ArticlePDF Available

Abstract and Figures

Most companies use codes of conduct, ethical training and regular communication to assure employees know about rules to follow to avoid misconduct. This paper focuses on the type of language used in codes of conduct, and shows that impersonal language (“employees/members”) and personal, communal language (“we”) lead to different behaviors because they change how people perceive the group or organization they are part of. Using multiple methods, including lab- and field-based experiments (N = 1,443) and a large dataset of S&P firms (i.e., U.S. publicly traded, large companies that are part of Standard & Poor’s stock market index), we robustly demonstrate that personal, communal language (as compared to impersonal language) influences perceptions of a group’s warmth, which, in turn, increase levels of dishonesty among the members.
Content may be subject to copyright.
1
The Ethical Perils of Personal, Communal Relations: A Language Perspective
Maryam Kouchaki
Kellogg School of Management, Northwestern University
Francesca Gino
Harvard Business School
Yuval Feldman
Faculty of Law, Bar-Ilan University
2
Abstract
Most companies use codes of conduct, ethical training and regular communication to assure
employees know about rules to follow to avoid misconduct. This paper focuses on the type of
language used in codes of conduct, and shows that impersonal language
(“employees/members”) and personal, communal language (“we”) lead to different behaviors
because they change how people perceive the group or organization they are part of. Using
multiple methods, including lab- and field-based experiments (N = 1,443) and a large dataset
of S&P firms (i.e., U.S. publicly traded, large companies that are part of Standard & Poor’s
stock market index), we robustly demonstrate that personal, communal language (as
compared to impersonal language) influences perceptions of a group’s warmth, which, in
turn, increase levels of dishonesty among the members.
Keywords: Dishonesty; Perception of Warmth; Communal Relation; Code of conduct;
Language Effects
3
Over the past three decades, numerous cases of corporate corruption and misconduct
have had a deep impact on the economic and political landscape, and have led organizational
members and the public at large to become more distrusting and cynical of leaders and firm
practices. In the wake of a large number of corporate scandals, from the fall of Enron in the
1990s to the more recent scandals at Wells Fargo and Volkswagen, policymakers have
responded with tighter regulations, and leaders in companies across the globe have vowed to
act more ethically (Feldman, 2018).
To make ethics salient within organizations and curb unethical behavior, regulators
and companies have primarily focused on codes of conduct based on a set of moral standards
to clarify appropriate and inappropriate behaviors, standards that are often similar across
companies (Paine, Deshpande, Margolis, Bettcher, 2005). Codes of conduct, as well as other
forms of corporate communication, reflect the company culture and shape what employees
believe is and is not important (Weick, 1979). Yet, past empirical evidence on the
effectiveness of codes of conduct in promoting ethical behavior in organizations is mixed
(e.g., Kish-Gephart, Harrison, Treviño, 2010). However, research suggests, the degree to
which an organization enforces its code of conductthat is, whether the organization
conveys to employees that it punishes code violationsdoes make a difference (Kish-
Gephart et al., 2010). A salient component of enforcement is the type of language used in
corporate communications, and codes of conduct more specifically.
At the extremes, language can be either personal and communal (i.e., “we”) or rather
impersonal (i.e., “employees”). For example, Wells Fargo’s code of conduct, entitled “Our
Code of Ethics and Business Conduct: Living Our Vision, Values & Goals,” uses the
communal language “we” (e.g., “We must never lose sight of putting our customers first and
helping them succeed financially”). By contrast, 3M’s code of conduct, “3M Compliance and
Business Conduct: Be 3M, uses the more impersonal term “3M People (e.g., “3M has a
4
longstanding and well-deserved reputation for doing business with uncompromising honesty
and integrity because 3M People obey the law and 3M’s Code of Conduct”).
We argue that these different types of language used in codes of conduct result in
different perceptions members have of the company they are part of, as the language signals
the type of relationship one has with others in one’s group or organization. Personal language
(i.e., “we”) highlights a communal relationship, while impersonal language (i.e.,
“employees”) refers to an exchange relationship (Clark & Mills, 1979; Mills & Clark, 1982).
Communal relationships are often exemplified by close relationships and family ones.
Exchange relationships, instead, are often exemplified by business or work-related relations,
and relations between acquaintances. Members of a communal relationship assume that “each
[individual] is concerned about the welfare of the other” (Clark & Mills, 1979, p.13), while
more distance exists among members in exchange relations. According to previous research,
individuals in a communal relationship, relative to those in an exchange relationship, for
example, tend to follow communal norms (Clark, 1984; Clark, Mills, & Powell, 1986), help
others more (Clark, Ouellette, Powell, & Milberg, 1987), and do not keep track of individual
inputs for rewards or punishments (Clark, 1984).
Language that is personal (i.e., “we”), then, signals a more communal environment in
which members are likely perceived to be easygoing, helpful, cooperative and forgiving of
one another. Instead, impersonal language (i.e., “employees”) signals a more exchange-based
environment in which members are more formal and distant. Perceptions like these are
captured by ‘warmth’—a fundamental dimension, with competence, of judgments we make
when forming impressions of others (Fiske, Cuddy, Glick, & Xu, 2002), including groups and
organizations (Aaker, Vohs, & Mogilner, 2010). Though some prior work has conceptualized
warmth as interchangeable with moral character traits, here we rely on the work by Goodwin,
Piazza and Rozin (2014) which shows that the two are separable. In this conceptualization,
5
warmth traits that are high in moral relevance, such as kindness, lovingness and generosity,
are considered “moral traits” that are distinct from “pure” warmth traits. These latter traits are
less relevant to morality, and include being easy-going, tolerant and forgiving. We argue that
companies that use personal “we” language in their corporate communications, as compared
to those using impersonal language, signal a communal rather than exchange relationship to
their members and thus signal that the group is warmer on pure warmth traits (i.e., tolerant,
easygoing, and forgiving). These warmth traits capture how members perceive the group in
terms of potential punishment due to behavior the group does not support. A more tolerant
and forgiving group, in fact, is likely to be perceived as less likely to punish its members for
misconduct or other forms of unacceptable behaviors.
We also argue that these perceptions are consequential, as they influence dishonest
behavior of organizational members. In many situations, whether at work or in life, external
pressures or our own selfish desires conflict with our moral values (Mazar, Amir, & Ariely,
2008). When balancing these competing motivations as we face the choice of whether to act
dishonestly or not, one aspect we consider is the cost of being caught, including the potential
punishment. In fact, according to standard economics and legal theorizing, the probability of
being caught and the magnitude of punishment are key determinants of dishonesty (Becker,
1968). Prior research in economics has mainly focused on explicit punishment expectations,
that is punishment that can be enforced. For instance, enforceable punishment has been found
to enhance a form of ethical behavior, namely cooperation, in public good games (Gachter,
Renner, & Sefton, 2008), though the presence of punishment does not always lead to less
unethical behavior (Sanchez-Pages & Vorsatz, 2007). Related research by Bernhard, Fehr and
Fischbacher (2006) found that participants’ expectations to be punished by a third party
affected how much money they offered to recipients in a dictator game, and that those
expectations varied depending on whether the third-party was an in-group or an out-group
6
member. Specifically, dictators expected to be punished much less by ingroup rather than
outgroup members for their violations (i.e., transferring less than 50% of their endowment to
the recipient, thus violating an egalitarian norm), and the dictators who expected harsher
punishment gave significantly more money to the recipient.
Though in studies in the economics literature punishment is generally enforceable,
there is ample evidence from social psychology that even implicit perceptions of punishment,
with no actual perceived threat of being caught or punished, can affect behavior. For instance,
Mazar et al (2008) found that participants in an experiment in which they could cheat acted
less dishonestly when they first read an honor code rather than not, even though, objectively,
the honor code had no implications for the potential of being punished since participants
played in an anonymous context with no identifiable information. In another study, Shariff
and Norenzayan (2011) found that people who believed in a comforting and forgiving God
(i.e., a God who is less likely to punish) were more likely to cheat than those who believed in
a punitive and angry God. Together, these lines of research suggest that a greater expected
severity of punishment if caught reduces the likelihood of cheating.
Here, we propose that perceptions of (pure) warmth—capturing a person’s or group’s
willingness to forgive and be easygoinginfluence moral behavior by altering such
expectations
1
. Specifically, we expect that perceptions of warmth, by indirectly influencing
the expected magnitude of punishment, will increase dishonest behavior. Thus, in situations
1
Although, to our knowledge, no research directly links warmth perceptions of a decision maker or group to
rule-breaking and other unethical behavior, some research has explored related questions. The most similar
study is one by Azevedo and colleague (2018) who manipulated the perceptions of warmth and competence of a
recipient in an anonymous dictator game that offered participants opportunities to lie. In this game, participants
had total control over both their own and the recipient’s pay-off. Participants were told that they were playing
against four opponents, each representing one of four categories, and described in a different way: high warmth
high competence [HW-HC], high warmthlow competence [HW-LC], low warmthlow competence [LW-LC],
or low warmthhigh competence [LW-HC]). For instance, the LC-LW opponent was described as a low-class
Eastern Europe migrant, while the HW-HC opponent was a friendly music student. The results showed
independent effects for warmth and competence such that individuals were less likely to deceive those
opponents seen as warm, and more likely to lie to highly competent ones. Thus, participants engaged in different
levels of deception based on the perceived characteristics of the person they were interacting with.
7
where employees face opportunities to behave unethically, we suggest that those in a
communal relationship, relative to an exchange relationship, perceive misconduct as less
costly because they see their group as more forgiving and tolerant, which would not impose
the potential negative consequences of rule-breaking.
Though in our research we mainly focus on how different types of language in codes
of conduct influence perceptions of warmth and, through them, dishonest behavior, we also
consider the role of group identification. Personal language may in fact result in higher group
identification as compared to impersonal language, and such higher group identification may
influence the choice of whether to act dishonestly, though prior research has often not found
evidence for this link (e.g., Hildreth, Gino, & Bazerman, 2016).
The Present Research
To shed light on our hypotheses, we conducted nine studies that employ a variety of
methods and sample populations. Importantly, in our studies, except in Study 2b, participants
completed tasks anonymously with no identifiable information, and thus with no explicit
expectation to be punished or potentially caught.
Study 1a
This study examines how the use of personal (i.e., “we”) versus impersonal (i.e.,
group members”) language in group communications affects members’ perceptions of the
group. In addition, to test an alternative mechanism, we included a measure of group
identification to examine whether the type of language used influences identification with the
group.
Method
Participants and design. Ninety-five individuals (47.4% male; Mage=30.3, SD=12.9)
at a university in the United States participated in this study for pay. Participants received $20
for completing an hour-long session. Study 1 employed a two-condition between-subjects
8
design (members-code, we-code). In this first study, we aimed for about 50 participants per
condition, consistent with the recommendations of Simmons et al. (2013). The sample size
was determined by the number of participants who showed up during the scheduled
laboratory sessions. Before conducting the study, we planned to stop data collection after the
scheduled sessions were over, hoping to recruit at least about 50 participants.
