Moral judgments and group membership appear to go hand in hand. Interestingly, the existing literature on how group membership specifically alters our perceptions of the morality of others is limited in scope. Intuitively, it stands to reason that people would be more likely to perceive members of their ingroup as morally superior than members of an outgroup, though evidence is needed to back such a claim. One of the best ways to study social perceptions is to examine them at the unconscious level. Automatic, unconscious processes occur without the perceiver’s awareness or intention, and as such are left at the mercy of his or her extant biases and stereotypes. Examining the influence of group membership on perceptions of a target’s morality through the lens of unconscious perception can thus reveal interesting insights into just how biased we are toward our own group members. While there is a breadth of research that has delved into each of these topics separately, little work has been conducted to combine them to examine how each influences the other.
Research on the role of morality in individual judgments has enjoyed an explosion in popularity in recent years. Through this growing line of research, it has been determined that moral evaluations typically hold sway over other evaluations (e.g., competence, intelligence) when an individual is making judgments of another individual or a group (Brambilla, Rusconi, Sacchi, & Cherubini, 2011; Brambilla, Sacchi, Rusconi, Cherubini, & Yzerbyt, 2012; Brambilla, Hewstone, & Colucci, 2013a; Brambilla, Sacchi, Pagliaro, & Ellemers, 2013b; Brambilla & Leach, 2014; Goodwin, Piazza, & Rosin, 2014;), though this work focused on conscious, controlled processes. Indeed, people typically attend more to moral information about a character than to either sociability or competence when determining whether or not that character represents a threat (e.g., Brambilla et al., 2013a; Willis & Todorov, 2006).
Moral attitudes held by individuals tend to be more stable and strongly held than nonmoral attitudes (Luttrell, Petty, Briñol, & Wagner, 2016). Furthermore, these morally-based attitudes have been found to result in more attitude-behavior consistency than nonmorally-based attitudes (Luttrell et al., 2016). Additional research has found that moral construal of an action results in more rapid, more extreme, and more universal judgments of the permissibility of that action than do either hedonic or pragmatic construals of the same actions (Van Bavel, Packer, Haas, & Cunningham, 2012a). This line of research gives a good indication of the strength that morality enjoys when predicting attitude stability and the corresponding behavior and judgments of those behaviors. Moreover, Gantman and Van Bavel (2014) found evidence for a moral pop-out effect, such that participants were more likely to recognize moral words over nonmoral words in a lexical decision task.
With regard to group evaluations, it has been shown that moral judgments of one’s ingroup are more important than judgments of competence or sociability (Leach, Ellemers, & Barreto, 2007). Perceiving one’s ingroup as moral has been shown to lead to more positive outcomes of a group’s self-concept, such that positive moral evaluations of one’s ingroup leads to less distancing from that group and greater group identification (Leach et al., 2007). This line of research further extends to the evaluation of outgroups, with the main finding that moral traits are weighted more heavily when members of one group form impressions about an outgroup (Brambilla et al., 2013a). A limitation of this line of research is its focus on conscious, controlled perceptions of morality. Unconscious perception enjoys an extensive influence on social behavior (e.g., Greenwald & Banaji, 1995), and as such studying morality at the unconscious level may reveal interesting differences in explicit versus implicit evaluations of outgroups.
