Exploring The Generalization Process From Past Behavior To Predicting Future Behavior

Journal of Behavioral Decision Making, J. Behav. Dec.
Making (2015)
Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/bdm.1889
Exploring the Generalization Process from Past Behavior to Predicting Future

Behavior
LASANA HARRIS1*, VICTORIA K. LEE1, ELIZABETH H. THOMPSON1 and RACHEL KRANTON2
1
Psychology and Neuroscience, Duke University, Durham, NC, USA
2
Economics, Duke University, Durham, NC, USA
ABSTRACT
Substantial evidence in social psychology documents that traits predict behavior. Research in behavioral economics establishes prior behav-
ioral information—the actual behavior of another person in the past—influences future decision making, suggestive of the role of traits in guid-
ing future behavior, but agnostic to the specific psychological mechanism. Yet the entire generalization process from past behavior to
predicting future behavior has not been fully explored. Additionally, previous paradigms do not adequately dissociate prediction from expla-
nation, and provide participants with trait information, or rely on participants to generate the appropriate trait. Here, we combine literature and
experimental approaches in social psychology and behavioral economics to explore the generalization process from prior behavior that guides
future decisions. Across three studies utilizing consequential economic game paradigms and online questionnaires, an initial group of partic-
ipants (employees) played a time estimation game and a charity donations game before a second group of participants (employers) viewed the
behavior of the first group, then decided whether to invest in employees in a trust game and rock guessing game. Although participants infer
trait warmth and competence from the behavioral information in the first two games, estimates of normative behavior predicted investment
decisions on the warmth-relevant games better than trait inferences. These results dissociate generalizations guided by warmth and competence
behavioral information, and question the extent to which traits always serve as heuristics to predict behavior. Copyright © 2015 John Wiley &
Sons, Ltd.
Additional supporting information may be found in the online version of this article at the publisher’s web-site
key words person perception; social decision making; trait inferences; norms
People somehow predict what others will do after observing Traits provide adequate explanation for behavior because
their prior behavior. Eighty years of social psychological they locate the person as the causal agent, holding him or her
research argues that trait inferences serve as heuristics— responsible for initiating the behavior. Therefore, traits are very
spontaneously generated mental shortcuts—to predict be- useful as abstract concepts that allow meaning making of social
havior, allowing people to generalize from prior behavior behavior. Certainly, on aggregate, traits describe or explain peo-
to predict future behavior (for early review, see Paunonen ple’s behavioral consistency over time (Beck, McCauley, Segal,
& Jackson, 1985; Pervin, 1985), often at the expense of & Hershey, 1988; Burke, Kraut, & Dworkin, 1984; Emmons &
base-rate normative information (Kahneman & Tversky, Diener, 1986; Funder & Colvin, 1991; Furr & Funder, 2004;
1973). However, it may be time to reconsider the extent to Hettema & Hol, 1998; Koestner, Bernieri, & Zuckerman,
which people prefer trait inferences to normative informa- 1989; Krahe, 1986; Leikas, Lonnqvist, & Verkasalo, 2012;
tion when making predictions. Prediction and explanation Lippa & Mash, 1981; Magnusson & Ekehammar, 1978;
are not different sides of the same coin, and although per- Moskowitz, 1994; Welbourne, 2001). But traits only correlate
sonality traits may be valuable explanatory devices, people with behavior across situations at r = +.30. Empirical results
may not actually use them to generalize from one behavior suggest methodological improvements could increase traits’
in service of predicting another. This may particularly be predictive power if they are used to predict behavior in a
the case when the other person’s behavior is consequential specific-enough social context; if the social context from which
for the perceiver. Explanations require understanding ab- the trait is inferred (previous context) and the social context that
stract concepts, a feature that is unnecessary for prediction was being predicted (future context) are similar enough, then
(Andrews, 2012). Instead, people may place more weight correlation coefficients rise above the modest mark of +.30,
on a heuristic about the context to guide future interactions, and traits become better predictors (Baird & Lucas, 2011;
as opposed to a more cognitively complex construct such as Hemmelgarn, James, Ladd, & Mitchell, 1995; Magnusson,
a personality trait inference. Stated differently, although per- 1976; Paunonen & Jackson, 1986; Van Mechelen, 2009).
sonality traits are useful as folk explanatory conceptions, However, traits may not be useful for generalizing from
personality trait inferences may not be as useful as norma- single or limited instances of behavior to predict another’s
tive information when predicting behavior. behavior in a less related social context. Traits could only
serve as useful tools for prediction if people always behaved
according to their traits. However, social psychology repeat-
edly demonstrates the power of the social context to influ-
*Correspondence to: Pieter de la Court Building, 52 Wassennaarseweg,
2223AK, Leiden, Netherlands. ence behavior. Therefore, when asked to make predictions
E-mail: [email protected] about another person’s behavior, people may also take into
Copyright © 2015 John Wiley & Sons, Ltd.

