ArticlePDF Available

The Ethical Perils of Personal, Communal Relations: A Language Perspective

September 2019
Psychological Science 30(12)

DOI:10.1177/0956797619882917

Authors:

Maryam Kouchaki

Northwestern University

Francesca Gino

Harvard University

Yuval Feldman

Bar Ilan University

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.

Rating of traits attributed to group across conditions in Study 1a.

…

Percentage of cheaters across conditions in Study 2b.

…

Rating of traits attributed to the group across conditions in Study 3a.

…

Multiple Estimates of Logistic Regression Models Predicting Corporate Illegality in Study 5.

…

Figures - uploaded by Yuval Feldman

Content may be subject to copyright.

Content uploaded by Yuval Feldman

Content may be subject to copyright.

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

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

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 conduct—that is, whether the organization

conveys to employees that it punishes code violations—does 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

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,

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

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 easygoing—influence moral behavior by altering such

expectations

. Specifically, we expect that perceptions of warmth, by indirectly influencing

the expected magnitude of punishment, will increase dishonest behavior. Thus, in situations

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 warmth–low competence [HW-LC], low warmth–low competence [LW-LC],

or low warmth–high 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.

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

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

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).

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.

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

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

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,

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

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.

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

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).

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

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

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

0.0

10.0

20.0

30.0

40.0

50.0

No-Code Member-Code We-Code No-Code Member-Code We-Code

Week 1 Week 2

Percentage of cheaters

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.

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

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

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

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,

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.

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).

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

=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

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

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).

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

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

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

another and playful around the office. Even when mistakes are made, the response is

forgiving.”

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

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.

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 firms’ codes 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 company’s 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

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.

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 illegality—that is, whether

or not a target firm engaged in any incident in a given year—these 873 incidents correspond

to 542 firm-year observations, coded as “1” in our dataset.

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

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.

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.