While previous research has provided a solid foundation for understanding just how important moral judgments are to individuals, more work needs to be done to fully examine how quickly moral judgments are made. Limited work has studied the role of implicit cognition in moral judgments, though there is reason to believe that moral judgments may be susceptible to nonconscious influences (e.g., Ma, Vandekerckhove, Baetens, Van Overwalle, Seurinck, & Fias, 2012; Willis & Todorov, 2006). Given that judgments of morality are deemed to be more relevant than other traits when judging whether a target represents a threat (Brambilla et al., 2013b; Willis & Todorov, 2006), we contend that research into the implicit attribution of moral personality traits is warranted to delineate whether morality is attributed automatically or through cognitive processes. This led to our first hypothesis, which predicts that participants will be more likely to recognize moral (versus nonmoral) traits
Spontaneous Trait Inferences
A spontaneous trait inference (STI) occurs when an individual makes a nonconscious, unintentional judgment about the character of another individual (Winter & Uleman, 1984). These inferences occur without the awareness of the individual making the judgment, and as such have become the hallmark for research into automatic judgments people make about the traits of others. Research into STIs typically involves false recognition paradigms in which participants are first shown a short sentence describing a behavior (encoding task), then are asked whether or not a target word was present in the previously displayed sentence (recognition task), with false recognitions indicating the formation of these inferences (e.g., Rim, Uleman, & Trope, 2009; Wells, Skowronski, Crawford, Scherer, & Carlston, 2011).
STI research has led to a number of interesting findings. One such finding is the interplay between STIs and spontaneous goal inferences (SGIs). Van Overwalle, Van Duynslaeger, Coomans, and Timmermans (2012) found that SGIs are often made faster than are STIs. Additionally, Van Overwalle et al. (2012) noted that while SGIs are often encoded more strongly in memory than STIs (as measured through reaction time), the inclusion of goal descriptions often expedites the process of STI formation.
It is important to note that there exists a dearth of evidence for the existence of STIs made about groups. Indeed, almost all STI research has been conducted about individuals (Hamilton, Chen, Ko, Winczewski, Banerji, & Thurston, 2015). It is important to include group-based research in this line of work, given the importance of group membership and belonging in social interactions (Hamilton et al., 2015). Otten and Moskowitz (2000) found that behaviors implying positive traits about ingroup members led to the formation of STIs more than either negative behavior descriptions or behavior descriptions of outgroup members. Hamilton et al. (2015) have found evidence for the existence of STIs about groups (dubbed STIGs). Importantly, they noted that these STIGs lay a framework for (a) stereotype formation about a group and (b) generalizations about the behavior of an individual based solely on his or her group membership.
In addition to the limited research involving groups, STI research has largely eschewed the study of how purported moral behaviors affect participants’ likelihood of inferring moral traits. In one such study, Ma et al. (2012) found that participants do generate STIs for moral and immoral behaviors, though a limitation of this work is the lack of a nonmoral group of traits to compare it to. Indeed, the lack of this variable makes it difficult to conclude whether moral behaviors increase STIs or immoral behaviors depress STIs. It is important to note that a host of research into impression formation has found a bias for negative behaviors over positive behaviors (for a review, see Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001; see also Skowronski & Carlston, 1989), leading to the intuition that perhaps immoral traits may be more readily inferred over moral traits, independent of the effect of group membership.
Membership in a group is one of the main features of social interaction. It has been established that membership in a group can alter one’s perception of other individuals, with this effect extending to both ingroup and outgroup members (Hackel, Looser, & Van Bavel, 2014). This includes having a skewed, positive outlook toward one’s ingroup members while inhibiting the extension of empathy and mind perception toward outgroup members (Hackel et al., 2014). Mind perception is the process of attributing a mind to another entity, and is an important mechanism for determining what is not only capable of agency (i.e., taking autonomous actions), but is also capable of feeling emotions, pain, and suffering and thus being afforded empathy (Gray, Gray, & Wegner, 2007).
Group membership can alter one’s perceptions of others in a number of ways. One such way is that membership in a group promotes a positive bias towards members of one’s ingroup over members of an outgroup (Lazerus, Ingbretsen, Stolier, Freeman, & Cikara, 2016; Tanis & Postmes, 2005; Van Bavel, Swencionis, O’Connor, & Cunningham, 2012b; Ziegler & Burger, 2011). Indeed, ingroup membership has been found to promote greater memory for ingroup faces (Van Bavel et al., 2012b). Furthermore, Tanis and Postmes (2005) found that participants afforded greater trust to anonymous individuals when they were told they were ingroup members. Lazerus and colleagues (2016) showed that individuals have a positivity bias when judging the emotional expression of ingroup members that did not emerge for outgroup members. Ziegler and Burger (2011) noted that ingroup membership can alter the amount of cognitive resources afforded to processing individuating information about an ingroup member versus an outgroup member depending on a target’s success (or failure) and the respondent’s mood.
