Psychological and philosophical discussions typically understand honesty as reporting truth with propositional statements. In this model, emotions are often seen as irrelevant or a hindrance to honesty, because they can bias our reports. In relational contexts, however, emotions can provide information about deep-seated convictions. We report the results of a study (N = 827) finding that individuals whose emotional responses are congruent with their explicitly stated egalitarian positions are judged as significantly more honest and less prejudiced than those with incongruent emotional responses. This is seen in judgments of white male targets who have negative emotional responses to a black man, a gay man, and a female supervisor. These results suggest that emotional reactions provide information used when judging the honesty and bias of an individual.
Honesty is typically defined as not intentionally distorting the truth (Miller, 2017). In the philosophical conception, honesty involves a deliberate decision to report or act on a propositional statement (e.g. “I did see the man in question on the night of the 13th”). In social situations, however, truth can involve emotional content not captured by explicit propositional statements. Illustrating this, Furtak (2017) reinterprets a philosophical scenario from Roberts (2003) in which an individual says he believes black and white people are equal, yet he feels strong negative emotions in reaction to seeing his sister dating a black man (Roberts, 2003). Roberts argues that this demonstrates how emotion and reason are separate. Furtak argues that the lack of congruence between the individual’s explicit statement and emotional responses reveals that at a deep level the individual does not really know that black and white people are equal. In the philosophical literature, therefore, there are two conflicting theories of the role of emotion. In Roberts’ theory, emotion is separate from propositional knowledge, and so emotional responses cannot undercut beliefs. In Furtak’s theory, the truth is constituted by both the explicit proposition and the learned, emotional association, so that both must be considered when judging whether an individual is making an honest statement; incongruence can indicate that the person does not fully understand himself.
These conflicting claims about the role of associative, emotional processes in truth relate to a theoretical point about bias. Expressing bias against others on the basis of race, gender, and sexual orientation influences judgments of character so strongly that stereotyping and prejudice researchers have suggested people are no longer willing to respond honestly to questions assessing explicit bias (Dowden & Robinson, 1993; Dunton & Fazio, 1997). Measures of implicit bias were developed in part to bypass this socially desirable responding (Fazio, Jackson, Dunton, & Williams, 1995; Greenwald, McGhee, & Schwartz, 1998). Implicit bias toward a group is thought to be an automatic, associative process that is facilitated by experiencing certain negative emotions, including anger, contempt, and disgust (Dasgupta, DeSteno, Williams, & Hunsinger, 2009; Matsumoto, Hwang, & Frank, 2017). Empirical results indicate that it is only weakly correlated with explicit bias and discriminatory behavior (Oswald, Mitchell, Blanton, Jaccard, & Tetlock, 2013). In modern society, people may truly believe they are non-biased and hold explicit egalitarian beliefs, but experience negative emotions in response to minority groups (Gaertner & Dovidio, 2005). As in the philosophical debate described above, this raises a question about whether someone with an explicit egalitarian belief but an implicit, emotional bias against members of a minority group is “really” unbiased or not.
Little is known about how people view and judge those who harbor biased emotional responses while maintaining strong explicit egalitarian beliefs and behavior. Are their explicit statements held out as the true barometer of bias, or do perceivers judge that these targets cannot fully grasp the truth that is embedded in their emotional responses? This incongruence between biased emotional reactions and egalitarian beliefs and behaviors is a serious issue, because it might influence an individual’s ability to appear honest and unbiased in others’ eyes.
Whether emotional reactions are associated with an individual’s core beliefs—such as whether that person is racist—is also closely related to questions that have been addressed by attribution theory in psychology (Jones & Davis, 1965; Heider, 1958). Central to attribution theory is the question of whether behavior can be attributed to internal or external causes (often recapitulated as person versus situation), and factors that lead to one categorization or the other. For example, Kelley’s covariation model proposes that consensus, distinctiveness, and consistency of behavior can be combined to make internal or external attributions about the cause of a given behavior (Kelley, 1973). Traditional attribution theory would appear to apply to the current investigation based on the social desirability of racist behaviors: in the case of the reformed racist, the negative emotional reaction might be considered an undesirable behavior and so should be especially informative about the man’s underlying dispositions. However, a further distinction made in the attribution literature—that of personal and impersonal causation—suggests a complication to the case of inferring honesty from an emotional reaction (Heider, 1958; Malle, 2011).
