extranet.sioe.org · web view2013 “efficient estimation of word representations in vector...
TRANSCRIPT
Under Pressure: Culture and Structure as Antecedents of Organizational Misconduct
AUTHORS
Andrea Cavicchini ([email protected])IESE Business School
Fabrizio Ferraro ([email protected])IESE Business School
Samspa Samila ([email protected])IESE Business School
Keywords: Organizational misconduct; Organizational Culture; Goal Setting; Incentives;
Organizational Structure
1
Under Pressure: Culture and Structure as Antecedents of Organizational Misconduct
Abstract
Does a toxic organizational culture focused on performance lead to misconduct? To address this
question, we build on Merton's strain theory and theorize the relationship between organizational
culture, structure and misconduct. We first theorize that organizations with cultures
characterized by strong performance pressure are more likely to engage in misconduct. Then we
consider how organizational structure moderates this relationship, which we hypothesize is
weaker in more formalized organizations and stronger in more decentralized organizations. To
test these hypotheses, we analyzed the regulatory and law violations of 880 publicly traded firms
in the United States and measured organizational culture and structure through a natural
language processing (NLP) analysis of the firms’ employee reviews on Glassdoor. The empirical
results lend support to our hypotheses. Organizations with high performance pressure are 68
percent more likely to be fined for misconduct than organizations with low performance
pressure. The moderation effect of organizational structure is fully supported for
decentralization, but only partially supported for formalization.
2
Introduction
Virtually every analysis of corporate misconduct in the last two decades, including Enron (Sims
and Brinkmann, 2003; Kuliks, 2005), Wells Fargo (Independent Directors of the Board of Wells
Fargo, 2017), Boeing (Bretton Putter, 2019), and Deutsche Bank (Enrich, 2020), points to toxic
organizational cultures with strong performance pressure as a key antecedent of misconduct. Yet
this evidence tends to be anecdotal, mostly originating from journalistic investigations of high-
profile scandals, and thus lacks generalizability. Building on Merton’s (1938) foundational strain
theory of social deviance and Sutherland’s (1939, 1949) work on white-collar crimes, the
theoretical foundations of the association between culture and misconduct were first developed
by sociologists and organization theorists (Merton, 1936; Sutherland, 1949; Vaughan, 1999). Yet,
as organization theorists moved from an earlier emphasis on intra-organizational processes
(incentives, structure, culture) towards broader institutional and network mechanisms, they all
but abandoned this topic in the misconduct literature (Greve, Palmer, and Pozner, 2010; Cooper,
Dacin, and Palmer, 2013) prematurely, in our view. As a consequence, the relationship between
organizational culture and misconduct remains undertheorized and its empirical support mixed.
In this paper, we build on Merton's (1938) strain theory to theorize the relationship between an
organization’s specific cultural and structural features specifically, the extent of performance
pressure, formalization, and decentralization—and organizational misconduct.
3
In his influential 1938 article on "social structure and anomie," Robert Merton directed
sociologists’ attention away from examining individual-level antecedents of deviance towards
discovering "how some social structures exert a definite pressure upon certain persons in society
to engage in nonconformist rather than conformist conduct" (Merton 1938:672). Society, he
suggested, defines aspirational references (goals, purpose, interests) that are not attainable by all
members. Individuals in lower social classes who do not have access to the same educational and
economic resources as individuals in more privileged ones are still expected to achieve wealth
and fame. This situation creates frustration and might translate into deviant and criminal
behavior. This theory was developed at the societal level, and specifically referred to class as the
relevant social structure shaping individual behavior, yet the logic of the theory extended to
other levels of analysis, including the organization and the individual (Agnew, 1992; Langton
and Piquero, 2007; Agnew, Piquero, and Cullen, 2009). Notably, in criminology Agnew’s
General Strain Theory (1992, 2001) defined and widened the range of possible sources of
pressure and strain in society, and has been employed to explain white-collars crimes (for
example, Langton and Piquero 2007) but this theory tend to be more focused at the individual
level, and does not clearly outline how organizational factors can lead to misconduct.
Furthermore, even in this field the empirical evidence for the theory is very thin (Agnew,
Piquero, and Cullen 2009:55).
4
In organization theory, and building on March and Simon (1958), Staw and Szwajkowski
(1975) first showed that organizations operating in less munificent environments were more
likely to commit illegal actions in the process of gaining resources. Other studies showed that
firms affected by declining profitability or competitive threats are more likely to engage in
misconduct (Wheeler and Mody, 1992; Vaughan, 1999). Yet, well performing firms are also
under constant pressure to maintain their performance, as their aspirations might be more
difficult to achieve, and failure to do so may translate into more risk taking (Cyert and March,
1963; Tversky and Kahneman, 1991; Audia and Greve, 2006) and even misconduct. Indeed,
research on the relationship between performance expectations and misconduct found that both
organizational results below (Harris and Bromiley 2007) or above expectations (Mishina et al.,
2010) were associated with a greater likelihood of misconduct. However, at the organizational
level, empirical evidence on the relationship between performance pressure and misconduct is
mixed (Hill, Kelley, and Agle 1992). At a more micro level, the literature on goal setting theory
had documented, both in experimental and field settings, the positive performance consequences
of setting challenging goals (Locke and Latham 1990; Locke and Latham 2013; 2002). However,
studies also have yielded evidence of how these goals can lead to unethical behavior
(Schweitzer, Ordóñez, and Douma, 2004; Ordóñez et al., 2009a; Ordóñez and Welsh, 2015). As
such, scholars have suggested that the interplay between organizational cultures and goal setting
can play an important role in influencing organizational outcomes (Locke and Latham, 2009;
5
Ordóñez et al., 2009b; Yip et al., 2021). Yet, the relationship between goal setting and
organizational culture remains undertheorized.
In this paper, we reconceptualize strain theory at the organizational level by teasing out its
cultural and structural mechanisms. At the societal level, strain theory leverages both a cultural
mechanism (i.e., the widespread adoption of cultural goals of success and the meritocratic ideal)
and a structural one (i.e, a lack of resources preventing some groups from achieving these
ideals). In this paper, we theorize strain at the organizational level as the result of both
organizational culture and organizational structure. We first develop the hypothesis that
organizations with a culture characterized by strong performance pressure are more likely to
engage in misconduct. This cultural hypothesis is consistent with the nature of the mechanism
suggested by Merton's strain theory and Agnew’s general strain theory. Yet, testing this cultural
hypothesis beyond individual case studies has not been not possible so far, given the lack of
systematic data on organizational culture and misconduct (for a recent exception on the
consequences of misconduct, see a paper on fraudulent firms’ investment behavior by Wang,
Stuart, and Li, 2020).
Building on the literature on bureaucracy (Weber 1922/2019; Merton 1940; Gouldner 1954;
Blau 1955; Adler and Borys 1996), and the literature on monitoring in economics (Nagin et al.,
2002; Duflo, Hanna, and Ryan, 2012; Pierce, Snow, and McAfee, 2015), we hypothesize that
6
more formalized organizations, which arguably have more controls in place, have fewer negative
effects of performance pressure. Likewise, more decentralized organizations, which tend to have
less effective monitoring mechanisms, provide more fertile ground for performance pressure to
have negative effects (McKendall and Wagner, 1997).
We test these hypotheses using data from a panel of 880 firms in the United States on
420,000 civil and criminal cases resolved with total penalty amounts of $616 billion between
2008 and 2019, as well as 2.6 million employer reviews gathered from Glassdoor.com, which we
used to peek inside corporations and unobtrusively observe the digital footprints (Golder and
Macy, 2014) of their cultures. Not surprisingly, this resource has enabled organization theorists
to reignite research on organizational culture (Luo, Zhou, and Shon 2016; Corritore, Goldberg,
and Srivastava 2020; Marchetti 2019). To disentangle the different topics and dimensions of
employee reviews, we used natural language processing (NLP) tools—specifically, unsupervised
topic modeling (i.e., latent Dirichlet allocation [LDA]; Campbell, Hindle, and Stroulia, 2015)
and word embedding (Mikolov et al., 2013).
Our findings show that, on average, performance pressure increases the likelihood of
organizational misconduct by up to 68 percent. At the same time, the number and dollar amounts
of organizational penalties are significantly higher (averaging $2.8 million) when performance
pressure is high. The negative effect of performance pressure is stronger for companies that rely
7
on a more decentralized structure, while the results for companies that rely on a more formalized
structure are mixed. Less formalized organizations with high performance pressure have a higher
likelihood of misconduct and a higher average number of violations per year than companies
with low performance pressure.
This article makes several contributions to literature on the antecedents of organizational
misconduct and strain theory (Agnew, 1992; Vaughan, 1999; Greve, Palmer, and Pozner, 2010),
as well as literature on the consequences of goal setting in organizational psychology
(Schweitzer, Ordóñez, and Douma, 2004; Ordóñez and Welsh, 2015). First, we provide
systematic, empirical evidence on the relationship between organizational culture and
misconduct in a large sample of organizations. To the best of our knowledge, we are the first to
empirically test this correlation systematically. From a theoretical point of view, we contribute to
the development of Merton's strain theory at the organizational level, articulating the role of
culture and structure in predicting misconduct. We show that decentralization creates the
conditions for misconduct only when mediated by organizational culture, and that formalization
might decrease the likelihood of misconduct, but also lead to higher penalties. Our findings also
contribute to the literature on the dark side of goal setting by showing the importance of culture
and structure in linking stretch goals with misconduct (Ordóñez et al., 2009b:86). Finally, our
methodological approach offers researchers new tools to revive research on organizational
culture and structure and offers practitioners a forensic tool to predict organizational misconduct.
8
THEORY AND HYPOTHESES
Organizational Misconduct and Strain Theory
Following Greve et al. (2010), we define organizational misconduct as "behavior in or by an
organization that a social control agent judges to transgress a line separating right from wrong"
(Greve et al. 2010: 56). Although our empirical test focuses on behaviors deemed illegal by U.S.
regulatory agencies or the Department of Justice, our theorizing need not be limited to illegal
activity. Corporate misconduct is widespread, with survey data, for instance, showing that 20%
of CFOs in U.S. public companies engage in earnings management practices (Dichev et al.,
2013), despite the increasing sophistication of internal and external control mechanisms.
Organization theorists have long sought to understand misconduct, as it is foundational to
the discipline's mandate to explore how organizations affect the societies in which they are
embedded (Boulding 1958; Stern, and Barley 1996). Under different labels (misconduct,
wrongdoing, illegality, white-collar crime, fraud, unethical behavior), scholars have explored the
antecedents and consequences of misconduct for decades (see Szwajkowski 1985; Vaughan
1999; Simpson and Weisburd 2009; Greve, Palmer, and Pozner 2010; Cooper, Dacin, and Palmer
2013; and more specifically on CEO misconduct, Schnatterly, Gangloff, and Tuschke 2018). Yet,
despite decades of scholarly inquiry, the development of research at the meso and macro levels
has lagged behind research at the individual level (Palmer, Smith-Crowe, and Greenwood 2016;
9
See Treviño, Den Nieuwenboer, and Kish-Gephart 2014 for a review of the micro literature).
Most recent advances in the macro level literature, furthermore, have drawn on institutional
theory (Gabbioneta et al., 2013) and theories based on the concepts of embeddedness and social
networks (Yenkey, 2018) to explain interorganizational processes of diffusion of misconduct
(Mohliver, 2019), devoting less attention to intra-organizational dynamics.
In a review of the literature on organizational misconduct, Diane Vaughan (1999: 273)
suggested: "Merton's thinking is the foundation of any consideration of the dark side of
organizations. He observed that any system of action inevitably generates secondary
consequences that run counter to its objectives." In this quote, she referred to his work on strain
theory (Merton, 1938) and the dysfunctional consequences of bureaucracy (Merton, 1940),
framing both within his general theoretical insight that the same mechanisms and processes that
generate positive outcomes can lead to negative consequences under certain conditions (Merton,
1936). Together with others (primarly Perrow 2011), she translated this insight into a research
program on the normalization of accidents, deviance and mistakes, stemming from her rich
analysis of the NASA Challenger disaster (Vaughan 1996). On the other hand, in assessing the
literature on misconduct based on strain theory, she found that the evidence on the role of
performance pressure for organizations as the antecedent of misconduct is not very conclusive.
She suggested that given the inherent competitiveness of business activity, “all organizations
may be structurally induced to violation, regardless of ranking in the organizational stratification
10
system” (Vaughan, 1999: 289). Ten years later, in another review, Greve et al. (2010) still
lamented this state of affairs but suggested that conflicting empirical evidence might be
explained by organizational expectations, which translate external pressures and aspirations into
specific goals for teams and individuals.
In line with this insight, we suggest a novel reconceptualization of Merton’s strain theory
that distinguishes the role of two different mechanisms: a cultural mechanism (performance
pressure perceived by employees), and a structural mechanism (the degree to which this pressure
is counterbalanced by formalization and centralization, two key components of organizational
structure). This conceptualization is consistent with the original theoretical emphasis on both a
cultural/symbolic mechanism (i.e., the widespread adoption of cultural goals of success and the
meritocratic ideal) and the role of broader structural features of society (i.e., stratification and a
lack of resources which prevents some groups from achieving those goals). This
reconceptualization is also consistent with the accounting literature on the “fraud triangle”
(Cressey, 1950, 1953), a conceptual framework used to understand the antecedents of financial
statement fraud based on three elements: pressure (i.e. the necessity to commit wrongdoing),
opportunity (i.e. the environmental conditions that leads to misconduct), and rationalization (i.e.
the moral justification for the fraud). This model had been applied to predict many cases of
organizational fraud, especially by professionals (American Institute of Certified Public
Accountants (AICPA), 2002) but academic attention on its effectiveness had shown mixed
11
results (for example, Skousen, Smith, and Wright 2009), and had been questioned on its absence
for individual and group-level explanations (Donegan and Ganon, 2008; Cooper, Dacin, and
Palmer, 2013; Power, 2013; Morales, Gendron, and Guénin-Paracini, 2014; Lokanan, 2015).
Organizational Culture, Performance Pressure and Organizational Misconduct
In line with strain theory (Merton, 1938), organizations might establish very ambitious formal or
informal goals for their employees, who might not have the resources to achieve them. Thus,
they feel the pressure to cut corners and ignore ethical guidelines, and sometimes engage in
misconduct. This pressure might operate at different levels of the organization: from top
managers feeling the pressure of shareholders and competitors (Bartov, Givoly, and Hayn 2002;
Bergstresser and Philippon 2006; Zhang et al. 2008; Harris and Bromiley 2007; Mishina et al.
