Abstract
Cognitive-dissonance theory was proposed in the mid-1950s (Festinger, 1957). Since then, the theory has generated research on the dynamic interplay between cognition (broadly defined), emotion, motivation, and behavior (e.g., E. Harmon-Jones, 2019). This research has led to an increased understanding of psychological processes and responses associated with expectancy violations, belief challenges/disconfirmations, difficult decisions, selective exposure to information, acting inconsistently with one’s beliefs/attitudes, and much more. Moreover, the theory has been tested in a range of species, from rats (Lawrence & Festinger, 1962; Lydall et al., 2010) to capuchin monkeys (Egan et al., 2007) to humans (E. Harmon-Jones, 2019). It has also been applied to marketing (Telci et al., 2011), management (Hinojosa et al., 2017), economics (Goetzmann & Peles, 1997), education (McFalls & Cobb-Roberts, 2001), politics (E. Harmon-Jones et al., 2020; Mullainathan & Washington, 2009), religion (Batson, 1975), health behavior (Stone & Fernandez, 2008), clinical psychology (e.g., motivational interviewing; Draycott & Dabbs, 1998a, 1998b), and many other domains.
According to the theory, cognitions, defined broadly as elements of knowledge, can be psychologically relevant or irrelevant to each other. If they are relevant, then they can be consonant (consistent) or dissonant (inconsistent) with each other. If they are relevant and dissonant, then they might evoke dissonance, an unpleasant state that motivates one to reduce the inconsistency. The magnitude of dissonance, this unpleasant motivational state, is a function of the ratio of dissonant to consonant cognitions, with each cognition weighted for its importance. Festinger (1957), who was a smoker, used the example of a smoker to illustrate the theory. Consider Leon, who smokes cigarettes regularly and knows that smoking is bad for his health. Assuming Leon wants to be healthy (and that both smoking and health are important to him), this knowledge is dissonant with his smoking behavior. Thus, he should experience dissonance, and he may reduce it by adding consonant cognitions (e.g., “Smoking keeps me thin”), subtracting dissonant cognitions (e.g., “Lots of things harm your health; I can’t do everything healthy”), increasing the importance of consonant cognitions (e.g., “Staying thin and beautiful is more important than living forever”), or decreasing the importance of the dissonant cognitions (e.g., “If you don’t smoke, you will live to 100, but it will sure seem like a helluva lot longer”).
One research paradigm used to test dissonance theory is the induced-compliance paradigm. This paradigm is based on Festinger and Carlsmith’s (1959) classic experiment in which participants were paid $1 or $20 to tell the next participant (who was actually a confederate) that boring tasks were interesting. Thus, their attitude toward the boring task was inconsistent with their behavior of saying the task was interesting (counterattitudinal behavior). As predicted by the theory, participants paid $1 for their counterattitudinal behavior came to have more positive attitudes toward the task than participants paid $20. Theoretically speaking, this occurred because the $1 group had fewer cognitions consonant with their behavior than the $20 group. In this situation, because the behavior is harder to undo than the attitude, the behavior is the cognition most resistant to change, and thus, it is the generative cognition (Beauvois & Joule, 2019). Thus, cognitions are considered consonant or dissonant with the behavior (not the attitude).
Subsequent experiments manipulated perceived choice (low or high) to engage in the counterattitudinal behavior rather than financial incentives. Theoretically, participants given low choice to engage in counterattitudinal behavior have more consonant cognitions to engage in counterattitudinal behavior than participants given high choice, and thus, they should experience less dissonance and change their attitudes less than participants given high choice. Dozens, if not hundreds, of experiments have provided evidence consistent with this prediction. As Vaidis et al. (2024) noted, “classic cognitive-dissonance designs were extensively criticized in the 1960s (see Chapanis & Chapanis, 1964), in which important methodological limitations and inadequacies in the statistical analyses were pointed out.” However, these flaws were addressed in subsequent research, as Wicklund and Brehm (1976) noted in their well-cited book reviewing the status of the theory and its research.