Procedure. To mirror the role of corporate communication with their members, in our
first study we recruited participants to become members of the research laboratory of one of
the authors. We told participants that researchers were currently looking to hire participants
to help the research group pilot-test different surveys. We randomly assigned participants to
one of two conditions and varied the content of the documents they read when joining the
research group, including a code of conduct. We had 47 participants in the we-code condition
(i.e., personal, communal language) and 48 in the members-code condition (i.e., impersonal
language). We introduced the manipulation at the beginning of the study session. The
instructions informed participants that the research group was interested in understanding the
factors that influence task performance. Next, participants received the code-of-conduct
manipulation and then completed a survey.
Code-of-conduct manipulation. The two different codes that participants received
were similar, except for the language used to introduce them to the group. In the members-
code version, the code of conduct was written in impersonal language from the group’s
perspective, referring to “[lab’s name]” and “members,” while the we-code condition had
words like “we” and “you.” For instance, in the members-code version, participants read,
“[name of group] emphasizes integrity, fairness, and respect. All staff including study
participants are expected to exercise his or her best judgment and behave according to
the highest ethical standards. [name of group] values integrity in all aspects of the
group’s work. Honesty and quality must come from every person. Only in this manner
9
can the group generate valid results which other organizations and teams can use
effectively.
In the we-code version, this passage read,
“At [name of group], we emphasize integrity, fairness, and respect. We expect each
other to exercise our best judgment and behave according to the highest ethical
standards. We value integrity in all aspects of our work. Honesty and quality stem
from each of us as members of this group. Only then can we generate valid results
which other organizations and teams can use effectively.
Perceptions of the group. After the manipulation, participants responded to a number
of questions about their group. As we explained earlier, even though perceptions of warmth
and morality are closely linked, recent work has made a clear theoretical and empirical
distinction between warmth perceptions (conceptualized as inclusive of highly morally
relevant traits, such as sincerity, as well as less morally relevant traits, such as friendliness)
and moral character traits that are low on warmth (e.g., principled, just, trustworthiness)
(Goodwin et al., 2014). Though we predicted that the language used in codes of conduct
would affect pure warmth traits, for completeness we use an inclusive list of attributes
capturing perceptions of both warmth and morality. Participants were provided with a list of
21 traits (adapted from Goodwin et al., 2014), six from the High Warmth/High Morality
category (forgiving, helpful, kind, empathetic, cooperative, tolerant; α = .91) and five from
the High Warmth/Low Morality category (agreeable, warm, sociable, easygoing, playful; α =
.91). There were four traits that captured moral character traits (Low Warmth/High Morality
category; fair, principled, responsible, trustworthy, α = .89) and six that measured
Competence category (competent, effective, efficient, innovative, organized, logical; α = .89).
Participants rated their perceptions on a 7-point scale of 1 (not at all) to 7 (very much).
10
We included three items (α = .94) to measure group identification (adapted from
Doosje, Ellemers, & Spears, 1995). Sample items are: “I feel a strong sense of belonging to
this group” and “I identify with this group.”
In addition, at the end of the study, we asked participants a few questions to see if the
language in the code was comprehended differently or not by condition. Specifically, we
asked, “How clear was the writing in the group’s goals and values document that you read?”
“How abstract would you rate the group’s goals and values document that you read?” and
“How well did you understand the group’s goals and values document that you read?”
Results
Perceptions of the group’s (pure) warmth. The language used (we vs. members)
increased perceptions of traits related to High Warmth/Low Morality (M we-code = 5.13, SD =
1.16, 95% CI[4.79,5.47] vs. M members-code = 4.47, SD = 1.31, 95% CI[4.09,4.85]; t(93) =2.62,
p = .010) and also those related to High Warmth/High Morality (M we-code = 5.26, SD = 1.17,
95% CI[4.91,5.60] vs. M members-code = 4.44, SD = 1.28, 95% CI[4.07,4.81], t(93) =3.23, p =
.002). See Figure 1.
Perceptions of the group’s morality and competence. However, the language used
did not significantly influence perceptions of Low Warmth/High Morality (M we-code = 5.84,
SD = 1.03, 95% CI[5.54,6.14] vs. M members-code = 5.58, SD = 1.28, 95% CI[5.21,5.95]), t(93)
=1.08, p = .284, nor those of Competence, (M we-code = 5.68, SD = .93, 95% CI[5.41,5.96] vs.
M members-code = 5.54, SD = 1.02, 95% CI[5.24,5.83]), t(93) =.74, p = .462.
Group identification. Group identification did not differ between conditions either
(M we-code = 4.03, SD = 1.59, 95% CI[3.56,4.49] vs. M members-code = 3.50, SD = 1.78, 95%
CI[2.98,4.02]), t(93) =1.53, p = .130.
11
Reactions to the code. We also found no significant differences between the two
codes in terms of how clear, abstract, and understandable participants found them to be (p’s >
.323).
Figure 1. Rating of traits attributed to group across conditions in Study 1a.
In this study, we found initial evidence that the use of personal (i.e., “we”) versus
impersonal (i.e., “group members”) language in group communications affects members’
perceptions of the group, leading them to view the group as warmer. However, we did not
find any difference on group identification between the two conditions (we-code and
members-code). It is possible that our direct request to join the research team in a laboratory
experiment was not a strong manipulation and, as such, the difference in identification based
on the language in the code of conduct was not salient enough to influence their level of
identification. Nevertheless, we include this measure in our next study to further investigate
3.0
4.0
5.0
6.0
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
High
Warmth/Low
Morality
High
Warmth/High
Morality
Low
Warmth/High
Morality
Competence Group
Identification
Ratings
12
the role of group identification, since group affiliation has been shown to affect willingness to
forgive egalitarian-norm violations and punishment expectations (Bernhard et al., 2006).
Study 1b
This preregistered study aims to replicate the effect of the use of personal versus
impersonal language on perceptions of warmth. We also directly measure group members’
perceptions of the magnitude of punishment if they were caught.
Method
Participants and design. Two hundred and three individuals (56.7.4% male;
Mage=36.1, SD=11.5) recruited from Amazon’s Mechanical Turk participated in this study in
exchange for $0.50. Study 1b employed a two-condition between-subjects design (members-
code, we-code). We aimed for approximately 100 participants per condition. This study was
pre-registered with materials available at
https://osf.io/epsva/register/5771ca429ad5a1020de2872e.
Procedure. We recruited participants to become members of the research laboratory
of one of the authors and varied the content of the code of conduct they read when joining the
research group. We used similar code-of-conduct instructions as in Study 1a. We randomly
assigned participants to one of the two conditions. We had 101 participants in the we-code
condition and 102 participants in the members-code condition.
Perceptions of the group. After the manipulation, participants responded to a number
of questions about their group in random order.
We measured expected magnitude of punishment if caught with two items (“To what
extent do you think a member of this group would be punished if they are caught making a
mistake?” and “How severe do you expect the punishment to be if a member of this group is
caught making a mistake?”; α = .88). In addition, we included two items to assess the
expected probability of being caught (How probable do you think it is in this group that your
13
actions would be scrutinized?” and “How probable do you think it is in this group to be
detected if you make a mistake?”; α = .70). These four items were measured on a 9-point
scale ranging from 1 (not at all) to 9 (extremely).
Similar to Study 1a, participants were provided with a list of 21 traits (adapted from
Goodwin et al., 2014) to rate on a 7-point scale of 1 (not at all) to 7 (very much): High
Warmth/High Morality category (forgiving, helpful, kind, empathetic, cooperative, tolerant; α
= .91); High Warmth/Low Morality category (agreeable, warm, sociable, easygoing, playful;
α = .87); Low Warmth/High Morality category (fair, principled, responsible, trustworthy, α =
.88); and Competence category (competent, effective, efficient, innovative, organized,
logical; α = .92). Finally, we included three items (α = .96) to measure group identification as
in Study 1a.
Results
Expected severity of punishment. Consistent with our theorizing, the language used
(we vs. members) lowered participants’ expectation of the magnitude of punishment for
violations (M we-code = 4.71, SD = 1.87, 95% CI[4.34,5.08] vs. M members-code = 5.73, SD = 1.74,
95% CI[5.39,6.08]), t(201) =4.03, p < .001.
Probability of punishment. However, there were no differences between conditions
on the perceived probability of being caught (M we-code = 6.39, SD = 1.66, 95% CI[6.06,6.72]
vs. M members-code = 6.60, SD = 1.62, 95% CI[6.28,6.92]), t(201) =.92, p = .358.
Perceptions of the group’s (pure) warmth. As shown in Figure 2, participants rated
the research group in the we-code condition higher on High warmth/Low morality traits (M
we-code = 4.53, SD = 1.16, 95% CI[4.30,4.76]) than they did in the members-code condition (M
member-code = 4.07, SD = 1.43, 95% CI[3.79,4.35]), t(201) = 2.51, p = .013. Perceptions of High
warmth/High morality also differed between the two conditions, (M we-code =5.05, SD = 1.13,
14
95% CI[4.82,5.28] vs. M member-code = 4.63, SD = 1.35, 95% CI[4.37,4.90]), t(201) =2.41, p =
.017.
Figure 2. Rating of traits attributed to group across conditions in Study 1b.
Perceptions of the group’s morality and competence. We found no significant
differences between conditions on perceptions of moral character (M we-code = 5.70, SD =
1.21, 95% CI[5.47,4.95] vs. M member-code = 5.68, SD = .99, 95% CI[4.48,4.87], t(201)<1), nor
on perceptions of competence (M we-code = 5.75, SD = 1.09, 95% CI[5.54,4.97] vs. M member-code
= 5.80, SD = .98, 95% CI[5.60,5.99], t(201) < 1).
Group identification. The language used did not produce differences on group
identification (M we-code = 4.51, SD = 1.51, 95% CI[4.21,4.81] vs. M member-code = 4.15, SD =
1.77, 95% CI[3.80,4.49], t(201) =1.58), p = .116.
3.0
4.0
5.0
6.0
7.0
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Expected
magnitude of
punishment
Expected
probability of being
caught
High Warmth/Low
Morality High Warmth/High
Morality Low Warmth/High
Morality Competence Group
Identification
Ratings
15
This preregistered study provided further evidence for the effect of personal versus
impersonal language on perceptions of warmth. Group identification, instead, was again not
significantly affected by our manipulation.
Study 2a
Having demonstrated that the use of personal (i.e., “we”) versus impersonal (i.e.,
group members”) language in codes of conduct affects perceptions of warmth of the group
with that code of conduct, this study examines how language also influence the members’
choice to act dishonestly.
Method
Participants and design. One hundred twenty individuals (66.7% male; Mage=31.9,
SD=10.0) recruited from Amazon’s Mechanical Turk participated in this study in exchange
for $1 and had the opportunity to earn up to $1 during the study. Study 2a employed a two-
condition between-subjects design (members-code, we-code). We aimed for approximately
50 participants per condition, consistent with the recommendations of Simmons et al. (2013).
Procedure. We used similar code-of-conduct instructions as in our previous studies.
We had 58 participants in the we-code condition and 62 in the members-code condition.
Dishonesty measure. Participants completed a performance task. They received 10
matrices (Kouchaki & Smith, 2014), each appearing for 15 seconds on their computer screen.