2010), to line employees working under ambitious supervisors and aggressive goal-setting
programs.
As high expectations for performance become part of employees' learning experience (i.e.,
their routine), the performance orientation of the organization will become part of its culture
(Schein, 1996; Chatman and O’Reilly, 2016), defined as "a system of shared values (that define
what is important) and norms that define appropriate attitudes and behaviors for organizational
members (how to feel and behave)" (O’Reilly and Chatman 1996: 160). Indeed, several studies
based on the Organizational Culture Profile (OCP) survey tool, have identified results
12
orientation as one of six key dimensions of organizational culture (together with adaptiveness or
innovation, detail-orientation, collaboration or teamwork, customer-orientation, and integrity
(O’Reilly, Chatman, and Caldwell, 1991; Chatman and O’Reilly, 2016). Results-orientation is
defined as an organization's tendency to have high expectations for achievement, goals and
individual performance (O'Reilly, Chatman, and Caldwell, 1991). In line with goal setting
theory, an organizational culture focused on results-orientation can have positive effects on
individual and organizational performance. Indeed, in decades of research in goal setting theory,
scholars have documented how ambitious goals can increase individual effort and performance
(Locke and Latham, 2002, 2013). However, considerable evidence has documented the negative
consequences of these aggressive goals on employee’s mental stress and anxiety (Karasek, 1979;
Demerouti et al., 2001; Fox, 2016; Yip et al., 2021), and unethical behavior (Schweitzer,
Ordóñez, and Douma, 2004; Ordóñez and Welsh, 2015).
Scholars studying these negative consequences suggest that the same psychological
mechanisms that explain the benefits of performance-oriented cultures (i.e., increased
performance), might also lead to unethical behavior under specific conditions (see Ordoñez, et
al. 2015 for a review). For instance, experimental evidence on goal-setting across multiple
rounds shows that people with unmet goals were more likely to engage in unethical behavior
(Schweitzer, Ordóñez, and Douma, 2004), and that the depletion of self-regulatory resources
mediates the relationship between goal structures (high-low) and unethical behavior (Ordoñez, et
13
al. 2015). Furthermore, individuals might engage in misconduct to restore their positive self-
evaluation: after experiencing a threat to their abilities, individuals who misrepresent their
performance as better than it is experience increased feelings of competence (Wakeman, Moore,
and Gino, 2019). These psychological mechanisms might translate into group norms that might
dominate broader moral concerns, as demonstrated in a series of studies showing that group
activities such as bribing officials, lying to investors, or cheating are compensated in various
ways (Wiley Wakeman and Moore, 2018), contributing to a general culture of moral
disengagement (Moore, 2008; Moore et al., 2012).
At the organizational level, several case studies, ethnographies, and journalistic
investigations have confirmed the insight that “toxic” performance cultures can create conditions
that facilitate misconduct. In a study of Enron's accounting fraud, Simms and Brinkmann (2003)
described how the adoption of performance rankings to evaluate employees created a culture
based on pressure and competition with peers. However, performance pressure can arise even in
the absence of rankings and performance bonuses, as the manager of the equity derivatives
trading room studied by Beunza (2019) explained: “Even if you tell a guy, ‘I’m going to pay you
$2 million no matter what,’ he will still like to show a $50 million profit and not $10, because
$50 million means he’s a rooster. He can walk through the hen house as the biggest rooster.”
(2019:229). Another recent example comes from the Wells Fargo accounting fraud of 2016,
whereby bank employees created millions of fraudulent savings and checking accounts on behalf
14
of clients. Findings from investigations show that individual workers and branches charged fees
and issued unexpected credit or debit cards to customers to meet the company's incentives to sell
multiple financial products. The company's strong emphasis on its sales goals "distorted the sales
model" and translated into a performance culture that fostered "an atmosphere that prompted
low-quality sales and improper and unethical behavior" (Shearman and Sterling LLP, 2017).
Organizational cultures might also convey limited concerns for following rules and regulations,
and turn a blind eye to unethical behavior. For instance, Kweku Adoboli, the UBS rogue trader
ultimately sentenced to 7 years in prison for illegal trading activities, revealed that his supervisor
had told him: "You don't know you are pushing the boundaries hard enough until you get a slap
on the wrist" (Fortado, 2012). On several occasions where he had pushed the boundaries through
lucrative trades, he had been congratulated rather than penalized (Moore et al., 2019).
Beyond case studies, the relationship between culture and misconduct has been established
at the macro level of analysis. Starting from national culture, Fisman and Miguel (2007) studied
parking violations of United Nations officials in Manhattan and found that officials from
countries with high levels of corruption accumulated significantly more unpaid parking
violations. Similarly, DeBacker, Heim, and Tran (2015) used the level of perceived corruption in
owners’ countries of origin and Internal Revenue Service (IRS) audit data to study corporate tax
evasion among U.S. companies with owners from foreign countries, and found that owners from
countries with higher corruption norms were more likely to engage in tax evasion. Similarly,
15
using U.S. Census data, Liu (2016) found that firms with executives whose ancestors came from
countries where corruption is less frowned upon were more likely to engage in fraud,
opportunistic insider trading, earnings management, and option backdating. Periods of economic
booms might also contribute to shaping a culture of overconfidence and risk taking among
executives, which in turn might translate into unethical behavior and misconduct. Bianchi and
Mohliver (2016) found that CEOs were more likely to backdate their stock options (an illegal
practice that maximizes financial gains), during prosperous economic times. They also showed
an imprinting effect, which hints at a cultural mechanism, as CEOs who began their careers in
prosperous times were also more likely to engage in misconduct. A similar socialization
mechanism, albeit in a different direction, was suggested in a study showing that CEOs who
served in the military were less likely to engage in financial misconduct (Koch-Bayram and
Wernicke, 2018).
The literature in finance, accounting, and management has documented the relationship
between aggressive financial targets and organizational misconduct (O’Connor et al. 2006;
Zhang et al. 2008; Johnson et al. 2009). For instance, executives who committed fraud had
greater financial incentives to do so (Johnson, Ryan, and Tian, 2009). CEOs are more likely to
misrepresent company earnings if their financial incentives are linked to company performance
(Bergstresser and Philippon, 2006; Zhang et al., 2008). Also, aggressive financial targets can be
the result of managerial aspiration levels, whether to outperform their peers (Harris and
16
Bromiley 2007), or to exceed shareholders’ expectations (Mishina et al.. 2010). In addition to
executives, middle managers can play a crucial role in translating high-level abstract
performance objectives into lower-level ones that induce deceptive performance, as shown in an
ethnographic study of sales at a telecom firm (den Nieuwenboer, da Cunha, and Treviño, 2017).
Regardless of whether misconduct originates from the top or lower levels of an organization
(Palmer, 2008), the literature on the diffusion of misconduct within organizations concur that
misconduct tends to become institutionalized in routines and embedded in organizational culture
(Brief, Buttram, and Dukerich, 2001; Ashforth and Anand, 2003). Thus, environmental pressures
on organizations, and specific incentive design decisions can translate into a culture of
performance for all employees, and in turn lead to unethical behavior and organizational
misconduct.
Another mechanism through which performance pressure can lead to misconduct is by
driving more risk taking. This mechanism is entirely consistent with strain, goal setting, and
prospect theories. Specifically, challenging goals serve as reference points that create a region of
perceived losses for outcomes below established thresholds (Kahneman & Tversky, 1979;
Tversky & Kahneman, 1992), which can cause employees to increase risk taking, as also
demonstrated in experimental settings (i.e., Larrick, Heath, and Wu 2009). For instance, banks
with more aggressive competitive cultures are more likely to engage in riskier lending practices
17
by extending more loans to borrowers with poorer ratings (Nguyen, Nguyen, and Sila, 2019).
Therefore, performance pressure likely increases risk taking among employees, which may
increase the likelihood of misconduct.
Building on these arguments, we expect:
Hypothesis 1. Organizations with cultures characterized by strong performance
pressure for employees are more likely to engage in misconduct.
Organizational Structure and Misconduct: Formalization
Despite its central role in the early development of the field (Talor, Weber, Blau, Gouldner,
March and Simon, Pugh et al), formal organizational structure has been neglected by
organization theorists for many years, in what some scholars refer to as a period of “collective
amnesia” in the field (McEvily, Soda, and Tortoriello 2014: 302). However, as predictions
regarding the disappearance of bureaucracy have been proven wrong, we have experienced a
(small) revival of studies of bureaucratic organizational structure (Adler and Borys 1996; Adler
2012; Du Gay 2005; Clement and Puranam 2018; Sandhu and Kulik 2019).
In line with the traditional Columbia approach to bureaucracy, which explored the dark side
of bureaucracy (Merton, Gouldner and Blau), sociologists initially tended to consider
bureaucratic organizational structure as a factor that makes organizations more prone to
18
misconduct or accidents (Vaughan, 1999; Perrow, 2011). However, empirical evidence on this
relationship remains flimsy and is limited to individual case studies. Furthermore, more recent
theorizing on bureaucracy has reinvigorated a more complex—and more positive—view of
bureaucracy, emphasizing its enabling properties (Adler and Borys, 1996) by developing a
hitherto underappreciated insight from the Columbia school that formal bureaucratic rules might
have different effects depending on the informal organizational and cultural practices used to
introduce them (Gouldner, 1954; Blau, 1995). Adler and Borys (1996) reconceptualized the
Toyota Production System as a form of enabling bureaucracy, given that formal rules were
coupled with workers’ control of the process. This enabling form contrasted with traditional
coercive ones wherein workers have no control over rules, perceive them as a control device,
and exhibit typical signals of alienation.
Building on Adler and Borys (1996), we theorize the relationship between organizational
structure and misconduct by distinguishing an organization’s level of formalization - “the extent
to which rules, procedures, instructions, and communications are written” (Pugh et al. 1968: 75)
- from its degree of decentralization and considering how each contributes to shaping the
organization’s culture by moderating the effect of performance pressure. In line with this more
positive view of bureaucratic organizing, more formalization should reduce the likelihood of
misconduct by making work processes more transparent; in turn, higher efficiency might reduce
pressure to engage in malfeasance. Notice that this does not necessarily mean that employees
19
have a positive perception of formal rules. Indeed, even when formalization is designed in
enabling ways that can lead to individual and collective improvements in productivity, individual
workers might still express ambivalence about the rules, as revealed in Adler’s (1993, 2012)
interviews with Nummi workers.
This expectation is also consistent with the economics literature on crime and deviance,
which models deviant behavior as the result of rational calculations wherein individuals weigh
the benefits of engaging in misconduct against the costs of sanctions and likelihood of getting
caught (Becker, 1968; Milgrom and Roberts, 1988). Empirical research on the topic has
confirmed this basic insight. A field experiment at a telephone call center, for instance, found
that employees engaged in malfeasance at a higher rate under reduced monitoring (Nagin et al.,
2002). Likewise, in a field experiment in Indian schools, teacher absenteism dropped from 43
percent to 21 percent in schools that introduced a system of attendance control and incentives
(Duflo, Hanna, and Ryan, 2012). Whereas higher formalization is not limited to control
mechanisms, it is likely that more formal organizations adopt stricter and more extensive
management control practices. For instance, restaurant chains that invest in technology-based
employee monitoring experience less theft and higher employee productivity (Pierce, Snow, and
McAfee, 2015). In line with our theoretical argument and the empirical evidence we described,
we expect:
20
Hypothesis 2. Formalization of organizational structure and processes negatively
moderates the relationship between performance pressure and organizational
misconduct.
Organizational Structure and Misconduct: Decentralization
Decentralization is another key structural feature of organizations and affects the locus of
decision-making power: in more decentralized organizational structures, employees participate
more in decisions, and have more autonomy in their jobs (Hage and Aiken, 1967; Pugh et al.,
1969). A decentralized, or flattened, structure can increase the benefits of a performance-oriented
culture by enhancing market responsiveness, morale, and employee accountability (Rajan and
Wulf, 2006). However, precisely the increased autonomy of employees in a decentralized
structure can also increase the likelihood of misconduct.
Decentralization fosters an entrepreneurial environment that enables quicker responses to
local demands. Although this autonomy, coupled with performance pressure, may yield
performance benefits (Hoskisson and Hitt, 1988; Williamson, 1975), it might also lead to
increased risk-taking by local managers to meet goals and expectations established by the
company. Furthermore, higher decentralization is usually more common in larger and more
diversified organizations. In these more complex organizations monitoring is usually more
difficult, and this might facilitate misconduct (Jamieson, 1994).
21
Evidence from previous studies on the effect of organizational structure on misconduct
suggests this structural feature of the organization is not a sufficient condition to explain
misconduct (Hill et al. 1992; Simpson and Koper 1997). Hill et al. (1992) studied misconduct in
the manufacturing sector by analyzing a range of different factors, from firm size to financial
strain and decentralization, using Environmental Protection (EPA) and Occupational Safety and
Health Administration (OSHA) data. They found that decentralization could not explain
misconduct in either category. Focusing on violations of environmental laws in the
manufacturing industry, McKendall and Warner (1997) found that while decentralization alone
could not explain an increased likelihood of misconduct, the interactions between structure and
organizational culture (i.e., ethical climate), and to a lesser extent, industry profitability,
decreased the likelihood of serious environmental violations.
Even if decentralization might not have an independent effect on misconduct, we propose
that decentralization would facilitate the development of a toxic performance culture in the
organization, and thus strengthen the positive relationship between performance pressure and
misconduct. For instance, in the Wells Fargo accounting fraud scandal of 2016, a decentralized
organizational structure coupled with aggressive sales targets at the regional level led to illegal
sales practices in multiple branches of the bank. Results of an independent investigation of Wells
Fargo show that the company's "decentralized corporate structure gave too much autonomy to
the Community Bank's senior leadership, who were unwilling to change the sales model or even
22
recognize it as the root cause of the problem" (Shearman and Sterling LLP, 2017: 2). Flatter
organizations might reduce the effectiveness of hierarchical control. At Wells Fargo, for instance,
the control function at the company was "constrained by the decentralized organizational
structure and culture of substantial deference to the business unit" (Shearman and Sterling LLP,
2017: 2).
