Process evidence for the question–behavior effect: Influencing socially normative behaviors

Abstract

Answering a question about performance of a behavior influences the probability of a person performing a target action in the future. Although this question–behavior effect has been shown across multiple contexts, several theoretical mechanisms have been suggested to drive the effect. While various explanations have been offered for the question–behavior effect, clear process evidence exists only for the theoretical perspectives of cognitive dissonance and attitude accessibility. In a series of experiments the current research tests the question–behavior effect from these two theoretical perspectives using a set of outcome measures not previously examined. Findings suggest that the act of self-prediction regarding socially normative behaviors generates cognitive dissonance, thereby obtaining greater compliance with subsequent requests to perform the behavior as a dissonance reduction strategy. Implications of the reported findings and directions for future research are provided.

Keywords:

• Question–behavior effect
• Self-prediction
• Normative influence
• Cognitive dissonance
• Attitude accessibility

It has been over three decades since Sherman (1980) published his seminal work demonstrating that asking someone about a behavior can alter the future performance of that behavior. This so-called “self-erasing nature of errors of prediction” (Sherman, 1980) demonstrated that participants making predictions about socially normative behaviors (such as volunteering for a charity) had an increased likelihood of subsequently engaging in the focal behavior. Since that time researchers have studied this effect primarily within two distinct streams of research in marketing: mere-measurement (e.g., Morwitz, Johnson, & Schmittlein, 1993) and self-prophecy (e.g., Spangenberg, Sprott, Grohmann, & Smith, 2003). Similarities between the two streams, across a wide variety of contexts and response frames, motivated convergence of these research traditions under the label of the “question–behavior effect” (Sprott, Spangenberg, Block, Fitzsimons, Morwitz, & Williams, 2006).

Broadly speaking, the question–behavior effect (QBE) finds that asking someone a behaviorally related question impacts the actual performance of said behavior in the future (Sprott et al., 2006). The question types can range from self-predictions to intention measures, with effects being demonstrated across a variety of behaviors (for a full review, see Dholakia, 2010). While mere-measurement researchers have typically demonstrated the effect for various consumer behaviors (e.g., durable and non-durable goods purchases, brand choice, choice of, and loyalty to, service providers, drug and alcohol consumption), self-prophecy demonstrations have most often been associated with socially normative actions (e.g., donating to a charity, recycling, attending a health club, cheating on an exam, voting, gender stereotyping). Although the strength and importance of the effect in terms of social influence have been demonstrated in the literature, process evidence for the effect is lacking. Following the call of researchers in the field (e.g., Spangenberg, Greenwald, & Sprott, 2008; Sprott et al., 2006; Williams, Block, & Fitzsimons, 2006), the current research helps address this gap by contributing to the growing body of evidence speaking to the processes underlying the question–behavior effect.

Although a variety of mechanisms have been suggested to underlie reported question–behavior effects (cf., Sprott et al., 2006), empirical support clearly favors attitude accessibility (as proposed by mere measurement researchers; e.g., Morwitz & Fitzsimons, 2004) and cognitive dissonance (put forth by self-prophecy researchers; e.g., Spangenberg et al., 2003). Both theoretical accounts share the common view that the question itself increases the salience of cognitions, which in turn influence future performance of the associated behavior. For attitude accessibility, the question is theorized to make behavior-relevant attitudes more accessible, thus increasing or decreasing likelihood to perform. In contrast, the cognitive dissonance account holds that questioning evokes social norms regarding the behavior while simultaneously reminding the respondent of past failures to adhere to these norms. This juxtaposition evokes dissonance, which in turn motivates behavior more adherent to social norms.

