behavior data produces a causal inference problem (Huebner, Neilands, Rebchook, & Kegeles, 2011; Webb & Sheeran, 2006; Weinstein, 2007). A cross-sectional intention-behavior correlation indicates the extent to which intention is consistent with people’s past behavior, and should not be interpreted as prediction of future behavior. Unfortunately, intention-behavior correlations obtained from cross-sectional designs are still being published as tests of behavioral prediction (e.g., de Bruijn, Kremers, Schaalma, Van Mechelen, & Brug, 2005; Keats, Culos-Reed, Courneya, & McBride, 2007; Kiviniemi, Voss-Humke, & Seifert, 2007).
Lagged measurement is challenging, both for methodological and budgetary reasons. It is therefore not surprising that cross-sectional studies greatly outnumber prospective studies. For example, Albarracín and colleagues (2001) collected 96 samples for their meta-analysis, but of these, only 23 could be used to test the theory’s ability to predict behavior. Similarly, Armitage and Conner (2001) obtained correlations from 185 samples, yet only 44 of these provided lagged intention-behavior correlations, and of the 33 samples that Cooke and French (2008) analyzed, 19 could be
used to test intention effects on behavior (but see Hagger et al., 2002, for a higher ratio). This means that although reasoned action theory was designed to predict behavior, it is primarily used to explain intention. This gives pause for reflection: Despite the thousands of reasoned action studies now in existence, only a fraction provides a convincing test of this key aspect of the theory.
Moderators of Intention Effects on Behavior
At least two factors determine the strength of intention-behavior relations. To begin, intention should affect behavior to the extent that intention is temporally stable. If between assessments of intention and behavior nothing happens that might change someone’s intention, then intention data should predict behavioral data. However, if intention changes between assessments because, for example, someone is exposed to a persuasive message, then the behavior data reflect an intention formed after intention data were obtained. The longer the gap between assessments of intention and behavior, the more likely it is that intention changes, thereby attenuating the intention-behavior correlation. Sheeran and colleagues (Sheeran & Orbell, 1998; Sheeran, Orbell, & Trafimow, 1999) found empirical support for this idea. For example, in a meta-analysis of 28 prospective condom use studies, Sheeran and Orbell (1998) found that intention-behavior relations were stronger when the time between measurement of intention and behavior was short rather than long. Note, however, that there is no gold standard for the optimal time lag between intention and behavior assessments, in part because it is near impossible to predict when people will be exposed to factors that influence their intention.
The SAGE Handbook of Persuasion Page 25