Vaidis et al. (2024) reported the results of a “multilab replication of the induced-compliance paradigm of cognitive dissonance.” The take-home message from their abstract is that dissonance created by a choice manipulation does not influence attitude change. To quote directly, “The primary analyses failed to support the core hypothesis: No significant difference in attitude was observed after writing a counterattitudinal essay under high choice compared with low choice.” Failures to replicate have the potential to unfairly discredit the field of psychology, in our opinion. Null effects are difficult to interpret because many factors may interfere with detecting a significant effect. We outline a few of these potential factors below. Although a null effect could represent a nonexistent effect, the thresholds set for a statistically significant effect may also lead to a Type 2 error or false negative (Braver et al., 2014; Cumming, 2008).
We do not always know what will cause a psychological experiment to work. But we do have a few thoughts about what might make an induced-compliance experiment more likely to work based on reading past research (e.g., Beauvois & Joule, 1996; Rabbie et al., 1959); learning from Jack Brehm (Brehm, 1956; Brehm & Cohen, 1962) and Jud Mills (Aronson & Mills, 1959; Mills, 1958), who were early pioneers in dissonance research (E. Harmon-Jones et al., 1996; E. Harmon-Jones & Mills, 1999); and conducting induced-compliance research (e.g., E. Harmon-Jones & Harmon-Jones, 2023).
We were informed that experts had approved of the methods that were used in Vaidis et al.’s (2024) replication. We did not provide any advice on the design of this replication. Experts have occasionally unwittingly provided bad advice throughout history (e.g., Titanic, USSR, Bay of Pigs, widespread use of leeches to treat health problems). In addition, dissonance-theory researchers disagree on what constitutes a good test of the induced-compliance effect on attitudes. For example, Jack Brehm advised us to use very short counterattitudinal statements because research had revealed that longer essays reduce the strength of the attitude change effect (e.g., Beauvois & Joule, 1996; Rabbie et al., 1959). In addition, short counterattitudinal statements help eliminate some alternative explanations (e.g., self-persuasion; Janis & King, 1954). In the replication, Vaidis et al. appear to have instructed participants to write longer essays (i.e., more than one sentence).
Considering Perceived Choice
What might have led to the replication results that both low- and high-choice counterattitudinal conditions caused attitude change and self-reported discomfort/conflict (compared with a high-choice/neutral essay condition)? One of the first things that caught our attention was the choice manipulation-check results. The mean on perceived choice for the low-choice condition was at the midpoint of scale; thus, these participants believed themselves to have a moderate freedom of choice. In the early days of induced-compliance research, this condition was often referred to as a “no-choice condition” (e.g., Linder et al., 1967). However, Vaidis et al.’s (2024) low-choice condition was more like a moderate-choice condition. Perhaps because these “low-choice” participants believed themselves to have choice to engage in the counterattitudinal behavior, they experienced dissonance-related discomfort and changed their attitudes. If these “low-choice” participants had rated themselves as having much lower choice, they may not have changed their attitudes.
To compare the Vaidis et al. (2024) manipulation-check perceived-choice ratings to those from past published studies that used different scales, we converted these mean ratings to percentages, where 0 = no choice and 100 = maximal choice. In the experiment by Croyle and Cooper (1983, Experiment 1) that was replicated by Vaidis et al., the rating of perceived choice in the low-choice condition was 31.93, which is almost 20% lower than the rating of perceived choice in the Vaidis et al. replication study (49.33). In several other induced-compliance studies, low-choice condition means were near the lowest point on the perceived-choice scale (e.g., Blanton et al., 1997; Elkin & Leippe, 1986; Friedman & Arndt, 2005; Gosling et al., 2006; C. Harmon-Jones et al., 2011; E. Harmon-Jones, 2000; E. Harmon-Jones et al., 1996, 2008, 2011; Scheier & Carver, 1980; Simon et al., 1995). Moreover, in some of these past experiments, the means for the high-choice condition’s choice manipulation check were lower than the means for the replication study’s low-choice condition. For example, one induced-compliance study had choice manipulation-check means of 17.86 and 35.71 for low- and high-choice conditions, respectively (E. Harmon-Jones et al., 2011), compared with the replication study’s choice manipulation-check means of 49.33 and 72.22 for low- and high-choice conditions, respectively (Vaidis et al., 2024). We can only speculate on what may have caused this higher perceived choice in the low-choice (and high-choice) conditions, but we do wonder if the modern lengthy consent forms that emphasize that participation is voluntary and that participants can withdraw at any time may have contributed to a greater sense of freedom of choice. In any event, replication researchers would be wise to follow the advice of Stroebe and Strack (2014), who suggested pretesting materials from older studies to make sure they still function as expected before using them directly because things may have changed over time.