Personal causality involves a desire to achieve an outcome—such as the desire to act in an egalitarian way— that can be dispositional or momentary, and is associated with intention. Impersonal causation involves elements that arise spontaneously and uncontrollably—such as having a sudden emotional reaction to seeing a member of a stigmatized group—and is associated with unintentional action. Indeed, Malle (2011) uses an emotional reaction—feeling sad—as an example of an impersonal cause. In modern theories of attribution, emotions are unintentional causes of behavior, considered separately from belief and desire (Malle & Knobe, 1997; Malle, 1999).
The role of emotion in judgments of honesty, therefore, leads to theoretical tension in attribution theory: the commonsense conception of honesty suggests that it is about intentional action only (e.g. someone cannot be accidentally dishonest), but attribution theory suggests that emotions should be categorized as unintentional actions. Assuming that honesty should be about intentional actions, current research in attribution theory would therefore suggest that emotion cannot play a role in determining honesty. If emotion does play a role in judgments of honesty, then, it suggests that current understanding of attribution needs to be updated. One theoretical implication would be that emotions have a special status in causal explanations, because they are unintentional in the moment, but are associated with a person’s history of behaviors and mental associations. Emotions may therefore need to be considered differently at different time scales in attribution theory.
Another implication would be that judgments of honesty are determined by the coherence of different sources of information about the individual. While attribution theory does consider personal history, it typically considers beliefs and emotions (as unintentional actions) to be separable causes without emphasizing the ways they may be integrated in an overall judgment of a person’s disposition. The degree to which emotional reactions change judgments of honesty will give a sense of the importance for coherence across belief, behavior, and emotion in making dispositional attributions. Examining people’s judgments of the philosophical scenario described by Furtak and Roberts therefore has the potential to give new insight into two areas of attribution theory: the status of emotions as momentary, unintentional causes, and the role of integration across information sources in making dispositional attributions.
It is further worth noting that the thought experiment—and scenario presented to participants in the experiment below—deals with the case of an individual who earnestly believes he is egalitarian, but still has incongruent emotional reactions based on his earlier life history. The target being judged is not presented as intentionally misleading others about his beliefs or emotions. He simply has not attended to information embedded in his own emotional responses. Our theoretical prediction is based on a conceptualization of emotions as containing rich information about a person’s perceptions of and judgments about the world. Having access to information embedded in one’s own emotions is important, but not always easy. In this conceptualization, perceivers are judging whether a person who doesn’t account for his own emotional reactions is truly honest.
This study examines the way that implicitly biased emotional responses influence judgments of honesty in the context of prejudice. The research addresses two core issues: (1) the role of emotions in judging honesty from a philosophical perspective, and (2) the role of emotional processes in judging whether someone is racist, sexist, or homophobic. Our preregistered hypothesis was that a mismatch between a person’s stated, pro-egalitarian beliefs and his emotional reaction would lead him to be seen as less honest and more biased.
Methods
Participants. 900 online workers from Mturk were recruited to participate in the approximately 10-minute study for $1 each using the service TurkPrime (Litman, Robinson, & Abberbock, 2017). Power calculations indicated that 500 participants were needed to detect a small effect (f2 = 0.025).1 Participants who spent less than one minute completing the survey—typically those who did not complete any items—were not included (N = 73), yielding a final sample of 827.
Procedures. Each participant read three narratives about individuals who were prejudiced and then changed their views. There were two versions of each narrative: one in which the individual in the narrative had emotional responses that were congruent with his changed, anti-prejudiced view, and one in which he had non-congruent emotional responses. The narratives examined prejudice against black people, against women, and against gay people. In all scenarios the individual being described was a white male. Participants were randomly assigned to see one version of each narrative; how often an individual saw a non-congruent version (0 to 3 times) was not controlled within participants, but was evenly distributed across participants. The narrative demonstrating prejudice against black people is reproduced below; the full text of all materials used in the study is provided in the Supplementary Materials.