Hackel et al. (2014) found that even rudimentary group membership following the use of a minimal group paradigm altered the extent to which minds were attributed to members of one group versus another, such that ingroup members were afforded mental states whereas outgroup members were not. A minimal group paradigm (MGP) is one method of splitting participants into arbitrary groups, and often involves false feedback to participants following some menial task (see Tajfel, Billig, Bundy, & Flament, 1971). Group-based research using an MGP has consistently found that even rudimentary group membership is enough to (a) satisfy our need to belong to a group and thus (b) alter our perceptions of ingroup members and outgroup members (e.g., Hackel et al., 2014; Tajfel et al., 1971; Van Bavel et al., 2012b). Thus, it is apparent that an MGP is sufficient for studying the ways in which group dynamics alter our perceptions and judgments of others.
Splitting participants into groups using an MGP may reveal interesting differences in how we perceive the morality of ingroup versus outgroup members, even on an unconscious level. Such work might hold important implications for how we interact with members of other groups, and why group differences are such a strong influence in our social lives.
The Present Research
While there is a growing body of evidence that implicates group membership in a number of social evaluations (Brambilla et al., 2013a; Hackel et al., 2014; Hamilton et al., 2015; Otten & Moskowitz, 2000; Tajfel et al., 1971), limited research has focused on how perceptions of another’s morality change as a function of whether they are an ingroup member or belong to an outgroup. Otten and Moskowitz’s (2000) work revealed promising effects for ingroup versus outgroup membership STI formation, however their study manipulated group membership as a between-subjects factor and they used RT as a measure of ingroup bias and not recognition rates. Furthermore, research into spontaneous trait inferences (STIs) has yet to examine whether they change when moral traits are involved. Indeed, it remains to be seen whether (a) STIs are more readily formed when the behaviors described are within the moral realm and (b) whether people are more likely to form negative moral STIs for outgroup members and more likely to form positive moral STIs for ingroup members.
We aim to fill these gaps in knowledge. Because of the strength of group membership, it stands to reason that people will view members of their own groups more favorably, especially with regard to morality (e.g., Leach et al., 2007). We hypothesize that moral traits will have a higher recognition rate than nonmoral traits, given the relative emotional salience of moral behaviors over nonmoral behaviors (e.g., Brambilla et al., 2013b; Brambilla & Leach, 2014; Goodwin et al., 2014; Willis & Todorov, 2006), and as such participants will form more moral STIs than nonmoral STIs. Furthermore, we predict that participants will form more STIs for positive traits (both moral and nonmoral) paired with ingroup members while forming more STIs for negative traits (both moral and nonmoral) paired with outgroup members, supporting the idea of ingroup favoritism (e.g., Hackel et al., 2014).
In order to further examine the effects of STI generation within a false recognition paradigm, a process dissociation procedure (PDP) analysis will be used, as described in detail by McCarthy and Skowronski (2011). In the PDP analysis, it is possible to account for both controlled and automatic processes that work in conjunction during the recognition of explicitly presented traits or in opposition for implied traits that are either matched or mismatched with the behavioral information presented. Importantly, the PDP allows one to control for rates of guessing that may arise due to the mismatched traits, allowing a clearer look at how automatic processes control the formation of STIs.
320 participants were recruited via Amazon’s Mechanical Turk (Nmale = 145, Nfemale = 174). Each participant was compensated $0.50 for completing the entire study. Participants had a mean age of 33.33 years, 88.2% spoke English first, 80.3% lived in the United States, and 63.6% were white. 237 participants were in the experimental task that involved group membership while 83 participants were in a control task that examined the effects of morality and trait valence independent of group membership.