Journal of Behavioral Decision Making
account normative information about the social context to prediction. This presents a potential confound that may have
guide their decisions. led to support for the idea that traits predict behavior.
Additionally, social neuroscience research is beginning to Another study in support of traits as predictive heuristics
show dissociations in trait inference processing in the brain, examined how people predict other people’s behavior using
suggesting that traits are not a homogeneous category. Most STIs (McCarthy and Skowronski, 2011). Participants saw
stable, enduring trait inferences activate a reliable pattern of pictures of people paired with a one-sentence description of
brain activity, including medial prefrontal cortex, superior behavior before rating the likelihood of the pictured person
temporal sulcus, precuneus, and temporoparietal junction performing three novel behaviors. Although this study did
(see for review Amodio & Frith, 2006; Frith & Frith, 2001; not provide participants with trait inferences (it demonstrates
Van Overwalle, 2009). These cortical brain regions also con- that traits could be inferred from picture–behavior pairings in
stitute nodes that contribute to other brain networks underly- a pilot study), it did not rule out perceived norms (frequency
ing cognitive processes such as economic valuation (Lee & of the behaviors) as sources of predictive information. Spe-
Harris, 2013) and cognitive control (Botvinick & Cohen, cifically, participants could just as easily have inferred the
2014). However, inferring a person’s trustworthiness de- likelihood that most people would perform each of the three
pends on the amygdala (Engell, Haxby, & Todorov, 2007; behaviors and used this estimate of behavior to guide their
Said, Baron, & Todorov, 2009; Todorov, 2008; Todorov, responses. The current study extends these results by asking
Baron, & Oosterhof, 2008), a subcortical brain structure im- participants to estimate normative information.
plicated in emotional learning, social biases, and vigilance
(Whalen & Phelps, 2009). This brain structure receives
almost immediate input from primary sensory cortex, The argument against trait inferences as heuristics
whereas the cortical structures receive their information via The oversight present in the studies described earlier and
a different, relatively longer, pathway from primary sensory present in theorizing about trait inferences as heuristics oc-
cortex, via many more synapses (Milner & Goodale, 2008). curs perhaps because investigators have long relied upon
Supporting evidence comes from research demonstrating that the domain of explanation by measuring self-reports or pro-
subliminal presentation of untrustworthy faces drives the viding trait information to participants. Traits represent ab-
amygdala (Winston, Strange, O’Doherty, & Dolan, 2002). stract concepts; the ability to understand such concepts
This suggests dissociation between different types of traits; develops later in life and should be a more controlled,
some of which may require more complex cognitive process- system-two process, compared with a heuristic, a prototypi-
ing, and others where such processing is unnecessary. Only cal system-one automatic process. Maybe researchers have
traits such as trustworthiness are inferences that may serve attempted to shoehorn trait inferences into a heuristic, largely
as heuristics. ignoring the difference between prediction and explanation.
Additionally, evidence is weaker that traits seem to be
inferred effortlessly; STIs depend on cognitive resources
The argument for trait inferences as heuristics (Ferreira et al., 1986; Van Overwalle, van Duynslaeger,
Social psychology has gathered evidence that supports the Coomans, & Timmermans, 2012; Wells, Skowronski,
hypothesis that trait inferences are heuristics (for review, see Crawford, Scherer, & Carlston, 2011). Indirect evidence
Booth-Butterfield, 1988; Furr, 2009). Within the thin-slice from the stereotyping literature suggests that traits constitute
and spontaneous trait inference (STI) literatures, studies have individuated representations that occur later, after time and
demonstrated that trait inferences can be fairly automatic. For motivation (Fiske & Neuberg, 1990; Fiske, Neuberg, &
instance, STIs of trustworthiness occur within a couple hun- Lin, 1999). There is also little evidence of trait inferences
dred milliseconds of face perception (Willis & Todorov, occurring without awareness because of the dependence on
2006), suggesting that some trait inferences are fast, a horse- self-report measures. Therefore, with only two out of four
man of automaticity (Bargh, 1994). Traits are also an efficient automaticity conditions satisfied, traits may not be heuristics
way of organizing information about a person (Fiske & given that they require cognitive resources and do not have
Taylor, 2013), a second condition of automaticity. apparent physiological or other implicit, non-self-report,
Given descriptions of two behaviors of fictitious people, components.
participants can identify the correct trait inference and predict Social psychological evidence suggests that cross-
the person’s behavior in a similar novel situation (Newman, situational behavior consistency occurs because of language
1996). However, this study used behavior of fictitious peo- —traits serve as linguistic concepts that suggest the same be-
ple, suggesting that the “correct” prediction is just one that havior is applicable in a novel social context (Semin &
the researchers identified as correct because participants are Krahne, 1988). This also supports the idea that traits are
not actually observing what the person did in the second sit- not heuristics. More damning evidence comes from a number
uation. Additionally, this study also provided participants of studies that failed to replicate classic cross-situational be-
trait inferences instead of simply having participants general- havior consistency effects (for instance, Chaplin &
ize from previous behavior. If traits probe explanatory mech- Goldberg, 1984; Eysenck & Wild, 1996), including cross-
anisms and explanatory mechanisms are not used for culturally (Church et al., 2008; Oishi, Diener, Napa Scollon,
prediction, then it is possible that asking people to make a & Biswas-Diener, 2004), although at least one paper finds
trait inference triggers an explanatory mechanism, making support for the folk intuition that traits serve as behavioral
this mechanism salient, and allowing it to masquerade as predictors cross-culturally (Norenzayan, Choi, & Nisbett,
Copyright © 2015 John Wiley & Sons, Ltd. J. Behav. Dec. Making (2015)
DOI: 10.1002/bdm
L. Harris et al. Predicting Future from Past Behavior
2002). However, other studies have questioned whether the experiment. As a result, they may employ one of a number
folk intuition exists (Heffler & Magnusson, 1979; Epstein of normative estimates of behavior to guide their decisions,
& Teraspulsky, 1986). such as the perceived ideal amount to donate, the estimated
Presumably, trait inferences facilitate generalizations that amount the person believes should be donated, or the esti-
can guide future behavior. But people may be required to mated amount on average that everyone will donate. This
generalize from behavior to broad traits in order to predict strategy allows us to determine which norms best match ac-
behavior in different contexts. For instance, someone who tual behavior. In Study 3, we employ a social psychological
engages in kind behavior in one context (helping an old lady approach to explicitly check whether the behavioral informa-
cross the street) may be considered a trustworthy person tion in our profiles leads to trait inferences about the em-
(returns extra money received in error from the cashier). ployees. We then use these trait ratings to predict Phase 2
Kindness and trustworthiness, though both warmth dimen- behavior in our first two studies.
sion traits, are separable and do not necessarily co-existent
in all people. As such, generalizations to the broad traits (like
morally good or warm in the example earlier) may lead to in- STUDY 1
accurate trait inferences, poor predictions, and suboptimal
decisions once the context shifts from crossing the street to Method
a monetary transaction with a cashier. Instead, participants Employees
may simply form a heuristic based on the person’s previous
Participants. Sixty-four individuals from Duke University
behavior that serves as information to guide their behavior. and the surrounding community contributed to the player da-
Stated differently, participants may judge how deviant the tabase. Four employees were removed from the sample, be-
person’s behavior is from other people’s behavior in the
cause of their wish not to have their picture taken (see
same social context (the norm) and use that information to succeeding text for significance of photographs), resulting
predict behavior (Kelley, 1967). However, when asked to ex- in 60 employees in the database. The sample’s mean age
plain the person’s behavior, they may then rely upon a trait
was 26.73 years (SD = 11.34), and 56.7% of the employees
inference. were female. All employees gave written consent prior to be-
ginning the experiment, in accordance with university stan-
dards, and were paid $10 for their participation.
Hypotheses
Here, we explore the entire generalization process from past
behavior to predicting future behavior across the two primary
person perception traits, warmth and competence (Asch, Employees’ tasks
1946; Fiske, Cuddy, & Glick, 2007). We created a paradigm Employees sequentially completed four games/tasks: a time
where participants make predictions about other people’s be- estimation game and charity donation game, with the order
havior during social interactions. We employ a behavioral counterbalanced across employees, and a trust game scenario
economics approach through the first two studies; we record task and rock estimation task, again with order
actual behavior from Phase 1 participants “employees” dur- counterbalanced across employees.
ing a trait warmth-relevant donation game, and a trait
competence-relevant time estimation game, before giving a Time estimation. We used the time estimation game to record
separate group of Phase 2 participants “employers” the op- competence behavioral information from the employees.
portunity to make trait inferences from this behavior to guide Employees viewed varying amounts of time (e.g., 7.2 sec-
future decisions regarding employees in a trait warmth- onds) and pressed a button to estimate the time interval.
relevant trust game and a trait competence-relevant rock esti- We scored accuracy as within 500 milliseconds of the desig-
mation game. As such, employers could choose to invest nated time. After each of 60 trials, employees were given ac-
with employees based on inferred warmth and competence. curacy feedback for that trial and their cumulative accuracy
Unlike previous paradigms, the participants predict how the score for the entire game.
other person would behave by behaving themselves, not
commenting on the person’s traits, or about what that person Charity donation. We used the charity donation game to re-
may do in a novel situation not involving the participant. cord warmth behavioral information from the employees.
We address alternate explanations for the pattern of re- Employees viewed 60 charities from within 10 different cat-
sults in Study 1 with Study 2; we balance the reward for em- egories, six charities per category. For each trial, employees
ployees across trust and rock estimation games, ruling out viewed a short description of a charity’s mission statement
motivational differences. Secondly, we collect normative es- and indicated what percentage of their participation fee
timates of donation game behavior from employers to deter- ($10) they would like to donate, if any, to the charity. Per-
mine alternate cutoff criteria for “good” investment decisions centages could range from 0% (i.e., $0) to 100% (i.e., $10).
in the trust game, and to test whether these normative esti- Employees were instructed that each decision should be in-
mates predicted participants’ decisions in Phase 2 better than dependent of the other donation decisions. Employees were
trait inferences. Additionally, participants would not be also informed that one trial would be randomly selected
aware of the actual social norm present among employees and realized at the end of the study for both the employee
during the game until they approached the last trial of the and charity.
DOI: 10.1002/bdm
Trust game scenario. We used the trust game scenario to re- Maze task. In order to allow employers ownership of the
cord warmth behavioral information from the employees. money used in the economic games, they completed a series
The player was assigned the role of trustee. An anonymous of mazes. For each maze completed in under 2 minutes, the
investor had the choice to either keep $10 or invest it with experimenter awarded the employer $5. The employers con-
the employee, at which point the money would be tripled, be- tinued completing mazes until they had earned $20 (every
coming $30. Employees were told that the investor had cho- employer except one accomplished this with the minimum
sen to invest with them. Employees were asked to indicate of four mazes). Therefore, each employer had earned $10
whether they would keep all of the profit or return half to each to play the trust and rock estimation games.
the investor. Employees only answered the question once.
Trust game. Phase 2 employers played the standard trust
Rock estimation game. We used the rock estimation game to game in the role of the investor. On each trial, employers
record competence behavioral information from the em- viewed an employee profile, followed by a decision screen
ployees. We asked employees to estimate the number of where they indicated whether or not they wanted to invest
rocks inside the jar. The correct number of rocks in the jar their $10. We then asked employers to rate their confidence
was 256. Employees estimated in a free response style in or- in their decision on a scale from 0 not at all confident to
der to avoid anchoring effects. Each player only made one 100 very confident. Employers did not receive trial-by-trial
estimate. We informed employees that they could receive feedback during the game in order to avoid learning effects.
money from us in the future based on their performance in We informed employers that we would choose one trial at
these latter two games. random and realize the decision for both the employer and
employee at the end of the experiment.
Rock estimation game. We modeled the rock estimation