Robert is a white man who grew up in a region of the United States where black people were regarded as sub-human. He believed that their being elevated to a status of equality was unjustifiable and threatened the very fabric of white civilization. His emotions towards black people were a mixture of fear, resentment, and contempt.
Years later, Robert has become convinced that his earlier beliefs were false. He no longer believes that black people are sub-human, and he believes they should be elevated to a status of equality.
His sister has started to date a black man, and when he sees them together he feels [Congruent: warmth and happiness]/[Non-congruent: revulsion and anxiety]. He always makes sure to speak to his sister’s boyfriend respectfully, asking about his family and trying to get to know him better.
After reading each narrative, participants rated their agreement, with a series of nine statements about the individual being described, using slider scales running from 1 to 100. Four of the statements pertained to the individual’s honesty: (1) asks whether the individual “truly believes” that people are equal (e.g. that black people and white people are equal); (2) whether a pro-equality statement attributed to him (e.g. “Robert says that he truly believes that black people and white people are equal”) is honest; (3) whether the individual is honest with himself; and (4) whether the individual is honest with others. Another statement (5) asks directly whether the individual is biased (e.g. “Robert is racist). In tables and figures, the statements are referred to using the following labels: (1) Belief in Egalitarian Statement, (2) Honesty of Statement, (3) Honesty to Self, (4) Honesty to Others, (5) Not Prejudiced.
Three statements to be used on an exploratory basis for planning future research assess dimensions of personality evaluation seen in ratings of strangers: warmth, competence, and morality. The statements describe the individual in the story as trustworthy (morality), intelligent (competence), and warm (warmth). One statement assessed whether the individual is able to “manage his emotions well.”
Results
Data were analyzed using multi-level modeling, with responses to narratives nested in individuals. Responses to questions regarding honesty were heavily skewed, with many respondents rating the target individual’s honesty as 100 out of 100. To deal with this non-normality, responses were converted to a 0 to 1 scale and a regression using the beta distribution was conducted.2 Beta regression is appropriate when data are bounded between 0 and 1, and accounts for clustering at one end of the scale (Ferrari & Cribari-Neto, 2004). We implemented the model in a Bayesian framework using the R package BRMS (Bürkner, 2016). The model was estimated using a log link, and so we report both the model-estimated regression coefficient in log units and the coefficient converted to raw units. Default priors were chosen for the model.
In all analyses, the effects of the particular narrative and of the order in which the narratives were seen were included as covariates, as well as all of the interactions between experimental condition, narrative, and order. Results indicate that the 95% credible interval for all of these covariates included zero, and so these effects are not considered further here. Descriptive data and correlations among outcome variables, not accounting for nesting within participants, are provided in Table 1. Means by condition for all outcome variables are provided in Table 2.
Descriptive Statistics and Correlations for Outcome Variables. Means, standard deviations, and correlations with confidence intervals.
| Variable | M | SD | 1 | 2 | 3 | 4 | ||||
|---|---|---|---|---|---|---|---|---|---|---|
| 1. Belief in Statement | 66.48 | 29.96 | ||||||||
| 2. Honesty: Statement | 67.07 | 30.47 | .90 | ** | ||||||
| [.90, .91] | ||||||||||
| 3. Honesty: To Self | 64.92 | 31.63 | .87 | ** | .90 | ** | ||||
| [.86, .88] | [.89, .91] | |||||||||
| 4. Honesty: To Others | 64.57 | 31.39 | .88 | ** | .90 | ** | .91 | ** | ||
| [.87, .89] | [.90, .91] | [.90, .91] | ||||||||
| 5. Not Prejudiced | 61.71 | 33.47 | .85 | ** | .83 | ** | .84 | ** | .85 | ** |
| [.83, .86] | [.81, .84] | [.83, .85] | [.83, .86] | |||||||
| Variable | M | SD | 1 | 2 | 3 | 4 | ||||
|---|---|---|---|---|---|---|---|---|---|---|
| 1. Belief in Statement | 66.48 | 29.96 | ||||||||
| 2. Honesty: Statement | 67.07 | 30.47 | .90 | ** | ||||||
| [.90, .91] | ||||||||||
| 3. Honesty: To Self | 64.92 | 31.63 | .87 | ** | .90 | ** | ||||
| [.86, .88] | [.89, .91] | |||||||||
| 4. Honesty: To Others | 64.57 | 31.39 | .88 | ** | .90 | ** | .91 | ** | ||
| [.87, .89] | [.90, .91] | [.90, .91] | ||||||||
| 5. Not Prejudiced | 61.71 | 33.47 | .85 | ** | .83 | ** | .84 | ** | .85 | ** |
| [.83, .86] | [.81, .84] | [.83, .85] | [.83, .86] | |||||||
Note. M and SD are used to represent mean and standard deviation, respectively. Possible values range from 0 to 100. Values in square brackets indicate the 95% confidence interval for each correlation. * indicates p < .05. ** indicates p < .01.