We employed a 2 x 3 x 2 x 2 within-subjects design. Group membership status (ingroup, outgroup), trial type (explicit, implicit-match, implicit-mismatch), trait morality (moral, non-moral), and trait valence (positive, negative) served as the independent measures. All of the independent variables were presented in random order to each participant. Trial type, the pairing of group labels with target faces, and the morality and valence of the stimuli were all counterbalanced across participants in a Latin Square design to remove any order effects that may have emerged during either encoding or recognition. In addition, a control task was developed with a 3 (trial type) x 2 (trait morality) x 2 (trait valence) within-subjects design to test the effects of trait morality and trait valence independently of participants’ group memberships. The dependent variable for this study was the rate at which participants indicated either hits (for the explicit trials) or false recognitions (for the implicit trials) for the presented trait words in the recognition task.
To determine the appropriate trait categories, two pretests were conducted. The first pretest concerned participants’ ratings of 336 character traits taken from Anderson’s (1968) list of 555 likeability-rated personality traits. Participants (N = 62) rated positivity on a 1-6 scale, with a rating of 1 indicating an extremely negative trait and a rating of 6 indicating an extremely positive trait. From these results, we selected the 80 most positive, 80 most negative, and 51 neutral traits and used them in the second pretest where participants (N = 63) rated their morality on a scale of -3 (extremely immoral) to 3 (extremely moral). The 30 most moral words and the 30 most immoral words were used to create the moral stimuli while 60 nonmoral valenced stimuli (30 positive and 30 negative) were drawn from those traits directly below the most extremely rated traits, yielding a total of 120 traits: moral, nonmoral positive, nonmoral negative, and immoral.
For the main study, participants were subjected to the minimal group procedure. In this task, each participant saw a pattern of circles presented on the screen and then were asked to estimate how many circles were present in each image (see Figure 1).
After five trials of the MGP task, participants were given randomly generated feedback that placed them into either the “overestimators” or the “underestimators” group, which they were then asked to validate via a forced-response question. Following this, participants were given the encoding task. Here, each participant saw stimuli that displayed a face that was paired with a group label and a behavioral sentence that either explicitly contained or implied a character trait (see Figure 2). In the encoding phase, 1/3 of the sentences explicitly stated traits while the other 2/3 implied them. Throughout the encoding task, participants also responded to three probe questions asking which group label a certain target face belonged to. These served as both attention checks and as manipulation checks to ensure that participants attended to the target faces’ group memberships.
After a series of 120 faces, each participant was given the distracter task. In the distracter task, participants were required to solve a series of 10 arithmetic problems. While the problems themselves were fairly straightforward, they each required a knowledge of the proper order of operations, a manipulation that was believed to be mentally taxing enough to remove any short term memory for the sentences. After completing the distracter task, participants were given the recognition task. In the recognition phase, participants again saw the same faces that were present during the encoding task, this time without the group label above the face. On each trial, participants indicated whether a particular trait was present in the behavioral sentence that appeared with that face (see Figure 3). For the implied traits, half matched the behavioral sentences seen at encoding while half were mismatched. For moral behaviors, positive non-moral traits were used as mismatches, while negative non-moral traits were mismatched with immoral behaviors, moral traits were mismatched with positive non-moral behaviors, and immoral traits were mismatched with negative non-moral behaviors.
This was done to avoid the possibility of participants easily recognizing the mismatched traits through valence alone. Participants also responded to similar group membership manipulation checks in the encoding task, this time with different faces belonging to different groups. F Following the recognition task, participants were asked a series of demographic questions, and then were debriefed. Following the requirements of the IRB, debriefing included both an overall study debriefing and a debriefing form specific to the deception in the MGP. Participants indicated whether they would allow their data to be used or not.
Group Attention Checks
Results for all six group attention checks revealed that participants performed significantly better than chance at determining which faces belonged to which groups, t(244) = 3.26, p = .001, Cohen’s d = .29. This indicates that participants were appropriately attending to and encoding the groups to which the target faces belonged.