Employee procedure game after the trust game. Employers decided to bet or not
The experimenter paid employees their $10 participation fee bet their previously earned $10 on whether the employees
upon arrival. Employees were informed that, at some point in would accurately estimate the number of rocks in a jar. On
the experiment, they would have the opportunity to donate a each trial, employers viewed the profile of an employee
portion of that money to charity if they so choose. The exper- (the same profile shown in the trust game), followed by a de-
imenter then explained the first two games (time estimation cision screen that asked whether or not they would like to in-
and donation games) with the help of a visual aid. Upon vest with that employee. Employers then rated their
completing the time estimation and donation games, em- confidence in their decisions.
ployees completed the trust game scenario task and the rock We matched the likelihood of payout for employers in
estimation task. both the rock estimation and trust games. That is, we chose
a range of rock estimations (50 rocks over or 50 under the ac-
tual amount, in other words estimates between 206 and 306)
to be considered correct in order to match the number of em-
Employers ployees who decided to share and keep in the trust game.
Participants. Thirty-two individuals from Duke University This resulted in 40 employees’ rock estimates deemed cor-
and the surrounding community participated as employers. rect and 20 employees’ rock estimates deemed incorrect to
The mean age was 21.86 years (SD = 4.56), and 56.3% of match the 40 employees that decided to share the money
the subjects were female. Two participants were excluded and 20 employees that decided to keep the money in the trust
from analysis because of incomplete data, for a total sample game scenario.
size of 30. Across both studies, participants earned $10 for
participating in the study and could earn up to an additional
$30 based on their decisions in two economic games. All
participants gave written consent prior to beginning the ex- Employers’ procedure
Employers performed the maze task until they had earned
periment, in accordance with university standards.
$20. The experimenter then explained the study in detail to
the employers, including a description of the donation and
time estimation games completed by the 60 employees from
Employers’ stimuli Phase 1, the information on the profiles, a description of the
Player profiles. We created a profile for each Phase 1 em- trust and rock estimation games to be completed in this ex-
ployee to be used as behavioral information by Phase 2 em- periment, and finally a description of the confidence ratings.
ployers on each trial, including the following: a photograph The experimenter additionally informed the employers that
of the employee; general demographic information, includ- one trial from each game would be randomly selected at
ing field of study, handedness, and age; and behavioral infor- the end of the study and realized for both parties. Then, em-
mation from their performance on the time estimation and ployers completed a short quiz to ensure complete under-
donation games, including overall average accuracy and gen- standing of the rules of the four games and the information
erosity scores, respectively, as well as round-by-round per- on the employee profiles. Employers were required to suc-
formance. This information appeared as percentages and cessfully complete the quiz (100% correct answers) before
bar graphs. proceeding.
DOI: 10.1002/bdm
We counterbalanced the order in which the employers expected value of investing, where 15x is the probability of
played the trust and rock estimation games. On each trial, winning the gamble multiplied by the outcome of winning
an employee profile was randomly selected without replace- the gamble and (1 x)*0 is the probability of losing the gam-
ment and displayed. Employers then indicated their decision ble multiplied by the outcome of losing the gamble. The
to invest or not and rated their confidence in their decision. value 10 represents the expected value of not investing be-
The trials in both games were visually identical (profile cause the probability of receiving $10 when not investing is
screen, decision screen, and confidence rating screen), with 1 given that participants keep the $10 they already earned.
the only difference being an instruction slide introducing Because (1 x)*0 will always be equal to 0, the equation
the game at the outset of the 60 trials (Supporting informa- can be simplified to
tion Figure 2). Each screen was self-paced. Finally, after 10¼15x
playing the trust and rock estimation games, employers com-
pleted the memory task to determine what information, if We defined a high competence employee as an employee
any, they could remember from the employee profiles. Em- whose average accuracy resulted in the statement 10 < 15x
ployers were not informed of the memory test until immedi- and a low competence employee as an employee whose aver-
ately before taking it to ensure that employers did not age accuracy resulted in the statement 10 ≥ 15x.
consciously commit any information to memory. We then dummy coded employer’s decisions as either
After all of the tasks had been completed, the experi- “good” or “bad,” based on these criteria. We defined good
menter randomly selected one trial from each game, that is, decisions as trials in which (a) the employee was high
realized one trial, paid the employers accordingly, and warmth or high competence and the employer did invest or
thanked them for their participation. At the end of the study, (b) the employee was low warmth or low competence and
the experimenters mailed the employees the money earned in the employer did not invest. The remaining two trial types
the randomly selected trials. were defined as bad decisions. We dummy coded good and
bad decisions as 1 and 0, respectively, allowing us to exam-
ine the amount of time employers used the trait inferences
Data analysis strategy from the profiles to guide investment decisions.
Investment decisions We performed the analysis described earlier for both match
To determine whether employers were using the employees’ and mismatch conditions created by our 2 × 2 factorial design.
behavioral information to guide their decisions, we dissoci- Match conditions are those in which behavioral information
ated employees with a high likelihood of returning the in- from a trait domain guides decisions in the game of that same
vestments from those with a lower likelihood of returning trait domain, and tell us whether trait inferences lead to gener-
investments. Furthermore, we used different approaches to alizations within person perception domains. Mismatch condi-
determine cutoff points for warmth and competence. tions are those in which behavioral information from a trait
Traditional economics would treat competence as an indi- domain guides decisions in the game of the other trait domain,
vidual trait and would measure competence by the individ- and tell us whether trait inferences lead to generalizations
ual’s ability to reach accurate conclusions. Measuring across person perception domains. We averaged the dummy-
warmth, on the other hand, would be possibly subjective; coded scores across the 60 trials within each of the four match
we adopted a framework where employees’ warmth is mea- and mismatch conditions, resulting in four scores for each em-
sured according to the distribution of generosity scores, cap- ployer indicating the amount of time they used behavioral in-
turing that individual generosity or warmth relative to a formation to guide their investment decisions. Numbers
social norm. To dissociate high from low warmth employees, closer to 1 indicate that the employer made decisions consis-
we used employees’ average generosity scores to calculate a tent with the hypothesis that behavioral information from em-
cutoff percentage one standard deviation above the mean. ployees would guide investment decisions. We performed a 2
For the sample of employees, this calculation resulted 20%. (trait) × 2 (decision) repeated measures analysis of variance
Therefore, we defined a high warmth employee as an em- (ANOVA) on confidence ratings to determine whether em-
ployee with an average generosity greater than or equal to ployers felt more or less confident when deciding to invest or
20% and a low warmth employee as an employee with an av- not. We computed paired sample t-tests with a Bonferroni
erage generosity less than 20%. corrected α = .0125 for all significant main effects and interac-
To dissociate high from low competence employees, we tions. We subtracted employees’ actual time estimation accu-
used employees’ average accuracy scores to calculate racy and donations from employers’ estimates of accuracy to
whether the expected value of investing with the employee create indices of memory accuracy. We first tested employers’
was greater than the expected value of not investing with recall accuracy by performing one-sample t-tests against the
the employee. We used the following expected value equa- value zero.
tion:
10¼15xþð1 xÞ*0
RESULTS
where x is the average accuracy score of a particular em-
ployee. This expected value equation sums up each possible Employee behavioral information statistics
outcome multiplied by the probability of that outcome. Mean accuracy scores from the time estimation game (which
Therefore, the right side of the equation represents the provided competence behavioral information) resulted in a
DOI: 10.1002/bdm
normal distribution (skew = 0.62 and kurtosis = 0.45), Employer investment decisions
ranging from 0% to 90% (M = 56.08, SD = 24.78; Figure 1b). We performed a 2 (behavioral information: warmth/compe-
The mean generosity scores from the donation game (which tence) × 2 (game: trust/rock) repeated measures ANOVA on
provided warmth behavioral information) resulted in a posi- employers’ investment decisions. There was a significant
tively skewed distribution (skew = 2.89 and kurtosis = 9.90) main effect of behavioral information, F (1, 29) = 22.32,
ranging from 0% to 72.67% (M = 6.94%, SD = 13.38%; p = 5.40 × 105, partial η2 = 0.44, Ω = 1.00, such that
Figure 1a). Employees’ estimates of behavior from the rock employers used warmth information (M = 78.50%,
estimation game (competence-relevant game) ranged from SD = 14.05%) more than competence information
guesses of 60 to 550, with a mean estimate of 197.77 (M = 66.39%, SD = 9.08%). This main effect was qualified
(SD = 93.63). They were normally distributed: skew = 1.41 by a significant behavioral information × game interaction,
and kurtosis = 3.57 (Figure 1d). Finally, 40 of the 60 em- F (1, 29) = 21.66, p = 6.60 × 105, partial η2 = 0.43, Ω = 0.99.
ployees (66.67%) indicated they would return the investment We followed up this interaction with paired sample t-tests
in the trust game scenario (warmth-relevant game; Figure 1c). with a Bonferroni corrected α = .0125. When comparing
We found no correlation between employees’ accuracy games within behavioral information, we found that warmth
scores and a correct rock estimate, r (58) = .12, p = .356, or behavioral information was used significantly more in the
between employees’ generosity scores and their decision to trust game than rock estimation game, t (29) = 3.57,
share the money in the trust game, rpbi (58) = .19, p = .141. p = .001, while competence information was used signifi-
Therefore, warmth and competence-relevant games used to cantly more in the rock estimation game than trust game, t
generate player behavioral information are unrelated to the (29) = 4.79, p = 4.53 × 105. This suggests that employers
warmth and competence-relevant games in which employers generalized from the corresponding trait-relevant behavioral
made decisions about said employees. Specifically, an em- information, using it more frequently to guide their decisions
ployee’s ability to estimate time and rocks is not related, nei- within the relevant than irrelevant game. Comparing behav-
ther is an employee’s willingness to donate money to charity ioral information within game, we found that in the trust
or return an investment in the trust game. Therefore, there is game, warmth behavioral information was used significantly
no added predictive value when employers use accuracy or more than competence, t (29) = 11.16, p = 5.17 × 1012, but in
generosity scores as behavioral information when making the rock estimation game, competence behavioral informa-
decisions. However, because of the poverty of other avail- tion was not used significantly more than warmth, t (29)
able information and the perceived relationship between the = 0.33, p = .075 (Figure 2). This suggests that while em-
pairs of behaviors along warmth and competence trait do- ployers generalized trait warmth from the relevant behavioral
mains, we hypothesize that employers will in fact use the be- information to more frequently guide decisions in the
havioral information to guide their investment decisions. warmth-relevant game, they generalized both trait warmth
Figure 1. Player behavioral information. Average behavioral information from the sample of employees on the (a) donation game, (b) time
estimation game, (c) trust game scenario, and (d) rock estimation game
DOI: 10.1002/bdm
Figure 2. Investment decisions in Study 1. Bar graphs depict the

proportion of behavioral information used in each game based on
actual norms in Study 1 as cutoff criteria. Error bars reflect standard
error of the mean. Different letters represent statistically significant
differences between means, that is, different letters are significantly
different from each other (e.g., the a’s are the same, but different
from b and c) Figure 3. Confidence ratings and memory for employees. Bar
graphs depict employers’ reported (a) confidence in their investment
and competence from the relevant behavioral information to decisions and employers’ (b) memory recall of employees’ perfor-
guide decisions in the competence-relevant game. mance in Study 1. Error bars reflect standard error of the mean
of memory accuracy. We first tested employers’ recall accu-

Employer confidence ratings racy by performing one-sample t-tests against the value zero.
We performed a 2 (game: rock/trust) × 2 (decision: invest/not For competence, we found that employers’ memory indices
invest) repeated measures ANOVA on confidence ratings to de- did not significantly differ for time estimation, t (29)
termine whether employers felt more or less confident when de- = 0.78, p = .444, suggesting that they were accurate in
ciding to invest or not in the warmth-related trust game and recalling employees’ competence information. We followed
competence-related rock estimation game. We found a signifi- up this finding by testing whether employers displayed mem-
cant game × decision interaction, F (1, 25) = 12.16, p = .002, ory differences for employees who were categorized as high
partial η2 = 0.33, Ω = 0.92. We then computed paired sample or low on the trait dimension, calculating memory accuracy in-
t-tests with a Bonferroni corrected α = .0125. We found a signif- dices separately for high and low competence employees. We
icant difference between confidence ratings in the trust game, t found that employers showed a significant difference for
(26) = 3.36, p = .002, such that confidence was higher when highly competent employees, t (29) = 6.48, p = 1.00 × 106,
employers decided not to invest (M = 72.09, SD = 17.66) than such that they significantly underestimated their time estima-
invest with an employee (M = 61.12, SD = 18.48). These results tion ability (Mdiff = 14.52, SD = 13.11). Employers also inac-
suggest that employers were least confident in their decisions to curately recalled low competence employees’ time estimation
trust the employees. Neither of the main effects revealed signif- ability, t (29) = 3.03, p = .005, such that they significantly
icant differences. overestimated (Mdiff = 6.71, SD = 12.12). Together, these re-
We then tested whether these confidence ratings differed sults suggest that although employers’ accurately recalled av-
for employees high or low on each trait dimension. For erage employee time estimation abilities, they did so by
warmth, we found a significant difference, t (26) = 5.25, overestimating the ability of less skilled employees and
p = 1.3 × 105, such that employers were more confident underestimating the ability of more skilled employees.
when making decisions regarding high warmth (M = 70.87, We performed a similar analysis for warmth, where we
SD = 20.49) rather than low warmth employees (M = 55.89, found that employers’ memory indices did not significantly
SD = 18.94). We found a similar effect for competence, t differ for donations to charity, t (29) = 1.08, p = .288, suggest-
(26) = 3.85, p = .001, where employers were also more confi- ing that they were accurate in recalling employees’ warmth
dent making decisions for high competence (M = 70.31, information. We followed this up by separately testing mem-
SD = 16.90) rather than low competence employees ory recall for high and low warmth employees. Unlike com-
(M = 57.86, SD = 19.76; Figure 3a). These findings suggest petence, we found no significant difference for recall of
that employers perhaps intuitively dissociated “good” from donation behavior for either high, t (29) = 1.16, p = .256, or
“bad” employees along both trait dimensions. low warmth employees, t (29) = 0.82, p = .417, suggesting
that employers accurately remembered warmth behavioral
information (Figure 3b).
Employer memory data We next tested whether there was enhanced memory for
We subtracted employees’ actual time estimation accuracy warmth relative to competence information by performing a
and donations from employers’ estimates to create indices paired-samples t-test using the memory indices. We did not
DOI: 10.1002/bdm
find a significant difference, t (31) = 1.31, p = .200, suggest- Method

ing that employers remembered warmth and competence in- Participants
formation equally well. Thirty individuals from Duke University and the surrounding
community participated in Study 2. The mean age was
22.37 years (SD = 6.36 years), and 66.7% of the subjects were
female.
CONCLUSION
Study 1 replicates previous behavioral economics studies