Cell Means and Standard Deviations of Outcomes by Group.
| Congruent | Non-Congruent | |||
|---|---|---|---|---|
| Mean | SD | Mean | SD | |
| Belief in Statement | 82.49 | 20.76 | 50.55 | 29.22 |
| Honesty: Statement | 82.06 | 21.54 | 52.17 | 30.74 |
| Honesty: To Self | 80.84 | 22.18 | 49.09 | 31.68 |
| Honesty: To Others | 81.11 | 21.67 | 48.07 | 30.91 |
| Not Prejudiced | 79.95 | 24.12 | 43.53 | 31.55 |
| Congruent | Non-Congruent | |||
|---|---|---|---|---|
| Mean | SD | Mean | SD | |
| Belief in Statement | 82.49 | 20.76 | 50.55 | 29.22 |
| Honesty: Statement | 82.06 | 21.54 | 52.17 | 30.74 |
| Honesty: To Self | 80.84 | 22.18 | 49.09 | 31.68 |
| Honesty: To Others | 81.11 | 21.67 | 48.07 | 30.91 |
| Not Prejudiced | 79.95 | 24.12 | 43.53 | 31.55 |
Note. Possible values range from 0 to 100.
The main test of our hypothesis was whether or not the emotion congruent condition led to a significant difference in judgments about honesty. Emotional congruence led to significantly higher judgments of honesty of the statement that a person is not prejudiced (b = 1.34, [0.94, 1.75], raw units: 29), honesty of the person to himself (b = 1.14, [0.71, 1.59], raw units: 13), and honesty of the person to others (b = 1.27, [0.84, 1.71], raw units: 24). These results are illustrated in Figure 1A.
A) Honesty Ratings by Emotion Congruence. B) Bias Ratings by Emotion Congruence. Notes: Ridge plots indicate the density distribution of responses in a given condition. The black line indicates the median of the distribution. These ridge plots were formed using the raw data, not the output of the model adjusting for order and narrative. The plots demonstrate that, although the mean differences were significantly different in these conditions, there was not a clear central tendency in the non-congruent condition.
A) Honesty Ratings by Emotion Congruence. B) Bias Ratings by Emotion Congruence. Notes: Ridge plots indicate the density distribution of responses in a given condition. The black line indicates the median of the distribution. These ridge plots were formed using the raw data, not the output of the model adjusting for order and narrative. The plots demonstrate that, although the mean differences were significantly different in these conditions, there was not a clear central tendency in the non-congruent condition.
We also tested whether the emotion congruence condition led to a significant difference in judgments about prejudice. Emotional congruence led to significantly higher ratings on the question asking if the person was not prejudiced (b = 1.26, [0.82, 1.70], raw units: 23), and the question asking if the person truly believed that people are equal (b = 1.34, [0.92, 1.75], raw units: 29). These results are illustrated in Figure 1B. Data, analysis scripts, and full output are included in the Supplementary Materials, hosted at https://osf.io/tfmjp/.