Table 1. Correlations and Descriptive Statistics

Variables

Mean

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

6.1993 Dummy

0.043

0.204

0.036

0.006

-0.056

-0.057

7.1994 Dummy

0.043

0.204

0.043

0.008

-0.057

8.1995 Dummy

0.043

0.204

0.042

0.008

-0.056

-0.057

9.1996 Dummy

0.043

0.204

0.038

0.009

-0.056

-0.057

-0.056

10.1997 Dummy

0.043

0.204

0.036

0.009

-0.056

-0.057

11.1998 Dummy

0.043

0.204

0.018

0.011

-0.056

-0.057

-0.056

-0.057

12.1999 Dummy

0.043

0.204

0.022

0.011

-0.055

-0.056

-0.055

-0.056

-0.055

13.2000 Dummy

0.043

0.204

0.009

0.011

-0.051

-0.052

-0.051

-0.052

-0.051

14.2001 Dummy

0.043

0.204

-0.014

0.007

-0.049

-0.050

-0.049

-0.050

-0.049

-0.045

15.2002 Dummy

0.043

0.204

-0.019

-0.011

-0.048

-0.049

-0.048

-0.047

-0.044

-0.042

16.2003 Dummy

0.043

0.204

-0.013

-0.005

-0.047

-0.048

-0.047

-0.048

-0.047

-0.043

-0.042

-0.041

17.2004 Dummy

0.043

0.204

-0.017

-0.007

-0.047

-0.048

-0.047

-0.048

-0.047

-0.043

-0.042

-0.040

18.2005 Dummy

0.043

0.204

-0.020

-0.008

-0.047

-0.046

-0.043

-0.041

-0.040

19.2006 Dummy

0.043

0.204

-0.026

-0.009

-0.046

-0.045

-0.042

-0.040

-0.039

-0.038

20.2007 Dummy

0.043

0.204

-0.023

-0.011

-0.045

-0.046

-0.045

-0.044

-0.041

-0.040

-0.038

-0.037

21.2008 Dummy

0.043

0.204

-0.044

-0.012

-0.044

-0.043

-0.040

-0.039

-0.037

-0.036

-0.035

22.2009 Dummy

0.043

0.204

-0.061

-0.010

-0.043

-0.042

-0.039

-0.038

-0.037

-0.036

-0.035

-0.034

23.2010 Dummy

0.043

0.204

-0.070

-0.019

-0.042

-0.043

-0.042

-0.043

-0.042

-0.039

-0.037

-0.036

-0.035

-0.034

-0.033

24.2011 Dummy

0.043

0.204

-0.074

-0.015

-0.042

-0.041

-0.038

-0.037

-0.036

-0.035

-0.034

-0.033

-0.032

25.2012 Dummy

0.043

0.204

-0.078

-0.016

-0.041

-0.042

-0.041

-0.042

-0.041

-0.038

-0.036

-0.035

-0.034

-0.033

-0.032

-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

0.010

-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

-0.024

-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

-0.039

-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

0.005

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

0.011

-0.007

30.Total # words

8277.827

5280.

827

0.092

0.310

0.013

0.006

0.010

0.008

0.012

0.003

0.004

0.008

0.016

0.002

-0.008

-0.004

-0.012

-0.008

-0.005

-0.017

-0.014

-0.016

0.111

0.198

0.030

0.003

Table 2. Multiple Estimates of Logistic Regression Models Predicting Corporate Illegality in

Study 5.

Variables

model 1

model 2

model 3

estimate

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

Total # of

words

0.000

*** p<0.001, ** p<0.01, * p<0.05

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

(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

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.

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

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-

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.

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.

Social Contagion and Goal Framing: The Sustainability of Rule Compliance

Chapter

Full-text available

Nov 2021

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. Second, Goal Framing Theory (GFT) (Lindenberg and Steg 2007) suggests that we should look not just at the contagion of concrete behaviour but at the process that governs the contagion of the very goal to comply with norms and legitimate rules. This is particularly important because it predicts that observed (non-) compliance regarding one rule also affects (non-)compliance with other rules (‘cross-norm effects’). Because the goal to comply tends to decay, it needs continuous support from the observation of other people’s respect for norms and legitimate rules. Compliance and non-compliance are thus both self- reinforcing mechanisms. This has clear implications for policy which are discussed in this chapter, most notably the importance of a focus on legitimizing rules, so that they are interpreted as norms by the general public.

Majority Nationalism Laws and the Equal Protection of Minorities: Experimental and Observational Evidence from Israel

Article

Full-text available

Jan 2021

(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.

Experiential Learning in an Online, Asynchronous Business Ethics Course

Chapter

Jul 2023

Sheldene Simola

The rapid pivot toward, and ubiquitous use of online learning associated with the COVID pandemic led to substantial questions about whether experiential learning (EL) could be implemented within online courses, and if so, how. This chapter considers the question of how EL can be integrated into online courses and describes the implementation of Kolb’s EL model within an online, asynchronous course on behavioral business ethics. Following a brief introduction to reflective narrative as method of course analysis, this chapter addresses five areas. First, an overview of Kolb’s EL theory is provided, including phases of the EL cycle; associated learning style types of students; corresponding teaching roles; and rationale for using the EL cycle. Second, the focus, content, and EL processes of the in-person version of the business ethics course that was rapidly transitioned into an online course are summarized. Third, the challenges associated with transitioning the in-person, EL course into an online, asynchronous, EL course are discussed. Fourth, technologies and methods used to implement EL in the online, asynchronous version of the course are described. Fifth, lessons learned and areas for future development are identified.KeywordsAsynchronousBehavioralBusinessEthicsExperientialKolbOnlineTeaching

Language, perceived warmth, and investors’ reactions to audit committee reports †