While evidence for the two leading theoretical processes has been provided in the literature (e.g., by demonstrating theory-relevant boundary conditions), a variety of other generative mechanisms have been suggested for the question–behavior effect and therefore the theoretical picture for what drives the effect is less than clear. One of the potentially limiting factors of prior research is an almost exclusive focus on behaviors as focal outcomes (a situation that is not too surprising given the nature, and name, of the effect). In the current research we address this issue by introducing a series of alternate dependent measures relevant to the QBE—measures suggestive of the processes underlying the effect, therefore providing new insights into theoretical explanation for the effect within the context of socially normative behaviors.

Theoretical Explanations For The Question–Behavior Effect

The question–behavior effect is a remarkably simple approach to behavioral influence: Simply ask a person a question about a future behavior and the likelihood of performing that behavior increases. While prior research has shown the QBE to be robust and of considerable practical importance, much less clear are the theoretical mechanisms underlying the effect. As noted by Sprott et al. (2006), there is considerable variability with regard to the theories that have been proposed and empirically supported for the effect.

Initially Sherman (1980) proposed impression management and script evocation for the effect. Since then, commitment and consistency (Cialdini & Trost, 1998), response fluency (Janiszewski & Chandon, 2007), ideomotor theory (Spangenberg et al., 2008), and implementation intentions (Gollwitzer, 1999) have been suggested as generative mechanisms for the QBE. Of the various theoretical explanations offered, two have received the most empirical support (Sprott et al., 2006) and both share the perspective that questioning activates behavioral cognitions that influence future performance of the focal behavior. These two most compelling theoretical explanations for the QBE are cognitive dissonance and attitude accessibility.

Cognitive dissonance

Making a self-prediction is proposed to cause psychological discomfort by making people aware of a discrepancy between held values (e.g., normative beliefs about performing the target behavior) and past behavior regarding the target action. Thus, by making a self-prediction, people simultaneously become aware of what they should do, in addition to what they have (or have not) previously done regarding the behavior. If these cognitions are discrepant (e.g., “I should recycle, but I throw away many recyclable materials in the trash, even though recycling services are available to me”), cognitive dissonance results. People making a self-prediction thus alleviate associated cognitive dissonance by performing the socially desirable action they would have been less likely to perform.

The pattern of effects reported in prior research supports the notion that a self-prediction can lead people to consider what they have done and what they should do regarding a specific behavior. In particular, a consistent finding in the literature is that control group participants behave in a less socially desirable manner than those who have completed a prediction request, suggesting that people perform socially normative behaviors in a sub-optimal manner, as compared to the norm. It is also clear from prior research that people tend to make self-predictions in the direction of the social norm (e.g., Spangenberg & Greenwald, 1999) indicating that they understand the norms associated with the focal behaviors. Finally, it is reasonable to assume that a behavioral self-prediction is likely to remind people of their failure to perform behaviors as they should. Taken together these research findings provide the necessary ingredients for dissonance to manifest at the time of prediction. More direct empirical evidence supporting a dissonance explanation for the QBE has also been provided by prior research (Spangenberg & Sprott, 2006; Spangenberg et al., 2003). For example, Spangenberg and Sprott (2006) demonstrated self-monitoring to moderate the QBE—a finding that is supportive of a dissonance-based process.

Attitude accessibility

An attitude accessibility explanation for the QBE has also been proposed and empirically supported in the literature (e.g., Williams et al., 2006). According to this theoretical mechanism, questioning makes behavior-relevant attitudes more accessible. These accessible attitudes then influence later performance of the questioned behavior. For example, Morwitz et al. (1993) demonstrated that measurement of purchase intentions increases actual purchase rates of consumer durables; attitude accessibility was these authors’ proposed explanation for this effect.

Empirical evidence for attitude accessibility has been provided in the literature (cf., Sprott et al., 2006). For example, Morwitz and Fitzsimons (2004) showed that prior held brand attitudes become more accessible when asked a general intention question. While attitude accessibility has been primarily put forth to account for non-normative consumer behaviors (in the mere-measurement literature), this theoretical view of the QBE could also account for changes in socially normative behaviors due to questioning. In particular, questioning someone about a socially normative behavior might increase the accessibility or salience of attitudes toward performing said behavior, which in turn could lead to greater consistency between attitudes and behavior (e.g., Fazio & Williams, 1986).