Recognition rates were calculated by taking the mean recognition for each within-measures cell, producing a set of recognition rates for each type of trait presented (e.g., explicitly presented moral overestimators, implied-matched nonmoral positive underestimators, etc.), yielding 24 rate variables. Tables 1 and 2 display summary statistics for each condition. To test for any differences that may have emerged for the counterbalanced study sets, we submitted all of the recognition rates to between-measures univariate ANOVA, which returned non-significant results (Fs < 1), indicating no effect for counterbalancing order.
We then submitted the data to a 3 (trial type: explicit, implicit-match, implicit-mismatch) x 2 (morality: moral, nonmoral) x 2 (valence: positive, negative) x 2 (group status: ingroup, outgroup) repeated-measures ANOVA. The analysis found main effects for trial type, F(2, 472) = 79.23, p < .001, ω²p = .283, R²p = .251, morality, F(1, 236) = 23.12, p < .001, ω²p = .085, R²p = .089, valence, F(1, 236) = 40.49, p < .001, ω²p = .142, R²p = .146, as well as a trial type by morality interaction, F(2, 472) = 14.28, p < .001, ω²p = .055, R²p = .057, and a trial type by valence interaction, F(2, 472) = 9.38, p < .001, ω²p = .035, R²p = .038. Figure 4 displays the plots for both the trial type by morality interaction and the trial type by valence interaction. No significant main effect or significant interactions were found for group status (Fs < 1).
To explore the effect of morality and valence on recognition rates independent of group membership, we submitted the control data to a 3 (trial type) x 2 (morality) x 2 (valence) repeated-measures ANOVA. This analysis revealed main effects for trial type, F(2, 164) = 38.23, p < .001, ω²p = .366, R²p = .318, morality, F(1, 82) = 14.39, p < .001, ω²p = .137, R²p = .15, valence, F(1, 82) = 12.26, p < .001, ω²p = .118, R²p = .130, as well as significant two-way interactions between trial type and morality, F(2, 164) = 9.19, p < .001, ω²p = .094, R²p = .101, trial type and valence, F(2, 164) = 17.36, p < .001, ω²p = .179, R²p = .173, morality and valence, F(1, 82) = 13.93, p < .001, ω²p = .133, R²p = .147, and a significant three-way interaction between trial type, morality, and valence, F(2, 164) = 4.39, p = .01, ω²p = .04, R²p = .05 (see Figure 5).
We then decomposed the three interactions revealed above. We first examined recognition rates across trial type for moral traits independent of valence. This revealed a significant effect of trial type, F(2, 472) = 23.08, p < .001, ω²p = .089, R²p = .089. This effect was largely driven by higher recognition rates for the explicit trials (M = .50, SD = .28). There were higher recognition rates for the implicit-mismatch trials (M = .44, SD = .29) than for the implicit-match trials (M = .43, SD = .28), though this difference was not significant, t(947) = 1.45, p = .15, Cohen’s d = .05, indicating no STI effect for moral traits. Our analysis of nonmoral traits also revealed a significant effect of trial type, F(2, 472) = 70.95, p < .001, ω²p = .257, R²p = .231. Again, recognition rates were highest in the explicit trials (M = .51, SD = .29), though this time recognition rates were higher for the implicit-match trials (M = .40, SD = .29) than for the implicit -mismatch trials (M = .37, SD = .29), with this difference reaching significance, t(947) = 2.234, p = .03, Cohen’s d = .07. This indicates that for nonmoral trials, there was some evidence for the formation of STIs.
We approached the interaction between valence and trial type in much the same way. We found a significant effect of trial type for our analysis of the positive traits, F(2, 472) = 27.87, p < .001, ω²p = .107, R²p = .106. Recognition rates were highest for the explicit trials (M = .52, SD = .28), though recognition rates were higher for the implicit-mismatch trials (M = .46, SD = .28) than for the implicit-match trials (M = .45, SD = .28), with this difference failing to reach significance (t < 1, p = .38), indicating no evidence for STIs. We found a significant effect for trial type when we examined the negative traits, F(2, 472) = 69.73, p < .001, ω²p = .253, R²p = .228. Recognition rates were highest for the explicit trials (M = .49, SD = .29), followed by the implicit-match trials (M = .38, SD = .28) and then the implicit-mismatch trials (M = .36, SD = .29). The difference between the implicit-match and implicit-mismatch trials only trended toward significance, t(947) = 1.81, p = .07, Cohen’s d = .06, indicating no effect for STIs.