Each matrix had 12 three-digit numbers (e.g., 2.56 and 7.44), and participants were asked to
find the two numbers that summed exactly to 10.00. Participants had an opportunity to cheat:
They were told to indicate that they found the matching pair by clicking the “found it” box on
the screen and were not asked to actually reveal the matching pair. To identify clear instances
of cheating, we provided participants with five matrices that were unsolvable (i.e., they did
not contain two numbers that summed to 10); the other five were solvable. Participants were
told that for each pair they found, they would receive 10 cents.
16
Results
A t-test on the number of all matrices reported solved revealed a significant difference
between the two code conditions, t(118) =2.52, p = .013 (M we-code = 6.36, SD = 2.89, 95%
CI[5.60,7.12] vs. M member-code = 5.06, SD = 2.76, 95% CI[4.36,5.76]). Participants in the we-
code condition reported solving a higher number of both unsolvable matrices (M = 2.83, SD =
1.85, 95% CI[2.34,3.13]) and solvable ones (M = 3.53, SD = 1.26, 95% CI[3.20,3.87]) as
compared to those in the members-code condition (unsolvable: M = 2.06, SD = 1.62, 95%
CI[1.65,2.48] and solvable: M = 3.00, SD = 1.38, 95% CI[2.65,3.35]), t(118) =2.41, p = .017.
and t(118) =2.21, p = .029, respectively.
Overall, the results of this study show that the use of personal (i.e., “we”) versus
impersonal (i.e., “group members”) language impacts dishonest behavior.
Study 2b
In Study 2b, to provide further evidence of the relationship between the use of
personal versus impersonal language and individual misconduct, we conducted a field
experiment on Upwork (formerly oDesk), an online labor market of registered freelancers.
Using this platform allowed us to hire people who’d be working for a few weeks in a real
setting. This was a natural field experiment method because, although participants were
informed they would be joining a research group, they did not know that their instructions
were part of an experiment. The oDesk platform allowed us to conduct targeted hiring of
workers with data-entry expertise at a specific base wage, which gave us the control over
recruitment needed for a field experiment.
Method
Participants and design. We hired freelancers for a data-entry task. All of them
identified themselves as administrative support professionals and classified themselves with a
data-entry specialty, according to their profiles. We restricted the recruitment to North
17
American freelancers, a subcategory that lists requested hourly wages of less than $12,
according to their profiles. To ensure that we included active workers in the sample, we
further restricted our sample to those who had logged into the platform within the last 30
days. We ended up with a total of 438 active workers in our sub-category at the time of data
collection (late 2014).
We notified all the workers that we had a job for them and invited them to accept the
job offer within the next five days. We told them that we were looking to hire a group of
people to help our research team with simple data entry or pilot tests over the course of three
weeks. They were informed that each week they would receive a link through the platform
that would direct them to an external website where they were to complete the task within
five days. Over the course of the three weeks, participants would complete 1.5 hours of work
and earn at least $20 (not including potential bonuses). They were informed that the entire job
consisted of an orientation survey (approximately 15 minutes in duration) for which they
could earn $6 and two tasks (approximately 30 minutes each in duration) for which they
could earn $14. We contacted 438 freelancers (28% male). One hundred and fifty-seven
workers (31.8% male) accepted our offer of employment within five days, an overall take-up
rate of 36%.
Procedure. Each Monday, we sent workers a message through the platform with a
link to an external website where they could complete a survey within five days. They were
paid at the end of each week. We randomly assigned workers who accessed the orientation
survey to one of the three conditions (members-code, we-code, no-code). Out of 157 workers
who were hired, 151 completed the orientation survey and were randomly assigned to
different code-of-conduct conditions (51 in no-code, 50 in the members-code, and 50 in the
we-code).
18
In the no-code condition, participants only were given instructions about their tasks to
be completed the following weeks and then filled out a short demographic survey. In the
members-code and we-code conditions, participants started with an orientation activity. As
part of this onboarding process, participants received similar code-of-conduct instructions as
in our previous studies. Then they completed a short demographic questionnaire.
All participants were informed that the link for the following week’s performance task
would be sent to them on Monday and were told they had five days to complete the task. On
Monday of Week 1, we sent each participant who completed the orientation (n = 151) a
message through the platform with a link based on his/her assigned condition to make sure
we controlled for the language used throughout the task. Based on the condition participants
had been assigned to, they were briefly reminded of the code of conduct they had read a few
days earlier.
Afterwards, they received instructions about the tasks they had to complete. They first
started with a filler typing task. Next, we gave them a CAPTCHA activity that involved
typing in letters and numbers displayed in an ambiguous image. They were informed that the
purpose of this task was to determine their average speed of visual processing and typing.
Given that we only recruited workers who listed data entry as a specialty on their oDesk
profiles, we chose a data entry task (entering CAPTCHAs) that is fairly common in online
labor markets. There was a reasonable demand on the platform for data entry, and specifically
for entering CAPTCHAs; thus, this was an ordinary request that would not arouse the
suspicion of workers that they were part of an experiment.
Dishonesty measure. All participants completed a first round of 65 CAPTCHAs.
Next, we measured cheating by giving participants the opportunity to overstate their
performance on a next round of data entry and thus earn undeserved money. Participants
were informed that in a second round they would have three minutes to complete a
19
CAPTCHA task again. However, in this round, their goal would be to complete 35
CAPTCHAs in three minutes, and they would be paid an additional $2 if they exceeded this
goal. They were further told that, given the complexity of the task, the online survey platform
could not check their work, so they would be the ones to provide the number of CAPTCHAs
they completed, and their bonuses would be paid based on their self-reports. Thus, we gave
workers an opportunity to lie about their performance and receive an additional $2.
Similarly, on Monday of Week 2, we sent a message through the platform with a link
to another survey (n = 141). We used the same task from the previous week with a new set of
CAPTCHAs and once again gave workers the opportunity to lie during their second round
about their performance on the CAPTCHA task and earn an additional $2.
Results
Out of 151 workers who were assigned to different code-of-conduct conditions, 141
of them completed the CAPTCHA task in the Week 1 (49 in no-code, 43 in members-code,
and 49 in the we-code). We compared the number of workers in each of the three conditions
in Week 1 who lied to earn a $2 bonus (i.e., they indicated that they had completed 35
CAPTCHAs when in reality they did not), and found a significant difference in unethical
behavior, χ2(2, 141) = 8.63, p = .013 (see Figure 3). More workers in the we-code condition
lied about their performance (20/49; 40.1%) as compared to those in both the members-code
condition (6/43; 14%, p = .004) and the no-code condition (12/49; 24.5%, p = .085), though
the difference between the we-code condition and the no-code condition was non-significant.
There was also no significant difference in lying between the members-code and the no-code
conditions (p = .20).
During Week 2, out of 141 workers who participated in Week 1, 123 workers (42 in
no-code, 39 in members-code, 42 in we-code) returned and entered a new set of CAPTCHAs.
We compared the number of workers across conditions who lied to earn a $2 bonus, and
20
found again a significant difference, χ2(2, 123) = 10.18, p = .006. More workers in the we-
code condition lied about their performance (8/42; 19%) as compared to both the members-
code condition (1/39; 2.6%, p = .018) and the no-code condition (1/42; 2.4%, p = .014).
There was no significant difference in lying between the members-code and the no-code
conditions (p = .96).
Figure 3. Percentage of cheaters across conditions in Study 2b.
We also performed repeated-measures analyses to test the effects of the code-of-
conduct manipulation on the level of dishonesty each week. Results showed a significant
main effect of time, F(1, 120) = 26.40, p < .001, and no significant interaction of condition by
time F(2, 120) = 1.55, p = .22. As expected, we found a significant main effect of condition,
F(2, 120) = 5.82, p = .004. The significant main effect of time demonstrates that participants
in all conditions were less likely to cheat the second week as compared to the first one. We
had no a-priori prediction about this, and thus we can only speculate as to why people
21
cheated less over time post-hoc. When looking more closely to the cheaters in Week 2, we
found that most of them (90%) also cheated during Week 1.
The results of this study provide further evidence for an effect of the use of personal
versus impersonal language on dishonest behavior. In this field experiment, participants were
not anonymous and therefore there was some threat of being caught or punished if caught.
However, we did not explicitly manipulate the expectations to be caught or punished.
Study 2c
In Study 2c, we attempt to replicate our findings in a controlled laboratory setting.
Although we did not find a difference in identification in Studies 1a-1b, we tested for group
identification once more as a possible mechanism explaining why personal, communal
language results in higher levels of dishonesty.
Method
Participants and design. One-hundred thirty-four students (33.8% male; Mage=20.5,
SD=1.8) at a university in the United States participated in this study for pay. We used the
same recruiting strategy as in Study 1a. The experimental session lasted up to 60 minutes,
and the students received $20 for participation with the possibility of earning another $10,
depending on their performance. Study 2c employed a three-condition, between-subjects
design (no-code, members-code, we-code). The sample size was determined by the number
of participants who showed up during the scheduled sessions. Before conducting the study,
we planned to stop data collection after the scheduled sessions were over, hoping to recruit at
least about 50 participants for each condition, similar to our earlier studies.
Procedure. We randomly assigned each session to one of the three conditions. We
had 47 in the we-code condition, 43 in the members-code condition, and 44 in the no-code
condition. We ran a total of 18 sessions with a minimum of three participants in each session.
The experimenter was blind to the session condition.
22
The code manipulation was identical to that used in previous studies. In the no-code
condition, participants did not receive a code of conduct and proceeded to complete the first
task. In the other two conditions, we introduced this manipulation at the beginning of the
session, informing participants that they would help pilot-test a number of performance tasks.
Each session started with two performance filler tasks, CAPTCHAs, and typing. Each task
was timed, for about 10 minutes. Afterward, we measured cheating by giving participants the
opportunity to overstate their performance on a problem-solving task and thus earn
undeserved money.
We gave participants an envelope that contained $10 (nine $1 bills and four quarters)
along with a sheet of paper. The paper was a worksheet with 20 matrices, each consisting of
12 three-digit numbers (e.g., 2.56 and 7.44) (Mazar et al., 2008). Participants had four
minutes to find two numbers in each matrix that added up to 10.00. For each correctly solved
matrix, they would earn $0.50. After four minutes, participants were asked to correct their
own answers, report the number of problems they correctly solved on an online survey, pay
themselves, recycle their worksheet, and leave the rest of money in the envelope.
In the eyes of the participants, the procedure seemed anonymous, since no identifying
information (e.g., name, participant number) was apparent on any of the forms or the
envelope. All participants received the same matrices to solve, except that a single number
was unique to each participant, so that we could calculate the difference between self-
reported and actual performance and use this difference as our main dependent variable,
dishonesty. Positive difference scores indicate that participants over-reported their
performance and cheated on the task.
At the end of the task, we assessed participants’ perceived identification with the
group they joined using the same three items (α = .94) as in Studies 1a-1b.
Results
23
Amount of dishonesty. A one-way analysis of variance on amount of dishonesty
revealed a significant difference among the three conditions, F(2, 131) = 3.49, p = .033 (see
Figure 4). Participants in the we-code condition over-reported their performance more (M =
1.85, SD = 3.29, 95% CI[.89,2.82]) than those in the members-code condition (M = .67, SD =
1.49, 95% CI[.22,1.13], p = .034) and those in the no-code condition (M = .80, SD = 1.72,
95% CI[.27,1.32], p = .061). The difference between the we-code and no-code conditions did
not reach significance (p = .061) but was consistent with the expected trend. There was no
significant difference in dishonesty between the members-code and the no-code conditions (p
= .84). In addition, we found no significant differences in actual performance across
conditions, F(2, 131) = .73, p = .483. None of the comparisons were significant.