Discussion
Participants judged that individuals who explicitly endorse egalitarian views—saying, for example, that black people and white people are equal—are less honest if they do not have emotional responses congruent with these views. This suggests that people do not judge honesty based solely on explicit propositional knowledge, but that the intuitive conception of honesty accounts for non-propositional, emotional information. This emotional information is used when making judgments about prejudice.
Our pre-registered hypothesis stated that “how honest an individual’s explicit statement is” would serve as our primary dependent variable, but that we were collecting data on related outcomes—including how honest the individual is to himself and to others. Our results found a large effect on our primary outcome of interest, but the pattern of results observed in these secondary variables can also provide insight into the underlying psychological processes. If the target was not judged as honest himself, but the individual was judged as being honest with others, this would suggest that participants saw the target as deluded—responding to the best of his ability, but not being aware of his own biases. If the target was not judged as honest with others, but was judged as honest with himself, this would suggest that participants saw the target as manipulative—knowing that he was biased but misleadingly claiming he was not. Instead, we found that participants judged the target as dishonest in his statement, to himself, and to others. Observed judgments were therefore not consistent with perceptions of the target as particularly deluded or manipulative, but as consistently dishonest.
Our results are in line with Furtak’s philosophical account of emotion as integral to knowledge (Furtak, 2017). An individual who says he believes black and white people are equal while having negative emotional responses to black people is not honest, in this account, because he ignores or has lost contact with his own emotional reactions. Our broader theoretical conceptualization of honesty as a psychological construct involves an important role for emotions. We suggest that honesty, as it is understood by perceivers, does not just involve the accurate reporting of reasoned statements; it must also involve a felt commitment to those statements (Furtak, 2017). True beliefs involve having the right kinds of emotional reactions in relevant situations, and perceivers use information about the consistency of these emotional reactions with propositional statements when judging honesty and character.
In reference to modern attribution theory, our results suggest that honesty is not just a willful act. Emotional responses, which are taken to be unintentional causes in the attribution literature, play into the judgment of honesty. An immediate, uncontrollable emotional response can therefore theoretically render an individual dishonest, despite the fact that the individual is trying to act honestly. This suggests that honesty might require the cultivation not just of a certain set of beliefs, but of habitual emotional responses. Futher, emotions might have different implications for attribution when considered at different time scales; at a larger time scale an individual may be able to become honest through the training of emotional associations. Coherence between emotion, behavior, and belief also have reasonably large effects on dispositional attributions of biased beliefs—23 or 29 points on a 100 point scale, depending on the outcome—suggesting that further research into the integration of information sources would be beneficial in attribution theory.
Emotional reactions are not changed simply through deliberate analysis; they must be cultivated through learned associations and patterns of thinking. There are different practices and discussion in ethics and moral education to enhance one’s self-awareness and to attain the congruence. For example, two of the focuses in Confucian ethical tradition are the practice of rituals (not necessarily religious) and of self-cultivation and reflection (considered jointly). Through ritual practice, including following a certain course of movements in music with other members of a community, the individual has an opportunity to wholeheartedly engage with her or his emotions. The practice also gives the individual a chance to adjust her or his beliefs with the emotions that arise, thereby encouraging frequent adjustment of beliefs in response to emotions. The practice of self-cultivation and reflection on the other hand, stresses the importance of understanding, meaning increasing self-knowledge and thereby changing oneself, through frequently accessing and managing one’s deeper emotions. An individual who is transparent to oneself, as well as to others, is seen as more straightforward and reliable; Confucius himself is an example. This is in line with our result that not being aware of and not being able to manage one’s emotions and the information embedded in them undercut perceptions of the individual’s honesty.
The initial Western philosophical analysis of this situation presented by Roberts (2003) suggests that belief and emotion are two separate spheres. There is no contradiction in the individual in the story saying that he is not racist, based on his reasoned beliefs, but still having emotional reactions consistent with racism. Participants in this study could have provided responses consistent with this understanding of honesty. This would have led to judgments of honesty in the consistent vs. inconsistent conditions to have no systematic difference, and would suggest that participants generally view emotions as unreliable signals of belief. Instead, results showed clear differentiation between the consistent and inconsistent conditions, suggesting that participants view emotions as inherently connected to propositional statements. The statement must be considered honest in light of the emotional responses the individual had.