The need for more evidence

Both leading theoretical views of the QBE have been supported in the literature, most often by demonstrating boundary conditions for observed effects (e.g., self-monitoring within the self-prophecy paradigm, Spangenberg & Sprott, 2006; or prior attitudes with mere-measurement, Morwitz & Fitzsimons, 2004). More direct evidence is lacking and has resulted in a call for more research clarifying the underlying processes associated with the effect (e.g., Sprott et al., 2006). The need for process-relevant evidence is particularly important given that the same behavioral outcomes may well manifest regardless of whether dissonance or attitude accessibility is the process underlying observed effects. The current research provides further insight to the QBE using a series of unique dependent measures.

A hallmark of QBE research is a focus on behavior as the primary dependent variable in experimental work. Indeed this dimension of question–behavior research is particularly noteworthy given that much research in the field focuses on cognitions and/or affect, as determinants of behavior. While the study of behavior is laudable, the focus on behavioral outcomes has limited one's ability to draw theoretical insights provided through the use of additional measures. We examine three different variables within the context of socially normative behaviors providing further understanding of the processes associated with the QBE. These variables include: recall of past behavior (Experiment 1); reporting on the behavior of others (Experiment 2); and response latencies using the implicit association test (Experiment 3).

Experiment 1: Recall Of Past Behavior

Experiment 1 tests whether self-prediction biases the report of one's past behavior. The dissonance-based explanation for the QBE clearly suggests that recall of prior behavior should play a particularly important role in manifestation of the effect. As previously noted, self-prediction may increase the salience of prior failings to perform a normative behavior (an outcome that is also plausible from an attitude accessibility perspective, assuming prior behavioral shortcomings are a component of attitudes). If attitude accessibility explains the effect, one would expect unbiased recall of past behavior for those making a prediction; that is, recall in report of past behaviors would not differ from that of a control group not making a prediction.

Prior cognitive dissonance theorizing, however, suggests an alternate pattern of effects likely to emerge regarding past behavior. In particular, Festinger (1957) describes the psychological discomfort experienced as a result of the proportion of dissonant elements. Thus one mechanism available for dissonance reduction would be the reduction of dissonant elements. When confronted by dissonance elicited through making a self-prediction, people could selectively recall past behavior that is consistent with the norm, thereby reducing the proportion of dissonant elements. From this perspective participants who make a self-prediction would subsequently report higher (not lower) levels of norm-relevant behavior than those not making a prediction.

Method

Participants and design

Participants were 56 undergraduates participating for course credit. Students were randomly assigned to one of two between-participants experimental treatments. Treatments included two prediction scenarios: donations to a charity and recycling of aluminum cans.

Materials

All participants completed two self-predictions, the first about job preference (a control prediction drawn from several published QBE studies) and the second was either a prediction about charity donation or recycling (the treatment). These two experimental self-predictions are associated with normative behaviors (i.e., one should donate to a worthy charity and recycle aluminum cans) and have been empirically demonstrated to result in question–behavior effects in published research. Further, participants were all likely to have had opportunity to perform these behaviors and failed to behave consistent with the social norm.

Procedures

The experiment was conducted in two stages. The first stage consisted of a one-page questionnaire with two prediction questions; (1) the job offer and (2) randomly assigned charity donation or recycling prediction. In all cases order of the two prediction choices was counterbalanced to preclude possibility of order effects. The second stage, which followed approximately 1 hour later, with a classroom assignment intervening, consisted of two self-report items, one for past recycling behavior and the other for past charity donation behavior. Thus each group of participants served as a control group for the focal prediction request completed by the other group.