Figure 4. Amount of cheating across conditions in Studies 2a and 2c.
Group identification. We also found a significant difference among the three
conditions on group identification, F(2, 131) = 7.73, p < .001; participants in the we-code
0.0
1.0
2.0
3.0
4.0
Members-Code We-Code No-Code Members-Code We-Code
Study 2a Study 2c
Amount of Cheating
24
condition felt stronger identification (M = 3.66, SD = 1.50, 95% CI[3.22,4.10]) as compared
to both those in the members-code condition (M = 2.67, SD = 1.09, 95% CI[2.33,3.00], p =
.001) and those in the no-code condition (M = 2.74, SD = 1.39, 95% CI[2.32,3.17], p = .003).
There was no significant difference in group identification between those in the members-
code and the no-code conditions (p = .78).
Though prior research has often not found a link between group identification and
dishonesty (e.g., Hildreth et al., 2016), strong identification with one’s group can encourage
unethical behavior that benefits the group. For example, prior work shows that individuals act
less ethically when they identify with their groups and hold strong reciprocity beliefs than
when they do not (Umphress, Bingham, & Mitchell, 2010). Given that we found differences
on identification in this study, we then tested whether group identification mediated the
relationship between the code-of-conduct condition and observed dishonesty. Based on
bootstrapping (with 10,000 iterations), the 95-percent bias-corrected confidence interval for
the size of the indirect effect included zero using the we-code condition as the condition of
reference. In sum, we found no indirect effect of our manipulation on dishonesty through
group identification.
Overall, in this study we found significant differences in cheating between the we-
code condition and both the members-code and no-code conditions. Even though participants
in the we-code condition identified more strongly with their group than those in other
conditions, their level of identification did not predict their dishonesty. Though we did not
find evidence for identification as a possible mechanism explaining why personal, communal
language encourages dishonesty, we explore the mediating role of perceptions of warmth as
well as group identification further in our next study.
Study 3a
25
In Study 3a, we examined whether the perception of a group’s warmth mediates the
effect of the language used in group communication and dishonesty. Once again, we included
group identification to examine its role as a potential alternative psychological mechanism.
Method
Participants and design. We recruited 250 individuals through Mechanical Turk to
participate in a two-part online study. Participants received $3 for completing both parts of
the study a week apart and had an opportunity to earn an additional $10. Following a decision
made prior to conducting the study, we excluded the responses of 17 participants who failed
an attention check in the first part. Out of the 234, 217 participants (57.1% male; Mage = 35.7
years, SD = 10.3) completed both parts. Study 3a employed a two-condition between-subjects
design (members-code, we-code). To ensure high power, we aimed for a sample size of 100
participants per cell and recruited more participants in Part 1.
Design and procedure. In Part 1, participants were randomly assigned to one of two
conditions (members-code, we-code). We had 112 in the we-code condition and 105 in the
members-code condition. The manipulation was identical to the one used in the previous
studies.
Perceptions of the group. After the manipulation, similar to Study 1, participants
responded to a number of questions in reference to the research group. They were provided
with a list of 14 traits (adapted from Goodwin et al., 2014). Three of these traits were from
the High Warmth/High Morality category (forgiving, kind, tolerant; α = .86), while two traits
were from the High Warmth/Low Morality category (warm, agreeable, α = .88). There were
three moral character traits from Goodwin et al. (Low Warmth/High Morality; fair,
principled, responsible, α = .77) and six traits that capture the Competence category
(competent, effective, efficient, innovative, organized, logical; α = .89). They rated their
perception of the “name of the research group” on each of these traits using a 7-point scale,
26
ranging from 1 (not at all) to 7 (very much). We also measured group identification with the
same three items (α = .96) from previous studies. Afterward, participants completed a brief
demographic questionnaire and then provided their email addresses to be contacted to
complete Part 2.
Dishonesty measure. One week later, an email was sent to all participants who
successfully completed Part 1 and provided a valid email address. At Time 2, participants
received a link based on their assigned condition to ensure we controlled for the language
used throughout the task. Based on the condition to which participants had been assigned,
they were briefly reminded of the code of conduct they had read a week earlier. Next,
participants completed a task where they could misreport their performance for extra money.
The task involved unscrambling 10 word jumbles in five minutes, for which they would
receive a $1 bonus for every jumble they reported to have solved correctly (adapted from
Kouchaki & Gino, 2016). They had to indicate which word jumbles they successfully
unscrambled without being asked to write down the unscrambled words (e.g., O L A R C).
The instructions notified them that they had to unscramble the word jumbles in the order they
appeared on the screen in order to be paid. They were informed that if they successfully
unscrambled the first six word jumbles, but not the seventh, they would be paid only for the
first six, even if they also successfully unscrambled the eighth word jumble.
Participants were given ten word jumbles; however, the fifth word jumble was
unsolvable. Given that solving the fifth one allowed participants to be paid for the rest of the
word jumbles (sixth word through the tenth word), they had an incentive to misreport their
performance on this word. We used participants reported number of correctly solved word
jumbles, starting from the fifth unsolvable word jumble, as the measure of cheating (a
number between 0 and 6). Participants earned $1 for every word, thus with each additional
word reported beyond the fourth word, participants earned more underserved money.
27
Results
Perceptions of the group’s (pure) warmth. On High warmth/High morality
attributes, participants rated the group in the we-code condition higher (M we-code = 5.43, SD =
1.14, 95% CI[5.22,5.64]) than the group in the members-code condition (M member-code = 4.92,
SD = 1.28, 95% CI[4.67,5.17]), t(215) =3.10, p = .002 (see Figure 5). The language used also
affected perceptions of High warmth/Low morality (M we-code = 5.44, SD = 1.23, 95%
CI[5.21,5.67] vs. M members-code = 5.09, SD = 1.23, 95% CI[4.85,5.32]), t(215) =2.13, p = .034.
Perceptions of the group’s morality and competence. However, perceptions of
moral character traits, i.e., Low Warmth/High Morality (M we-code = 6.34, SD = .70, 95%
CI[6.21,6.37] vs. M members-code = 6.21, SD = .80, 95% CI[6.06,6.37]), t(215) =1.27, p = .20,
and Competence (M we-code = 6.06, SD = .82, 95% CI[5.91,6.22] vs. M members-code = 6.06, SD =
.79, 95% CI[5.91,6.21]), t(215) < 1, were not significantly different.
Group identification. Group identification also did not differ between conditions (M
we-code = 4.94, SD = 1.47, 95% CI[4.67,5.22] vs. M member-code = 5.03, SD = 1.37, 95%
CI[4.76,5.29] t(215) < 1).
28
Figure 5. Rating of traits attributed to the group across conditions in Study 3a.
Cheating. We compared the number of word jumbles that participants reported to
earn money, starting with the fifth unsolvable word jumble, as our measure of unethical
behavior. We found that participants in the we-code condition reported a higher number of
word jumbles to be paid (M = 4.13, SD = 1.94) than those in the members-code condition (M
= 3.44, SD = 2.38), t(215) =2.34, p = .020.
Mediation analyses. We first tested whether perceptions of High Warmth/High
Morality mediated the relationship between our code-of-conduct conditions and observed
dishonesty. Based on bootstrapping (with 10,000 iterations), we estimated the direct and
indirect effects of the condition via these perceptions on the dependent variable, cheating. In
support of our prediction, the results suggested that perceptions of High Warmth/High
Morality mediated the relationship between code condition and cheating (indirect effect
3.0
4.0
5.0
6.0
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
High
Warmth/Low
Morality
High
Warmth/High
Morality
Low
Warmth/High
Morality
Competence Group
Identification
Ratings
29
=0.168, SE=0.09, 95% CI = [0.026, 0.376]). Similarly, perceptions of High Warmth/Low
Morality mediated the relationship between code condition and cheating (indirect effect
=0.085, SE=0.06, 95% CI = [0.001, 0.240]). To test for the potential role of perceptions of
moral character traits (Low Warmth/High Morality), competence, or identification, we ran
additional mediation analyses. None mediated the link between the code-of-conduct
condition and observed dishonesty.
In sum, this study provided direct evidence for the mediating role of perceptions of
warmth (both High Warmth/low Morality and High Warmth/High Morality), but no evidence
for group identification as a potential mediator. Even though we have examined the role of
group identification in four studies, we have not found reliable differences in identification
across conditions and no evidence for mediation. One reason for the lack of empirical support
may be that, in our experiments, we asked every participant to join a research team and to
complete an onboarding process in which we varied the language used to refer to the group
participants thought they’d be joining. The effect of communal language might not have been
strong enough to influence identification, as all participants felt equally identified with the
group they joined. In real-world settings, though, when complemented by other forms of
communication and behaviors, communal language may in fact result in higher levels of
group or organizational identification. Future research could examine this possibility as group
identification may in fact influence the decision to act dishonestly, particularly within natural
groups and organizations.
Study 3b
To comprehensively test whether perceptions of warmth contribute to dishonesty, we
conducted a preregistered, high-power study that provided participants with an opportunity to
behave dishonestly by self-reporting performance to earn money. We measured perceptions
30
of warmth as the result of the code of conduct condition and then tested whether it predicted
subsequent dishonest behavior.
Method
Participants and design. Three hundred one individuals (48.7% male; Mage = 34.3
years, SD = 13.0) at a university in the United States participated in this study for pay.
Participants received $25 for completing an hour-long session and had an opportunity to earn
an additional $5. We employed a two-condition, between-subjects design (members-code,
we-code). Before conducting the study, we preregistered it on the Open Science Framework
(https://osf.io/69t4u/register/5771ca429ad5a1020de2872e). Thus, we ran laboratory sessions
to achieve the pre-determined number of participants (i.e., 300 of them).
Design and procedure. Participants were randomly assigned to one of two conditions
(members-code, we-code). We had 151 in the we-code condition and 150 in the members-
code condition. The code of conduct manipulation was similar to previous studies (see SOM
for exact wording used).
Perceptions of the group. After the manipulation, participants responded to a number
of questions about their group on a 7-point scale, ranging from 1 (not at all) to 7 (very much).
They were provided with the list of attributes from Studies 1a-1b. Six of these traits were
from the High Warmth/High Morality category (forgiving, helpful, kind, empathetic,
cooperative, tolerant; α = .89), and five from the High Warmth/Low Morality category
(agreeable, warm, sociable, easygoing, playful; α = .88). Four traits captured moral character
traits (Low Warmth/High Morality category; fair, principled, responsible, trustworthy, α =
.87) and six measured the Competence category (competent, effective, efficient, innovative,
organized, logical; α = .92).
Dishonesty measure. Afterwards, they completed a matrix task similar to the one
from Study 2c that allowed participants to self-report their performance and gave them an
31
opportunity to overreport and be dishonest. Given that we previously found no difference in
actual performance, participants completed 10 solvable matrices presented to them one by
one. They were informed that they would receive $0.50 for each matrix they reported having
solved.