We suspect that this greater perceived choice in the low-choice condition of the replication study may be primarily responsible for the attitude change that was observed in the low-choice condition. Why the low-choice condition did not produce less attitude change than the high-choice condition is still a mystery. One possibility is that perceived choice may not be linearly related to dissonance arousal/discomfort and reduction, particularly when choice is measured with a single item that is confusingly worded (e.g., Vaidis et al., 2024, used “How free did you feel to decline to write the essay?,” whereas Scheier & Carver, 1980, used a more directly worded question: “How much choice did you have over the content of your essay?”). Perhaps choice may have more of a dichotomous or categorical effect on dissonance processes. That is, once individuals pass some threshold on perceived choice, they are committed to that action. Brehm and Cohen (1962) described choice as being critical for commitment, and commitment was necessary for dissonance reduction in the form of attitude change to occur in the induced-compliance paradigm. These ideas about choice and commitment were pursued in much research by Kiesler (1971) and Beauvois and Joule (1996), and all of these explorations seemed to consider commitment as a binary variable (a person is either committed or not committed). Past research on the induced-compliance paradigm, for the most part, did not report correlations of perceived choice and attitude change. If it is true that commitment/choice is a binary variable, then the choice ratings from Vaidis et al. (2024) may indicate that participants in both the low- and high-choice conditions in Vaidis et al. perceived themselves as having high choice to write the essay.
The Replicated Experiment Worked Only 50% of the Time
However, if one assumes a linear effect of choice on attitude change, then other explanations may exist for the lack of difference between the two counterattitudinal conditions. First, Vaidis et al. (2024) replicated the method of Croyle and Cooper’s (1983) Experiment 1. Experiment 1 of Croyle and Cooper found an attitude-change effect when low choice was compared with high choice. However, Experiment 2 in the Croyle and Cooper article did not find a significant effect of choice on attitude change. Thus, in the original experiments, only one out of two experiments worked according to predictions. Croyle and Cooper attributed the lack of an attitude-change effect in Experiment 2 to the fact that participants wore physiological recording equipment (for measuring skin conductance and heart rate) in Experiment 2 but not in Experiment 1. They wrote, In the second experiment, subjects in the high-choice counterattitudinal condition did not display any overall attitude change. Since the induced compliance procedure in Experiment 2 was a replication of that used in Experiment 1, the mere presence of the physiological recording device may have played a direct role in the prevention of attitude change. It seems likely then that subjects misattributed their heightened arousal to some aspect of the physiological recording situation. (p. 788)
However, several dissonance experiments have had participants wear physiological recording equipment and still observed the predicted induced-compliance attitude-change effect (e.g., E. Harmon-Jones et al., 2008, 2011; Martinie et al., 2013). So, it is possible that other variables contributed to Croyle and Cooper failing to replicate themselves in the original publication. Note that conceptual variables such as “counterattitudinal behavior” and “choice” can be operationalized in myriad different ways. We recommend that researchers who want to test whether a conceptual effect is replicable choose an operationalization that has been found to be robust.
COVID Precautions and Dissonance Processes in the Lab
Second, the Vaidis et al. (2024) study was run during the midst of the COVID pandemic, and folks were wearing masks, practicing social distancing, and so on. Vaidis et al. argued, however, that these precautions had already become a norm for almost a year when the data collection began in most labs. Mask wearing specifically might even have served to some extent in neutralizing facial expression variations during interactions. Thus, it is challenging to explain the results and the absence of effect as due to any of the above circumstances. (p. 40)
In contrast to Vaidis et al. (2024), we suspect that it is possible that these COVID times may have prevented the observation of the predicted dissonance effect. For example, some participants may have misattributed their dissonance arousal to the COVID pandemic, and as already noted, several experiments have found that misattribution to another source of arousal decreases attitude change in the induced-compliance paradigm (e.g., Losch & Cacioppo, 1990; Zanna & Cooper, 1974). In addition, contrary to what Vaidis et al. speculated, several studies have reported that negative affect increased over 1 year after the onset of the pandemic (e.g., Benke et al., 2023). Thus, the wearing of face masks, social distancing, and so on may have made these anxious concerns more prevalent in the lab and thus more likely to act as misattribution cues.