Stimuli
showing that previous behavior affects future decisions using
We used the same employee profiles from Study 1. At the
a novel paradigm that utilizes actual behavior (i.e., finan-
end of Study 2, employers completed a post-study question-
cially consequential decisions without the aid of deception)
naire that asked the following:
and highlights the primacy of warmth behavioral informa-
tion. In particular, we demonstrated that competence behav- 1. If you played the donation game, how much would you
ioral information generalizes within person perception donate to charity from your money?
domain, and warmth behavioral information generalizes both 2. In general, how much do you think people should donate
within and across person perception domains. Thus, em- to charity from their money?
ployers are using warmth information to help guide decision 3. Before playing the game, what did you think the average
making even in situations in which warmth information donation to charity would be?
should be irrelevant. This finding is consistent with the halo
The first question assesses employers’ personal norms for
effect literature where traits and behaviors of one kind can
donating, while the second and third assess estimates of ideal
carry over and influence traits and behaviors of another kind
norms and actual norms, respectively.
(Dion, Berscheid, & Walster, 1972; Landy & Sigall, 1974;
Moore, Filippou, & Perrett, 2011). The fact that employers
were significantly more confident when making good deci-
Procedure
sions in the warmth-relevant game (trust game) than in the
The procedure was the same as Study 1, with one exception:
competence-relevant game (rock estimation game) lends ad-
employees were also rewarded for correct rock guesses if
ditional support; warmth information in particular seems
employers invested with them, and we realized that outcome.
both reliable and predictive. Finally, employers also better
remembered warmth behavioral information than compe-
tence behavioral information, lending further support for
Data analysis strategy
the notion that warmth behavioral information is prioritized
The analysis strategy was the same as Study 1. In addition,
relative to competence behavioral information.
we also determined the cutoff criterion for warmth informa-
However, this study did not assess whether employers ac-
tion individually for each employer based on his or her re-
tually made trait inferences, so it cannot rule out the possibil-
sponse to each question on the post-study questionnaire.
ity that trait inferences are the mechanism that guides
decision. Additionally, employees’ payout structure for the
rock estimation game was different from the trust game such
RESULTS
that employers may infer a lack of effort from employees on
the rock estimation game. Finally, the criteria used for char-
Employee norms
acterizing high and low warmth employees relied on the
On average, employers reported that they would donate
mean and standard deviation of the labor market, which em-
16.07% (SD = 13.37%) of their earnings to charity (personal
ployees would not have an accurate estimation of until the
norms), that people should donate 12.59% (SD = 7.91%) of
end of the study. We addressed these methodological issues
their earnings to charity (perceived ideal norms), and that be-
in Study 2.
fore playing the game, they believed the average donation by
an employee would be 10.72% (SD = 7.58%) of their earn-
ings to charity (perceived actual norms). Only the reports
STUDY 2 of personal norms and perceived actual norms statistically
significantly differed, t (29) = 2.63, p = .013, while personal
Study 2 is a replication of Study 1, with two changes to ad- norms and perceived ideal norms marginally differed,
dress potential flaws. First, we rewarded employees for accu- t (28) = 1.93, p = .064. These findings suggest that each of
rate rock estimations, and second, we asked employers to these reports could yield different results when used as
indicate their estimation of normative donation behavior to estimates to distinguish high from low warmth employees
determine alternate cutoff criteria for “good” investment de- (warmth behavior cutoff criteria).
cisions. This latter addition also allows us to test the extent to
which perceived norms may have guided behavior in the
warmth domain. This study used the same database of behav- Employer investment decisions
ioral information from the employees in Study 1, but re- We performed a series of 2 (behavioral information) × 2
cruited a new sample of participants as employers. (game) repeated measures ANOVAs on the investment
DOI: 10.1002/bdm
behavior using the four different estimates of employers’ game, competence behavioral information was not used signif-
normative behavior ascertained from the employee sample icantly more than warmth, t (29) = 0.52, p = .606. There was
and the post-experiment questionnaire. We report each sepa- no significant main effect for game, suggesting that em-
rately later. ployers did not differ in their use of behavioral information
across both games. These findings replicate Study 1.
Actual norms
We first replicated the analysis performed in Study 1 using
the actual mean and standard deviation of the employees’ do- Personal norms
nation behavior to generate cutoff criteria for warmth infor- We next used the response to the question asking employers
mation (Figure 4a). This tested whether we could replicate to report their own donation behavior to generate cutoff
Study 1. There was a significant main effect of behavioral in- criteria (Figure 4b). There was a significant behavioral
formation, F (1, 29) = 10.39, p = .003, partial η2 = 0.26, information × game interaction, F (1, 29) = 22.87,
Ω = 0.88, such that employers used warmth more than com- p = 4.6 × 105, partial η2 = 0.44, Ω = 1.00. We computed
petence behavioral information. This main effect was quali- paired sample t-tests with a Bonferroni corrected α = .0125.
fied by a significant behavioral information × game When comparing games within behavioral information, we
interaction, F (1, 29) = 27.47, p = 1.3 × 105, partial found that warmth behavioral information was used signifi-
η2 = 0.47, Ω = 1.00. We computed paired sample t-tests with a cantly more in the trust than rock estimation game, t (29)
Bonferroni corrected α = .0125. When comparing games = 3.24, p = .003, while competence information was used sig-
within behavioral information, we found that warmth behav- nificantly more in the rock estimation game than trust game,
ioral information was used significantly more in the trust t (29) = 5.40, p = 8.0 × 106. This suggests that employers
than rock estimation game, t (29) = 4.02, p = 3.75 × 104, generalized from the corresponding trait-relevant behavioral
while competence information was used significantly more in information, using it more frequently to guide their decisions
the rock estimation game than trust game, t (29) = 5.40, within the relevant than irrelevant game. Comparing behav-
p = 8.0 × 106. This suggests that employers generalized from ioral information within game, we found no significant dif-
the corresponding trait-relevant behavioral information, using ferences. There was no significant main effect for game or
it more frequently to guide their decisions within the relevant behavioral information, suggesting that employers did not
than irrelevant game. Comparing behavioral information differ in their use of behavioral information across both
within game, we found that in the trust game, warmth behav- games or in their use of warmth or competence behavioral in-
ioral information was used significantly more than compe- formation. These findings do not replicate the pattern of re-
tence, t (29) = 8.25, p = 1.0 × 107, but in the rock estimation sults in Study 1 or the actual norms as cutoff criteria,
Figure 4. Investment decisions in Study 2. Bar graphs depict the proportion of behavioral information used in each game based on actual
norms in (a) Study 2, and (b) personal norms, (c) perceived ideal norms, and (d) perceive actual norms as cutoff criteria. Error bars reflect
standard error of the mean. Different letters represent statistically significant differences between means, that is, different letters are signifi-
cantly different from each other (e.g., the a’s are the same, but different from b and c)
DOI: 10.1002/bdm
suggesting that personal norms serve as less suitable substi- Employer confidence ratings
tutes for actual norms. We performed a 2 (game) × 2 (decision) repeated measures
ANOVA on confidence ratings to determine whether em-
ployers felt more or less confident when deciding to invest
Perceived ideal norms or not in the trust and rock estimation games. We found a sig-
We next used the response to the question asking employers nificant decision main effect, F (1, 26) = 6.23, p = .019, par-
to report their perceived ideal norm to generate cutoff tial η2 = 0.19, Ω = 0.67, such that employers felt more
criteria (Figure 4c). There was a significant behavioral confident when not investing (M = 73.13, SD = 11.12) rather
information × game interaction, F (1, 28) = 27.12, than investing (M = 66.09, SD = 17.21). We computed paired
p = 1.6 × 105, partial η2 = 0.49, Ω = 1.00. We computed sample t-tests with a Bonferroni corrected α = .0125. This
paired sample t-tests with a Bonferroni corrected main effect was driven by a significant difference for trust
α = .0125. When comparing games within behavioral infor- game decisions, t (28) = 2.91, p = .007, such that confidence
mation, we found that warmth behavioral information was was higher when employers decided to not trust an employee
used significantly more in the trust than rock estimation (M = 75.88, SD = 13.03) than when they trusted an employee
game, t (28) = 4.07, p = 3.47 × 104, while competence in- (M = 66.67, SD = 17.76; Figure 5a). These results suggest that
formation was used significantly more in the rock estimation employers were least confident in their decisions to invest in
game than trust game, t (28) = 5.37, p = 1.0 × 105. This sug- the employees. Neither the game main effect nor the interac-
gests that employers generalized from the corresponding tion revealed significant differences. These results perfectly
trait-relevant behavioral information, using it more fre- replicate Study 1.
quently to guide their decisions within the relevant than ir- We then tested whether these confidence ratings differed for
relevant game. Comparing behavioral information within employees rated as high or low on each trait dimension. For
game, we found that in the trust game, warmth behavioral competence, we found a significant difference, t (27) = 5.79,
information was used significantly more than competence, p = 4.0 × 106, where employers were also more confident mak-
t (28) = 6.89, p = 1.0 × 107, but in the rock estimation game, ing decisions for high (M = 75.20, SD = 10.10) rather than low
competence behavioral information was not used significantly competence employees (M = 64.24, SD = 14.58). For warmth,
more than warmth, t (28) = 1.69, p = .119. There was no signif- we tested for differences in confidence ratings using each of
icant main effect for game, suggesting that employers did not the four cutoff criteria for a “good” decision.
differ in their use of behavioral information across both games.
These results replicate the findings for actual norms as well as
the result in Study 1, suggesting that perceived ideal norms
serve as suitable substitutes for actual norms. Actual norms
We found a significant difference, t (27) = 2.43, p = .022,
such that employers were more confident when making deci-
Perceived actual norms sions regarding high (M = 76.91, SD = 11.91) than low
Finally, we used the response to the question asking warmth employees (M = 63.30, SD = 28.59).
employers to report their perceived actual norms of the
employees to generate cutoff criteria (Figure 4d). There
was a significant behavioral information × game interaction,
F (1, 28) = 23.65, p = 4.0 × 105, partial η2 = 0.46, Ω = 1.00.
We computed paired sample t-tests with a Bonferroni
corrected α = .0125. When comparing games within behav-
ioral information, we found that warmth behavioral informa-
tion was used significantly more in the trust than rock
estimation game, t (28) = 3.61, p = .001, while competence
information was used significantly more in the rock estima-
tion game than trust game, t (28) = 5.30, p = 1.2 × 105. This
suggests that employers generalized from the corresponding
trait-relevant behavioral information, using it more fre-
quently to guide their decisions within the relevant than irrel-
evant game. Comparing behavioral information within
game, we found no significant differences. There was no
significant main effect for game or behavioral information,
suggesting that employers did not differ in their use of
behavioral information across both games or in their use of
warmth or competence behavioral information. These find-
ings do not perfectly replicate the pattern of results in Study
Figure 5. Confidence ratings and memory for employees. Bar
1 or using actual norms as cutoff criteria, suggesting that graphs depict employers’ reported (a) confidence in their investment
perceived actual norms serve as less suitable substitutes for decisions and employers’ (b) memory recall of employees’ perfor-
actual norms. mance in Study 2. Error bars reflect standard error of the mean
DOI: 10.1002/bdm
Behavioral intentions results of Study 1, demonstrating that employers accurately