Results
Perceptions of the group’s (pure) warmth. Participants rated the group they were
joining higher on High Warmth/Low Morality traits in the we-code condition (M we-code =
4.49, SD = 1.24, 95% CI[4.29,4.69]) than in the members-code condition (M member-code = 4.10,
SD = 1.43, 95% CI[3.87,4.33]), t(299) = 2.54, p = .012. Similarly, perceptions of High
Warmth/High Morality differed between the code conditions (M we-code =5.01, SD = 1.06,
95% CI[4.84,5.19] vs. M member-code = 4.60, SD = 1.28, 95% CI[4.39,4.81], t(299) =3.05), p =
.002 (see Figure 6).
Perceptions of the group’s morality and competence. Once again, we found no
significant differences on perceptions of moral character (M we-code = 5.67, SD = 1.04, 95%
CI[5.50,5.84] vs. M member-code = 5.49, SD = 1.16, 95% CI[5.30,5.68], t(299)=1.41, p = .160)
nor on perceptions of competence (M we-code = 5.49, SD = 1.07, 95% CI[5.31,5.66] vs. M
member-code = 5.41, SD = 1.17, 95% CI[5.23,5.60], t(299) < 1).
32
Figure 6. Rating of traits attributed to the group across conditions in Study 3b.
Cheating. When comparing cheating on the number of matrices reported, we found
no differences between conditions (M we-code = 5.99, SD = 2.79, 95% CI[5.55,6.44] vs. M
member-code = 5.81, SD = 2.83, 95% CI[5.36,6.27]), t(299) =.56, p = .579.
Mediation analyses. Since the lack of a direct effect does not preclude an indirect
effect (see MacKinnon, Krull, & Lockwood, 2000; Rucker, Preacher, Tormala, & Petty,
2011; Zhao, Lynch & Chen, 2010), we next conducted mediation analyses to test whether
each category of traits mediated the relationship between the language used in the code of
conduct and cheating. Based on bootstrapping (with 10,000 iterations), supporting our
predictions, the results suggested that perceptions of High Warmth/Low Morality mediated
the relationship between code condition and cheating (indirect effect =0.098, SE=0.06, 95%
CI = [0.004, 0.242]). Similarly, perceptions of High Warmth/High Morality mediated the
3.0
4.0
5.0
6.0
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
Members-Code
We-Code
High Warmth/Low
Morality High Warmth/High
Morality Low Warmth/High
Morality Competence
Ratings
33
relationship between code condition and cheating (indirect effect =0.114, SE=0.07, 95% CI =
[0.002, 0.278]).
However, we did not find evidence of mediation by perceptions of Low Warmth/High
Morality (indirect effect =0.018, SE=0.04, 95% CI = [-0.036, 0.116]) nor by perceptions of
competence (indirect effect =0.008, SE=0.03, 95% CI = [-0.034, 0.077]).
Thus, consistent with the findings of Study 3a and with our theorizing, Study 3b
provided further evidence for the role of perceptions of warmth (both High Warmth/Low
Morality and High Warmth/High Morality) in explaining the relationship between personal,
communal language and dishonest behavior. Perceptions of moral character (Low
Warmth/High Morality) and perception of competence, instead, did not mediate this effect.
Unlike previous studies (Studies 2a-2c and 3a), we did not find a direct effect of our
manipulation on dishonesty measure in current study. Though emerging perspectives
question the requirement that a total X on Y effect need to be present before assessing
mediation (MacKinnon et al., 2000; Rucker et al., 2011; Zhao et al., 2010), we believe our
measurement of mediator (perceptions) right before the dependent variable (cheating
measure) may have contaminated responses to the dependent measure, or drawn attention to
the purpose of the study. Indeed, Sigall and Mills (1998) show that sometimes the inclusion
of a measure of mediator or manipulation check can undermine the direct effect.
Study 4
So far, we relied on the language used in codes of conduct to infer the type of
relationship perceivers presumably believe they have with the group they are joining. In this
study, we aim to provide direct evidence for the link between perceptions of (pure) warmth
(specifically the perceptions of being forgiving, tolerant, agreeable, kind, and easy-going) and
dishonesty by manipulating perceptions of (pure) warmth directly. We expected that
34
perceptions of warmth (and not morality) would influence the expected magnitude of
punishment and consequently increase the likelihood of misconduct.
Method
Participants and design. Two hundred individuals (52.5% male; Mage = 35.9 years,
SD = 11.1) recruited through Mechanical Turk participated in an online study and received
$0.50 for completing it. The study employed a two-condition, between-subjects design
(warm, neutral). Prior to conducting the study, we planned to stop data collection after
recruiting about 100 participants for each condition. This study was pre-registered with
materials available at https://osf.io/jdbcu/register/564d31db8c5e4a7c9694b2c0.
Design and procedure. Participants were randomly assigned to one of two conditions
(warm, neutral). We had 99 in the warm condition and 101 in the neutral condition. They
were asked to imagine themselves in a described situation as if they were actually
experiencing it. All participants read that a few months ago, they started a new job as an
office assistant at a large company in Chicago, where they are responsible for answering the
front desk phone and taking care of administrative tasks for a few of the office managers.
They further read, After a week of job training, you feel confident and comfortable in your
position at the office. Although it is not your dream job, the pay is good for an assistant
position and you enjoy the people and culture at the company. You feel like you could see
yourself staying in this job for a while. Your workload is manageable; often, you find
yourself with free time after your daily tasks are complete. In the warmth condition, the
participants then read some additional information intended to manipulate perceptions of
warmth: The office managers are very agreeable and easy-going. Everyone is kind to one
35
another and playful around the office. Even when mistakes are made, the response is
forgiving.
2
After reading the scenario, participants in both conditions read a description of an
unethical act and were asked to indicate the likelihood of them engaging in it. They read,
The company policy warns against personal internet use. Although the company does not
restrict or monitor any websites to your knowledge, it is prohibited in the company code of
conduct. You find yourself getting bored towards the end of each day. You have holiday
shopping to get done and you figure you could online shop to save time. How likely are you
to engage in this behavior? They responded to this question on a 7-point scale (1 =
extremely unlikely to 7 extremely likely).
At the end, as a manipulation check, we asked participants to rate the group on 21
traits as in our previous studies (High Warmth/High Morality, α = .93; High Warmth/Low
Morality, α = .92; moral traits, α = .87; competence α = .92).
Results
Participants in the warmth condition (M = 3.88, SD = 1.96, 95% CI[3.50,4.26]) were
more likely to engage in the morally questionable behavior as compared to those in the
neutral condition (M = 3.22, SD = 1.93, 95% CI[2.83,3.61]), t(198) =2.40, p = .017.
The warmth condition was rated higher on High Warmth/Low Morality traits (M warm
= 5.34, SD = 1.24, 95% CI[5.10,5.59])) than the neutral condition (M neutral = 4.39, SD = 1.21,
95% CI[4.15,4.63])), t(198) = 5.53, p < .001. Similarly, perceptions of High Warmth/High
2
Before conducting the study, we ran a pilot study to make sure individuals viewed the two groups differently.
After presenting the description, we asked participants to rate the group with a list of 21 traits from our previous
studies (High Warmth/High Morality, α = .92; High Warmth/Low Morality, α = .90; moral traits, α = .87;
competence α = .92). As we expected, the warmth condition was rated higher on High warmth/Low morality
traits (M warm = 5.77, SD = 1.02) than the neutral condition (M neutral = 5.37, SD = 1.04), t(198) = 2.75, p = .006.
Similarly, perceptions of High Warmth/High Morality differed between conditions (M warm = 5.72, SD = .93 vs.
M neutral = 5.37, SD = 1.00, t(198) = 2.52, p = .013). We found no significant differences between conditions on
perceptions of moral character (M warm = 5.38, SD = .95 vs. M neutral = 5.52, SD = 1.00, t(198) = .96, p = .337)
and perceptions of competence (M warm = 5.21, SD = .94 vs. M neutral = 5.36, SD = 1.06, t(198) = 1.03, p = .304).
The findings from this pilot study confirm that we were able to successfully manipulate perceptions of the
group’s warmth and not its competence or moral character.
36
Morality differed between conditions (M warm = 5.23, SD = 1.20, 95% CI[4.99,5.46] vs. M
neutral = 4.53, SD = 1.12, 95% CI[4.30,4.75], t(198) = 4.24, p < .001).
However, we found no significant differences between conditions on perceptions of
moral character (M warm = 5.33, SD = 1.06, 95% CI[5.12,5.54] vs. M neutral = 5.23, SD = 1.08,
95% CI[5.01,5.44], t(198) = .67, p = .501) nor competence (M warm = 5.19, SD = 1.02, 95%
CI[4.929,5.39] vs. M neutral = 5.30, SD = 1.07, 95% CI[5.08,5.51], t(198) = .70, p = .482).
These results confirm that we were able to successfully manipulate perceptions of the group’s
warmth and not its competence or moral character. The findings from this study provide
direct evidence that exclusively manipulating warmth perceptions can increase morally
questionable behaviors.
Study 5
Focusing on a large sample of companies, we aim to examine the relationship
between the language used (“we” vs. “employees”) in codes of conduct and corporate
illegality. Most firmscodes of conduct consist of a company vision and a letter from the
CEO. Our conversations with ethics officers led us to believe that employees typically have
access to the code of conduct and that it is the basis for additional forms of communication,
such as ethics training. As such, it is an important document that exemplifies the type of
relationship that exists between the firm and its members as well as the corporate culture.
Pilot study. Based on our conversations with ethics officers and employees across a
few different organizations, we believe that most employees read their companys code of
conduct at some point during their tenure, although to varying degrees. We conducted a
survey to test this assumption by recruiting a large sample (N = 1,916) of full-time employees
(55.6% male, mean age = 36.9, SD = 10.7) across the United States who received $5 for
completing the survey. Among the 1,916 full-time employees we recruited, 1,445 (75%)
indicated that their organization had a written code of conduct. Among these 1,445
37
employees, 1% indicated that they did not know how employees could access the code of
conduct or indicated that it was not available. Among the rest, 50.7% indicated that they had
received a hard copy of the code of conduct, 48.1% indicated it was on the organization’s
internal server, 34.1% noted its availability online, and 31.1% reported that portions of the
code were posted in public spaces in the company. As for when they last read or reviewed
their company’s code of conduct, 57.4% reported that they read/reviewed the code within that
last year, 22.6% reported reviewing it within the last five years, 10% reported they had not
read/reviewed it when they were hired, 1.4% noted that they reviewed it more than five years
ago, 6% said they did not recall, and only 2.4% said they had not read/reviewed it. Given
these responses from a sample of working adults, we feel confident in our assumption that
employees have access to their company’s code of conduct and generally have been exposed
to its language at some point during their employment. Thus, consistent with the results if our
previous studies, we expect the use of personal, communal (i.e., “we”) versus impersonal
(i.e., “group members”) language in a company’s code of conduct to affect employees’
perceptions of their organization and their unethical behavior, operationalized in this study as
corporate illegality.
Method
Our sample of companies consisted of all manufacturing firms that were part of the
S&P 500 between 1990 and 2012. Standard & Poor (S&P) provides financial information and
analysis and is one of the most commonly followed equity indices, and many consider it to be
one of the best representations of the U.S. stock market. Of the 210 firms in our sample, we
could not readily identify a code of conduct for 22 firms and did not receive any e-mail
response after querying investor relations. Thus, our S&P 500 manufacturing firm sample
consists of 188 firms for which we had access to a copy of their code of conduct.