To decompose the three-way interaction between trial type, morality, and valence we found in our analysis of the control data, we first examined recognition rates for the negative traits with simple main effects tests. This revealed main effects for trial type, F(2, 164) = 45.88, p < .001, ω²p = .426, R²p = .359, morality, F(1, 82) = 27.28, p < .001, ω²p = .238, R²p = .25, and a significant interaction between trial type and morality, F(2, 164) = 3.22, p = .04, ω²p = .026, R²p = .038. Within this interaction, we examined the simple effects of trial type by morality. Our analysis of moral traits revealed a significant effect of trial type, F(2, 164) = 23.2, p < .001, ω²p = .236, R²p = .22. Recognition was highest for the explicit trials (M = .59, SD = .27), while participants recognized more traits in the implicit-match trials (M = .47, SD = .26) than in the implicit-mismatch trials (M = .43, SD = .27). This difference did achieve significance, t(82) = 2.05, p = .04, Cohen’s d = .23, indicating an effect for STIs for negative moral traits. Our analysis for negative nonmoral traits also found a significant effect, F(2, 164) = 29.9, p < .001, ω²p = .297, R²p = .267. Again, recognition rates were highest for the explicit trials (M = .56, SD = .27), though rates were the same in both the implicit-match (M = .35, SD = .25) and implicit-mismatch (M = .35, SD = .28) trials, indicating no effect for STIs (t < 1, p = .93). In our analysis of the positive traits, we found a main effect for trial type, F(2, 164) = 11.42, p < .001, ω²p = .118, R²p = .123, and an interaction between trial type and morality, F(2, 164) = 10.27, p < .001, ω²p = .106, R²p = .112. We decomposed this interaction by first examining the effect of morality on positive traits, which revealed no significant effect (F = 1.18, p = .31). We did find a significant effect for nonmoral positive traits, F(2, 164) = 22.87, p < .001, ω²p = .233, R²p = .219. Again, while recognition rates were highest for the explicit trials (M = .59, SD = .24), we found a significant difference between recognition rates for the implicit-match trials (M = .51, SD = .25) and the implicit-mismatch trials (M = .43, SD = .27), t(82) = 3.18, p = .002, Cohen’s d = .35, indicating that participants were able to form STIs for nonmoral positive traits.
Before proceeding with any PDP analyses, we examined whether there was an overall STI effect by comparing the recognition rates in the implicit-match trials (M = .41, SD = .28) with those in the implicit-mismatch trials (M = .40, SD = .29). This analysis revealed no significant difference (p = .5), indicating that for the experimental group-level analysis, STIs had not been formed. A similar analysis was done for the control data which revealed only slight differences in recognition rates between implicit-matched trials (M = .46, SD = .26) and implicit-mismatched trials (M = .44, SD = .27), t(331) = 1.67, p = .10, Cohen’s d = .09. Thus, neither the experimental group data or the control data found any substantial evidence for the formation of STIs. Due to this, we concluded that a PDP analysis would reveal only spurious effects and was not necessary.