Measures

The dependent variable was a self-reported measure of respective participants’ past actions regarding a behavior for which they had made a prediction and a behavior for which no prediction had been made—charity donation and recycling. Both measures employed 10-point scales (ranging from 0 to 100%) anchored with “absolutely never” and “absolutely every time”.

Results

Analysis consisted of a simple mean comparison of the reported behavior for those who had made a prediction versus those who had not. Participants who had made a self-prediction about recycling in the future reported recycling (67.1%) more in the past year than those who had not made such a prediction (51.4%), t(55) = 2.54, p = .01. Participants who made predictions about donating to a charity reported more charitable donation in the past year (40.3%) than those who had not made this prediction (20.7%), t(55) = 3.33, p = .002.

Discussion

Results of the first experiment demonstrate that a self-prediction biases self-report of past behavior in a normative direction. Our interpretation of this finding is that participants reduced their psychological discomfort (i.e., dissonance evoked by the question) by biasing recall of, or selectively recalling, past behavior consistent with the social norm. These findings are consistent with Aronson's (1968) view of dissonance wherein people view themselves as morally competent. Although the prediction is likely to have reminded participants of their past failings in terms of the behavior, in an effort to reduce the dissonance and restore a positive view of the self, recall was positively biased toward the social norm.

One might argue that findings of Experiment 1 are the result of impression management (cf., Sherman, 1980), but the anonymous nature of the predictions and responses, and the fact that all participants reported past behavior for both recycling and donating, preclude such an interpretation. An attitude accessibility explanation for observed effects is also untenable given that such a view of the QBE would not predict biased recall of prior behavior after being questioned about the focal action. Further, all participants reported past behavior, which should have made attitudes toward the behavior equally accessible across conditions.

Experiment 2: Behavior Of Others

The design of previous QBE experiments focusing on participants’ own behavior was modified by exploring the effects of self-prediction on a person's perception of other people's behavior, following Spangenberg et al. (2003). In contrast to Spangenberg et al. (2003) who examined the dissonance-reducing effects of predicting other people's behavior on a measure of psychological discomfort, the current research explores the direct effects of self-prediction on perceptions of others’ performance of the questioned behavior. The focal behaviors in this experiment (recycling in Experiment 2A and cheating in Experiment 2B) have been shown to be affected by the question–behavior technique in prior research, but in opposite norm-consistent directions. In other words, while the temptation to cheat has been reduced by behavioral questioning, recycling behavior has been increased (see Spangenberg & Greenwald, 2001).

Depending on the process underlying the QBE, a different pattern of results should emerge. If a self-prediction merely increases the salience of attitudes and norms, this salience should not only affect one's own behavior, but also correspondingly affect how that same person predicts the behavior of others. From this theoretical perspective people (once having predicted their own future behavior) should be more likely to predict that others will perform the behavior in a manner consistent with attitudes and social norms (as compared to a control condition), since the accessible cognitions will bias the estimate of others’ behavior.

If dissonance accounts for the effects of self-prediction, however, a different outcome is expected. A dissonance explanation would predict that the self-concept of those making a self-prediction is threatened as they are confronted with their own history of norm violation. From this perspective people making a self-prediction should be motivated to perceive others to have performed in a manner inconsistent with socially normative prescriptions (via downward social comparisons), as compared to a control condition. In response to evoked dissonance, self-predictors can justify their (non-normative) history by concluding that others would have acted similarly. Therefore people making a self-prediction should be more likely to predict that others have (also) performed the behavior in a manner inconsistent with social norms, as compared to a control condition.

While at first glance expectations for this study may seem incongruous with the results of the first experiment; however, these two studies are actually consistent from a dissonance perspective given that responses to both outcome variables can help reduce dissonance for those making a self-prediction. In Experiment 1, a biased self-report of prior behavior helps to avoid a threat to one's self-concept. Similarly, predicting others to act in a less normative fashion (in Experiment 2) also serves as a way to protect the self from dissonance evoked by the prediction.