Thus, in line with our theoretical arguments, and the evidence emerging from multiple case
studies of misconduct, we hypothesize:
Hypothesis 3. Decentralization of organizational authority positively moderates the
relationship between performance pressure and organizational misconduct.
DATA AND METHODS
Sample
We tested our hypotheses using a panel dataset of yearly firm observations for a sample of firms
that were covered in both Compustat and evaluated on Glassdoor to which we subsequently
added the variables gathered from Good Jobs First.
The variables measuring organizational misconduct were gathered from Good Jobs First
and, specifically, by the “Violation Tracker” dataset which covers 71 types of legal and
regulatory violations addressed by more than 50 federal regulatory agencies, including 412,000
23
civil and criminal cases with total penalties of $616 billion and a minimum penalty of $5,000. In
our sample, we restricted our sample to publicly traded companies with matched CIK codes in
Compustat, resulting in 1,293 companies affected by penalties for 34,203 violations.
We used Glassdoor employee reviews to construct our measures of performance pressure,
formalization, and decentralization. Glassdoor aggregates millions of reviews and company
ratings, CEO approvals, salary and benefits reports, and interview reviews. The service counts
more than 70 million reviews and covers 250 thousand companies only in the US. We provide an
example of a Glassdoor employee review in Figure 1 (see Online Appendix A for more
information on Glassdoor). The initial sample included 2.6 million reviews of 3,423 companies
with performance data on Compustat. Companies were matched through a Python fuzzy name-
matching algorithm.
------------------------------------------
Insert Figure 1 about here.
-------------------------------------------
We used Compustat as the source of corporate financial information and as the starting point for
the matching process of both Glassdoor and Good Job First datasets.
Matching the three datasets resulted in an initial sample of 3,688 companies. Following recent
literature using Glassdoor data (i.e., Corritore et al., 2019) and using a similar filtering approach,
24
we excluded all observations with less than 100 reviews per year, resulting in a sample of 925
firms, between 2008 and 2019, yielding an unbalanced panel dataset of 4,296 firm-year
observations.
Dependent Variable
We operationalized organizational misconduct using official resolutions from U.S. courts
provided by Good Jobs First. Due to the nature of these data, our examination is limited to cases
of misconduct discovered by official agencies. Nevertheless, our data cover a broader spectrum
of illegal activities, have not yet been used in academic research, and are more accurate than data
from media coverage, which tends to be focused only on the largest cases of misconduct. We
measured organizational misconduct in three ways. For each firm-year observation, we: created
a binary variable, penalties, assigning a value of 1 if the company received a penalty due to
misconduct, and 0 otherwise; counted the number of penalties; and calculated the total dollar
amount of penalties.
Independent Variables
The dataset includes all employer reviews written by employees in the United States from 2008
to 2019 on the website Glassdoor.com. Glassdoor reviews are divided into two sections: a
quantitative section where users are asked to rate their firms on some attributes, and a qualitative
25
open textual section divided into two sentiment-coded categories: pros (i.e., positive aspects of
the company), and cons (i.e., negative aspects of the company), and an additional text section
(optional) in which employees can submit suggestions regarding how to improve the company
(advice to management). Since our analysis focuses on the negative perception of the
performance orientation norm, we only focused our analyses on the cons section of the reviews.
To grasp the specific dimension that might refer to the performance orientation norm in a
wide sample of reviews, we used NLP tools and, specifically, unsupervised topic modeling
(LDA; Blei, Ng, and Jordan 2003). Our model was based on 26 topics which resulted from
coherence score metrics (Newman et al., 2010). For technical details on the number of topics and
the methodology, see Online Appendix B. Prior studies in management research used topic
modelling in multiple areas. For example, Kaplan and Vakili (2015) applied LDA on patent text
to identify inventions that originated new topics in the body of knowledge, Corritore, Goldberg,
and Srivastava (2020) used topic modelling on Glassdoor reviews to create measures of cultural
heterogeneity, and, Di Maggio et al. (2013) used this methodology to study the evolution of
press coverage regarding public funding of the arts.
Following Giorgi and Weber (2015), we assigned unique labels to the topics resulting from
the LDA process to better contextualize them, as shown in Table 2. Topic 1 is the topic under
26
study, defined by words that connote a negative perception of the company’s pressure on
performance (i.e., sale, goal, number, expectation, pressure, commission).
------------------------------------------
Insert Table 1 about here.
-------------------------------------------
The final stage of the textual analysis involved inferring a score representing the
standardized probability from 0 to 100 for every topic discussed across the entire corpus of data.
Thus, the higher the score, the higher the probability that the topic would be discussed in an
employee’s review. Finally, we created a binary variable (performance pressure), assigning a
value of 1 if the weight of the topic was in the 95 th percentile or above among firm-year
observations. This value is arbitrary and represents approximately 3 standard deviations from the
mean (Mean = 3.13; SD = 1.61; Max = 14.85). We used this arbitrary value because it is a more
stable representation of a year in which the performance pressure topic is one of the most
discussed by employees. Because the computed model includes 26 topics, a single topic is
unlikely to appear in more than 50% of reviews in a given year. The values found for the topic
under study follow a distribution that is roughly similar to those of the other 25 topics.
Therefore, we posit that analyzing the tail of the semantic distribution (i.e., 95 th percentile)
assures that the topic under study is prevalent in the focal year.
27
Organizational Structure
Researchers have studied organizational structure primarily by analyzing proprietary data
(Guadalupe and Wulf, 2010), interview and observational data (Soderstrom and Weber, 2020), or
survey data (i.e. World Management Survey (WMS); Bloom and Van Reenen, 2008). In this study
we measured formalization and decentralization through an unobtrusive NLP technique: linear
text classification of words in employee discourse (i.e., company reviews on Glassdoor.com). To
do so, we used word embedding and FastText as a pre-trained dictionary (see Appendix B for
details on the methods). This pre-trained dictionary enabled us to classify every word of a review
into a semantic space and identify word similarities between target vectors (i.e., words related to
formalization and decentralization) in our dataset. FastText enable us to identify and analyze less
common and more specific topics that would not appear as often and were discarded by our topic
modelling analysis. Our use of this tool is similar to Cao, Koning, and Nanda (2020), who
studied biased sampling of early users in entrepreneurial experiments and used FastText to train
a sematic model to measure product’s focus on female customers on a large sample descriptions
of new technology products.
We derived our target vectors through additive compositionality (Gittens, Achlioptas, and
Mahoney, 2017), whereby arithmetic operations are applied to vectors to refine their word
28
compositions. For example, the nearest vector to the word “king” can be identified via an
arithmetic operation (Mikolov et al., 2013):
vman+vroyal=vking,
with v representing the vector with the related word. In the same way, the target vectors used in
our study for formalization and decentralization, respectively, are represented by:
v target−formalization=vbureaucratic+v formalization+vred−tape ; and
v target−decentralization=vdecentralization+vdecentralized−vcentralization.
Table 2 shows the list of words most similar to our target vectors. Finally, we calculated cosine
similarity between our target vectors and every word of each review, and retained the maximum
score for each vector. Thus, for every review, we obtained two scores (from -1 to 1) representing
the maximum similarity of the review to our target vectors for formalization and
decentralization.
------------------------------------------
Insert Table 2 about here.
-------------------------------------------
We considered only the cons section of each review to measure the high formalization
variable as the words in the target vector related to formalization tend to have negative
29
connotations, and thus are less likely to appear in the pro section of the review. We treated high
formalization as a binary variable, assigning a value of 1 when the value of formalization for the
company exceeded the industry (by two-digits SIC) median, and 0 otherwise. We analyzed both
the pros and cons sections of each review to measure the high decentralization variable, as the
words in the target vector are neutral. We treated high decentralization as a binary variable,
assigning a value of 1 if the value of decentralization exceeded the median for of our sample, 0
otherwise. Decentralization does not include industry, as we expected decentralized structure to
be more concentrated in some industries (i.e., manufacturing) more than others (i.e., mining).
Control Variables
We included several control variables in our analysis. First, we controlled for the number of
reviews in the focal year (in logarithmic form). Second, we operationalized firm size as the
number of employees reported in Compustat for the focal year, using natural logarithm values to
reduce the potential effects of extreme values. Third, we included sales growth (%), and ROA
(both winsorized at the 1% and 99% level) to account for firm performance in the focal year as a
possible explanation that could increase performance pressure on employees. Fourth, we
measured market competition as the Hirschman-Herfindahl index representing the distribution of
sales among firms in a given industry (based on 2-digit SIC code), as a higher concentration of
sales could lead to misconduct whether the thread represented by competition brings to a decline
30
in status for lower-performing firms (Wilmot and Hocker, 2001), or to lose customers to
competitors (Bennett et al., 2013). Finally, we controlled for other quantitative variables
measured in the reviews on Glassdoor. We controlled for CEO approval by assigning a value of -
1 (negative), 0 (neutral), or 1 (positive) representing average employee ratings for each firm-year
observation. We also controlled for senior management approval, compensation and benefits,
and work-life balance using average employee ratings on a 1 to 5 scale for each firm-year
observation.
Estimation techniques
Because the dependent variable is penalties incurred for organizational misconduct, we
calculated all independent and control variables using values from the end of the prior year. We
specified various models to test our hypotheses using three dependent variables representing
organizational misconduct to offer wider evidence for the proposed effects.
First, we used fixed-effects logistic regression analysis for the penalty variable because the
variable is dichotomous and takes the value of 1 if the company received a penalty, and 0
otherwise. Second, we used a fixed-effects Poisson regression analysis for the number of
penalties variable as it represents a count of events, accounting for both year and industry (2-
digit SIC code) fixed-effects. Third, we used a fixed-effects linear regression analysis for the
amount of penalties, using the dollar amount as a continuous variable. We specified fixed effects
31
on both industry (2-digit SIC code) and year, and transformed the variable to represent millions
of dollars. For all of the models, we excluded outliers in the dependent variable (i.e. winsorized
at the 1st and 99th percentiles of the distribution) to eliminate the potential for coefficients to be
influenced by a small group of observations with a high number and/or amount of penalties.
Also, we clustered standard errors by organization to control for potential heteroscedasticity and
provide a more conservative test of the hypotheses (White and White, 1980).
To test Hypotheses 2 and 3, we ran three separate analyses, dividing the sample by high/low
formalization and high/low decentralization. By using different subsets in our models we
described formalization and decentralization (whether high or low) as different environments,
therefore assuming that the effect of performance pressure would differ among them. As
Venkatraman (1989) put it, these types of analyses reflect the strength of moderation. A test of
differences confirmed that subgroup analysis is appropriate, as the two groups are significantly
different for all the variables under study. Finally, we ran collinearity diagnostics to check for the
presence of multicollinearity in our models, and all values were below the threshold of 30
(Bollinger et al., 1981). Finally, as the analyses uses fixed-effects to estimate the results, the
number of observations will not strictly follow the initial specification of observations as fixed-
effects models drops all the observations without within-group variance of the dependent
variable.
32
RESULTS
Table 3 provides pairwise correlations and descriptive statistics for each of the variables in our
study. The predicted likelihood of receiving a penalty for misconduct, of $5000 or more, is 28.6
percent for our sample. The probability of receiving a penalty with a dollar amount over the
mean is 8.6 percent; this likelihood is similar to that reported in other studies on corporate
misconduct (e.g., 15 percent in Mishina et al., 2010). As shown in Table 3, the correlation
between performance pressure and the dependent variable is not high, and is actually negative
between the binary penalty variable and number of penalties, although the correlation is very
close to zero.
------------------------------------------
Insert Table 3 about here.
-------------------------------------------
The regressions in Table 4A show that performance pressure increases both the likelihood of
a company incurring a penalty in the subsequent year (model 1). The result for performance
pressure holds adding financial controls (model 2) and the quantitative measures of the reviews
(model 3), thus supporting Hypothesis 1. Companies with result pressure have a probability to
receive a penalty to subsequent year around 68 percent. In addition, Tables 4b and 4c showing,
respectively, the effect of results pressure on the number of penalties and the amount in dollars,
33
provide more support for Hypothesis 1: companies with results pressure receive a higher number
of penalties and pay 2.8 million dollars more for their misconduct the subsequent year compared
to firms without such pressure.
-----------------------------------------
Insert Table 4 about here.
-------------------------------------------
Hypothesis 2 is only partially supported. The sub-sample analyses for the full models 1 and
2 in Tables 5A and 5B show that performance pressure affects companies with low formalization
by increasing both the likelihood of misconduct (83 percent higher than companies with low
performance pressure) and the number of violations. However, even though the effect of
performance pressure is stronger in the low formalization subsample, model 2 of Table 5A and
5B show that performance pressure is significant also in the high formalization subsample.
Furthermore, results for the dollar amount of penalties in Table 5C (model 1) show that
performance pressure does not have a significant effect on misconduct in companies with low
formalization (p=0.110).
Finally, Hypothesis 3 is supported. The results for models 3 and 4 in Table 5A show that
when performance pressure is high, the likelihood of a company with a decentralized structure
incurring a penalty the subsequent year is 86 percent; In centralized companies, the effect of
34
performance pressure is not significant. The results for models 3 and 4 in Table 5B reveal the
effect of performance pressure on the number of penalties incurred by companies with different
levels of decentralization. Companies with high performance pressure in the decentralized
subsample receive on average 1 penalty more than companies in the same group without high
performance pressure. Finally, the results for models 3 and 4 in Table 5C support the hypothesis
that the effect of performance pressure is stronger in companies with a decentralized structure,
which receive higher average fines of about 5 million dollars.
------------------------------------------
Insert Table 5 about here.
-------------------------------------------
These results suggest that, overall, performance pressure increases the likelihood of
organizational misconduct. When dividing the companies by structure, the result holds for
companies with decentralized structures, but are mixed for companies with highly formalized
structures.
ROBUSTNESS CHECKS
We conducted 3 additional analyses to assess the robustness of our findings. First, to test whether
the results might be driven by CEO-specific characteristics, we replicate our analyses including
35
CEO compensation, tenure, and gender finding nearly identical results. These analyses are
presented in Online Appendix D.