Article

Feb 2023

An audit committee (AC) report is the primary channel through which investors learn about the responsibilities and activities of an AC. AC members may use personal language (“we”) or impersonal language (“the audit committee”) in an AC report. Psychology research suggests that personal (versus impersonal) language signals that the language user has greater warmth and sense of communion (i.e., being part of a larger group). Applying this theory, we predict that an AC's use of personal (versus impersonal) language leads investors to perceive a warmer and more communal AC, and that an AC's perceived warmth/communion (cued by personal language) positively impacts investor judgments. We further posit that the positive effect of personal language is stronger when AC compensation is largely short term than long term. This is because investors need more assurance of AC oversight effectiveness when AC compensation is short term, which makes investors rely more on heuristic cues such as AC language to make judgments. Consistent with this prediction, we find that when AC compensation is largely short term, nonprofessional investors (proxied by MBA students) react more positively to an AC's use of personal language than impersonal language. The effect of AC language is insignificant when AC compensation is largely long term, as the long‐term compensation structure already provides assurance about the AC's oversight effectiveness, and thus, investors rely less on heuristic cues. Further, we find that the perceived warmth of AC members explains the effect of AC language. Finally, our interviews with nonprofessional investors corroborate some of our main findings and validate their practice implications. This article is protected by copyright. All rights reserved.

Majority Nationalism Laws and the Equal Citizenship of Minorities: Experimental, Panel, and Cross-Sectional Evidence from Israel

Article

Jun 2022

The orchestrated digital inequalities of the IoT: How vendor lock-in hinders and playfulness creates IoT benefits in every life

Article

Full-text available

Nov 2022
NEW MEDIA SOC

As the Internet of Things (IoT) is becoming the next frontier for the digital divide, we argue that access to most of the IoT and its benefits are hindered by vendor lock-in. Yet, people with an advanced set of operational skills can find creative ways to overcome vendor lock-in. In this article, we study socio-contextual dispositions to why some people are better able to capitalize on IoT benefits creatively than others. We use interview data and a novel approach to a diary study using a mobile application among 30 households. By focusing on vendor lock-in against the interplay between operational skills and creative IoT uses, we find those with the least access to IoT benefits are users who mainly task-oriented and consider IoT tools to solve specific problems. Those with a more play orientation are better positioned to access IoT benefits and further develop their operational skills.

Typical corrupt practices in the criminal prosecution of businessmen

Article

Full-text available

Sep 2022

Gennadii Porosenkov

The subject. The study is devoted to the study of corruption aspect of criminal prosecution of entrepreneurs in Russia. The aim of this paper is to identify typical corrupt practices in the criminal prosecution of entrepreneurs, to investigate their causes, as well as to formulate recommendations for minimizing the identified practices of abuse. The methodology. The author analyzes the cases of criminal prosecution of entrepreneurs who have filed appeals to the Center for Public Procedures "Business Against Corruption", the Center "Protection of Business". These public platforms act as filters of appeals for unwarranted criminal prosecution. In case of confirmation of the facts of illegal use of criminal law instruments in relation to an entrepreneur, the appeal with the attached expertise is sent to the Commissioner for the Protection of Entrepreneurs' Rights. The main results, scope of application. To corrupt practices in the field of criminal prosecution of entrepreneurs the author included the following: unreasonable use of preventive measures in the form of detention in contravention of the direct prohibition of pt. 1.1 of the art. 108 of the Code of Criminal Procedure of the Russian Federation; ignoring the fact of committing a crime in the field of business and as a consequence qualification of the crime not by special business, but by common crimes; "superficial" preliminary investigation, resulting in criminal proceedings without due cause; ignoring the prejudicial facts established by arbitration courts in criminal proceedings; unreasonable imputing the commission of an intentional act. The reasons for the corrupt practices highlighted and investigated by the author are such factors as the wording of the law and its interpretation, limitations and imprecision of the law, allowing discretion on the part of law enforcers; the established KPIs of law enforcement agencies that encourage "the pursuit of performance" and accusatory bias; declarative ethical standards adopted by public authorities and companies, i.e. the lack of effectively implemented instruments of integrity management; low level of public control over criminal proceedings. In order to minimize the above practices, it is necessary to further improve regulations, to stimulate ethical regulation in organizations; to modernize the metrics for evaluating the effectiveness of law enforcement agencies, focusing on non-departmental parameters; to significantly increase the role of the institution of business ombudsman as an additional guarantor of legal rights and interests of entrepreneurs, and to actively introduce digitalization tools, in particular, digital records of criminal cases. It is important to emphasize again that the author does not claim to cover all possible corrupt practices in the prosecution of entrepreneurs and their causes. For example, there is a vicious practice of reclassifying a witness in a criminal case after testifying as a suspect and subsequently as a defendant. Further research could look in more detail at each of these corrupt practices, identify and analyze additional practices, as well as explore opportunities for their mitigation. From the point of view of the applied implementation of the results of the study, it seems that they can be useful, on the one hand, to decision makers when improving criminal policy towards entrepreneurs, but also to entrepreneurs themselves to understand the potential criminal law risks that they may face and must minimize.