Method

Participants and design

Undergraduates (Experiment 2A, N = 155; Experiment 2B, N = 113) participated in the experiment for course credit; they were randomly assigned to one of two experimental treatments. The treatments, similar to prior research, included individually administered question–behavior and control prediction conditions.

Materials

In the control prediction condition participants were asked to complete the same job preference question as used in Experiment 1. In the self-prediction condition, participants were asked to predict their own behavior regarding recycling in Experiment 2A and cheating in Experiment 2B.

Procedures

In both replications participants first completed the self-prediction followed by perceptions of other people's behavior. The order of the two prediction choice alternatives was counterbalanced to preclude possibility of order effects.

Measures

The main dependent variable was the perception of other people's recycling and cheating behaviors (i.e., “other students in your college”) using a dichotomous choice option of “yes they will recycle [cheat] or no they will not recycle [cheat]”. After participants had completed these measures, social norms regarding the focal behaviors were collected from them for both behaviors using two 9-point Likert-type scale items: “Students I know make it a priority to recycle”, and “Students I know think it's important to recycle” (recycling items correlated at r = .68); “Students I know resist the temptation to cheat”, and “Students I know think it's important to avoid cheating” (cheating items correlated at r = .64). These items, drawn from Sprott, Spangenberg, and Fisher (2003), were averaged for each behavior and included as covariates in analyses.

Results

As expected, the prediction of one's own behavior had a significant directional influence on how participants perceived the behavior of others regarding both recycling and cheating. Specifically, when asked to make a prediction about their own recycling and cheating behavior, a chi-square analysis found that participants subsequently stated that other people were less likely to recycle 82.1%, χ²(1, N = 155) = 9.36, p = .002, and more likely to cheat 76.3%, χ²(1, N = 113) = 3.75, p = .053, compared to control participants not making a self-prediction, 59.7% and 59.3% respectively.

Logistic regression was used to further assess the role of social norms in the experiment. For both behaviors the perception of other people's behaviors was regressed on the self-prediction factor (predict vs no predict) and social norms regarding the focal behavior. The logistic regression models were significant for recycling, χ²(2A, N = 155) = 30.61, p < .001, and cheating, χ²(2B, N = 113) = 26.79, p < .001. In both models social norms had a significant influence on the perception of others’ behaviors (ps < .001) such that people's beliefs about others matched their own normative beliefs about performing the behavior. As with the chi-square difference tests, there was a similar influence of question–behavior for both behaviors (recycling, β = 1.21, p _1-tail = .001; cheating, β = −.78, p _1-tail = .045). The question–behavior by social norms interaction was not significant for either behavior and therefore not included in either logistic regression model.

Discussion

Results of Experiments 2A and 2B provide additional support for a cognitive dissonance explanation for the QBE—counter to what an accessibility perspective would suggest. In particular, when research participants were asked about other people's behavior after predicting their own behavior, they indicated that others are not likely to avoid cheating or engage in recycling in an apparent attempt to alleviate cognitive discomfort. In other words, participants appear to be engaging in a form of cognitive self-affirmation: “I know the right thing to do, but I’m not too bad of a person if I haven’t done it in the past because that is how most people behave.” Self-esteem is bolstered by lowering one's estimation of others. The non-significant interaction between social norms and self-prediction stands in contrast to the results of Sprott et al., (2003) who demonstrated that norms moderated the QBE. The lack of such an interaction in the current research may be due to the fact that those making a self-prediction actively avoided consideration of norms, which would be difficult to reconcile with their biased perceptions of others who are behaving in a non-normative fashion.