We found a significant difference, t (27) = 4.40, p = 1.53 × 104, remembered warmth behavioral information for employees.
such that employers were more confident when making deci- We next tested whether there was enhanced memory
sions regarding high (M = 76.45, SD = 13.64) than low warmth for warmth or competence information by performing a
employees (M = 60.37, SD = 17.08). paired-samples t-test using the memory indices. There
was no significant difference, t (31) = 1.83, p = .078, repli-
cating Study 1.
Perceived ideal norms
We found a significant difference, t (27) = 4.77, p = 5.7 × 105,
such that employers were more confident when making deci- CONCLUSION
sions regarding high (M = 77.33, SD = 12.62) than low warmth
employees (M = 63.17, SD = 16.60). There was no significant change in investment behavior in
the rock estimation game after instituting the equivalent pay-
off paradigm. Additionally, we found little difference be-
Perceived actual norms tween our results when using the mean generosity score
We found a significant difference, t (26) = 4.56, cutoff criterion for ‘good’ decisions employed in Study 1,
p = 1.08 × 104, such that employers were more confident and using the employers’ estimates of donation norm. This
when making decisions regarding high (M = 75.80, was particularly true when we used estimates of ideal norms,
SD = 13.31) than low warmth employees (M = 58.99, and less true when we used estimates of personal norms and
SD = 18.14). Together, these results suggest that again em- perceived actual norms. This suggests that norms may be
ployers were intuitively aware of “good” and “bad” em- used to predict warmth-relevant behavior. Therefore, Study
ployees regardless of the cutoff criteria, replicating the 2 replicated our results from Study 1 using norm estimates
findings of Study 1. as determinants of high and low warmth employees. Study
2 also adds reliability to the findings in Study 1 while ruling
out perceived player motivational differences between the
Employer memory data competence and warmth games.
As in Study 1, we subtracted employees’ actual time estima-
tion accuracy and donations from employers’ estimates to
create indices of memory accuracy before performing one- STUDY 3
sample t-tests against the value zero. For competence, we
found that employers’ memory indices did significantly The first two studies demonstrate that people use behavioral
differ for time estimation, t (29) = 2.09, p = .045, information to predict behavior, guiding consequential deci-
Mdiff = 4.76, SD = 12.44, such that they underestimated em- sions. However, we have not yet tested whether trait infer-
ployees’ time estimation ability. We followed up this finding ences predict behavior. We intentionally avoided asking for
by testing whether employers displayed memory differences self-reports of traits or measuring trait inferences directly
for employees who were categorized as high or low on the during either of the first two studies because we did not want
trait dimension, calculating memory accuracy indices sepa- to engage an explanatory system. Therefore, we collected
rately for high and low competence employees. We found data in Study 3 to determine (1) whether it is possible to infer
that employers showed a significant difference for highly trait warmth and competence from the behavioral informa-
competent employees, t (29) = 6.50, p = 1.00 × 107, such tion in employee profiles and (2) whether trait warmth infer-
that they significantly underestimated their time estimation ences better predict behavior than trait inferences in Studies 1
ability (Mdiff = 17.79, SD = 14.99). However, employers ac- and 2.
curately recalled low competence employees’ time estima-
tion ability, t (29) = 1.55, p = .131 (Figure 5b). Together,
these results suggest that employers’ accurately recalled Method
low competence employees’ time estimation abilities, Participants
underestimating the ability of more skilled employees. These Three hundred seventeen individuals recruited from Ama-
findings partially replicate the results of Study 1, demonstrat- zon’s Mechanical Turk completed Study 3 online. The sam-
ing that employers’ memory for highly skilled employees ple consisted of the following age breakdown: 34.7% were
underestimated their competence. 18 to 29 years old, 27.1% were 30 to 39 years old, 14.8%
We performed a similar analysis for warmth, where we were 40 to 49 years old, 15.8% were 50 to 59 years old,
found that employers’ memory indices did not significantly and 7.6% were over 60 years old; 51.1% of the subjects were
differ for donations to charity, t (29) = 0.98, p = .327. We female.
followed this up by separately testing memory recall for high
and low warmth employees. We found no significant differ-
ence for recall of donation behavior for both high, t (29) Stimuli
= 0.98, p = .336 and low warmth employees, t (29) = 0.85, We replaced the employees’ photographs with computer-
p = .405, suggesting that employers accurately remembered generated neutral faces (Todorov et al., 2013) to control for
warmth behavioral information. These findings replicate the the impact of identity on any trait inferences or heuristic
DOI: 10.1002/bdm
processes. All other information in the 60 employee profiles Reliability estimates for scale warmth and competence
remained unchanged from the first two studies. We then averaged the traits along the two factors, creating
composites. A reliability analysis of the warmth composite
resulted a Cronbach’s alpha = .992, while the competence
Procedure composited resulted a Cronbach’s alpha = .987, suggesting
The procedure for employees was the same as the first two that the items in the composite formed a reliable scale. The
studies, except that it took place online. Participants still warmth and competence scales were significantly positively
viewed employee profiles, but instead of making investment correlated, r (58) = .575, p = 2.0 × 106.
decisions, they rated 10 profiles on 10 warmth-related traits
(helpful, sincere, trustworthy, moral, sociable, caring, un-
friendly [reverse scored], insensitive [reverse scored], gener- Trait–behavior correlations
ous, and warm) and 10 competence-related traits (intelligent, We next ran correlations between the behavioral information
skillful, creative, efficacy, capable, lazy-reverse scored, dis- in each profile and the inferred traits warmth and competence
organized [reverse scored], punctual, precise, and compe- to test whether trait inferences corresponded to actual
tent). Therefore, 50–55 participants rated each profile. behavior. We found significant correlations between trait
warmth inferences and donation behavior (r (58) = .839,
p = 1.0 × 108), while trait competence correlated with both
donation (r (58) = .329, p = .010) and time estimation behav-
Data analysis strategy ior (r (58) = .847, p = 1.0 × 108). This suggests that behav-
We first perform a cluster analysis with varimax rotation on ioral information from both games underlie their respective
the traits to reduce the number of traits to the warmth and trait inferences, although competence is also inferred from
competence dimensions. We then performed reliability anal- generous behavior.
yses on each dimension before averaging across the ratings
to create overall warmth and competence composites. We
then compute correlation coefficients to test the association Traits predicting behavior
between trait inference and behavior. Finally, we wondered whether trait inferences served as heu-
Finally, we use the average trait rating across the raters as ristics in Studies 1 and 2. To test this, we relied on the pattern
each profile’s cutoff criteria to re-analyze the data collected of results in the first two studies to derive a number of ax-
in Studies 1 and 2. This strategy informs us whether actual ioms that traits should satisfy when examined with similar
ratings of trait warmth and competence based on the em- statistical analyses. These axioms concern the pattern of
ployees’ behavioral information (independent of the em- main effect and interactions that trait inferences should result
ployee’s identity) guides participants’ real consequential if they serve as heuristics:
decisions. This allows us to test whether trait inferences serve
as heuristics without relying on self-report from participants 1. A significant main effect of trait inference, such that
during the task. warmth trait inferences are more frequent than compe-
tence trait inferences.
2. A significant trait inference × game interaction, such that
RESULTS a. no difference exists between the occurrence of warmth and
competence trait inferences in the rock estimation game,
Principle components factor analysis b. warmth trait inferences occur more frequently than compe-
We first completed a principle components factor analysis tence trait inferences in the trust game,
with varimax rotation to reduce the number of trait items. Be- c. warmth trait inferences occur more frequently in the trust
cause of our a priori hypothesis, we set criteria for conver- game than rock estimation game, and
gence at two factors, which explained 96.49% of the d. competence trait inferences occur more frequently in the
variance. The screen plot confirmed a two-factor solution. rock estimation game than trust game.
A first warmth factor (listed in order of decreasing factor
3. No significant main effect of game, suggesting no differ-
loadings) included items generous, helpful, caring, insensi-
ence between trait inferences used in the rock estimation
tive (reverse coded), warm, unfriendly, moral, sociable, sin-
game compared with trust game.
cere, and trustworthy, with eigenvalue that ranges from
.968 to .832. A second competence factor emerged, We employed a similar analysis as in Studies 1 and 2,
consisting of items (listed in order of decreasing factor load- characterizing good and bad decisions this time using the
ings) precise, skillful, punctual, capable, intelligent, efficacy, mean trait inferences from the profiles as cutoff criteria for
competent, disorganized (reverse coded), lazy (reverse high and low warmth and competence. We calculated mean
coded), and creative, with eigenvalue that ranges from .982 trait ratings for all behavioral information; profiles with aver-
to .741. Interestingly, generous and precise both loaded the age trait ratings that fell at or above the mean were consid-
highest on warmth and competence factors, respectively, ered high on the trait dimension, while those that fell below
suggesting that they best captured the specific warmth and the mean were considered low. We coded a good investment
competence traits inferred from the donation and time esti- decision if employers invested in the rock or trust game and
mation behaviors. the employee was rated as high on the relevant traits or if the
DOI: 10.1002/bdm
employer did not invested in the employee in either game sub-traits: generous and punctual, and trustworthy and pre-
and the employee was rated as low on the relevant traits. cise. The first pair of traits tests the specific trait inference
We coded bad investment decisions when trait and behavior that may occur from the behavioral information contained
mismatches occurred. in the employee profiles—generosity may be inferred from
We thus ran a series of 2 (trait inference) × 2 (game) re- the charity donations game, and punctuality may be inferred
peated measures ANOVAs collapsed across employers’ be- from the time estimation game. The second pair of traits tests
havior in Studies 1 and 2, with study as a between-subjects the specific trait inference appropriate for the decision facing
covariate (the study main effect was not significant and did the employer in the investment games—trustworthiness is
not interact with any other main effect or interaction across relevant for the trust game, and precision is relevant for the
all our analyses; as such, it is no longer discussed, and results rock estimation game. We employ a Bonferroni correction
are collapsed across Studies 1 and 2). If employers were for multiple comparison across the four ANOVAs, resulting
using trait inferences as heuristics to predict behavior, then α = .0125 for significance.
these ANOVAs should replicate the pattern of results ob-
served in the first two studies, satisfying all three axioms.
We ran this analysis first using broad trait inferences of Scale warmth and competence trait inferences
warmth and competence resulting from the averages of the We replicated the analysis performed in Studies 1 and 2
10 warmth-related traits and the 10 competence-related traits. using the average warmth and competence ratings, collapsed
This allows us to test how a scale measure of warmth and across all warmth and competence traits, to generate cutoff
competence fairs in our test of axioms. Specifically, this test criteria for high and low warmth and competence. There
informs us whether a general or broad sense of warmth or was a significant main effect of game, F (1, 58) = 147.69,
competence is estimated. We then test subsets of the specific p < 1.0 × 106, partial η2 = 0.72, Ω = 1.00, such that
trait attributions that make up the broad trait warmth and employers used trait inferences more in the rock estimation
competence inferences. This second test assesses the same game than trust game (Figure 6a). This main effect was
question as the first but allows dissociation; specifically, test- qualified by a significant trait inference × game interaction,
ing scale warmth and competence does not rule out whether a F (1, 58) = 43.55, p < 1.0 × 106, partial η2 = 0.43, Ω = 1.00.
specific warmth or competence sub-trait (e.g., precision or There was no main effect of trait inference, F (1, 58)
generosity) is driving our findings. Testing how endorsement = 0.18, p = .674.
of the specific words “warm” and “competent” replicates this We computed follow-up paired sample t-tests with a
initial test without the influence of the nine other specific Bonferroni corrected α = 3.125 × 103. When comparing
sub-traits. Furthermore, we perform two additional tests to games within trait inferences, we found that warmth trait in-
examine the effects of specific warmth and competence ferences (t (59) = 5.00, p = 5.0 × 106) and competence trait
Figure 6. Investment decisions using trait inferences to predict behavior. Bar graphs depict the proportion of trait inferences used in each game
based on trait inferences about (a) warmth and competence (scale), (b) warmth and competence, (c) trustworthy and precise, and (d) generous
and punctual. Error bars reflect standard error of the mean
DOI: 10.1002/bdm
inferences (t (59) = 12.02, p < 1.0 × 106) were used signifi- inferences (t (59) = 11.77, p < 1.0 × 106) were used signifi-
cantly more in the rock estimation game than trust game. cantly more in the rock estimation game than trust game.
This suggests that both inferred traits more frequently guided This suggests that both inferred traits more frequently guided
decisions within the rock estimation game. Comparing trait decisions within the rock estimation game. Comparing trait
inferences within game, we found that in the trust game, inferences within game, we found that in the trust game,
warmth trait inferences were used significantly more than trustworthiness trait inferences were used significantly more
competence, t (59) = 7.06, p < 1.0 × 106, and in the rock es- than precision, t (59) = 5.80, p < 1.0 × 106, but in the rock
timation game, competence trait inferences were used signif- estimation game, precision trait inferences were not used
icantly more than warmth, t (59) = 3.56, p = .001. These significantly more than trustworthy, t (59) = 2.55, p = .013.
results satisfy axioms 2, 2b, and 2d. These findings satisfy axioms 2, 2b, and 2d.
Specific warmth and competence trait inferences