38
Independent variable. Two research assistants blind to our research questions and
hypotheses coded each code of conduct on a dichotomous variable: “1” if the focal firm’s
code of conduct primarily used “we” language and “0” if it mainly used member/employee
language. The research assistants were trained by one of the authors about the distinctions
between the two categories. They completed a sample of codes of conduct from the Fortune
Top 50 companies as practice and discussed the coding with one of the authors. For all the
firms in our sample, the research assistants read the code and then categorize it. There was
very high (>87%) agreement between the research assistants’ ratings. Inconsistencies were
resolved through discussion among the two research assistants and one of the authors.
Dependent variable. We closely followed all the procedures reported by Mishina and
colleagues (Mishina, Dykes, Block, & Pollock, 2010) to identify acts of corporate illegality
by searching various media sources for any type of illegal acts, such as environmental
violations, anticompetitive actions, false claims, and fraudulent actions. We took the same
approach and created a dichotomous variable and coded it as “1” if we identified any incident
of corporate illegality in a given year and “0” otherwise. Our search identified 873 incidents
of corporate illegality with settlement or conviction between 1990 and 2012. No cases were
excluded. Given that we used a dichotomous measure of corporate illegalitythat is, whether
or not a target firm engaged in any incident in a given yearthese 873 incidents correspond
to 542 firm-year observations, coded as “1” in our dataset.
3
Control variables. Following Mishina et al. (2010), we controlled for firms’ size and
levels of slack resources, which may affect the propensity to engage in corporate illegality. In
addition, we included year indicators to control for systematic differences in the incidence of
3
To assess the robustness of our results, we also created a measure of corporate illegality with the sum of the
number of violations every year in each of the categories coded (environmental violations, anticompetitive
actions, false claims, and fraudulent actions). This variable reflects the total number of illegal incidents per year.
We conducted additional analyses, the results of which are available in the SOM. The results are similar to the
findings reported here.
39
corporate illegality. Firm size was operationalized as the natural logarithm of number of
employees annually. We controlled for three types of slack resources, since firms with more
slack resources have less need to pursue illegal activities, which pose the greatest risk to
sustained performance. The first one, absorbed slack, was measured as the ratio of
administrative expenses to sales; the second one, unabsorbed slack, was measured as the ratio
of cash and marketable securities to liabilities; and the third, potential slack, was measured as
the ratio of debt to equity.
Finally, we included the total number of words in the code of conduct as another
control variable to account for potential systematic differences and to provide a more
conservative test of our hypothesis.
Results
Table 1 provides correlations and descriptive statistics for each of the variables in our
study. Table 2 presents the results of our analyses predicting corporate illegality. We predict
that the “we” version (coded as 1), relative to the “member/employee” version (coded as 0),
will be positively related to a firm’s propensity to engage in corporate illegality. In fact, the
code-of-conduct variable was positively related to corporate illegality. Even after controlling
for the number of words and other control variables, we found a firm’s code of conduct to be
positively related to its propensity to engage in corporate illegality.
40
Table 1. Correlations and Descriptive Statistics
Variables
Mean
SD
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
1.Corporate
illegality
0.128
0.334
2.We-Code
0.368
0.482
0.095
3.1990 Dummy
0.043
0.204
0.047
0.007
4.1991 Dummy
0.043
0.204
0.042
0.009
-0.056
5.1992 Dummy
0.043
0.204
0.031
0.006
-0.056
-0.056
6.1993 Dummy
0.043
0.204
0.036
0.006
-0.056
-0.057
-0.057
7.1994 Dummy
0.043
0.204
0.043
0.008
-0.057
-0.057
-0.057
-0.057
8.1995 Dummy
0.043
0.204
0.042
0.008
-0.056
-0.056
-0.056
-0.057
-0.057
9.1996 Dummy
0.043
0.204
0.038
0.009
-0.056
-0.056
-0.056
-0.057
-0.057
-0.056
10.1997 Dummy
0.043
0.204
0.036
0.009
-0.056
-0.057
-0.057
-0.057
-0.057
-0.057
-0.057
11.1998 Dummy
0.043
0.204
0.018
0.011
-0.056
-0.056
-0.056
-0.057
-0.057
-0.056
-0.056
-0.057
12.1999 Dummy
0.043
0.204
0.022
0.011
-0.055
-0.055
-0.055
-0.056
-0.056
-0.055
-0.055
-0.056
-0.055
13.2000 Dummy
0.043
0.204
0.009
0.011
-0.051
-0.051
-0.051
-0.052
-0.052
-0.051
-0.051
-0.052
-0.051
-0.051
14.2001 Dummy
0.043
0.204
-0.014
0.007
-0.049
-0.049
-0.049
-0.050
-0.050
-0.049
-0.049
-0.050
-0.049
-0.049
-0.045
15.2002 Dummy
0.043
0.204
-0.019
-0.011
-0.048
-0.048
-0.048
-0.048
-0.049
-0.048
-0.048
-0.048
-0.048
-0.047
-0.044
-0.042
16.2003 Dummy
0.043
0.204
-0.013
-0.005
-0.047
-0.047
-0.047
-0.048
-0.048
-0.048
-0.047
-0.048
-0.048
-0.047
-0.043
-0.042
-0.041
17.2004 Dummy
0.043
0.204
-0.017
-0.007
-0.047
-0.047
-0.047
-0.048
-0.048
-0.047
-0.047
-0.048
-0.047
-0.047
-0.043
-0.042
-0.040
-0.040
18.2005 Dummy
0.043
0.204
-0.020
-0.008
-0.047
-0.047
-0.047
-0.047
-0.047
-0.047
-0.047
-0.047
-0.047
-0.046
-0.043
-0.041
-0.040
-0.040
-0.040
19.2006 Dummy
0.043
0.204
-0.026
-0.009
-0.046
-0.046
-0.046
-0.046
-0.046
-0.046
-0.046
-0.046
-0.046
-0.045
-0.042
-0.040
-0.039
-0.039
-0.039
-0.038
20.2007 Dummy
0.043
0.204
-0.023
-0.011
-0.045
-0.045
-0.045
-0.045
-0.046
-0.045
-0.045
-0.045
-0.045
-0.044
-0.041
-0.040
-0.038
-0.038
-0.038
-0.038
-0.037
21.2008 Dummy
0.043
0.204
-0.044
-0.012
-0.044
-0.044
-0.044
-0.044
-0.044
-0.044
-0.044
-0.044
-0.044
-0.043
-0.040
-0.039
-0.037
-0.037
-0.037
-0.037
-0.036
-0.035
22.2009 Dummy
0.043
0.204
-0.061
-0.010
-0.043
-0.043
-0.043
-0.043
-0.043
-0.043
-0.043
-0.043
-0.043
-0.042
-0.039
-0.038
-0.037
-0.036
-0.036
-0.036
-0.035
-0.034
-0.034
23.2010 Dummy
0.043
0.204
-0.070
-0.019
-0.042
-0.042
-0.042
-0.043
-0.043
-0.043
-0.042
-0.043
-0.043
-0.042
-0.039
-0.037
-0.036
-0.036
-0.036
-0.036
-0.035
-0.034
-0.033
-0.033
24.2011 Dummy
0.043
0.204
-0.074
-0.015
-0.042
-0.042
-0.042
-0.042
-0.042
-0.042
-0.042
-0.042
-0.042
-0.041
-0.038
-0.037
-0.036
-0.035
-0.035
-0.035
-0.034
-0.034
-0.033
-0.032
-0.032
25.2012 Dummy
0.043
0.204
-0.078
-0.016
-0.041
-0.041
-0.041
-0.042
-0.042
-0.041
-0.041
-0.042
-0.041
-0.041
-0.038
-0.036
-0.035
-0.035
-0.035
-0.035
-0.034
-0.033
-0.032
-0.032
-0.031
-0.031
26.log number of
employees
3.146
1.010
0.252
0.112
-0.009
-0.010
-0.013
-0.016
-0.018
-0.022
-0.020
-0.014
-0.015
-0.004
-0.001
-0.001
0.010
-0.008
-0.003
-0.005
0.005
0.016
0.030
0.036
0.032
0.036
0.033
27.Absorbed slack
0.230
0.140
-0.073
0.159
-0.037
-0.028
-0.015
-0.020
-0.020
-0.024
-0.024
-0.020
-0.009
0.000
0.016
0.010
0.021
0.017
0.024
0.015
0.016
0.021
0.024
0.021
0.029
0.014
0.010
-0.136
28.Unabsorbed
slack
0.324
0.568
-0.053
0.050
-0.045
-0.045
-0.039
-0.048
-0.048
-0.076
-0.073
-0.068
-0.074
-0.078
-0.019
0.007
0.015
0.043
0.058
0.089
0.083
0.061
0.044
0.043
0.119
0.105
0.093
-0.226
0.241
29.Potential slack
-0.126
35.35
0.009
0.015
0.002
0.003
0.000
0.004
0.004
0.005
0.002
0.003
0.007
0.003
0.005
0.007
0.011
-0.001
-0.029
0.020
0.003
0.002
0.031
-0.035
-0.019
-0.041
0.000
0.000
0.011
-0.007
30.Total # words
8277.827
5280.
827
0.092
0.310
0.013
0.006
0.010
0.008
0.012
0.012
0.003
0.004
0.008
0.016
0.002
-0.008
-0.008
-0.008
-0.004
-0.004
-0.012
-0.012
-0.008
-0.005
-0.017
-0.014
-0.016
0.111
0.198
0.030
0.003
41
Table 2. Multiple Estimates of Logistic Regression Models Predicting Corporate Illegality in
Study 5.