Experiment 3: Attitude Accessibility

Given the results of Experiments 1 and 2 are supportive of dissonance as explanation for the QBE, Experiment 3 was designed to directly examine attitude accessibility as an explanation for the effect in the context of socially normative behaviors. If attitude accessibility underlies the QBE, attitudes toward the behavior would be more accessible after exposure to a prediction request than for those in a control condition, and attitude accessibility after self-prediction should not differ from a condition inducing greater attitude accessibility. Such a pattern of effects would not emerge if cognitive dissonance underlies the effect. From a dissonance-based view of the QBE, levels of attitude accessibility should be lower following a self-prediction as compared to an attitude accessibility condition. A dissonance explanation predicts no difference in attitude accessibility between self-prediction and control conditions. This is because attitudes are not being made more accessible in either of these two conditions relative to an attitude accessibility manipulation. In other words, attitudes are not made more accessible by a self-prophecy manipulation (or in a control condition)—observed QBE effects are driven by alternative cognitions. The design of Experiment 3 directly compares the two main competing accounts for the QBE by directly measuring attitudes toward recycling.

Method

Participants and design

Study 3 included undergraduate marketing students participating in the study for course credit (N = 99). The study employed a between-participants design with three conditions: a prediction request condition (n = 34), an attitude accessibility condition (n = 35), and a control condition (n = 30). Participants were randomly assigned to conditions, completing the study in groups of up to six people.

Materials

Following published QBE work (e.g., Sprott, Spangenberg, & Perkins, 1999), the question–behavior treatment was administered in the form of a survey in which a prediction request appeared second in a series of four prediction scenarios. The other predictions pertained to a variety of innocuous behaviors (i.e., deciding whether to watch a documentary on health care reform; subscribing to a local paper or the New York Times, and choosing between Pepsi or bottled water). The focal prediction asked whether “you will” or “will not recycle” package materials (responses were counterbalanced to reduce the possibility of order effects).

Modeled after published work (e.g., Fazio, Chen, McDonel, & Sherman, 1982; Schuette & Fazio, 1995), the attitude accessibility condition consisted of a survey measuring attitudinal information concerning the same issues focused on in the prediction scenarios. A brief introductory statement and the name of the attitude object (i.e., recycling) preceded a 9-point attitude measure anchored with “bad” and “good”. As with the question–behavior condition, recycling appeared second. The control condition consisted of a survey measuring attitudinal information on a number of issues not related to recycling.

The experiment consisted of two ostensibly unrelated sessions. In the first session research participants completed the survey containing the appropriate experimental treatment. Upon completion of the survey participants were escorted to an adjacent room and seated at desktop computer stations for the test of attitude accessibility. Different experimenters conducted the two sessions to support in participants’ minds that the sessions were unrelated.

Measure

The criterion variable for Study 3 was accessibility of recycling attitudes represented as an index based on response latencies in milliseconds. Inquisit software (Greenwald, McGhee, & Schwartz, 1998) was used to collect response latencies. In particular, the computer program presented participants with a series of trials, each of which included an issue about which participants were asked to express their opinions as quickly and accurately as possible. The issues included in the study were vegetarianism, politics, civil rights, and recycling. During the trials each issue was randomly presented on the screen with either a positive or negative adjective e.g., “Civil Rights: good?” or “Civil Rights: bad?” opinion. Positive adjectives were comprised of “beneficial”, “good”, and “positive”; negative adjectives included “harmful”, “bad”, and “negative”. If the participant disagreed (agreed) with the stated position they were instructed to press the key labeled “NO” (“YES”) on the keyboard. As a reminder of these instructions, the words “NO” and “YES” appeared respectively at the top left and top right corner of the computer screen during each trial.

The dependent variable was an index constructed by subtracting the average response latency for the three non-recycling issues from the average response latency for recycling. This indexing approach has been used in prior research (e.g., Fazio & Williams, 1986) and controls for individual-level variation in response times. The index was based on normalized data (i.e., response latencies were reciprocally transformed). Participants with response latencies greater than 4 seconds were excluded to remove those not focused on the task; participants with response latencies faster than 850 milliseconds were also excluded as most people would be unable to read the screen's content in such a brief period of time.