Generous and punctual trait inferences
We conducted the ANOVA for specific warmth and compe-
Next, we replicated the analysis performed in Studies 1 and 2
tence trait attribution (ratings of the terms “warm” and “com-
using generous and punctual trait attribution (ratings of the
petent”) to generate cutoff criteria for high and low warmth
terms “trustworthy” and “precise”) to generate cutoff criteria
and competence. There was a significant main effect of
for high and low trustworthiness and precision. There was a
game, F (1, 58) = 139.37, p < 1.0 × 106, partial η2 = 0.76,
significant main effect of game, F (1, 58) = 153.15,
Ω = 1.00, such that employers used trait inferences more in
p < 1.0 × 106, partial η2 = 0.73, Ω = 1.00, such that em-
the rock estimation game than trust game (Figure 6b). There
ployers used trait inferences more in the rock estimation
was also a marginally significant main effect of trait infer-
game than trust game (Figure 6d). There was also a main ef-
ence, F (1, 58) = 4.99, p = .029, partial η2 = 0.08, Ω = 0.59,
fect of trait inference, F (1, 58) = 8.46, p = .005, partial
such that employers used warmth more than competence trait
η2 = 0.13, Ω = 0.82, such that generous trait inferences guided
inferences to guide behavior. These main effects were
decisions more than punctual trait inferences. Both main ef-
qualified by a significant trait inference × game interaction,
fects were qualified by a significant trait inference × game in-
F (1, 58) = 42.58, p < 1.0 × 106, partial η2 = 0.42, Ω = 1.00.
teraction, F (1, 58) = 22.07, p = 1.7 × 105, partial η2 = 0.28,
We computed follow-up paired sample t-tests with a
Ω = 1.00.
Bonferroni corrected α = 3.125 × 103. When comparing
We computed follow-up paired sample t-tests with a
games within trait inferences, we found that warmth trait in-
Bonferroni corrected α = 3.125 × 103. When comparing
ferences (t (59) = 5.45, p = 1.0 × 106) and competence trait
games within trait inferences, we found that generous trait
inferences (t (59) = 12.34, p < 1.0 × 106) were used signifi-
inferences (t (59) = 4.67, p = 1.8 × 105) and punctual trait in-
cantly more in the rock estimation game than trust game.
ferences (t (59) = 11.86, p < 1.0 × 106) were used signifi-
This suggests that both inferred traits more frequently guided
cantly more in the rock estimation game than trust game.
decisions within rock estimation game. Comparing trait in-
This suggests that both inferred traits more frequently guided
ferences within game, we found that in the trust game,
decisions within the rock estimation game. Comparing trait
warmth trait inferences were used significantly more than
inferences within game, we found that in the trust game, gen-
competence, t (59) = 8.82, p < 1.0 × 106, but in the rock es-
erous trait inferences were used significantly more than
timation game, competence trait inferences were not used
punctual, t (59) = 5.80, p < 1.0 × 106, but in the rock estima-
significantly more than warmth, t (59) = 1.72, p = .091.
tion game, punctual trait inferences were not used signifi-
These findings satisfy axioms 2, 2b, and 2d.
cantly more than generous, t (59) = 1.20, p = .235. These
results satisfy axioms 1, 2, 2b, and 2d.
Trustworthy and precise trait inferences
We next replicated the analysis performed in Studies 1 and 2
using trustworthy and precise trait attribution (ratings of the CONCLUSION
terms “trustworthy” and “precise”) to generate cutoff criteria
for high and low trustworthiness and precision. There was a Study 3 demonstrated that people do infer traits from the be-
significant main effect of game, F (1, 58) = 152.87, havioral information in the profiles. However, these trait in-
p < 1.0 × 106, partial η2 = 0.73, Ω = 1.00, such that ferences were informed by both warmth and competence
employers used trait inferences more in the rock estimation behavioral information. It further demonstrated that these
game than trust game (Figure 6c). This main effect was trait inferences only partially replicate the pattern of behavior
qualified by a significant trait inference × game interaction, observed when using behavioral information or norm esti-
F (1, 58) = 22.07, p = 1.7 × 105, partial η2 = 0.28, Ω = 1.00. mates to predict warmth-related behavior. All trait infer-
There was no main effect of trait inference, F (1, 58) ences, whether to broad traits warmth and competence or
= 0.03, p = .873. specifically to traits that more narrowly described the con-
We computed paired sample t-tests with a Bonferroni text, failed to satisfy all three axioms. In particular, they all
corrected α = 3.125 × 103. When comparing games within failed to satisfy axioms 2a, 2c, and 3. Only inferences to
trait inferences, we found that trustworthiness trait infer- the specific traits (generous and punctual) from the initial be-
ences (t (59) = 9.15, p = 1.0 × 106) and precise trait havioral information satisfied axiom 1, suggesting that these
DOI: 10.1002/bdm
traits most likely served as heuristics. However, the data sug- Our findings also indicate that warmth behavioral informa-
gest that none truly did. tion is used significantly more overall than competence be-
havioral information, and this holds true even across person
perception domains. This is consistent with recent brain-
GENERAL DISCUSSION imaging evidence (Lee & Harris, 2014).
It should be noted that we are not contesting the general
This research combines literature in social psychology and premise that traits predict behavior: Indeed, we the authors
behavioral economics to demonstrate that people generalize hold this point of view because of the overwhelming data
to broad primary person perception dimensions trait warmth supporting this assertion. What the current data contest is that
and competence from single instances of behavior, but those traits predict behavior even when explanatory mechanisms
trait inferences do not guide future warmth-relevant deci- are not engaged. Indeed, we did not measure explanation in
sions better than estimates of normative information. People our samples along with prediction, so we cannot rule out
often infer traits from single acts of behavior. They then gen- the presence of explanation in our paradigm; future studies
eralize to broad trait domains to predict behavior. For in- will address this question. However, our paradigm is one of
stance, an observer witnesses an act of generosity, which the few where participants make consequential predictions
leads to a trait inference that the person is generous. That ob- and are not asked to report trait inferences, controlling for ex-
server may generalize across the warmth domain, thinking planatory mechanisms. People make spontaneous trait attri-
that the person may also be trustworthy, guiding the ob- butions when observing behavior when they are asked to
server’s decision to trust that person. Here, we demonstrate report on the specific attribution; our paradigm does not ask
that such generalizations across broad trait domains warmth participants to make such a report and finds that norms do
and competence exist, yet fail to predict warmth-relevant be- a better job.
havior as well as normative estimates of behavior. Additionally, our interest is not in addressing whether
We find that people depend on estimated norms relative to people ignore base-rate information. In such studies, partici-
the social context to predict warmth-relevant behavior. When pants are often provided with this information, not asked to
making decisions in a social context, people often have little generate it themselves. Here, we simply use normative infor-
or no information about how a specific person might behave. mation as a subjective assessment of the social context by
They do, however, have much more information available each individual employer, not as an objective truth regarding
about the behavior of other people in that social context more behavior. Therefore, our use of normative information devi-
generally. Because people are social agents, we often try to ates from the use of the term “base-rates” commonly opera-
predict what others will do, relying on heuristics to guide tionalized in the literature; the norm may not be based on
those decisions. While previous studies have identified trait the past behavior, but on estimates of behavior based on an
inferences as generalizations that guide these predictions, interpretation of the social context (this is a donation context,
our results show that although participants are capable of so people should donate a lot, but probably will donate a lit-
making broad trait inferences, they are in fact more likely tle). Employers may then evaluate the observed employee
using previous behavior and estimated norms to generate behavior in light of this standard, which may serve as a cut-
heuristics when predicting warmth-relevant behavior. off. Therefore, each prediction about future behavior may
All of the behaviors in the first two studies were the result rely on where the employee’s decision falls relative to the
of real people’s consequential decisions. Interestingly, broad cutoff. Hence, the normative information we describe, pro-
trait inferences occur across unrelated tasks—the ability to vided not by the experimenter but self-generated by the em-
estimate time is not related to estimating rocks, and the ployers, serves to create these arbitrary cutoffs, which are
amount of money donated to charity is not related to the like- used to predict behavior. Interestingly, there is still a trait
lihood of sharing money in the trust game—suggesting that component to this process perhaps, where employers view
broad trait inferences are not actually predictive of people’s an employee who comes in below the cutoff as less generous
actual behavior in our experimental paradigm. Psychology for instance, relative to other people, and therefore may be
has long ignored the question of accuracy of trait inferences less trustworthy relative to other people. In fact, our use of
(Zaki & Ochsner, 2011). By using consequential behavior the term normative information can be considered an individ-
in our task design, we are able to demonstrate that broad in- ual difference measure because we are examining behavioral
ferences are inaccurate and participants avoid using such in- deviation from these estimated normative means for the be-
ferences in predicting behavior: a reasonable occurrence. havior. However, it is difficult to test this assertion using
Additionally, the dissociation of the two person percep- the current paradigm. Future students can determine whether
tion domains suggests that generalizations do not occur uni- estimated norms are used differently for prediction than base-
formly for all types of social information. Trait warmth rate information framed as true or accurate information about
behavior affects investment decisions equally in both the the social context.
warmth and competence domains, while competence behav- Our results beg the following questions: Why does
ior affects decisions more in its own domain than the warmth warmth information hold a position of primacy, and why
domain. Participants were also significantly more confident are estimates of ideal norms closer to actual norms than per-
when making decisions in the warmth domain, and more ac- sonal norms or estimates of actual norms? Also, we referred
curately remembered warmth behavior, further suggesting a to the competence games as the investment and gambling
dissociation of processing between warmth and competence. games. There is often a stigma associated with “gambling.”
DOI: 10.1002/bdm
It is therefore possible that participants might have been bi- REFERENCES