Second, our organizational misconduct variable measured corporate illegal actions that had
been identified and sanctioned by the relevant authorities. To test whether performance pressure
might affect broader measures of misconduct, we replicate our analyses using negative media
coverage on environmental, social, and governance (ESG) issues as our dependent variable
(Online Appendix E). We measure this variable with RepRisk, a data provider that collect news
articles from various national and local sources that criticize organizations on ESG issues. The
results of the analyses show that the likelihood to receive negative media coverage is positively
correlated with performance pressure, though the result is not significant (p= 0.124). The sub-
sample analysis divided by type of organizational structure shows that the likelihood to receive
negative media coverage is higher for low formalized and decentralized organizations. Overall,
the results for the analyses on negative ESG media coverage show that the effect of performance
pressure is consistent with our theory.
Finally, our performance pressure variable only focused on the tail of the distribution for the
performance pressure topic. We operationalized the variable in this way to identify the
companies where performance pressure characterizes the culture of the organization.
Nevertheless, we tested whether our results are consistent with different cutoffs for our measure
36
of performance pressure. Specifically, we replicated the analysis at the 50th, 75th, 90th percentile
of the continuous variable for performance pressure resulting from our topic analysis (not shown
here). The distribution for the relationship between performance pressure on the dollar amount
of penalties shown in Figure F1 of Online Appendix F shows that our results are only confirmed
at the 95th percentile.
DISCUSSION
In this study, we reconceptualized Merton’s strain theory at the organizational level and
theorized that organizations with cultures characterized by strong performance pressure are more
prone to engage in misconduct. Furthermore, in line with prior research on the contingency
effects of organizational structure on misconduct (Hill et al., 1992; McKendall and Wagner,
1997), we argued that formalization and decentralization, two key structural features of
organizations, can dampen or amplify this effect by reducing the effectiveness of control
systems. Unlike other studies in which researchers measured culture through surveys (e.g.,
McKendall and Wagner 1997; Bloom, Sadun, and Van Reenen 2010), we used machine learning
and NLP tools to measure employees’ perceptions of the culture (i.e., high vs. low performance
pressure) and structure (formalization and decentralization) of each organization in our sample.
Results of our empirical analysis generally support our hypotheses. Companies with high
performance pressure are 68 percent more likely to receive a penalty in the subsequent year than
37
companies with low performance pressure. We also found that performance pressure has a
positive effect on both the number and dollar amount of penalties incurred by a firm in the
subsequent year. On average, companies with high performance pressure paid $2.8 million more
per year in penalties for misconduct. The results on the relationship between formalization and
performance pressure provide limited support for our hypotheses. On the one hand, our results
confirm that when performance pressure is high, organizations with low formalization are more
likely to engage in misconduct and incur more penalties. On the other hand, results for the
monetary value of penalties were not found to be significant.
The results on decentralization, in contrast, completely support our hypotheses. Consistent
with the idea that the delegation of responsibilities in decentralized organizations leads to a lack
of control compared to centralized organizations, we found that decentralized organizations
experience harsher consequences when performance pressure is high, with an increased
likelihood to commit misconduct, as well as a greater number and average amount of penalties.
Taken together, our results show that a decentralized organizational structure can create
conditions that enable the development of a toxic culture, with negative consequences for the
organization. This result extends McKendall and Warner’s (1997) findings that an ethical climate
negatively mediates the effects of decentralization on environmental misconduct by
demonstrating that performance-oriented cultures could create conditions that increase the
likelihood for misconduct.
38
Limitations
Given the nature of the data we analyzed, this study has some limitations that present
opportunities for future research. Our measures of organizational culture and structure were
based on employee reviews. Although our method enabled us to measure performance pressure,
formalization and decentralization, using a corpus of data consisting entirely of employee
reviews might have biased our results in ways we could not fully control. For instance,
performance pressure is obviously much more salient than an organization’s structural features
when writing a review as an employee. Thus, our approach might have led to oversampling the
former and under sampling the latter. Nevertheless, we would expect these effects to hold across
types of companies and industries. In the case of organizational decentralization, we confirmed
our measure in a random sample of companies via a media search of newspaper articles
describing the companies’ organizational structures, and were satisfied by the results (not shown
here). The wider range of terms and sentiments associated with formalization, however, make it
difficult to confirm these data in a similar way. In future studies, scholars could validate these
measures through, for example, surveys or interviews. Despite these limitations, our method
could help researchers revive empirical research on organizational structure, a key concept for
organization theorists that until now, has been very difficult to study empirically. For instance,
drawing on Adler and Borys’s (1996) conceptualization of formalization as coercive or enabling,
39
researchers could explore the difference between these two concepts empirically in large sample
of organizations, thereby advancing our understanding of their antecedents and consequences.
Another limitation of our paper stems from two types of selection bias inherent in our data.
First, like other studies that build on voluntary employee reviews (e.g., Corritore et al., 2019) our
results might have been affected by the different propensities of employees from different firms
to contribute reviews. To address this limitation, researchers could apply the same NLP methods
to executive discourse (i.e., quarterly earnings calls) to identify references to organizational
decision-making processes and control mechanisms. Second, because we studied official legal
penalties, we did not observe any misconduct cases that were not discovered by official
agencies. This is a common limitation in the organizational misconduct research (Vaughan,
1999). Yet, given the scope of our data (which include many different types of violations) and
our methods, we are confident that our study offers what we believe to be the first empirical
evidence of links between performance pressure, structure, and misconduct over a large sample
of organizations.
Contributions
Our study makes several contributions to the literature on antecedents of organizational
misconduct (e.g., Greve, Palmer, and Pozner 2010; Agnew 1992; Vaughan 1999b). First, from a
theoretical point of view, we contribute to the operationalization of strain theory at the
40
organizational level, and suggest that the strain generated by aggressive performance cultures
can be dampened or amplified by structural features of the organization. Second, from an
empirical point of view, whereas it is not novel to point to culture as an antecedent of
misconduct (Vaughan, 1999), to our knowledge we are the first to empirically test this
relationship systematically with a large sample of organizations. Similar to Corritore et al.’s
(2020) approach, we collected data on organizational culture—specifically, performance
pressure—by analyzing employee discourse. This approach enabled us to measure culture in a
large sample of organizations and control for a wide range of potential confounders. Finally,
whereas in prior studies researchers focused on strain or performance expectations for top
managers (e.g. Harris and Bromiley 2007; Mishina et al. 2010), we studied performance pressure
as experienced by employees.
This study also contributes to research in organizational psychology on the dark side of goal
setting (Ordóñez et al., 2009a; Ordóñez and Welsh, 2015) and on the relationship between goal
setting and organizational culture (Ordóñez et al., 2009b). Beyond the existing evidence on the
dark side of goal setting (Ordóñez et al., 2009a; Ordóñez and Welsh, 2015), our results suggest
that aggressive, ambitious targets can translate into durable characteristics of organizational
culture. Although goal setting strategies can be easily adjustable, organizational culture as a
socially constructed reality (Berger and Luckmann 1967) is resistant to change. Cultural change
can be a slow-moving process (Meyerson and Martin, 1987), and toxic norms may persist many
41
years after incentives have changed, as demonstrated lately by the process of culture change at
Wells Fargo after the cross-selling scandal (Flitter and Cowley, 2019).
Finally, we also hope our study contributes to a revival of research on culture and
organizational structure from a methodological perspective. By studying formalization and
decentralization through unobtrusive observation and measurement of employee discourse (via
NLP and LDA), we suggest a novel approach (and methods) to identify structural properties of
organizations for larger samples of companies. Our methodological approach could also be
fruitfully employed by practitioners to predict the risk of organizational misconduct, either from
the inside (e.g., compliance officers, internal auditors) or from the outside (e.g., auditors,
investors, members of the media, regulators, law enforcement officers).
42
References
Adler, P S, and B. Borys
1996 “Two types of bureaucracy: Enabling and coercive.” Administrative Science Quarterly, 41: 61–89.
Adler, Paul S.
2012 “The Sociological Ambivalence of Bureaucracy: From Weber via Gouldner to Marx.” Organization Science, 23: 244–266.
Agnew, R.
1992 “Foundation for a General Strain Theory of Crime and Delinquency.” Criminology, 30: 47–88.
2001 “Building on the Foundation of General Strain Theory: Specifying the Types of Strain Most Likely to Lead to Crime and Delinquency.” Journal of Research in Crime and Delinquency, 38: 319–361.
Agnew, R., N. L. Piquero, and F. T. Cullen
2009 “General strain theory and white-collar crime.” The Criminology of White-Collar Crime: 35–60.
American Institute of Certified Public Accountants (AICPA)
2002 “New Statements on Auditing Standards.” Consideration OfFraud in a Financial Statement Audit, 1719–1770.
Ashforth, B. E., and V. Anand
2003 “THE NORMALIZATION OF CORRUPTION IN ORGANIZATIONS.” Research in Organizational Behavior. doi:10.1016/S0191-3085(03)25001-2
Audia, P., and H. Greve
2006 “Less likely to fail: Low performance, firm size, and factory expansion in the shipbuilding industry.” Management Science, 52: 83–94.
Bartov, E., D. Givoly, and C. Hayn
2002 “The Rewards to Meeting or Beating Earnings Expectations.” Journal of Accounting and Economics, 33: 173–204. Elsevier.
43
Becker, G. S.
1968 “Crime and Punishment: an Economic Approach.” The Economic Dimensions of Crime: 13–68. Palgrave Macmillan UK.
Bennett, V. M., L. Pierce, J. A. Snyder, and M. W. Toffel
2013 “Customer-driven misconduct: How competition corrupts business practices.” Management Science, 59: 1725–1742.
Bergstresser, D., and T. Philippon
2006 “CEO incentives and earnings management.” Journal of Financial Economics, 80: 511–529.
Beunza, D.
2019 Taking the Floor: Models, Morals, and Management in a Wall Street Trading Room. Princeton, NJ: Princeton University Press.
Bianchi, E. C., and A. Mohliver
2016 “Do good times breed cheats? Prosperous times have immediate and lasting implications for CEO misconduct.” Organization Science, 27: 1488–1503. INFORMS Inst.for Operations Res.and the Management Sciences.
Blau, P.
1955 “Consultation Among Colleagues.” Dynamics of Bureaucracy: 99–116.
Bloom, N., and J. Van Reenen
2007 “Measuring and explaining management practices across firms and countries.” Quarterly Journal of Economics, 122: 1351–1408.
Bloom, N., R. Sadun, and J. Van Reenen
2010 “Does product market competition lead firms to decentralize?” American Economic Review, 100: 434–438.
Bollinger, G., D. A. Belsley, E. Kuh, and R. E. Welsch
1981 “Regression Diagnostics: Identifying Influential Data and Sources of Collinearity.” Journal of Marketing Research, 18: 392. New York : Wiley.
Bretton Putter
44
2019 “What Boeing Has Taught Us About Not Neglecting Company Culture.” Forbes. Retrieved from https://www.forbes.com/sites/brettonputter/2019/12/18/what-boeing-has-taught-us-about-not-neglecting-company-culture-culturegene/?sh=43faf741496a
Brief, A., R. Buttram, and J. Dukerich
2001 “Collective Corruption in the Corporate World: Toward a Process Model.” Groups at Work: 485–514. Psychology Press.
Campbell, J. C., A. Hindle, and E. Stroulia
2015 “Latent Dirichlet Allocation: Extracting Topics from Software Engineering Data.” The Art and Science of Analyzing Software Data, 139–159.
Cao, R., R. Koning, and R. Nanda
2020 “Biased Sampling of Early Users and the Direction of Startup Innovation.” SSRN Electronic Journal. Elsevier BV. doi:10.2139/ssrn.3736119
Chatman, J. A., and C. A. O’Reilly
2016 “Paradigm lost: Reinvigorating the study of organizational culture.” Research in Organizational Behavior, 36: 199–224. Elsevier Ltd.
Clement, J., and P. Puranam
2018 “Searching for structure: Formal organization design as a guide to network evolution.” Management Science, 64: 3879–3895. INFORMS Inst.for Operations Res.and the Management Sciences.
Cooper, D. J., T. Dacin, and D. Palmer
2013 “Fraud in accounting, organizations and society: Extending the boundaries of research.” Accounting, Organizations and Society, 38: 440–457.
Corritore, M., A. Goldberg, and S. B. Srivastava
2020 “Duality in Diversity: How Intrapersonal and Interpersonal Cultural Heterogeneity Relate to Firm Performance.” Administrative Science Quarterly, 65: 359–394.
Cressey, D. R.
1950 “The Criminal Violation of Financial Trust.” American Sociological Review. doi:10.2307/2086606
1953 Other People’s Money: A Study in the Social Psychology of Embezzlement. New York: Free Press.
45
Cyert, R. M., and J. C. March
1963 A Behavioral Theory of the Firm. Englewood Cliffs Vol. 1. Englewood Cliffs: Prentice Hall.
DeBacker, J., B. T. Heim, and A. Tran
2015 “Importing corruption culture from overseas: Evidence from corporate tax evasion in the United States.” Journal of Financial Economics, 117: 122–138. Elsevier.
Demerouti, E., A. B. Bakker, F. Nachreiner, and W. B. Schaufeli
2001 “The job demands-resources model of burnout.” Journal of Applied Psychology, 86: 499–512.
Dichev, I. D., J. R. Graham, C. R. Harvey, and S. Rajgopal
2013 “Earnings quality: Evidence from the field.” Journal of Accounting and Economics, 56: 1–33.
Donegan, J. J., and M. W. Ganon
2008 “Strain, Differential Association, and Coercion: Insights from the Criminology Literature on Causes of Accountant’s Misconduct.” Accounting and the Public Interest, 8: 1–20.
Duflo, E., R. Hanna, and S. P. Ryan
2012 “Incentives work: Getting teachers to come to school.” American Economic Review.
Enrich, D.
2020 Dark Towers: Deutsche Bank, Donald Trump, and an Epic Trail of Destruction.
Fisman, R., and E. Miguel
2007 “Corruption, norms, and legal enforcement: evidence from diplomatic parking tickets.” Journal of Political Economy, 115: 1020–1048.
Flitter, E., and S. Cowley
2019 “Wells Fargo says its culture has changed. Some employees disagree.” The New York Times.
Fortado, L.
2012 “Adoboli says bonus was lower due to hidden profits.” Bloomberg. Retrieved from https://www.bloomberg.com/news/articles/2012-10-29/adoboli-says-bonus-was-lower-due-to-
46
profits-hidden-in-umbrella
Fox, M. L.