Results

Pairwise comparisons were conducted to test for differences in attitude accessibility across conditions (i.e., prediction request, control condition, and attitude accessibility condition). A significant difference emerged between the question–behavior and attitude accessibility conditions (t ₆₇ = 2.07, p = .02; r = .25), such that indexed response latencies were faster in the attitude accessibility condition (M = .157 × 10⁻³) than in the prediction condition (M = .228 × 10⁻³). There was no difference in response latencies between question–behavior and control conditions (M = .233 ×10⁻³; t ₆₂ = .17, p = .43) and, as one would expect, attitudes were more accessible in the attitude accessibility condition than in the control condition (t ₆₃ = 1.91; p = .056; r = .23).

Discussion

The findings of Experiment 3 support the hypothesis that dissonance underlies the observed effects. In particular, recycling attitudes were more accessible (as indicated by faster response latencies) for people completing a measure of recycling attitudes than for those making a prediction request. This pattern of results cannot be accounted for by an attitude accessibility explanation for the QBE.

General Discussion

Three experiments were conducted directly testing both attitude accessibility and cognitive dissonance as process explanations for the QBE. The current studies’ focus on socially normative behaviors is most consistent with the literature underlying a dissonance-based view of the QBE. That said, a main contribution of these studies is the use of unique (non-behavioral) dependent measures that have not been used in prior QBE research. The pattern of findings using these alternate dependent measures speaks to the process underlying the QBE, serving as a test between dissonance and attitude accessibility explanations for the effect. Results of the three experiments compellingly support dissonance as the theoretical mechanism for the QBE.

Experiment 1 showed a biased self-report of a socially normative behavior for those questioned about future performance of that same behavior. In particular, when participants are asked to make a self-prediction about a behavior, both the prediction and the reporting of past behaviors are consistent with the societal norm, as compared to those who did not make such a prediction. The prediction arguably reminded participants of their past failings with regard to the behavior in question. In order to reduce the cognitive discomfort associated therewith (i.e., the dissonance), participants restored a positive self-view by positively biasing their recall consistent with the social norm. Attitude accessibility does not explain the observed effects of Experiment 1; such an explanation would not predict biased recall of prior behavior after being questioned about the focal action.

In the second study participants were asked to make a self-prediction and then report perceived behavior of others (not their own behavior). If attitude accessibility underlies the QBE, then the reporting of other's behavior should be biased in a normative direction after making a prediction because attitude and associated norms would be activated. In contrast, this experiment found that those making a self-prediction reported others to engage less often in the normative behavior. This pattern of results supports a dissonance interpretation as participants making a self-prediction could reduce their own dissonance by reporting a downward bias regarding others’ behaviors.

Arguably the most convincing process-related evidence for the QBE in this series of studies is contained in Experiment 3. In this study a relatively unique dependent variable—response latency—was used to better understand the nature of the observed effects. Use of response latencies made it possible to determine how the accessibility of attitudes toward a target behavior (recycling) varied across treatment conditions. The question–behavior manipulation did not make recycling attitudes more accessible to participants in contrast to a typical attitude accessibility manipulation. In fact, attitude accessibility for question–behavior participants also did not differ from that of control participants. These findings support the proposition that a cognitive process other than attitude accessibility drives the QBE. While not interpretable through an attitude accessibility explanation, results of this study are supportive of a cognitive dissonance explanation for the QBE.

The weight of evidence from the current research clearly supports cognitive dissonance as the theoretical process mechanism underlying observed effects for the socially normative behaviors studied. The dissonance framework proposes that making a behavioral self-prediction invokes inconsistent cognitive elements. On the one hand is the norm—what one should do. On the other hand is self-knowledge about conformity to the norm, which may take the form of memories of specific episodes or generalization about prior behavioral performance (e.g., “As a rule, I usually…”). Given that few of us act in normatively appropriate ways all of the time, most people likely experience at least some dissonance when made aware of the difference between self-knowledge and the social norm. When subsequently confronted with the opportunity to perform the relevant behavior, this dissonance motivates norm-appropriate behavior.