Despite some evidence for the formation of STIs revealed in the simple main effects tests, the fact that we failed to find evidence for STI generation overall gives us little to compare these effects to. Thus, we conclude that much more work is needed to determine whether moral traits are more readily inferred than nonmoral traits. While we did find an effect for nonmoral traits, we cannot compare this effect against the rest of the data, so we can only conclude that there may be a limitation in participants’ willingness to infer moral behavior on the basis of a single behavioral sentence. This conclusion is limited, since participants in the group manipulation condition only inferred nonmoral traits while participants in the nongroup control condition inferred negative (but not positive) moral traits. Despite this, we argue that perhaps morality is a strongly held concept that requires information about an actor’s intent or the behavior’s outcome. Indeed, many studies assessing the effects of moral behavior on impression formation have given participants a host of information from which to form impressions (e.g., Brambilla et al., 2012; Brambilla et al., 2013a; Goodwin et al., 2014). This allows participants to have an in-depth look at the personality characteristics of the targets they are forming impressions of. Conversely, in our study, participants only had one brief behavioral description from which to infer a trait. Moreover, studies assessing perceptions of morality not only include descriptions of the target individual, but also convey (either explicitly or implicitly) that character’s intentions and the outcomes of the behavior. Additional research has led to the idea of a moral pop-out effect (Gantman & Van Bavel, 2014), where it is argued that participants more readily recognize moral words than nonmoral words. While our results ostensibly conflict with this theory, we argue that it is not that simple. The moral pop-out effect is tied to more rapid recognition of a host of moral words, but it does not account for one’s willingness to attribute moral traits on the basis of a single behavior.
Theories of morality often advocate for the necessity of intent when a person is determining whether a given behavior belongs in the moral realm (e.g., Fiske, Cuddy, & Glick, 2008; Gray, Young, & Waytz, 2012; Luo Nakic, Wheatley, Richell, Martin, & Blair, 2006). As is often argued in the literature, while harm (an outcome) is indeed necessary for an action to be deemed immoral, intent is as important, if not more so. For example, a violent storm and a violent serial killer may both involve harm, often in the form of the death of otherwise innocent victims. However, people tend to shy away from attributing a moral violation to the storm that they would to the serial killer, even if the two had a disparate number of victims. This is simply because a storm possesses no agency (Gray et al., 2007). A storm does not intentionally harm someone as a serial killer would. Consequently, because the behavioral descriptions we employed did not contain references to intent, only descriptions of actions, ascriptions of intent were left at the mercy of participants’ assumptions. A person may have returned a lost wallet with all of the money in it because he or she intended to be a good person, or because they simply forgot to check for the presence of cash. Similarly, someone may have threatened to hit another person because they intended to cause them fear and harm or because they believed they were finally taking a stand against injustice. A person’s moral personality is not constructed from any single deed, but instead requires multiple behavioral examples before a moral ascription can be made (Aquino & Reed, 2002). Because morality is such a strongly held concept within each individual’s mind (e.g., Haidt, 2001), it may be the case that single behaviors are not sufficient for an individual to determine a moral character attribute. Instead, perhaps multiple behavioral examples are required for a person to generate a stable inference of another’s morality. As such, moral judgment may be reserved for targets that are socially closer to an individual, as he or she would have a wealth of behavioral information from which to draw upon when attributing traits. Because we did not find an effect for morality within the context of STI generation, perhaps morality is a superordinate category of behavior whose weight dictates that an individual must have a host of information available about a target prior to inferring moral traits.
Another issue complicating the morality angle of our study is the multifaceted nature of morality itself. Research into moral versus nonmoral traits has largely attempted to distinguish moral traits from others (e.g., warmth, sociability, competence; Brambilla et al., 2011; Brambilla et al., 2012; Goodwin et al., 2014). The results of our pretest indicate that morality may not be a subordinate aspect of the warmth dimension, as it is often argued to be (e.g., Goodwin et al., 2014). Instead, our data indicate that morality permeates into aspects of warmth, sociability, and even competence. Perhaps then, morality is a superordinate dimension of personality traits from which the other dimensions are derived. This is not a definite conclusion, however, and more research is needed to delineate the true nature of morality as a personality dimension.