Variables
model 1
model 2
model 3
estimate
SE
estimate
SE
estimate
SE
Intercept
0.212
***
0.026
-0.093
*
0.037
-0.107
**
0.038
We-Code
0.052
***
0.011
0.043
**
0.013
0.034
*
0.014
1991 Dummy
-0.009
0.036
-0.007
0.039
-0.006
0.039
1992 Dummy
-0.025
0.036
-0.024
0.039
-0.023
0.039
1993 Dummy
-0.017
0.036
-0.013
0.039
-0.013
0.039
1994 Dummy
-0.010
0.036
-0.002
0.039
-0.002
0.039
1995 Dummy
-0.010
0.036
0.002
0.039
0.002
0.039
1996 Dummy
-0.018
0.036
-0.004
0.039
-0.003
0.039
1997 Dummy
-0.023
0.035
-0.012
0.039
-0.011
0.039
1998 Dummy
-0.051
0.035
-0.038
0.039
-0.037
0.039
1999 Dummy
-0.043
0.035
-0.035
0.039
-0.035
0.039
2000 Dummy
-0.084
*
0.036
-0.063
0.041
-0.061
0.041
2001 Dummy
-0.123
***
0.036
-0.107
*
0.041
-0.104
*
0.042
2002 Dummy
-0.138
***
0.036
-0.120
**
0.042
-0.117
**
0.042
2003 Dummy
-0.131
***
0.036
-0.105
*
0.043
-0.103
*
0.043
2004 Dummy
-0.136
***
0.036
-0.116
**
0.043
-0.114
**
0.043
2005 Dummy
-0.135
***
0.036
-0.127
**
0.043
-0.126
**
0.043
2006 Dummy
-0.153
***
0.036
-0.144
***
0.044
-0.141
**
0.044
2007 Dummy
-0.153
***
0.036
-0.142
**
0.044
-0.139
**
0.044
2008 Dummy
-0.182
***
0.036
-0.190
***
0.045
-0.189
***
0.046
2009 Dummy
-0.200
***
0.036
-0.231
***
0.045
-0.229
***
0.045
2010 Dummy
-0.218
***
0.036
-0.261
***
0.046
-0.257
***
0.046
2011 Dummy
-0.224
***
0.036
-0.273
***
0.046
-0.270
***
0.046
2012 Dummy
-0.230
***
0.036
-0.281
***
0.046
-0.278
***
0.046
Log number of employees
0.106
***
0.007
0.104
***
0.007
Absorbed slack
-0.123
*
0.048
-0.144
**
0.049
Unabsorbed slack
0.056
***
0.013
0.055
***
0.013
Potential slack
0.000
0.000
0.000
0.000
Total # of
words
0.000
*
0.000
*** p<0.001, ** p<0.01, * p<0.05
42
General Discussion
Our study joins an emerging literature that has begun to explore factors responsible
for curbing and reducing unethicality in organizations and society more broadly. Research in
behavioral ethics has suggested revisiting and revising traditional enforcement instruments,
such as codes of conduct and corporate culture, to see if they can be improved to better
address the dishonesty of “good” people. Using multiple methods and settings, we have
shown that perceptions of group warmth through subtle changes in the language used in
group communications (“we” vs. “employees”) can have a sizable impact on individuals’
ethical behavior. Thus, in contrast to the current trend of looking for nudges that might
enforce ethical behavior (e.g., priming the Ten Commandments), our study shows that
placing an emphasis on enforcement could go a long way toward improving ethical conduct.
Our research contributes to the literature on moral psychology and ethics in important
ways. First, we demonstrate a direct link between perceptions of warmth and dishonesty.
Relatively few studies have examined the perceptions of warmth and competence of a
recipient or evaluators in moral decision making. Our paper highlights an important challenge
that groups face: being perceived as warm and accepting while still signaling to members that
any misbehavior will have consequences. Second, instead of focusing on increasing the
salience of morality (the adoption of codes of conduct or triggering morality thoughts more
generally), we focus on the perceptions of enforcement (the content of these codes of conduct
or the emphasis on strong enforcement) to better understand what type of language is most
effective at deterring unethical behavior.
Third, this paper also adds insight to the broader debate in enforcement theory of
whether to address people’s moral or extrinsic motivations. The classic argument against the
stricter approach that focuses on punishment as its main tool of enforcement is that it crowds
out people’s intrinsic motivations and hence reduces the quality of their moral behavior
43
(Bohnet, Frey, & Huck, 2001; Gneezy & Rustichini, 2000). Our focus on language effects
provides an important new way to think about how to optimize the use of sanctions. We have
shown that language that sends subtle cues to employees (e.g., using employees” instead of
“we”) that an organization is serious about enforcing ethical behavior might avoid some of
the negative consequences of the rigid use of sanctions discussed by scholars. In other words,
our findings suggest that subtle cues communicating a group/community culture, beyond
enforcement practices and the size of sanctions, could undermine the ethicality of people’s
decisions. This effect should be considered in theoretical and policy discussions of ethical
decision making in the workplace.
The conclusions drawn from our results should consider the limitations of the studies.
Even though we employed data using multiple methods, and diverse sample populations, our
study designs potentially limit the generalizability of the findings. The use of controlled
experiments helped us establish causality and provide an understanding of the psychological
mechanisms explaining our proposed effects but did not allow us to explore the role of
alternative mechanisms as closely. For instance, in our studies, we examined the role of
group identification, but did not manipulate it explicitly. Future research could test our
proposed relationships using other methodological approaches and strategies to promote
identification. It remains unclear what’s the effect of language on identification and how
identification influence dishonest behavior. Additionally, in our experimental studies, we
used the same manipulation of personal versus impersonal language. This potentially limits
the contribution of our work. Future research could extend our findings by using other
manipulations of the type of language used not only in codes of conduct but also in other
forms of communications with people who are about to join groups or organizations.
Another limitation of our studies may be the lack of explicit expectations to be caught
or punished. In most our studies, except Study 2B, similar to previous psychological
44
paradigms (Mazar et al., 2008) participants completed tasks anonymously with no
identifiable information, and thus with no explicit expectations for punishment that could be
enforced. Thse extent to which the effects we find hold under explicit punishment
expectations needs further investigation. In fact, research has shown that cheating is affected
when all concerns about being exposed as a cheater are eliminated (Kajackaite & Gneezy,
2017). This could have important implications for the ecological validity of our findings and
should be further examined. Finally, we tested our predictions across many studies and
contexts, and the effect of the language manipulation we used was rather weak. This can be
attributed to our subtle manipulations. Though not very strong, we believe our effects to be
meaningful and of practical importance. In fact, as we found in our field experiment,
differences in the language used in codes of conduct changed individuals’ behavior. Future
research could investigate whether stronger language manipulations would lead to stronger
results, thus advancing our understanding of the potential perils of communal language.
45
References
Aaker, J., Vohs, K. D., & Mogilner, C. (2010). Nonprofits are seen as warm and for-profits as
competent: Firm stereotypes matter. Journal of Consumer Research, 37(2), 224-237.
Azevedo, R. T., Panasiti, M. S., Maglio, R., & Aglioti, S. M. (2018). Perceived warmth and
competence of others shape voluntary deceptive behaviour in a morally relevant
setting. British Journal of Psychology, 109(1), 25-44.
Becker, G. S. (1968). Crime and punishment: An economic approach. In The economic
dimensions of crime (pp. 13-68). Palgrave Macmillan, London.
Bernhard, H., Fehr, E., & Fischbacher, U. (2006). Group affiliation and altruistic norm
enforcement. American Economic Review, 96(2), 217-221.
Bohnet, I., Frey, B. S., & Huck, S. (2001). More order with less law: On contract
enforcement, trust, and crowding. American Political Science Review, 95(1), 131-144.
Clark, M. S. (1984). Record keeping in two types of relationships. Journal of Personality and
Social Psychology, 47(3), 549.
Cark, M. S., & Mills, J. (1979). Interpersonal attraction in exchange and communal
relationships. Journal of Personality and Social Psychology, 37(1), 12.
Clark, M. S., Mills, J., & Powell, M. C. (1986). Keeping track of needs in communal and
exchange relationships. Journal of Personality and Social Psychology, 51(2), 333.
Clark, M. S., Oullette, R., Powell, M. C., & Milberg, S. (1987). Recipient's mood,
relationship type, and helping. Journal of Personality and Social Psychology, 53(1),
94.
Doosje, B., Ellemers, N., & Spears, R. (1995). Perceived intragroup variability as a function
of group status and identification. Journal of Experimental Social Psychology, 31(5),
410-436.
Feldman, Y. (2018). The law of good people: Challenging states' ability to regulate human
46
behavior. Cambridge University Press.
Fiske, S. T., Cuddy, A. J., Glick, P., & Xu, J. (2002). A model of (often mixed) stereotype
content: competence and warmth respectively follow from perceived status and
competition. Journal of Personality and Social Psychology, 82(6), 878.
Gneezy, U., & Rustichini, A. (2000). A fine is a price. The Journal of Legal Studies, 29(1), 1-
17.
Goodwin, G. P., Piazza, J., & Rozin, P. (2014). Moral character predominates in person
perception and evaluation. Journal of Personality and Social Psychology, 106(1), 148.
Gächter, S., Renner, E., & Sefton, M. (2008). The long-run benefits of punishment. Science,
322(5907), 1510-1510.
Hildreth, J. A. D., Gino, F., & Bazerman, M. (2016). Blind loyalty? When group loyalty
makes us see evil or engage in it. Organizational Behavior and Human Decision
Processes, 132, 16-36.
Kajackaite, A., & Gneezy, U. (2017). Incentives and cheating. Games and Economic
Behavior, 102, 433-444.
Kish-Gephart, J. J., Harrison, D. A., & Treviño, L. K. (2010). Bad apples, bad cases, and bad
barrels: meta-analytic evidence about sources of unethical decisions at work. Journal
of Applied Psychology, 95(1), 1.
Kouchaki, M., & Gino, F. (2016). Memories of unethical actions become obfuscated over
time. Proceedings of the National Academy of Sciences, 113(22), 6166-6171.
Kouchaki, M., & Smith, I. H. (2014). The morning morality effect: The influence of time of
day on unethical behavior. Psychological Science, 25(1), 95-102.
MacKinnon, D. P., Krull, J. L., & Lockwood, C. M. (2000). Equivalence of the mediation,
confounding and suppression effect. Prevention Science, 1(4), 173-181.
Mazar, N., Amir, O., & Ariely, D. (2008). The dishonesty of honest people: A theory of self-
47
concept maintenance. Journal of Marketing Research, 45(6), 633-644.
Mills, J., & Clark, M. S. (1982). Exchange and communal relationships. Review of
Personality and Social Psychology, 3, 121-144.
Mishina, Y., Dykes, B. J., Block, E. S., & Pollock, T. G. (2010). Why “good” firms do bad
things: The effects of high aspirations, high expectations, and prominence on the
incidence of corporate illegality. Academy of Management Journal, 53(4), 701-722.
Paine, L., Deshpande, R., Margolis, J. D., & Bettcher, K. E. (2005). Up to code: Does your
company's conduct meet world-class standards? Harvard Business Review, 83(12),
122-33.
Rucker, D. D., Preacher, K. J., Tormala, Z. L., & Petty, R. E. (2011). Mediation analysis in
social psychology: Current practices and new recommendations. Social and
Personality Psychology Compass, 5(6), 359-371.
Sánchez-Pagés, S., & Vorsatz, M. (2007). An experimental study of truth-telling in a sender
receiver game. Games and Economic Behavior, 61(1), 86-112.
Shariff, A. F., & Norenzayan, A. (2011). Mean gods make good people: Different views of
God predict cheating behavior. The International Journal for the Psychology of
Religion, 21(2), 85-96.
Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2013, January). Life after P-hacking. Paper
presented at the Meeting of the Society for Personality and Social Psychology, New
Orleans, Louisiana. Retrieved from
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2205186
Umphress, E. E., Bingham, J. B., & Mitchell, M. S. (2010). Unethical behavior in the name
of the company: The moderating effect of organizational identification and positive
reciprocity beliefs on unethical pro-organizational behavior. Journal of Applied
Psychology, 95(4), 769.
48
Weick, K. E. (1979). Cognitive processes in organizations. Research in Organizational
Behavior, 1(1), 41-74.
Zhao, X., Lynch Jr, J. G., & Chen, Q. (2010). Reconsidering Baron and Kenny: Myths and
truths about mediation analysis. Journal of Consumer Research, 37(2), 197-206.