Academic theorizing suggests that racism is defined primarily by a system of beliefs (Garner, 2010). Beliefs may be assessed by examining the explicit statements made by individuals about their own views. At issue in this study, then, is whether lay observers judge the honesty of a belief statement as separable from an emotional reaction, or as influenced by emotional reactions. Our results suggest that people assume racism and similar oppressive systems of beliefs inherently have an emotional component. Under this interpretation it is not surprising that manipulating emotional reactions leads to changes in judgments of racism; these emotional reactions partially constitute what it is to have that belief. Inconsistent with these results is the idea that, for U.S. participants, beliefs and emotional reactions are separable constructs. Our results support the position that coherence with emotional reactions is at least part of how people judge the honesty of beliefs.
This study provides insight into specific judgments, but many questions remain. This study does not address emotional incongruence with negative or hateful statements (for example, an online troll who posts negative explicit statements but has positive emotional reactions to a diverse friend group). This study also holds constant the identity of the prejudiced person, choosing not to explore biases of people in less powerful groups (e.g., biases a woman might hold against men). The study also does not make normative claims about how people should be perceived.
These conditions provide constraints on the generality of our findings. Our results do not provide a test of emotional congruence when hateful statements are undercut by positive emotion, nor do they provide a test of emotion-to-statement congruence when the speaker is a member of a minority group, nor do they provide evidence for what is right or ought to be done morally. Further, while we find qualitatively similar results when examining judgments of racism, sexism, and homophobia, other forms of bias—such as those against overweight or obese individuals—are not addressed by these results.
It would be theoretically consistent with our results to find that emotional congruence of negative statements with positive emotions are judged as less honest; that judgements of minority speakers would also be influenced by the congruence of their emotional reactions with their statements; and that congruence between emotion and statements would be important when judging people’s character in other situations, such as regards prejudice against overweight others. However, these related questions have not been directly addressed by this research. A failure to find an effect of emotion/statement congruence on judgements of honesty in these contexts would establish a constraint on the generality of our theory, but would not constitute a failure to replicate the core effect we have reported.
We here demonstrate that emotional associations, which are thought to underlie implicit associations, are combined with information about explicit bias when judging an individual’s overall moral character. More broadly, these results suggest an important role for emotion in judging an individual’s honesty.
Data Accessibility Statement
Materials, data, analysis scripts, and outputs in R Markdown are available at https://osf.io/tfmjp/.
The preregistration specifies a sample size of 500, but due to an error more responses were collected. The first author was attempting to collect data on two studies for separate research projects (both dealing with judgments of honesty) on the platform TurkPrime simultaneously. Instead of providing two unique links for each of the two studies, the author put the same study link into the online form used to set up the two studies. Participants who signed up for either study were therefore all routed to the study reported on here. The first author could not identify the source of the additional participants being added to the study immediately, and therefore did not stop data collection until 900 participants (before data trimming) were collected. Analyses using only the first 500 individuals yield identical results, and are presented in Supplementary Materials. All participants were included in the analyses presented here to yield the best parameter estimates for the literature.
The distribution of the data made a regression model based on the normal distribution inappropriate. When conducted using analyses assuming an underlying normal distribution, the results were identical in significance. We chose to report only the results using the beta distribution, despite the fact that they represent a deviation from our preregistered analysis plan, because this model yields more accurate parameter estimates given the observed response distribution.
Funding Informations
AFD and JH acknowledge support from a John Templeton Foundation grant for their postdoctoral fellowships at the Institute for the Study of Human Flourishing at the University of Oklahoma.
Competing Interests
The authors have no competing interests to declare.
Author Contributions
A. Danvers and J. Hu developed the study concept. A. Danvers collected and analyzed data. J. Hu, A. Danvers, and M. O’Neil wrote the manuscript. All authors approve the final version of the manuscript for submission.
Peer review comments
The author(s) of this paper chose the Open Review option, and the peer review comments are available at: http://doi.org/10.1525/collabra.178.pr
University of California Press