ased against betting in the gambling game. People also infer
traits about others from photographs and faces (Todorov, Amodio, D. M., & Frith, C. D. (2006). Meeting of minds: the me-
Pakrashi, & Oosterhof, 2009; van’t Wout & Sanfey, 2008). dial frontal cortex and social cognition. Nature Reviews Neuro-
Although we used photographs of faces of employees in science, 7(4): 268–277.
Andrews, K. (2012). Do Apes Read Minds? Toward a new folk psy-
our initial two studies, we did not assess the extent to which chology. MIT Press, Cambridge: MA.
employees’ faces or perceived social identity influenced trait Asch, S. E. (1946). Forming impressions of personality. Journal of
inferences that guided behavior; future research could ad- Abnormal & Social Psychology, 41, 258–290.
dress the extent to which trait generalization from these im- Baird, B. M., & Lucas, R. E. (2011). “… And how about now?”: Ef-
ages interact with the prior behavior of the person to guide fects of item redundancy on contextualized self-reports of per-
sonality. Journal of Personality, 79(5), 1081–1112.
decisions. Finally, we did not test whether norm estimates Bargh, J. A. (1994). The four horsemen of automaticity: Intention,
predict competence-related behavior. Future research is awareness, efficiency, and control as separate issues.
needed to rule out alternate hypotheses presented by these Beck, L., McCauley, C., Segal, M., & Hershey, L. (1988). Individ-
potential limitations. ual differences in prototypicality judgments about trait catego-
Another limitation surrounds the trait ratings in Study 3; ries. Journal of Personality and Social Psychology, 55(2), 286.
Booth-Butterfield, S. (1988). Inhibition and student recall of in-
the mean trait ratings determine the employee’s standing on structional messages. Communication Education, 37(4),
the trait. This fairly crude method of testing “inferred traits” 312–324.
suggests that the traits will not always match the decision Botvinick, M. M., & Cohen, J. D. (2014). The computational and
maker’s own trait inference, and thus, this may not provide neural basis of cognitive control: Charted territory and new fron-
a very sensitive test. Consequently, it may not be highly sur- tiers. Cognitive Science, 38, 1249–1285.
Burke, P. A., Kraut, R. E., & Dworkin, R. H. (1984). Traits, consis-
prising or informative to find that trait inferences (based on tency, and self- schemata: what do our methods measure? Jour-
means from another sample) are not all that predictive of de- nal of Personality and Social Psychology, 47(3), 568.
cisions. However, this is a consequence of the paradigm and Chaplin, W. F., & Goldberg, L. R. (1984). A failure to replicate the
the attempts to make trait inferences less salient. Future re- Bem and Allen study of individual differences in cross-
search can require participants to also generate trait infer- situational consistency. Journal of Personality and Social Psy-
chology, 47(5), 1074.
ences of the past behavior, although at a time other than Church, A. T., Anderson-Harumi, C. A., del Prado, A. M., Curtis,
when participants make predictions. G. J., Tanaka-Matsumi, J., Valdez Medina, J. L., & Katigbak,
In conclusion, these studies suggest that a general as- M. S. (2008). Culture, cross-role consistency, and adjustment:
sumption about trait inferences as heuristics that guide be- Testing trait and cultural psychology perspectives. Journal Of
havior may be inaccurate. Given low cross-situational Personality And Social Psychology, 95(3), 739–755.
Dion, K., Berscheid, E., & Walster, E. (1972). What is beautiful is
behavior consistency, people may be intuitively aware of good. Journal of Personality and Social Psychology, 24,
the problems associated with using traits as predictive tools 285–290.
and rely instead on normative information and previous be- Emmons, R. A., & Diener, E. (1986). A goal-affect analysis of ev-
haviors to generate heuristics that may guide future behavior. eryday situational choices. Journal of Research in Personality,
This study by no means provides definitive evidence that 20(3), 309–326.
Engell, A. D., Haxby, J. V., & Todorov, A. (2007). Implicit
norms guide future behavior exclusively and traits play no trustworthiness decisions: Automatic coding of face properties
role. Perhaps traits also interact with norms to guide behav- in human amygdala. Journal of Cognitive Neuroscience, 19,
ior. Nonetheless, this research does highlight the difference 1508–1519.
between explanation and prediction and suggests there is Epstein, S., & Teraspulsky, L. (1986). Perception of cross-
value in conducting research that investigates prediction ex- situational consistency. Journal of Personality and Social Psy-
chology, 50(6), 1152.
clusively (without relying on self-report and tapping into ex- Eysenck, M. W., & Wild, J. (1996). Trait anxiety, social desirability
planatory mechanisms). Moreover, this research utilizes two and cross-situational variability. Personality and Individual Dif-
experimental approaches across the same basic paradigm, ferences, 21(2), 231–234.
shedding new light on an old phenomenon. Ferreira, M. B., Garcia-Marques, L., Hamilton, D., Ramos, T.,
Uleman, J. S., & Jerónimo, R. (2012). On the relation between
spontaneous trait inferences and intentional inferences: An infer-
ence monitoring hypothesis. Journal of Experimental Social
Psychology, 48(1), 1–12.
ACKNOWLEDGEMENTS Fiske, S. T., Cuddy, A. J. C., & Glick, P. (2007). Universal dimen-
sions of social cognition: Warmth and competence. Trends in
Start-up funds and a research award from Duke University to Cognitive Science, 11, 77–83.
Fiske, S. T., Lin, M., & Neuberg, S. L. (1999). The continuum
L. T. Harris and E. H. Thompson, respectively, funded this model: Ten years later. In S. Chaiken, Y. Trope, S. Chaiken,
research. This work comprised the senior honor thesis for Y. Trope (Eds.), Dual-process theories in social psychology
E. H. Thompson. These data were presented at the Society (pp. 231–254). New York, NY, US: Guilford Press.
for Neuroeconomics Meeting in Lausanne Switzerland, Sep- Fiske, S. T., & Neuberg, S. L. (1990). A continuum of impression
tember 2013. We would like to thank Naomi Ellermers and formation, from category-based to individuating processes: In-
fluences of information and motivation on attention and interpre-
Yael Coen for their comments on earlier versions of this tation. Advances in Experimental Social Psychology, 23, 1–74.
manuscript. L. T. Harris is now at the Department of Social Fiske, S. T., & Taylor, S. E. (2013). Social Cognition: From brains
and Organisational Psychology, Leiden University. to culture. Sage, London: UK.
DOI: 10.1002/bdm
Funder, D. C., & Colvin, C. R. (1991). Explorations in behav- Newman, L. S. (1996). Trait impressions as heuristics for predicting
ioral consistency: Properties of persons, situations, and future behavior. Personality And Social Psychology Bulletin,
behaviors. Journal of Personality and Social Psychology, 22(4), 395–411.
60(5), 773. Norenzayan, A., Choi, I, & Nisbett, R. E. (2002). Cultural similari-
Furr, R. M., & Funder, D. C. (2004). Situational similarity and be- ties and differences in social inference: Evidence from behav-
havioral consistency: Subjective, objective, variable-centered, ioral predictions and lay theories of behavior. Personality and
and person-centered approaches. Journal of Research in Person- Social Psychology Bulletin, 28(1), 109–120.
ality, 38(5), 421–447. Oishi, S., Diener, E., Napa Scollon, C., & Biswas-Diener, R.
Furr, R. M. (2009). Personality psychology as a truly behavioural (2004). Cross-situational consistency of affective experiences
science. European Journal of Personality, 23(5), 369–401. across cultures. Journal of Personality and Social Psychology,
Frith, U., & Frith, C. (2001). The biological basis of social interaction. 86(3), 460.
Current Directions In Psychological Science, 10(5), 151–155. Paunonen, S. V. & Jackson, D. N. (1985). Idiographic measurement
Heffler, B., & Magnusson, D. (1979). Generality of behavioural strategies for personality and prediction: Some unredeemed
data: IV. Cross-situational invariance of objectively measured promissory notes. Psychological Review, 92(4), 486.
behaviour. Perceptual and Motor Skills, 48(2), 471–477. Paunonen, S. V. & Jackson, D. N. (1986). Nomothetic and
Hemmelgarn, P. L., James, P. L., Ladd, P. L., & Mitchell, D. K. idiothetic measurement in personality. Journal of Personality,