2016 “Effects of Stressful Job Demands and Control on Physiological and Attitudinal Outcomes in a Hospital Setting Author ( s ): Marilyn L . Fox , Deborah J . Dwyer and Daniel C . Ganster Published by : Academy of Management Stable URL : http://www.jstor.org/st,” 36: 289–318.
Gabbioneta, C., R. Greenwood, P. Mazzola, and M. Minoja
2013 “The influence of the institutional context on corporate illegality.” Accounting, Organizations and Society, 38: 484–504.
Du Gay, P.
2005 “The values of bureaucracy.” The Values of Bureaucracy: 1–13.
Giorgi, S., and K. Weber
2015 “Marks of Distinction: Framing and Audience Appreciation in the Context of Investment Advice.” Administrative Science Quarterly, 60: 333–367.
Gittens, A., D. Achlioptas, and M. W. Mahoney
2017 “Skip-Gram - Zipf + Uniform = Vector additivity.” ACL 2017 - 55th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference (Long Papers), 1: 69–76.
Golder, S. A., and M. W. Macy
2014 “Digital footprints: Opportunities and challenges for online social research.” Annual Review of Sociology. Annual Reviews Inc.
Gouldner, A. W.
1954 Patterns of Industrial Bureaucracy. Free Press Vol. 15. Retrieved from https://psycnet.apa.org/record/1955-01682-000
Greve, H. R., D. Palmer, and J. E. Pozner
2010 “Organizations gone wild: The causes, processes, and consequences of organizational misconduct.” Academy of Management Annals, 4: 53–107.
Guadalupe, M., and J. Wulf
2010 “The flattening firm and product market competition: The effect of trade liberalization on corporate hierarchies.” American Economic Journal: Applied Economics, 2: 105–127.
47
Hage, J., and M. Aiken
1967 “Relationship of Centralization to Other Structural Properties.” Administrative Science Quarterly, 12: 72.
Harris, J., and P. Bromiley
2007 “Incentives to cheat: The influence of executive compensation and firm performance on financial misrepresentation.” Organization Science, 18: 350–367.Hill, C. W. L., P. C. Kelley, B. R. Agle, M. A. Hitt, and R. E. Hoskisson
1992 “An Empirical Examination of the Causes of Corporate Wrongdoing in the United States.” Human Relations, 45: 1055–1076. Sage PublicationsSage CA: Thousand Oaks, CA.
Independent Directors of the Board of Wells Fargo
2017 “Independent Directors of the Board of Wells Fargo & Company: Sales practices investigation report.” Sales Practice Investigation Report, 10: 113.
Jamieson, K.
1994 The Organization of Corporate Crime: Dynamics of Antitrust Violation. (I. SAGE Publications, ed.). Retrieved from https://www.ncjrs.gov/App/abstractdb/AbstractDBDetails.aspx?id=149752
Johnson, S. A., H. E. Ryan, and Y. S. Tian
2009 “Managerial incentives and corporate fraud: The sources of incentives matter.” Review of Finance, 13: 115–145.
Kaplan, S., and K. Vakili
2015 “The double-edged sword of recombination in breakthrough innovation.” Strategic Management Journal, 36: 1435–1457. John Wiley and Sons Ltd.
Karasek, R. A. J.
1979 “Job Demands , Job Decision Latitude , and Mental Strain : Implications for Job Redesign Author ( s ): Robert A . Karasek , Jr . Published by : Sage Publications , Inc . on behalf of the Johnson Graduate School of Management , Cornell University Stable URL.” Administrative Science Quarterly, 24: 285–308.
Klappholz, K., and K. E. Boulding
1960 “Principles of Economic Policy.” Economica, 27: 275.
48
Koch-Bayram, I. F., and G. Wernicke
2018 “Drilled to obey? Ex-military CEOs and financial misconduct.” Strategic Management Journal, 39: 2943–2964. John Wiley and Sons Ltd.
Kuliks, B. W.
2005 “Agency theory, reasoning and culture at enron: In search of a solution.” Journal of Business Ethics, 59: 347–360.
Langton, L., and N. L. Piquero
2007 “Can general strain theory explain white-collar crime? A preliminary investigation of the relationship between strain and select white-collar offenses.” Journal of Criminal Justice, 35: 1–15.
Lapin, H.
2010 Economy and Society. Proceedings of the British Academy Vol. 165. Berkeley: University of California Press. doi:10.4324/9781351245463-7
Larrick, R. P., C. Heath, and G. Wu
2009 “Goal-induced risktaking in negotiation and decision making.” Social Cognition, 27: 342–364.
Light, D. W., P. L. Berger, and T. Luckmann
1967 “The Social Construction of Reality: A Treatise in the Sociology of Knowledge.” Sociological Analysis, 28: 55. Garden City, N.Y., N.Y.: Doubleday.
Liu, X.
2016 “Corruption culture and corporate misconduct.” Journal of Financial Economics, 122: 307–327. Elsevier B.V.
Locke, E. A., and G. P. Latham
1991 “A Theory of Goal Setting and Task Performance.” The Academy of Management Review, 16: 480.
2002 “Building a Practically Useful Theory of Goal Setting and Task Motivation A 35-Year Odyssey.” Psycnet.Apa.Org, 57: 705–717. American Psychological Association Inc.
2009 “Has Goal Setting Gone Wild, or Have Its Attackers Abandoned Good Scholarship?” Academy of Management Perspectives, 23: 17–23. Academy of Management.
2013 New Developments in Goal Setting and Task Performance. New Developments in Goal
49
Setting and Task Performance. doi:10.4324/9780203082744
Lokanan, M. E.
2015 “Challenges to the fraud triangle: Questions on its usefulness.” Accounting Forum, 39: 201–224. Elsevier Ltd.
Luo, N., Y. Zhou, and J. J. Shon
2016 “Employee satisfaction and corporate performance: Mining employee reviews on glassdoor.com.” 2016 International Conference on Information Systems, ICIS 2016.
March, J. G., and H. A. Simon
1958 Organizations. New York, NY: Wiley.
Marchetti, A.
2019 “Firms of a Feather Merge Together: The Coordination Benefits of Compatible Cultures.” SSRN Electronic Journal. doi:10.2139/ssrn.3712838
McEvily, B., G. Soda, and M. Tortoriello
2014 “More Formally: Rediscovering the Missing Link between Formal Organization and Informal Social Structure.” Academy of Management Annals, 8: 299–345.
McKendall, M. A., and J. A. Wagner
1997 “Motive, Opportunity, Choice, and Corporate Illegality.” Organization Science, 8: 624–647.
Merton, R. K.
1936 “The Unanticipated Consequences of Purposive Social Action.” American Sociological Review, 1: 894.
1938 “Social Structure and Anomie.” American Sociological Review, 3: 672–682.
1940 “Bureaucratic structure and personality.” Social Forces, 18: 560–568.
Meyerson, D., and J. Martin
1987 “Cultural Change: An Integration of three Different Views.” Journal of Management Studies, 24: 623–647.
Mikolov, T., K. Chen, G. Corrado, and J. Dean
2013 “Efficient estimation of word representations in vector space.” 1st International
50
Conference on Learning Representations, ICLR 2013 - Workshop Track Proceedings.
Milgrom, P., and J. Roberts
1988 “An Economic Approach to Influence Activities in Organizations.” American Journal of Sociology, 94: S154–S179. University of Chicago Press.
Mishina, Y., B. Dykes, E. Block, and T. Pollock
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: 701–722.
Mohliver, A.
2019 “How Misconduct Spreads: Auditors’ Role in the Diffusion of Stock-option Backdating.” Administrative Science Quarterly, 64: 310–336. SAGE Publications Ltd.
Moore, C.
2008 “Moral Disengagement in Processes of Organizational Corruption.” Springer, 80: 129–139.
Moore, C., J. R. Detert, L. Klebe Treviño, V. L. Baker, and D. M. Mayer
2012 “Why Employees Do Bad Things: Moral Disengagement and Unethical Organizational Behavior.” Personnel Psychology, 65: 1–48.
Moore, C., D. M. Mayer, F. F. T. T. Chiang, C. Crossley, M. J. Karlesky, and T. A. Birtch
2019 “Leaders matter morally: The role of ethical leadership in shaping employee moral cognition and misconduct.” Journal of Applied Psychology, 104: 123–145. American Psychological Association Inc.
Morales, J., Y. Gendron, and H. Guénin-Paracini
2014 “The construction of the risky individual and vigilant organization: A genealogy of the fraud triangle.” Accounting, Organizations and Society, 39: 170–194. Elsevier Ltd.
Nagin, D. S., J. B. Rebitzer, S. Sanders, and L. J. Taylor
2002 “Monitoring, motivation, and management: The determinants of opportunistic behavior in a field experiment.” American Economic Review, 92: 850–873.
Newman, D., J. H. Lau, K. Grieser, and T. Baldwin
2010 “Automatic evaluation of topic coherence.” NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association
51
for Computational Linguistics, Proceedings of the Main Conference, 100–108.
Nguyen, D. D., L. Nguyen, and V. Sila
2019 “Does Corporate Culture Affect Bank Risk-Taking? Evidence from Loan-Level Data.” British Journal of Management, 30: 106–133.
den Nieuwenboer, N. A., J. V. da Cunha, and L. K. Treviño
2017 “Middle managers and corruptive routine translation: The social production of deceptive performance.” Organization Science, 28: 781–803. INFORMS Inst.for Operations Res.and the Management Sciences.
O’Connor, J. P., R. L. Priem, J. E. Coombs, and K. M. Gilley
2006 “Do CEO stock options prevent or promote fraudulent financial reporting?” Academy of Management Journal, 49: 483–500. Academy of Management.
O’Reilly, C. A., and J. A. Chatman
1996 “Culture as social control: Corporations, cults, and commitment.” Research in Organizational Behavior, 18: 157–200.
O’Reilly, C. A., J. Chatman, and D. F. Caldwell
1991 “People and Organizational Culture: a Profile Comparison Approach To Assessing Person-Organization Fit.” Academy of Management Journal, 34: 487–516.
Ordóñez, L. D., M. E. Schweitzer, A. D. Galinsky, and M. H. Bazerman
2009a “Goals gone wild: The systematic side effects of overprescribing goal setting.” Academy of Management Perspectives, 23: 6–16.
2009b “On Good Scholarship, Goal Setting, and Scholars Gone Wild.” Academy of Management Perspectives, 23: 82–87.
Ordóñez, L. D., and D. T. Welsh
2015 “Immoral goals: how goal setting may lead to unethical behavior.” Current Opinion in Psychology, 6: 93–96. Elsevier.
Palmer, D.
2008 “Extending the process model of collective corruption.” Research in Organizational Behavior.
Palmer, D., K. Smith-Crowe, and R. Greenwood
52
2016 “The imbalances and limitations of theory and research on organizational wrongdoing.” Organizational Wrongdoing: Key Perspectives and New Directions: 1–16. Cambridge University Press.
Perrow, C.
2011 Normal Accidents: Living with High Risk Technologies. Normal Accidents: Living with High Risk Technologies. doi:10.5465/amr.1985.4278477
Pierce, L., D. C. Snow, and A. McAfee
2015 “Cleaning house: The impact of information technology monitoring on employee theft and productivity.” Management Science, 61: 2299–2319. INFORMS Inst.for Operations Res.and the Management Sciences.
Power, M.
2013 “The apparatus of fraud risk.” Accounting, Organizations and Society, 38: 525–543. Elsevier Ltd.
Pugh, D. S., D. J. Hickson, C. R. Hinings, and C. Turner
1969 “The Context of Organization Structures.” Administrative Science Quarterly, 14: 91.
Rajan, R. G., and J. Wulf
2006 “The flattening firm: Evidence from panel data on the changing nature of corporate hierarchies.” Review of Economics and Statistics, 88: 759–773.
Rasche, A., and A. Scherer
2014 “Jurgen Habermas and organization studies: Contributions and future prospects.” In Paul S. Adler, P. du Gay, G. Morgan, and M. Reed (eds.), Oxford Handbook of Sociology, Social Theory and Organization Studies: Contemporary Currents: 158–181. Oxford, UK: Oxford University Press.
Sandhu, S., and C. T. Kulik
2019 “Shaping and Being Shaped: How Organizational Structure and Managerial Discretion Co-evolve in New Managerial Roles.” Administrative Science Quarterly, 64: 619–658.
Schein, E. H.
1996 “Culture : The Missing Concept in Organization Studies Author ( s ): Edgar H . Schein Source : Administrative Science Quarterly , Jun ., 1996 , Vol . 41 , No . 2 , 40th Anniversary Published by : Sage Publications , Inc . on behalf of the Johnson Graduate ,” 41: 229–240.
Schnatterly, K., K. A. Gangloff, and A. Tuschke
53
2018 “CEO Wrongdoing: A Review of Pressure, Opportunity, and Rationalization.” Journal of Management, 44: 2405–2432. SAGE Publications Inc.
Schweitzer, M. E., L. D. Ordóñez, and B. Douma
2004 “Goal setting as a motivator of unethical behavior.” Academy of Management Journal, 47: 422–432.
Simpson, S. S., and C. S. Koper
1997 The Changing of the Guard: Top Management Characteristics, Organizational Strain, and Antitrust Offending. Journal of Quantitative Criminology Vol. 13.
Simpson, S. S., and D. Weisburd
2009 The Criminology of White-Collar Crime. The Criminology of White-Collar Crime. Springer. doi:10.1007/978-0-387-09502-8
Sims, R. R., and J. Brinkmann
2003 “Enron Ethics (Or: Culture Matters More than Codes).” Journal of Business Ethics, 45: 243–256.
Skousen, C. J., K. R. Smith, and C. J. Wright
2009 Detecting and Predicting Financial Statement Fraud: The Effectiveness of The Fraud Triangle and SAS No. 99 in Corporate Governance and Firm Performance. International Journal of Quality & Reliability Management Vol. 32.
Soderstrom, S. B., and K. Weber
2020 “Organizational Structure from Interaction: Evidence from Corporate Sustainability Efforts.” Administrative Science Quarterly, 65: 226–271.
Staw, B. M., and E. Szwajkowski
1975 “The Scarcity-Munificence Component of Organizational Environments and The Commission of Illegal Acts.” Administrative Science Quarterly, 20: 345–354.