In addition, the presence of the pandemic may have increased the likelihood that participants would use other dissonance-reduction methods (that were not measured). One possible one is trivialization, or mentally reducing the importance of the dissonant cognitions (Simon et al., 1995). We can imagine some participants thinking something like “Who cares about this counterattitudinal behavior; we are in the midst of a global pandemic!”
Experimenter Behavior and Dissonance Processes
In addition to the pandemic and its precautions likely increasing anxiety, experimenter behavior can increase anxiety (e.g., Dimitroff et al., 2017), potentially masking an effect of choice. We viewed a video tutorial by a replication experimenter who talked very fast and seemed a bit nervous. Fast-talking nervous experimenters may provide another misattribution cue and other possible sources of error variance. These cues may evoke unwanted emotions, motivations, and responses, which may be enough to wipe out a small and delicate experimental effect such as the one obtained by Croyle and Cooper (1983). We believe that these possible experimenter and lab effects should be investigated in replication studies to help understand psychological effects. For example, assessments of experimenters’ personality and behavior and participants’ perceptions of experimenters could be collected to test the effects of the social interaction on outcomes. Such would move away from a judicial decision of “The effect exists versus the effect does not exist” to a more nuanced and likely accurate evaluation of moderator variables that increase versus decrease the effect size.
Considering Events Before the Induced-Compliance Experiment
Another difference between the Vaidis et al. (2024) replication study and most past induced-compliance studies is that Vaidis et al. had participants complete another study before engaging in the induced-compliance experiment, whereas most past studies did not. Vaidis et al. described their method as follows: “The study was presented as a set of smaller studies, starting with a neutral filler task that was alleged to be the main study. After the filler task, participants completed the essay task.” As noted above, Croyle and Cooper’s (1983) Experiment 1 found an effect of choice on attitude change, whereas their Experiment 2 did not. Paralleling these attitude effects, Croyle and Cooper’s Experiment 1 did not have participants complete another study before engaging in the induced-compliance experiment, whereas their Experiment 2 did. These methodological differences could explain why Croyle and Cooper did not replicate their own attitude-change effect. The same may have occurred in the Vaidis et al. replication study. It is not clear why the first “filler” study may have eliminated the choice effect on attitudes, but one possibility is that the filler study may have caused some participants to perceive the relevant cognitions (e.g., attitude, behavior) as less important. Dissonance theory predicts that the magnitude of dissonance is reduced when the relevant cognitions are perceived as less important.
Is the Induced-Compliance Effect on Attitudes Weak?
All of these above concerns may make it seem like the induced-compliance effect is weak. It may be. However, we note that the paradigm is designed to place individuals in a state of motivational conflict between their attitude and behavior such that they may be looking for easy ways out of this conflict that does not involve attitude change. As Festinger (2019) noted in his 1987 address at the 95th Annual Convention of the American Psychology Association, The early experiments emphasizing predictions from the theory that were counterintuitive generally blocked off every conceivable avenue of dissonance reduction that we could block off, so that whatever effect there was would show itself in attitude change. But in the ordinary world and if the experimenter is not very careful, a little bit sloppy, there are lots and lots of avenues of dissonance reduction, and those have never been explored. (p. 291)
Since these early experiments, dissonance researchers have explored the wide variety of ways in which participants may reduce dissonance (e.g., trivialization, distraction, self-affirmation; E. Harmon-Jones, 2019).
Conclusion
In the end, Vaidis et al. (2024) still found an effect of “counterattitudinal” behavior on attitude change, and this may have been the result of both low- and high-choice conditions feeling some freedom of choice. Along with Festinger (1957), we believe that choice is a moderator of dissonance, not a necessary condition. As Festinger explained, the ratio of dissonant to consonant cognitions and their importance determines the magnitude of dissonance. Even when individuals are forced to kill others as part of their job, many of these individuals experience severe distress (Maguen et al., 2009); we suspect that this could be due in part to dissonance.