Regarding our findings on group membership and trait attribution, our hypothesis for ingroup bias and outgroup derogation was not supported. Previous research has found a bias toward one’s ingroup for a range of psychological processes, including positive trait attribution (Otten & Moskowitz, 2000), recognition of positive emotional expressions (Lazerus et al., 2016), mind perception (Hackel et al., 2014), reward allocation (Tajfel et al., 1971), higher rates of trust (Tanis & Postmes, 2005), overall rates of face recognition (Van Bavel et al., 2012b), as well as the recognition of individuating information (Ziegler & Burger, 2011). Given such robust results regarding ingroup bias, it is perplexing that we did not find a similar effect of ingroup bias on STI rates. Similarly, Giannakakis and Fisher (2011) found that a salient group identification led to greater ingroup praise and outgroup derogation, though this effect was ablated by making group membership a part of a superordinate category (e.g., recategorizing English speakers as Europeans). Perhaps the large amount of information made group assignment less salient than in other studies, though the fact that participants scored above chance in recognizing random actors’ group assignments indicates that this may not be the case. Instead, perhaps future studies may want to incorporate a more strongly held group categorization that is important to participants’ identities to more effectively delineate the role of group membership on the generation of STIs.
Despite our limited findings, we did find mixed evidence for a negativity bias, though this was inconsistent, as the effect of negative traits only trended toward significance in the group manipulation condition, was significant for the negative moral traits in the nongroup control condition, and disappeared for the positive nonmoral traits in the nongroup control condition. Previous work (e.g., Pratto & John, 1991) shows that bad trait terms attract more attention in automatic processes, though this was not due to increased diagnosticity of the trait or behaviors, but rather to the negativity of the trait itself. This argument is supported by the results of Skowronski and Carlston (1987), who found that bad behaviors promote increased recall that suggests a memory bias for unpleasant information. Indeed, a host of research results report a bias for negative traits over positive traits (for a review see Skowronski & Carlston, 1989). This position is further supported by the results reported in a review paper by Baumeister et al. (2001). Specifically, Baumeister and colleagues review work that indicates a strong attentional bias toward negative behaviors over positive ones. This negativity bias can help explain why we failed to find an overall effect for morality in every condition except for negative moral traits in the nongroup control study. Perhaps the negative moral behaviors described in the sentences overrode participants’ reluctance to infer moral traits, though the lack of an interaction between morality and valence in the group manipulation condition strongly limits this argument. In sum, our (inconsistent) results point to a possible mechanism for an unconscious bias toward negative traits, though additional work is needed to support the existence of this phenomena in the arena of implicit trait attribution.
As with any study, there are a few limitations that must be noted. First, our use of Amazon’s Mechanical Turk may have been a more limiting factor than we anticipated. Indeed, there has been much debate around the use of MTurk for psychological research, with some studies concluding that platform can be a viable replacement for in-lab participants (e.g., Casler, Bickel, & Hackett, 2013), while others argue that there is a threat to data quality based on the likelihood of false responses and misrepresentation on the part of the participants (Kan & Drummey, 2018). Other work has argued that researchers should exercise extreme caution when attempting to collect data from MTurk workers, as the wide variability among the participants leaves it unknown if workers have taken similar studies (thus invalidating their responses), how their cognitive abilities compare to university participant pools, how their prior experience influences their willingness to put forth adequate effort, and their actual level of attention dedicated to the task at hand (Paolacci & Chandler, 2014). Buhrmester, Talaifar, and Gosling (2018) argue that tasks that collect the best-quality data are those that do not require large amounts of devoted attention, as workers have the ability to become distracted without the benefit of attentive research assistants present to keep them on track. Despite this claim, there is evidence that MTurk workers actually outperform university pool participants on tasks that measure attention to instructions (Hauser & Schwarz, 2016). One possible reason for our failure to find reliable STI effects may be the high attentional demand of our false recognition paradigm. While the task itself was relatively quick, the sheer amount of information participants were exposed to, coupled with the freedom of not performing the task in a lab setting and the relatively low compensation rate, demotivated participants from fully attending to the task. Buhrmester, Kwang, and Gosling (2011) argue that the rate of compensation and the task length has a large effect on data quality, though interestingly enough this does not mean the compensation rates need to be realistic with regard to employment pay. Rather, the pay needs only to be commensurate with the length of the task.
...(download the rest of the essay above)