Stern, R. N., and S. R. Barley
1996 “Organizations and social systems: Organization theory’s neglected mandate.” Administrative Science Quarterly, 41: 146–162.
Sutherland, E.
1939 “The white collar criminal.” Presidential Address Delivered to the AmericanSociological and Economic Societies. Philadelphia, PA.
54
1949 White Collar Crime. (R. and W. New York : Holt, ed.). New York: Holt Rinehart and Winston.
Szwajkowski, E.
1985 “Organizational Illegality: Theoretical Integration and Illustrative Application.” Academy of Management Review, 10: 558–567.
Treviño, L. K., N. A. Den Nieuwenboer, and J. J. Kish-Gephart
2014 “Unethical behavior in organizations.” Annual Review of Psychology.
Tversky, A., and D. Kahneman
1991 “Loss Aversion in Riskless Choice: A Reference-Dependent Model.” The Quarterly Journal of Economics, 106: 1039–1061.
Vaughan, D.
1996 The Challenger Launch Decision. Risky Technology, Culture and Deviance at NASA. University of Chicago Press. doi:10.1177/1350507698292011
1999 “The dark side of organizations: Mistake, misconduct, and disaster.” Annual Review of Sociology, 25: 271–305.
Venkatraman, N.
1989 “The Concept of Fit in Strategy Research: Toward Verbal and Statistical Correspondence.” Academy of Management Review, 14: 423–444.
Wakeman, S. W., C. Moore, and F. Gino
2019 “A counterfeit competence: After threat, cheating boosts one’s self-image.” Journal of Experimental Social Psychology, 82: 253–265. Academic Press Inc.
Wang, Y., T. Stuart, and J. Li
2020 “Fraud and Innovation.” Administrative Science Quarterly. SAGE Publications Ltd. doi:10.1177/0001839220927350
Wheeler, D., and A. Mody
1992 “International investment location decisions. The case of U.S. firms.” Journal of International Economics, 33: 57–76.
White, H.
1980 A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for
55
Heteroskedasticity. JSTOR Vol. 48.
Wiley Wakeman, S., and C. Moore
2018 “Rewarding deviants: Unethical behavior as a signal of one’s communal value.” 78th Annual Meeting of the Academy of Management, AOM 2018. Academy of Management. doi:10.5465/AMBPP.2018.240
Wilmot, W. W., and J. Hocker
2001 “Power: The structure of conflict.” McGraw Hill. McGraw Hill.
Yenkey, C. B.
2018 “Fraud and Market Participation: Social Relations as a Moderator of Organizational Misconduct.” Administrative Science Quarterly, 63: 43–84.
Yip, J. A., E. E. Levine, A. W. Brooks, and M. E. Schweitzer
2021 “Worry at work: How organizational culture promotes anxiety.” Research in Organizational Behavior, 100124. Elsevier Ltd.
Zhang, X., K. M. Bartol, K. G. Smith, M. D. Pfarrer, and D. M. Khanin
2008 “Ceos on the edge: Earnings manipulation and stock-based incentive misalignment.” Academy of Management Journal, 51: 241–258. Academy of Management.
56
FIGURE 1
Example of a Glassdoor Review
57
TABLE 1
LDA Analysis of Glassdoor Reviews – Main Topics
The following table shows the keywords, and labels, for the topic modelling analysis on Glassdoor reviews. To generate the topics, we started from our complete sample of Glassdoor reviews (2.6 millions) and selected a random 20% of the reviews (CONS section) for the LDA analysis. Please see Appendix A for more details on the method.
Topic Topic label Top topic words1 Performance pressure sale, goal, number, expectation, push, pressure, meet, set, commission2 Senior management manager, store, department, supervisor, depend, manage, associate3 Corporate strategy change, make_things_right, decision, policy, direction4 Type of position job, worker, full_time, position5 Leadership culture, leadership, corporate, communication6 Teamwork team, lead, leader, performance, individual, role7 Work hours work, day, hour, schedule, shift, family, holiday8 Details of the position year, month, first, end
9 Respectful treatment of employees employee, treat, respect, value, good, better, fair
10 Job opportunities opportunity, growth, grow, advancement, career, promotion, advance, limited
11 Work hours hour, long, pay, time, horrible, terrible12 Pay and benefits pay, benefit, salary, raise, increase, bonus, compensation13 Open communications communication, open, idea, meeting, bring, information14 Technology system, issue, technology, old, back, side, tech, software15 N.D. sell, follow, break, rule16 Listen to employees employee, listen, talk, feedback, concern, review, hear, speak17 Career opportunities hire, care, people, promote, experience, talent, position, attention18 Training training, staff, need, train, learn, support, provide, improve, better,19 Customer relationship customer, time, help, deal, call, ask, rude, task20 Stressful work conditions environment, anything, stressful, negative, create, stress
21 Product/service quality focus, product, quality, market, service, customer, long_term, build, profit
22 Employee turnover people, stop, leave, let, start, fire, stay, look, quit, rid23 Company characteristics company, big, large, lose, layoff, location, due, industry, future, go24 Work-life balance work, life_balance, place, none, life, appreciate, time, home, load25 Management management, level, upper, senior, middle, advice, need, top, executive26 Work processes process, project, lot, organization, move, bit, engineer, internal, time
58
TABLE 2
Target vectors’ most similar words (FastText)
Target: formalization Similarity1 Target: decentralization Similarity2
bureaucratic 0.8804816 decentralized 0.8884735bureaucracy 0.80162853 decentralised 0.7864013bureacratic 0.78376913 de-centralized 0.6914408beaurocratic 0.77073193 decentralization 0.6866374redtape 0.73887277 Decentralized 0.6521732bureacracy 0.73699331 decentral 0.6317272beauracracy 0.72877342 non-centralized 0.6103989beurocratic 0.71758479 decentralizing 0.6012831red-tapism 0.71712649 decentralize 0.5883033beauracratic 0.71091759 de-centralised 0.5882939beauracracy 0.67835552 decentralizes 0.5851191beurocracy 0.6729213 Decentralization 0.5676084burocratic 0.66317499 decentralisation 0.5537003burocracy 0.64025646 Decentralizing 0.5480962Bureaucratic 0.64011562 nonhierarchical 0.5362982bureaucracies 0.63829845 Decentralised 0.5333222hoop-jumping 0.60826135 self-organizing 0.5269514bureacracies 0.60292083 non-hierarchical 0.5261458tapism 0.59983265 self-organized 0.5193331paper-pushing 0.58549082 decentralism 0.5161116bureaucratically 0.58070523 censorship-resistant 0.5156702Bureaucracy 0.56742257 self-governed 0.5148664bureaucratization 0.56587136 self-organised 0.511211bureaucratism 0.56382871 bottom-up 0.5067362paper-work 0.55870563 participatory 0.5027277inefficiency 0.55810237 Decentral 0.502153over-regulation 0.55697292 decentralise 0.4997308paper-shuffling 0.55400884 decentralising 0.499221bureaucratisation 0.55029672 centrally-controlled 0.4977726bureaucratized 0.54976827 peer-production 0.4971926overregulation 0.54810959 decentralist 0.4971693formalization 0.54752803 citizen-centered 0.4952511rigmarole 0.54262245 self-organising 0.4930755bureaucrats 0.54181415 participative 0.4858297micro-management 0.53989881 micro-grids 0.4832599bureaucratese 0.5398289 community-driven 0.4828814bureacrats 0.53723311 non-authoritarian 0.4798779
1 Refers to cosine similarity with our target vector. 2 Refers to cosine similarity with our target vector.
59
TABLE 3
Descriptive Statistics and Correlation Table
60
Variables Mean s.d. 1 2 3 4 5 6 7 8 9 10 11 12 13 14
1.Organizational Misconduct: Penalty (1/0) 0.285871 0.451859 -
2.Organizational Misconduct: No of Penalties 0.965328 2.29512 0.631
8 -
3.Organizational Misconduct: Penalty ($) 3.043791 12.12639 0.365
2 0.4014 -
4.Performance pressure n-1 0.051604 0.221257 -0.006 -0.023 0.022
8 -
5.Decentralized n-1 0.547559 0.497802 -0.119 -0.089 0.054
6 -0.057 -
6.Formalized n-1 0.458577 0.498351 -0.031 -0.066 0.013
3 0.0255 0.1856 -
7.Number of reviews (log) n-1 5.585789 0.833584 0.221
6 0.2213 0.1622 0.0045 -0.197 -0.033 -
8.Number of employees (log) n-1 3.442483 1.4669 0.391 0.4258 0.217
6 -0.053 -0.141 -0.043 0.4451 -
9.ROA n-1 0.13468 0.098014 0.0681 0.0148 -0.064 -0.002 -0.125 -0.064 0.1013 0.211 -
10.Sales growth n-1 0.063432 0.187245 -0.06 -0.05 -0.034 0.0102 0.0112 -0.053 -0.058 -0.127 -0.078 -
11.Market Competitionn-1 0.102186 0.094971 0.162
7 0.24 0.0022 0.0581 -0.244 -0.046 0.1607 0.153 0.0627 -0.026 -
12.Sen. MGMT Approval n-1 3.026037 0.582549 0.044
4 0.0786 0.0633 -0.042 0.2649 -0.174 -0.014 0.084 -0.009 0.138
5 0.0154 -
13.Compensation & Benefitsn-1 3.298355 0.485537 0.006 0.0386 0.092
3 -0.045 0.3113 -0.048 0.0241 0.01 -0.114 0.0953 -0.238 0.463 -
14.CEO Approval n-1 3.844842 0.441321 -0.033 -0.052 0.014
9 -0.005 0.1343 -0.205 0.1373 -0.016 -0.008 0.2048 -0.089 0.568 0.609 -
15.Work-Life Balance n-1 3.262898 0.431379 -0.132 -0.161 0.025
9 -0.089 0.4134 -0.064 0.0001 -0.109 -0.151 0.0687 -0.221 0.589 0.605 0.595
TABLE 4A
Regressions on Penalties (Likelihood of Organizational Misconduct)
VariablePenalties
(1) (2) (3)Performance pressure n-1 0.619* 0.762** 0.783**
(0.343) (0.367) (0.375)Number of reviews (log)n-1 0.138 0.0286 -0.00488
(0.207) (0.213) (0.216)Number of employees (log)n-1 0.824** 1.106*** 1.121***
(0.337) (0.365) (0.378)ROA n-1 1.453 1.454
(2.211) (2.221)Sales growth n-1 -0.929 -0.961
(0.599) (0.628)Market competitionn-1 -1.782 -2.213
(4.348) (4.441)Senior management approval n-1 0.0192
(0.271)Compensation and benefits n-1 0.0734
(0.557)CEO approval n-1 -0.0590
(0.319)Work-life balance n-1 0.409
(0.592)
Observations 1,680 1,436 1,436Number of firms 289 254 254Year FE YES YES YESRobust standard errors in parentheses.*** p < 0.01, ** p < 0.05, * p < 0.1
61
TABLE 4B
Regressions on Number of Penalties
VariableNumber of penalties
(1) (2) (3)Performance pressure n-1 0.222* 0.257* 0.287**
(0.120) (0.131) (0.130)Number of reviews (log)n-1 -0.0195 -0.0252 -0.0511
(0.0748) (0.0794) (0.0798)Number of employees (log)n-1 0.550*** 0.570***
(0.0880) (0.0860)ROA n-1 0.856** 0.908**
(0.431) (0.457)Sales growth n-1 -0.167 -0.178
(0.160) (0.158)Market competitionn-1 0.162 0.152
(1.043) (1.035)Senior management approval n-1 -0.0258
(0.0473)Compensation and benefits n-1 -0.152
(0.163)CEO approval n-1 -0.183*
(0.100)Work-life balance n-1 0.221
(0.210)Constant -0.618 -1.136** -0.486
(0.533) (0.473) (0.753)
Observations 2,281 1,995 1,995Number of firms 393 359 359Year FE YES YES YESIndustry FE YES YES YESRobust standard errors in parentheses.*** p < 0.01, ** p < 0.05, * p < 0.1
62
TABLE 4C
Regressions on Amount of Penalties (in Dollars)
VariableAmount of penalties (in dollars)
(1) (2) (3) Performance pressure n-1 2.663* 2.882* 2.894*
(1.498) (1.574) (1.588)Number of reviews (log)n-1 1.430 1.606* 1.397
(0.883) (0.888) (0.881)Number of employees (log)n-1 1.754 2.255* 2.310**
(1.126) (1.170) (1.176)ROA n-1 -1.758 -0.530
(6.199) (6.124)Sales growth n-1 -1.555 -1.289
(2.351) (2.351)Market competitionn-1 -14.59 -14.83
(13.43) (13.22)Senior management approval n-1 -1.130
(1.277)Compensation and benefits n-1 5.108**
(2.074)CEO approval n-1 -2.163*
(1.240)Work-life balance n-1 0.811
(2.057)Constant -9.169 -10.42* -17.35**
(5.769) (6.002) (8.271)
Observations 3,094 2,986 2,985Number of firms 663 637 637 R-squared 0.464 0.424 0.426Year FE YES YES YESIndustry FE YES YES YESRobust standard errors in parentheses.*** p < 0.01, ** p < 0.05, * p < 0.1
63
TABLE 5A
Regressions on Penalties (Likelihood of Organizational Misconduct), by Type of
Organizational Structure
Variable
Penalties(1) (2) (3) (4)
Low formalization
High formalization
Low decentralization
High decentralization
Performance pressure n-1 1.645* 1.271* 0.406 1.816**
(0.857) (0.766) (0.365) (0.712)Number of reviews (log)n-1 -0.101 0.285 0.288 -0.318
(0.317) (0.241) (0.231) (0.360)Number of employees (log)n-1 0.565 1.640** -0.0835 2.147***
(0.733) (0.713) (0.624) (0.547)ROA n-1 6.627* 0.930 3.235 5.536
(3.631) (4.284) (2.920) (3.921)Sales growth n-1 -2.161* -1.856 0.368 -3.555***
(1.148) (1.271) (0.625) (1.106)Market competitionn-1 -2.788 -0.258 -10.09 3.365
(9.381) (8.346) (6.280) (9.373)Senior management approval n-1 1.509* 0.103 0.760 -0.522
(0.898) (0.300) (0.540) (0.465)Compensation and benefits n-1 -0.596 0.109 -0.483 -0.0672
(1.111) (0.849) (0.927) (0.818)CEO approval n-1 -0.263 -0.225 -0.553 -0.0340
(0.613) (0.527) (0.528) (0.597)Work-life balance n-1 -0.466 1.628 0.135 1.731*
(1.186) (0.994) (0.950) (0.955)
Observations 453 488 691 492Number of firms 149 143 152 99Year FE YES YES YES YESRobust standard errors in parentheses.*** p < 0.01, ** p < 0.05, * p < 0.1
64
TABLE 5B
Regressions on Number of Penalties, by Type of Organizational Structure
Variable
Number of Penalties(1) (2) (3) (4)
Low Formalization
High Formalization
Low Decentralization
High Decentralization
Performance pressure n-1 0.329** 0.256* 0.290 1.113***
(0.159) (0.147) (0.204) (0.135)Number of reviews (log)n-1 0.00698 -0.172 -0.0264 0.0399
(0.114) (0.132) (0.122) (0.117)Number of employees (log)n-1 0.421* 0.880*** 0.160 1.091***
(0.241) (0.299) (0.142) (0.239)ROA n-1 1.203** 0.745 0.255 3.093**
(0.566) (1.059) (0.524) (1.309)Sales growth n-1 -0.0155 -0.494** -0.0266 -0.287
(0.191) (0.228) (0.178) (0.233)Market competitionn-1 0.768 -0.885 -0.420 -1.040