... Third, our study adds to the emerging literature on organizational culture by highlighting some of the downsides to modern cultures that emphasize communal bonding and a family-like atmosphere (e.g., Kouchaki et al. 2019). Often, organizations actively foster a "team" dynamic at the workplace so that people can collaborate in their pursuit of a common goal in a positive emotionally engaged atmosphere (see Green et al. 2017). ...
... In this case, explicitly incorporating language in code of conducts that espouses the company's values (e.g., placing fairness over loyalty) could be beneficial. Moreover, clear ramifications for members of the organization who knowingly conceal wrongdoing and fail to consider victim protections could help serve to counteract the perception that warm and positive workplaces, as those that are highly entitative may be, are associated with leniency when lapses occur (Kouchaki et al. 2019). ...
Article
Full-text available
In the aftermath of shocking workplace scandals, people are often baffled when individuals within the organization were aware of clear-cut wrongdoing yet did not inform authorities. The current research suggests that moral concern for the suffering that a transgressor might face if a crime were reported is an under-recognized, powerful force that shapes whistleblowing in organizations, particularly when transgressors are fellow members of a highly entitative group (i.e., a group that is perceived as highly unified). Two experiments show that group entitativity heightens concern about possible consequences that the transgressor would face if a crime were to be reported, and that this concern reduces the likelihood of reporting wrongdoing in organizations to authorities. Further, the studies identify a mechanism through which concern about the transgressor is heightened in highly entitative groups: potential reporters perceive that the transgressor felt remorse for their crime. Thus, when fellow members of highly entitative organizations commit crimes, people are more likely to imagine that these transgressors felt anxiety or guilt about their actions, and this prompts greater concern for transgressors in ways that encourage people to let them “off the hook.” We discuss the implications of these findings for how reporting to authorities can be encouraged within highly entitative organizations.
... In fact, such a message conveys that the rule might not be really serious and that compliance is not expected by the legitimizing authorities. Recent research (Kouchaki, Gino and Feldman 2019) showed that such messages convey 'warmth' rather than legitimacy and are likely to even increase non-compliance. ...
Chapter
Full-text available
Rule compliance (in organizations or society at large) may be strengthened or weakened by social contagion processes. Observing others' (non-)compliance with rules influences one's own likelihood of compliance. Extant literature shows two social contagion theories that can explain this phenomenon. First, the theory of normative conduct (TNC) (Cialdini et al. 1990) suggests that people interpret the observed behaviour of others (i.e. the descriptive norm) as adaptive for that context, resulting in rational imitation.
... The expressive function of the law lies in its ability to signal which views and behaviors are socially accepted and desirable and which are socially forbidden (Adler 1999(Adler , p. 1404Anderson andPildes 2000, p. 1506). Thus, the law influences peoples' beliefs about what others will do (Bohnet and Cooter 2003; see also Cooter 1998Cooter , 2000Feldman 2009;McAdams 1997McAdams , 2015McAdams andNadler 2005, 2008;Tyran and Feld 2002;Kouchaki et al 2019). Scholars have hypothesized about the processes through which laws affect attitudes and behaviors in various contexts, including voting rights Electronic copy available at: https://ssrn.com/abstract=3816988 ...
Article
Full-text available
(Forthcoming The Journal of Legal Studies 2022) Western societies are increasingly enacting majority nationalism laws to strengthen majority culture. We propose that these laws may alter public attitudes about minorities’ equal citizenship with varied impact on majorities and minorities. To explore this issue, we examine the impact of Israel’s recently enacted Nation Law on the Jewish majority and the Arab minority. Experimental evidence from before the Law’s enactment reveals that both minority and majority respondents perceive the passage of the Law as permitting discrimination against the minority in housing, employment, and voting, but the effect on minority perceptions is larger. Panel and cross-sectional data from before and after the Law’s enactment reveal that the effect on majority respondents was fleeting, whereas the negative impact on minority respondents was stronger and more durable. These findings expose the troubling effects of majority nationalism laws and suggest that law may operate as a prism, expressing different messages to different groups.
Article
In many spheres of life, from applying for a job to participating in an athletic contest to vying for a date, we face competition. Does the size of the competition pool affect our propensity to behave unethically in our pursuit of the prize? We propose that it does. Across four studies, we found that a larger (vs. smaller) number of competitors led participants to cheat more in a performance task to earn undeserved money. We also explored the psychological mechanisms of competition pool size to explain why and how being in a larger competition pool increases cheating. Our findings reveal a serial mediation pathway whereby having a larger number of competitors increases expectations of the absolute number of cheaters in the competition group, which heightens perceptions that cheating is an acceptable social norm, which leads to more cheating. We also examined and ruled out various alternative psychological mechanisms for this effect. We discuss the theoretical and practical implications of our finding that being in a large group of competitors can increase people’s propensity to cheat for personal gain.
Article
In contrast to the view that businesses corrupt and bring out the worst in people, we propose that people's experiences at work can also lead to profound ethical learning and growth, and that organizations can be designed to help workers strive to become their best moral selves. To this end, we selectively review literature from the fields of organizational behavior, moral psychology, and behavioral ethics to identify five key barriers to ethical learning at work: defensiveness, overconfidence, selfishness, inexperience, and negative social influence. Then, to help leaders develop an organizational environment that helps workers overcome these barriers and promotes ethical learning and character development, we present five guiding principles (i.e., create an ethical culture, develop psychological safety, promote moral humility, foster prosocial motivation, and institutionalize moral reflection) and offer specific recommendations related to each.
Article
In four studies, we evaluated how corporate misconduct relates to language patterns, perceptions of immorality, and unethical behavior. First, we analyzed nearly 190 codes of conduct from S&P 500 manufacturing companies and observed that corporations with ethics infractions had more linguistically obfuscated codes than corporations without ethics infractions. Next, we tested perceptions of a company based on values statements modified by obfuscation (Study 2). Participants perceived low-obfuscation companies as more moral, warmer, and more trustworthy than high-obfuscation companies. Finally, behavioral experiments (Studies 3a and 3b) revealed that group members cheat more after reading a high-obfuscation values statement than a low-obfuscation values statement. The results provide evidence of a potentially troublesome cycle: corporate unethicality has linguistic traces, can affect how people appraise a company, and can change ethical behavior.
Book
Full-text available
Cambridge Core - Macroeconomics - The Law of Good People - by Yuval Feldman
Article
Full-text available
What sorts of trait information do people most care about when forming impressions of others? Recent research in social cognition suggests that "warmth," broadly construed, should be of prime importance in impression formation. Yet, some prior research suggests that information about others' specifically moral traits-their moral "character"-may be a primary dimension. Although warmth and character have sometimes been conceived of as interchangeable, we argue that they are separable, and that across a wide variety of contexts, character is usually more important than warmth in impression formation. We first showed that moral character and social warmth traits are indeed separable (Studies 1 and 2). Further studies that used correlational and experimental methods showed that, as predicted, in most contexts, moral character information is more important in impression formation than is warmth information (Studies 2-6). Character information was also more important than warmth information with respect to judgments of traits' perceived fundamentalness to identity, their uniquely human quality, their context-independence, and their controllability (Study 2). Finally, Study 7 used an archival method to show that moral character information appears more prominently than warmth information in obituaries, and more strongly determines the impressions people form of the individuals described in those obituaries. We discuss implications for current theories of person perception and social cognition. (PsycINFO Database Record (c) 2013 APA, all rights reserved).
Article
Stereotype research emphasizes systematic processes over seemingly arbitrary contents, but content also may prove systematic. On the basis of stereotypes' intergroup functions, the stereotype content model hypothesizes that (a) 2 primary dimensions are competence and warmth, (b) frequent mixed clusters combine high warmth with low competence (paternalistic) or high competence with low warmth (envious), and (c) distinct emotions (pity, envy, admiration, contempt) differentiate the 4 competence-warmth combinations. Stereotypically, (d) status predicts high competence, and competition predicts low warmth. Nine varied samples rated gender, ethnicity, race, class, age, and disability out-groups. Contrary to antipathy models, 2 dimensions mattered, and many stereotypes were mixed, either pitying (low competence, high warmth subordinates) or envying (high competence, low warmth competitors). Stereotypically, status predicted competence, and competition predicted low warmth.
Chapter
IntroductionBasic AnalysisOptimality ConditionsShifts in the Behavioral RelationsFinesSummary and Concluding Remarks
Article
The temptation to deceive others compares to a moral dilemma: it involves a conflict between the temptation to obtain some benefit and the desire to conform to personal and social moral norms or avoid aversive social consequences. Thus, people might feel different levels of emotional and moral conflict depending on the target of the deception. Here we explored, in a morally relevant setting, how social judgements based on two fundamental dimensions of human social cognition - 'warmth' and 'competence' - impact on the decision to deceive others. Results revealed independent effects for warmth and competence. Specifically, while people are less inclined to deceive for self-gain those individuals they perceive as warm, they also tend to lie more to highly competent others. Furthermore, the perceived warmth and competence modulated the general tendency to reduce deceptive behaviour when there was a risk of disclosure compared to when the lying was anonymous, highlighting the importance of these judgements in social evaluation processes. Together, our results demonstrate that the emotional costs and personal moral standards that inhibit engagement in deceptive behaviour are not stable but rather malleable according to the target and the consequences of the deception.
Article
We study how cheating behavior is affected by incentives. After replicating the finding in the cheating game literature that lying does not increase with incentives, we show that this insensitivity is not a characteristic of the intrinsic lying cost, but rather a result of concern about being exposed as a liar. In a modified “mind” game in which this concern is eliminated, we find that people lie more, and in particular lie more when the incentives to do so increase. Thus, our results show that for many participants, the decision to lie follows a simple cost-benefit analysis: they compare the intrinsic cost of lying with the incentives to lie; once the incentives are higher than the cost, they switch from telling the truth to lying.
Article
Despite our optimistic belief that we would behave honestly when facing the temptation to act unethically, we often cross ethical boundaries. This paper explores one possibility of why people engage in unethical behavior over time by suggesting that their memory for their past unethical actions is impaired. We propose that, after engaging in unethical behavior, individuals' memories of their actions become more obfuscated over time because of the psychological distress and discomfort such misdeeds cause. In nine studies (n = 2,109), we show that engaging in unethical behavior produces changes in memory so that memories of unethical actions gradually become less clear and vivid than memories of ethical actions or other types of actions that are either positive or negative in valence. We term this memory obfuscation of one's unethical acts over time "unethical amnesia." Because of unethical amnesia, people are more likely to act dishonestly repeatedly over time.
Article
Loyalty often drives corruption. Corporate scandals, political machinations, and sports cheating highlight how loyalty's pernicious nature manifests in collusion, conspiracy, cronyism, nepotism, and other forms of cheating. Yet loyalty is also touted as an ethical principle that guides behavior. Drawing on moral psychology and behavioral ethics research, we developed hypotheses about when group loyalty fosters ethical behavior and when it fosters corruption. Across nine studies, we found that individuals primed with loyalty cheated less than those not primed (Study 1A and 1B). Members more loyal to their fraternities (Study 2A) and students more loyal to their study groups (Study 2B) also cheated less than their less loyal counterparts due to greater ethical salience when they pledged their loyalty (Studies 3A and 3B). Importantly, competition moderated these effects: when competition was high, members more loyal to their fraternities (Study 4) or individuals primed with loyalty (Studies 5A and 5B) cheated more.