(1995). Testing for cross-situational-consistency: A confirma- 54(2), 447–459.
tory factor analytic approach. Journal of Management, 21(1), Pervin, L. A. (1985). Personality: Current controversies, issues, and
121–139. directions. Annual review of psychology, 36(1), 83–114.
Hettema, J., & Hol, D. P. (1998). Primary control and the consis- Said, C. P., Baron, S., & Todorov, A. (2009). Nonlinear amygdala
tency of interpersonal behaviour across different situations. Eu- response to face trustworthiness: Contributions of high and low
ropean journal of personality, 12(4), 231–247. spatial frequency information. Journal of Cognitive Neurosci-
Kahneman, D., & Tversky, A. (1973). On the psychology of predic- ence, 21, 519–528.
tion. Psychological Review, 80(4), 237. Semin, G. R., & Krahé, B. (1988). Explaining perceived cross-situ-
Kelley, H. H. (1967). Attribution theory in social psychology. Ne- ational consistency: Intuitive psychometrics or semantic media-
braska Symposium On Motivation, 15, 192–238. tion? European Journal of Personality, 2(4), 239–252.
Koestner, R., Bernieri, F., & Zuckerman, M. (1989). Trait-specific Todorov, A. (2008). Evaluating faces on trustworthiness: An exten-
versus person-specific moderators of cross-situational consis- sion of systems for recognition of emotions signaling approach/
tency. Journal of Personality, 57(1), 1–16. avoidance behaviors. In A. Kingstone & M. Miller (Eds.), The
Krahé, B. (1986). Similar perceptions, similar reactions: An idio- Year in Cognitive Neuroscience 2008, Annals of the New York
graphic approach to cross-situational coherence. Journal of Re- Academy of Sciences, 1124, 208–224.
search in Personality, 20(3), 349–361. Todorov, A., Baron, S., & Oosterhof, N. N. (2008). Evaluating face
Landy, D., & Sigall, H. (1974). Beauty is talent: Task evaluation as trustworthiness: A model based approach. Social, Cognitive, &
a function of the performer’s physical attractiveness. Journal of Affective Neuroscience, 3, 119–127.
Personality and Social Psychology, 29, 299–304. Todorov, A., Pakrashi, M., & Oosterhof, N. N. (2009). Evaluating
Lee, V. K., & Harris, L. T. (2013). How social cognition can inform faces on trustworthiness after minimal time exposure. Social
social decision-making. Frontiers in Neuroscience, 1, 1–13. Cognition, 27, 813–833.
Lee, V. K., & Harris, L. T. (2014). Sticking with the nice guy: Trait Todorov, A., Dotsch, R., Porter, J., Oosterhof, N., & Falvello, V.
warmth information impairs learning and modulates person per- (2013). Validation of Data-Driven Computational Models of So-
ception brain network activity. Cognitive and Behavioral Neuro- cial Perception of Faces. Emotion, 13, 724–738.
science, 14(4), 1420–1437. van’t Wout, M., & Sanfey, A. G. (2008). Friend or foe: The effect of
Leikas, S., Lönnqvist, J. E., & Verkasalo, M. (2012). Persons, situ- implicit trustworthiness judgments in social decision-making.
ations, and behaviors: Consistency and variability of different Cognition, 108, 796–803.
behaviors in four interpersonal situations. Journal of Personality Van Mechelen, I. (2009). A royal road to understanding the mech-
and Social Psychology, 103(6), 1007. anisms underlying person-in-context behavior. Journal of Re-
Lippa, R., & Mash, M. (1981). The effects of self-monitoring and search In Personality, 43(2), 179–186.
self-reported consistency on the consistency of personality judg- Van Overwalle, F., Van Duynslaeger, M., Coomans, D., &
ments made by strangers and intimates. Journal of Research in Timmermans, B. (2012). Spontaneous goal inferences are often
Personality, 15(2), 172–181. inferred faster than spontaneous trait inferences. Journal of Ex-
Magnusson, D. (1976). The person and the situation in an interna- perimental Social Psychology, 48(1), 13–18.
tional model of behavior. Scandinavian Journal of Psychology, Welbourne, J. (2001). Changes in impression complexity over time
17(1), 253–271. and across situations. Personality And Social Psychology Bulle-
Magnusson, D., & Ekehammar, B. (1978). Similar situations—Sim- tin, 27(9), 1071–1085.
ilar behaviors?: A study of the intraindividual congruence be- Wells, B. M., Skowronski, J. J., Crawford, M. T., Scherer, C. R., &
tween situation perception and situation reactions. Journal of Carlston, D. E. (2011). Inference making and linking both re-
Research in Personality, 12(1), 41–48. quire thinking: Spontaneous trait inference and spontaneous trait
McCarthy, R. J., & Skowronski, J. J. (2011). What will Phil do transference both rely on working memory capacity. Journal of
next? Spontaneously inferred traits influence predictions of be- Experimental Social Psychology, 47(6), 1116–1126.
havior. Journal Of Experimental Social Psychology, 47(2), Willis, J., & Todorov, A. (2006). First impressions making up your
321–332. mind after a 100-ms exposure to a face. Psychological Science,
Milner, A. D., & Goodale, M. A. (2008). Two visual systems re- 17(7), 592–598.
viewed. Personality And Neuropsychologia, 46(3), 774–785. Winston, J. S., Strange, B. A., O’Doherty, J., & Dolan, R. J.
Moore, F. R., Filippou, D., & Perrett, D. (2011). Intelligence and at- (2002). Automatic and intentional brain responses during eval-
tractiveness in the face: Beyond the attractiveness halo effect. uation of trustworthiness of faces. Nature neuroscience, 5(3),
Journal of Evolutionary Psychology, 9, 205–217. 277–283.
Moskowitz, D. S. (1994). Cross-situational generality and the inter- Zaki, J., & Ochsner, K. (2011). Reintegrating accuracy into
personal circumplex. Journal Of Personality And Social Psy- the study of social cognition. Psychological Inquiry, 22,
chology, 66(5), 921–933. 159–182.
DOI: 10.1002/bdm
Authors’ biographies: interests focus on examining how social cognition and person
perception affect decision making in economic contexts.
Lasana T. Harris is a senior lecturer at University College
London in Experimental Psychology and a guest lecturer in So- Elizabeth Thompson earned her bachelors degree from Duke
cial and Organisational Psychology at Leiden University. He re- University. She is currently a strategist at a branding and design
ceived his PhD from Princeton University, where he studied consultancy. Her research interests include social neuroscience
both social psychology and cognitive neuroscience. His research and neuroeconomics.
interests explore the boundary conditions of social cognition
(dis)engagement. Authors’ addresses:
Rachel E. Kranton is James B. Duke Professor of Economics at Lasana T. Harris, Psychology and Neuroscience, Duke Univer-
Duke University. She earned her PhD in Economics at the Uni- sity, Durham, NC, USA.
versity of California, Berkeley. She is a fellow of the Economet-
ric Society and has held fellowships at the Russell Foundation Victoria K. Lee, Psychology and Neuroscience, Duke University,
and the Institute for Advanced Study. Her research brings social Durham, NC, USA.
interactions, networks, and identity into formal economic models
of decision making and strategic interaction. Elizabeth Thompson, Psychology and Neuroscience, Duke Uni-
versity, Durham, NC, USA.
Victoria K. Lee is a PhD candidate in the Social Psychology
program at Duke University. She received her BA from Rutgers Rachel E. Kranton, Economics, Duke University, Durham, NC,
University, where she majored in psychology. Her research USA
DOI: 10.1002/bdm

Exploring The Generalization Process From Past Behavior To Predicting Future Behavior

Uploaded by

Copyright:

Available Formats

Exploring The Generalization Process From Past Behavior To Predicting Future Behavior

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Exploring The Generalization Process From Past Behavior To Predicting Future Behavior

Uploaded by

Copyright:

Available Formats

Journal of Behavioral Decision Making, J. Behav. Dec.

Exploring the Generalization Process from Past Behavior to Predicting Future

Copyright © 2015 John Wiley & Sons, Ltd.

Rock estimation game. We modeled the rock estimation

Figure 2. Investment decisions in Study 1. Bar graphs depict the

of memory accuracy. We ﬁrst tested employers’ recall accu-

ﬁnd a signiﬁcant difference, t (31) = 1.31, p = .200, suggest- Method

Study 1 replicates previous behavioral economics studies

Behavioral intentions results of Study 1, demonstrating that employers accurately

Speciﬁc warmth and competence trait inferences

It is therefore possible that participants might have been bi- REFERENCES

You might also like