(1.208) (2.246) (1.491) (1.706)Senior management approval n-1 0.0331 -0.103 -0.0911 -0.0739
(0.0513) (0.0895) (0.0716) (0.112)Compensation and benefits n-1 0.0419 -0.607* 0.151 -0.188
(0.211) (0.321) (0.227) (0.241)CEO approval n-1 -0.362** 0.0139 -0.209 -0.331*
(0.167) (0.177) (0.132) (0.189)Work-life balance n-1 0.349 0.308 0.0369 0.682**
(0.244) (0.372) (0.237) (0.294)Constant -0.783 -0.265 1.368 -4.214***
(1.497) (1.770) (1.032) (1.484)
Observations 966 823 919 900Number of firms 217 204 209 194Year FE YES YES YES YESIndustry FE YES YES YES YESRobust standard errors in parentheses.*** p < 0.01, ** p < 0.05, * p < 0.1
65
TABLE 5C
Regressions on Amount of Penalties (in Dollars), by Type of Organizational Structure
Variable
Amount of penalties (in dollars)(1) (2) (3) (4)
Low formalization
High formalization
Low decentralization
High decentralization
Performance pressure n-1 7.271 0.293 1.125 5.402**
(4.536) (1.533) (2.428) (2.685)Number of reviews (log)n-1 -0.329 1.685 1.667 1.954*
(1.210) (1.505) (1.650) (1.081)Number of employees (log)n-1 4.904*** -0.339 1.103 6.347***
(1.683) (2.013) (1.899) (2.299)ROA n-1 4.378 -7.975 -17.33* 13.30
(8.603) (10.08) (9.855) (9.552)Sales growth n-1 -2.791 3.323 1.804 -2.199
(2.729) (5.162) (3.830) (4.008)Market competitionn-1 -20.85 -24.50 0.175 -46.78
(20.00) (18.29) (16.29) (30.30)Senior management approval n-1 1.739 -5.615*** 1.508 -0.441
(1.538) (1.983) (2.075) (2.046)Compensation and benefits n-1 5.166 3.566 3.605 5.436*
(3.317) (3.518) (3.349) (2.844)CEO approval n-1 -1.305 -1.954 -0.788 -5.442***
(1.945) (1.768) (2.040) (2.041)Work-life balance n-1 -0.742 5.253 2.094 1.715
(2.888) (3.942) (2.639) (3.810)Constant -24.61** -4.701 -26.08* -26.23*
(11.43) (14.01) (13.76) (14.44)
Observations 1,494 1,242 1,278 1,507Number of firms 381 347 328 379R-squared 0.463 0.488 0.450 0.469Year FE YES YES YES YESIndustry FE YES YES YES YESRobust standard errors in parentheses.*** p < 0.01, ** p < 0.05, * p < 0.1
66
ONLINE APPENDIX
Appendix A: Glassdoor
Glassdoor is a job board service that can both help employees look for new job positions and,
especially, gather information on the organization’s culture, climate, and job conditions. To
access content on the website, users have to anonymously rate their employers in what is called a
give-to-get model3, a service through which the user can have free access to the service by
leaving a review on his/her actual, or former, jobs. This mechanism had been shown to reduce
polarization in evaluations (Marinescu et al., 2018).
Glassdoor reviews are divided into two sections: the quantitative section asks users to rate
their firms on a scale ranging from 1 to 5 on a series of organizational attributes (i.e. general
evaluation, culture and values, senior management, work-life balance, career opportunities, and
compensation and benefits). Glassdoor users can also optionally state opinions ranging from -1
to 1 about the CEO, the outlook (i.e. how do the short-term future of the organization will be),
and whether they would recommend the job to a friend.
The qualitative section is divided into two sentiment-coded categories: pros (i.e., positive
aspects of the company), and cons (i.e., negative aspects of the company). Furthermore,
3 See https://help.glassdoor.com/article/Using-Glassdoor/en_US/Glassdoor_Basics
67
Glassdoor reviews provide an additional text section in which employees can submit suggestions
on how to improve the company (advices to management).
References
Marinescu, I., N. Klein, A. Chamberlain, and M. Smart.
2018 “Incentives can reduce bias in online employer reviews”. Academy of Management
Proceedings Vol. 2018. Cambridge, MA. doi:10.5465/ambpp.2018.11189abstract
68
Appendix B: Detecting Organizational Culture Topics Using Latent Dirichlet Allocation
(LDA)
The Latent Dirichlet Allocation (LDA) algorithm starts from the assumption that a document
has its own topic distribution from which a topic is randomly selected. Moreover, each randomly
selected topic has its own word distribution, from which a word is randomly selected. Repeating
these two steps word by word generates a document. The LDA algorithm discovers the topic
distribution for each document and the word distribution of each topic iteratively by fitting this
two-step generative model to the observed words in all documents until it finds the best set of
variables to describe the topic and word distributions. Essentially, like cluster analysis or
principal component analysis, LDA reduces the extraordinary dimensionality of linguistic data
from words to topics based on word co-occurrences in the same document, and provides a
reliable and replicable classification of topics.
Like any other generative model, LDA adopts a training model to infer the words contained
in the topics and their distributions. For computational reasons, we created the training model
starting from a random sample comprising 20% of the full sample of reviews (cons section
only). Other researchers (Corritore, Goldberg, and Srivastava, 2020) created a training model
using all sentences containing the word “culture” or synonyms (i.e., atmosphere, attitude,
climate, value, philosophy, belief) to reduce the number of “non-culture” topics that might come
69
up in the reviews. Unlike these researchers who analyzed similarity and heterogeneity using all
topics from the training model, we focused on a single topic: performance pressure.
LDA is an unsupervised model. The only thing the researcher must choose is the number of
topics. To select the optimal number of topics, we followed the computational linguistics
literature (Newman et al., 2010) and calculated the coherence score of the LDA model computed
with different numbers of topics (i.e. from 2 to 200). Topic coherence is a standardized measure
that evaluates the cohesion or accuracy of a set of topics in a model using a co-occurrence
measure based on pointwise mutual information. Figure A1 shows the distribution of cohesion
scores based on the number of topics selected. We decided to use the topic model with the
highest coherence score, which was the 26 topics model.
As suggested by DiMaggio et al. (2013) even though tests like log-perplexity or coherence
scores help to find the best model that “fits” the data, there is no statistical test yet for the
semantically optimal number of topics. Another concern related to topic modeling regards the
semantic validation of the topics (Chang et al., 2009; DiMaggio, 2015) as models that might
have high coherence scores, for example, might be semantically confusing (Hannigan et al.,
2019). To address this concern we coded the topics focusing on the words that were both
frequent and exclusive for every topic, which, as demonstrated by Bischof and Airoldi (2012), is
a more rigorous way to characterize topical content. Figure A2 shows the words related to our
70
topic of interest (performance pressure) in terms of both frequency and exclusivity using
PyLDA, a Python package that permits us to visualize topics in a two-dimensional space. This
tool helps researchers identify semantic similarities between topics and macro categories inside
the topic model. In figure A2, we report the words that best characterize the topic (Figure A2).
71
FIGURE B1
Coherence Scores for the Topic Models
0 20 40 60 80 100 120 140 1600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Number of Topics
72
FIGURE B2
Performance Pressure Topic Words Filtered on Exclusivity
73
References
Bischof, J. M., and E. M. Airoldi
2012 “Summarizing topical content with word frequency and exclusivity.” Proceedings of the 29th International Conference on Machine Learning, ICML 2012, 1: 201–208.
Chang, J., J. Boyd-Graber, S. Gerrish, C. Wang, and D. M. Blei
2009 “Reading tea leaves: How humans interpret topic models.” Advances in Neural Information Processing Systems 22 - Proceedings of the 2009 Conference, 288–296.
Corritore, M., A. Goldberg, and S. B. Srivastava
2020 “Duality in Diversity: How Intrapersonal and Interpersonal Cultural Heterogeneity Relate to Firm Performance.” Administrative Science Quarterly, 65: 359–394.
DiMaggio, P.
2015 “Adapting computational text analysis to social science (and vice versa).” Big Data and Society, 2: 1–5.
DiMaggio, P., M. Nag, and D. Blei
2013 “Exploiting affinities between topic modeling and the sociological perspective on culture: Application to newspaper coverage of U.S. government arts funding.” Poetics, 41: 570–606. Elsevier B.V.
Newman, D., J. H. Lau, K. Grieser, and T. Baldwin
2010 “Automatic evaluation of topic coherence.” NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference, 100–108.
74
Appendix C: Word Embedding and FastText
Word embedding (Turian, Ratinov, and Bengio, 2010; Mikolov et al., 2013) is a set of modeling
techniques for mapping words or sentences to a vectorial space. Whereas other methods (e.g.,
LDA) use a similar operation to transform words inside a vectorial space, word embedding
methods maintain syntactic and semantic information from the text. Unlike LDA, however, word
embedding methods are supervised: to create a neural network with the ability to classify text
into vectorial spaces, word embedding needs to “learn” from training inputs based on a series of
ideal examples. Training a dictionary in machine learning is a long and computationally
intensive process that requires various iterations to find the best fit for the semantic model that
represents the data.
In this paper, we decided to skip the training process by using FastText to represent the text in
vectorial space. FastText (Joulin et al., 2017) is a library that increases the efficiency of prior
word representation models and sentence classifications developed by Facebook AI Research
and trained using the entire Wikipedia dataset. Joulin et al. (2017) calculated that FastText had
better performances regarding accuracy, scaling, and prediction compared to any other existing
text classifiers. In a nutshell, FastText is a pre-trained dictionary that offers researchers that
adopt machine learning operations such as text tagging, or sentiment analysis, to quickly set up a
baseline semantic model. Furthermore, by using a pre-trained dictionary we reduce the concerns
related to the replicability of the semantic model.
75
References
Joulin, A., E. Grave, P. Bojanowski, and T. Mikolov.
2017 “Bag of tricks for efficient text classification.” 15th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2017 - Proceedings of Conference, 2: 427–431.
Turian, J., L. Ratinov, and Y. Bengio.
2010 Word representations: A simple and general method for semi-supervised learning. Aclweb.Org. Association for Computational Linguistics.
Mikolov, T., K. Chen, G. Corrado, and J. Dean.
2013 “Efficient estimation of word representations in vector space.” 1st International Conference on Learning Representations, ICLR 2013 - Workshop Track Proceedings.
76
Appendix D: Robustness Checks - Regressions with CEO controls
In this robustness check, we control for various characteristics of CEOs that have been
associated with higher pressure on results. Executive compensation, for instance, can lead he
CEO to exert more pressure on results, and thus might be leading to misconduct (Zhang et al.
2008; Bergstresser and Philippon 2006). CEO tenure can also lead to misconduct as the length of
tenure might lead to more decision-making autonomy, as demonstrated by Altunbaş, Thornton,
and Uymaz (2018) in the banking industry, while Simpson and Koper (1997) had shown that
new CEOs significantly decreased the probability of misconduct. Finally, women CEOs have
been associated with more conservative earnings management and ethical leadership (Ho et al.,
2015), and women employees have a lower propensity for white-collar crimes (Daly, 1989).
We gathered data on CEOs from Execucomp and created two variables: CEO Salary
measuring CEO’s Total Compensation (Salary + Bonus + Other Annual + Restricted Stock
Grants + LTIP Payouts + Value of Option Grants), and CEO Bonus Fraction measuring how
much of the CEO’s salary is paid as a bonus (bonus/salary).
Including the controls for CEO compensation, tenure and gender did not materially change
our results as shown by tables D1, D2, D3, and D4. Hypothesis 1 is mostly confirmed by the
analyses. One exception is presented by model 1 of D1 in which the inclusion of these controls
altered the significance of results pressure. We believe this result depends on the significant
77
restriction in the sample due to the inclusion of the controls (i.e. a drop of 500 observations).
Hypothesis 2 was rejected as results pressure is significant for low formalization organizations
only in the case of Table D3 regarding the number of violations. Hypothesis 3 was confirmed
even after adding CEO controls: all the models for high decentralization firms show significant
effects for performance pressure on organizational misconduct. Overall, despite the data
limitations. we believe that the results from our analyses with CEO controls is consistent with
our our main models.