“Many others are doing it, so why shouldn't I?”: How being in larger competitions leads to more cheating

Article

May 2021

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.

Pledging to Behave Honestly

Article

Jan 2020

Ethical Learning: The Workplace as a Moral Laboratory for Character Development

Article

Nov 2020

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.

Keeping Track of Needs in Communal and Exchange Relationships

Article

Full-text available

Aug 1986

Keeping track of needs in communal and exchange relationships was investigated in two experiments. In both, subjects could check to see if another person needed help. The first experiment tested the hypotheses that (a) When there is no opportunity for the other to reciprocate in kind, keeping track of the other's needs will be greater if the person desires a communal relationship with the other than if the person desires an exchange relationship with the other. (b) If the person desires an exchange relationship with the other, keeping track of the other's needs will be greater when an opportunity for the other to reciprocate in kind exists than when it does not. (c) If a communal relationship is desired, the existence of an opportunity for the other to reciprocate in kind will not influence keeping track of the other's needs. The results of the first experiment supported all three hypotheses. The second experiment tested and found support for the hypothesis that even when nothing can be done to help the other, keeping track of the other's needs will be greater if a communal relationship is desired with the other than if an exchange relationship is desired.

A Model of (Often Mixed) Stereotype Content: Competence and Warmth Respectively Follow From Perceived Status and Competition

Article

Full-text available

Jun 2002

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.

The Law of Good People: Challenging States' Ability to Regulate Human Behavior

Book

Full-text available

Jun 2018

Yuval Feldman

Cambridge Core - Macroeconomics - The Law of Good People - by Yuval Feldman

Crime and Punishment: An Economic Approach

Chapter

Dec 2007

Gary S. Becker

IntroductionBasic AnalysisOptimality ConditionsShifts in the Behavioral RelationsFinesSummary and Concluding Remarks

Perceived warmth and competence of others shape voluntary deceptive behaviour in a morally relevant setting

Article

Mar 2017

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.

Incentives and cheating

Article

Feb 2017

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.

Memories of unethical actions become obfuscated over time

Article

May 2016

Significance We identify a consistent reduction in the clarity and vividness of people’s memory of their past unethical actions, which explains why they behave dishonestly repeatedly over time. Across nine studies using diverse sample populations and more than 2,100 participants, we find that, as compared with people who engaged in ethical behavior and those who engaged in positive or negative actions, people who acted unethically are the least likely to remember the details of their actions. That is, people experience unethical amnesia: unethical actions tend to be forgotten and, when remembered, memories of unethical behavior become less clear and vivid over time than memories of other types of behaviors. Our findings advance the science of dishonesty, memory, and decision making.

Blind loyalty? When group loyalty makes us see evil or engage in it

Article

Jan 2016
ORGAN BEHAV HUM DEC

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.

Cognitive processes in organizations, Research in Organizational Behavior

Article