Aronson's (1968, 1992) characterization of cognitive changes brought about by classic induced compliance dissonance experiments accords nicely with the QBE. Specifically, Aronson described the typical person as thinking, “I am a decent truthful human being” and “I have misled a person … and I cannot set him straight” (1968, p. 24); opinion change then serves to restore consistency by providing a self-concept-preserving justification for the induced actions. In the typical QBE experiment participants might engage in a similar self-dialogue: “I know what a decent human being should do. I know I have failed to do so in the past. Now that I have an opportunity, I will do the right thing.” In the induced compliance situation people are manipulated into performing a behavior that runs counter to prior attitudes. The QBE twist is that counter-normative behaviors have already been performed, and it is the norm that is induced.

Sherman (1980) favored evocation of norm-related scripts or norm salience as an explanation for the effect. Under this explanation, self-prediction presumably evokes the attendant social norm, associated attitude, and well-learned behavioral response, thereby biasing the prediction. Subsequently, when the opportunity for the behavior is encountered, the norm and scripted response were more available in memory because of their recent access. No one (including Sherman) has provided empirical evidence for this theoretical account. Our experiments, on the other hand, show that merely making the norm salient or accessible does not sufficiently explain the QBE. A norm salience explanation would suggest that those making a self-prediction would have biased the perception of others’ behavior in a socially normative direction as compared to a control condition, an effect we did not find. Consistent with the general question–behavior finding that people show a positive bias in self-predictions for positively normed behaviors, Epley and Dunning (2000) found that people consistently estimate that they are more likely than others to do good things. Further, the overestimation is due more to an upward bias predicting their own behavior, not a downward bias with respect to others.

It is important to note that the focus of the current research is on socially normative behaviors, the same category of behaviors that has been the focus of the self-prophecy researchers originally proposing dissonance as theoretical explanation for the QBE. Other QBE research has demonstrated associated effects for less normative behaviors (such as the purchase of consumer durables; Morwitz et al., 1993). It should be noted that no research has been conducted that demonstrates evidence for QBE-evoked dissonance for less- or non-normative behaviors. Thus future research could usefully replicate the current studies with less-normative behaviors to determine if effects generalize to such behavioral categories. This idea is promising given that researchers have recently shown normative, as compared to non-normative, behaviors to have an impact on manifestation of the QBE (Chandon, Smith, Morwitz, Spangenberg, & Sprott, 2011). Given that most consumer behaviors have some type of normative dimension, it is plausible that a QBE, dissonance-based outcome will manifest for less-normative behaviors.

Further, most (if not all) prior QBE research has focused on behaviors that are under the conscious control of the person completing the prediction request. Therefore future research exploring behaviors beyond control of the predictor could be particularly instructive. Other useful avenues for future research would explore the processes associated with regard to the QBE. One approach would be to use Elliot and Devine's (1994) measure of psychological discomfort as a mediational variable in a typical QBE experiment. Spangenberg et al. (2003) used this approach in a limited fashion, but additional research using this measure of dissonance could produce important results (particularly, if the measure were incorporated within the experimental designs reported herein. For example, if our interpretation of the findings holds true, one would expect that the variables reported in the current studies (e.g., recalled prior behavior) would serve as mediators for the effect of questioning on behavior.

In summary, our series of experiments shows that the cognitive dissonance explanation for the QBE first suggested by Spangenberg and Greenwald (1999) remains compelling. Additionally, the current research is unique in that it significantly differs from previous question–behavior work with regard to dependent measures. While prior research has looked extensively at how questioning changes future behavior or intentions to perform future behavior, these studies used alternate dependent variables allowing us to more clearly understand the process(es) underlying previously observed question–behavior effects. Introduction herein of variables including recall of past behaviors, the behavior of others, and response latencies of behavioral attitudes provides unique and compelling insights regarding how questioning influences people's behavior.