78
TABLE D1
Robustness Checks - Regressions with CEO controls
VARIABLESPenalties Number of
Penalties
Amount of penalties
(in dollars)(1) (2) (3)
Performance pressure n-1 0.634 0.269** 3.836**
(0.413) (0.136) (1.932)Number of reviews (log)n-1 -0.0466 0.00240 1.862
(0.242) (0.0786) (1.241)Number of employees (log)n-1 1.068** 0.551*** 2.550
(0.427) (0.0819) (1.647)ROA n-1 2.745 0.926* -1.302
(2.392) (0.478) (9.712)Sales growth n-1 -0.651 -0.0863 0.126
(0.629) (0.159) (3.555)Market competitionn-1 -0.264 0.411 -27.14
(5.641) (1.084) (19.01)Senior management approval n-1 0.0542 0.00622 -0.923
(0.301) (0.0443) (1.370)Compensation and benefits n-1 0.509 0.0252 7.899***
(0.655) (0.169) (2.647)CEO approval n-1 -0.0216 -0.186* -2.608*
(0.374) (0.103) (1.466)Work-life balance n-1 0.0741 0.222 1.235
(0.706) (0.228) (2.458)CEO Salary n-1 9.50e-06 1.97e-06 -2.10e-05
(8.00e-06) (2.40e-06) (3.70e-05)CEO Bonus Fraction n-1 -0.000235 -0.000357*** 0.00152
(0.000232) (6.59e-05) (0.00141)CEO Gender n-1 -0.171 0.202 6.587**
(0.534) (0.153) (3.114)CEO Tenure n-1 -0.0231 -0.00219 -0.0288
(0.0202) (0.00430) (0.0741)Constant - -1.444** -28.96**
- (0.704) (12.01)
Observations 1,171 1,669 2,181Number of firms 205 295 438R-squared - - 0.394Year FE YES YES YESIndustry FE NO YES YESRobust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
79
TABLE D2
Robustness Checks - Regressions with CEO controls on Penalties (Likelihood of
Organizational Misconduct), by Type of Organizational Structure
Variable
Penalties(1) (2) (3) (4)
Low High Low High
formalization formalization decentralization decentralization
Performance pressure n-1 0.476 0.912 0.551 1.457*(0.713) (0.648) (0.374) (0.762)
Number of reviews (log)n-1 0.0716 -0.310 0.165 -0.320(0.233) (0.235) (0.247) (0.429)
Number of employees (log)n-1 0.960 0.906 0.0900 2.250***(0.584) (0.580) (0.644) (0.654)
ROA n-1 1.303 2.829 4.034 7.543*(3.342) (3.088) (2.995) (4.464)
Sales growth n-1 -1.503 -0.147 -0.131 -3.721***(1.126) (0.714) (0.631) (1.340)
Market competitionn-1 7.113 -23.50*** -14.41* 5.843(6.586) (8.056) (8.406) (10.45)
Senior management approval n-1 0.0181 -0.0701 0.515 -0.544(0.401) (0.407) (0.586) (0.582)
Compensation and benefits n-1 -0.227 0.142 0.522 0.274(0.880) (1.080) (1.160) (1.007)
CEO approval n-1 -0.0792 -0.273 -0.305 0.369(0.672) (0.409) (0.598) (0.745)
Work-life balance n-1 -0.0886 1.430 -0.712 1.465(0.952) (1.068) (1.126) (1.107)
CEO Salary n-1 5.52e-06 1.25e-05 4.72e-05* -2.75e-05(2.17e-05) (1.04e-05) (2.45e-05) (2.37e-05)
CEO Bonus Fraction n-1 -0.000256 -0.155 0.189 -0.000150(0.000248) (0.150) (0.186) (0.000318)
CEO Gender n-1 -1.478 0.0809 0.664 -1.314(1.104) (0.629) (0.740) (1.390)
CEO Tenure n-1 -0.0341 -0.0107 -0.0318 -0.0255(0.0315) (0.0257) (0.0308) (0.0344)
Observations 468 478 593 378Number of firms 102 109 130 78Year FE YES YES YES YESRobust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
80
TABLE D3
Robustness Checks - Regressions with CEO controls on Number of Penalties, by Type of
Organizational Structure
Variable
Number of Penalties(1) (2) (3) (4)
Low High Low Highformalization formalization decentralization decentralization
Performance pressure n-1 0.315* 0.185 0.313 1.064***(0.170) (0.136) (0.206) (0.120)
Number of reviews (log)n-1 0.0534 -0.0781 0.0386 0.120(0.114) (0.153) (0.118) (0.118)
Number of employees (log)n-1 0.383 0.950*** 0.152 1.164***
(0.243) (0.302) (0.145) (0.238)ROA n-1 1.297** 0.816 0.428 3.467**
(0.550) (1.105) (0.541) (1.368)Sales growth n-1 0.0463 -0.464* 0.125 -0.311
(0.190) (0.253) (0.157) (0.260)Market competitionn-1 0.486 -0.137 0.0462 -0.610
(1.218) (2.482) (1.634) (1.778)Senior management approval n-1 0.0469 -0.133 -0.0609 -0.0373
(0.0520) (0.0891) (0.0664) (0.116)Compensation and benefits n-1 0.0501 -0.245 0.380 -0.0128
(0.208) (0.338) (0.251) (0.272)CEO approval n-1 -0.361** 0.112 -0.185 -0.474***
(0.176) (0.165) (0.137) (0.179)Work-life balance n-1 0.345 0.191 -0.0428 0.664**
(0.255) (0.394) (0.234) (0.333)CEO Salary n-1 6.08e-07 2.38e-06 7.05e-07 -3.79e-07
(3.42e-06) (4.78e-06) (3.54e-06) (4.69e-06)CEO Bonus Fraction n-1 -0.000245** 0.00583 0.0789 -2.28e-05
(0.000114) (0.0612) (0.0698) (0.000155)CEO Gender n-1 0.287 0.198 0.751** -0.0103
(0.365) (0.258) (0.319) (0.206)CEO Tenure n-1 0.00757* -0.0145 0.00461 0.0101
(0.00459) (0.0121) (0.00349) (0.0119)Constant -0.944 -2.401 0.162 -5.311***
(1.522) (1.836) (0.923) (1.636)
Observations 844 667 782 731Number of firms 189 165 175 157Year FE YES YES YES YESIndustry FE YES YES YES YESRobust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
81
TABLE D4
Robustness Checks - Regressions with CEO controls on Amount of Penalties (in Dollars),
by Type of Organizational Structure
Variable
Amount of penalties (in dollars)(1) (2) (3) (4)
Low High Low Highformalization formalization decentralization decentralization
Performance pressure n-1 8.021 -0.502 2.102 8.064**(5.081) (2.073) (2.702) (3.531)
Number of reviews (log)n-1 0.606 1.110 3.145 2.524(1.653) (2.122) (2.188) (1.605)
Number of employees (log)n-1 6.001*** -0.909 -0.488 9.645***(2.296) (2.447) (2.217) (3.707)
ROA n-1 6.612 -9.432 -19.76* 19.82(12.87) (14.00) (11.82) (17.75)
Sales growth n-1 -4.027 6.126 4.971 -1.809(4.266) (6.926) (5.390) (6.294)
Market competitionn-1 -55.46* -20.44 5.903 -53.13(30.54) (21.75) (26.81) (38.41)
Senior management approval n-1 1.889 -6.899*** 2.284 -0.239(1.616) (2.167) (2.139) (2.340)
Compensation and benefits n-1 8.835** 3.205 3.524 8.866**(3.847) (4.755) (4.360) (4.234)
CEO approval n-1 -2.809 -1.725 -1.711 -6.344**(2.644) (2.132) (2.313) (2.635)
Work-life balance n-1 -0.0224 7.585 3.829 2.283(3.166) (5.084) (3.455) (4.961)
CEO Salary n-1 2.60e-05 -1.62e-05 -6.21e-05 -3.47e-05(7.30e-05) (5.13e-05) (0.000102) (6.01e-05)
CEO Bonus Fraction n-1 0.00178 -1.403** -0.928 0.00230(0.00204) (0.550) (0.871) (0.00238)
CEO Gender n-1 12.43 -0.0576 2.973* 4.514(9.014) (2.797) (1.548) (4.301)
CEO Tenure n-1 0.0825 -0.229 -0.0514 -0.0915(0.114) (0.159) (0.0939) (0.139)
Constant -41.24** 38.95 -31.76* -52.84**(16.82) (23.98) (17.75) (24.04)
Observations 1,142 886 1,000 1,037Number of firms 240 281 244 254R-squared 0.464 0.455 0.446 0.428Year FE YES YES YES YESIndustry FE YES YES YES YESRobust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
82
References
Altunbaş, Y., J. Thornton, and Y. Uymaz
2018 “CEO tenure and corporate misconduct: Evidence from US banks.” Finance Research Letters, 26: 1–8. Elsevier Ltd.
Bergstresser, D., and T. Philippon
2006 “CEO incentives and earnings management.” Journal of Financial Economics, 80: 511–529.
Daly, K.
1989 “Gender and Varieties of White‐Collar Crime.” Criminology, 27: 769–794.
Ho, S. S. M., A. Y. Li, K. Tam, and F. Zhang
2015 “CEO Gender, Ethical Leadership, and Accounting Conservatism.” Journal of Business Ethics, 127: 351–370. Kluwer Academic Publishers.
Simpson, S. S., and C. S. Koper
1997 The Changing of the Guard: Top Management Characteristics, Organizational Strain, and Antitrust Offending. Journal of Quantitative Criminology Vol. 13.
Zhang, X., K. M. Bartol, K. G. Smith, M. D. Pfarrer, and D. M. Khanin
2008 “Ceos on the edge: Earnings manipulation and stock-based incentive misalignment.” Academy of Management Journal, 51: 241–258. Academy of Management.
83
Appendix E: Robustness Checks - Regressions with Negative ESG Media Coverage
Our analysis of misconduct relies on data on official penalties received after legal and regulatory
proceedings. An alternative approach to measure misconduct is to rely on media’s account of
potential misconduct, which might both capture misconduct that does not lead to legal penalty,
but might also be noisier and more skewed towards larger, more visible companies. In any case,
in this set of analyses, we created two new dependent variables based on data provided by
RepRisk, a business intelligence provider specialized in environmental, social, and governance
(ESG) risk analytics and metrics. RepRisk identifies news items that criticize companies for
misconduct on ESG issues such as corruption, fraud, environmental degradation, and human
rights abuses. In this analysis, we will follow prior literature that used RepRisk data (e.g. Kölbel,
Busch, and Jancso, 2017) and use a similar scope of consideration of the ESG issues.
RepRisk search methodology is guided by a scope of 28 pre-defined issues, divided into 5
categories: environmental footprint, community relations, employee relations, corporate
governance, general. Issues in the last category are only used in conjunction with another
category. RepRisk analysts collect and code this media information as a professional service for
banks and investors, and thus we believe they provide a good level of consistency and reliability.
We define negative ESG media coverage as the number of articles covering the
organization for their misconduct on ESG issues. The variable was winsorized (1st and 99th
84
percentile) and, as it represents media coverage which is different from official penalties coming
after a trial, we did not use lagged independent variables. The analyses represent panel OLS
regressions with year and industry (SIC 2-digits) fixed-effects.
The results in Table E1 show the analyses testing the effects of performance pressure on
negative ESG media coverage. The result of model 1 is consistent with our main models on
misconducts regarding the direction of the result for performance pressure, though the
coefficient is not significant (p= 0.124).
Table E2 show the split sample analyses by organizational structure. The results of Table E3
are consistent with the logic of both Hypothesis 2 on the increased effect of performance
pressure regarding low formalized organizations and of Hypothesis 3 on the higher negative
effects of performance pressure for decentralized organizations.
Overall, these results are broadly consistent with the logics of our hypotheses, even if the
main effect of performance pressure on ESG media converage was not significant.
85
TABLE E1
Robustness Checks - Regressions with Reputational Risk
Variable
Negative ESG Media Coverage
(1) Performance pressure n-1 1.290
(0.836)Number of reviews (log)n-1 0.340
(0.619)Number of employees (log)n-1 3.498***
(0.648)ROA n-1 -1.008
(4.485)Sales growth n-1 -2.173**
(1.011)Market competitionn-1 21.86***
(6.815)Senior management approval n-1 0.431
(0.478)Compensation and benefits n-1 -0.188
(1.441)CEO approval n-1 -0.373
(0.738)Work-life balance n-1 0.450
(1.283)Constant -13.00**
(5.605)
Observations 1,145Number of firms 370R-squared 0.738Year FE YESIndustry FE YESStandard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
86
TABLE E2
Robustness Checks - Regressions with Number of Articles, by Type of Organizational
Structure
Variable
Negative ESG Media Coverage(1) (2) (3) (4)
Low High Low Highformalization formalization decentralization decentralization
Performance pressure n-1 0.984** -0.299 0.104 0.529*(0.479) (0.331) (0.395) (0.280)
Number of reviews (log)n-1 0.882*** 0.198 0.697** 0.150(0.265) (0.393) (0.351) (0.280)
Number of employees (log)n-1 0.714** 1.303** 0.358 1.375**(0.349) (0.530) (0.506) (0.590)
ROA n-1 -0.872 -2.010 -3.797** -1.370(1.370) (1.709) (1.745) (1.848)
Sales growth n-1 -1.425*** -0.976** -0.681* -1.127**(0.514) (0.457) (0.352) (0.500)
Market competitionn-1 8.421** 6.291 4.148 12.27***(3.605) (3.927) (4.786) (4.316)
Senior management approval n-1 0.0843 -0.209 0.341 0.207(0.441) (0.427) (0.354) (0.528)
Compensation and benefits n-1 -0.544 0.650 -1.595** 1.102(0.689) (0.737) (0.776) (0.686)
CEO approval n-1 0.348 0.743* 1.672*** -0.184(0.388) (0.421) (0.485) (0.417)
Work-life balance n-1 -0.203 -0.732 -0.132 -0.654(0.621) (0.643) (0.708) (0.550)
Constant -5.623** -6.121* -5.603* -5.613**(2.564) (3.165) (3.287) (2.819)
Observations 1,812 1,530 1,726 1,665Number of firms 406 306 276 276R-squared 0.693 0.690 0.658 0.753Year FE YES YES YES YESIndustry FE YES YES YES YESStandard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
87
References
Kölbel, J. F., T. Busch, and L. M. Jancso
2017 “How Media Coverage of Corporate Social Irresponsibility Increases Financial Risk.” Strategic Management Journal, 38: 2266–2284.
88
Appendix F: Robustness Checks – Distribution of Performance Pressure on Dollar Amount
of Penalties
FIGURE F1
Robustness Checks – Distribution of Performance Pressure (Continuous variable) on Amount of Penalties (in Dollars)
89
FIGURE F2
Robustness Checks – Distribution of Performance Pressure (Continuous variable) on Amount of Penalties (in Dollars) by Type of Organizational Structure
90