Public-Sector Honesty and Corruption: Field Evidence from 40 Countries

Abstract

This study presents a theoretical model of honest behavior in the public sector (public-sector honesty) and its relationship with corruption. We test this model empirically by utilizing and extending a unique data set of honest behavior of public- and private-sector workers across 40 countries, gathered in a field experiment conducted by Cohn et al. (N = 17,303). We find that public-sector honesty is determined by country-level societal culture and public-sector culture; public-sector honesty predicts corruption levels, independently from the effect of incentive structures—in line with the Becker–Stigler model. We find no support for a global mean difference in honest behavior between public- and private-sector workers, alongside substantive cross-country variation in sector differences in honest behavior. The emphasis assigned to honesty of public-sector workers within each country appears to be locally determined by the prevailing public-sector culture. These results imply that beyond cross-national variation in the scope of publicness, it is very content may vary across countries. Lastly, the results of this study consistently fail to support the selection thesis, and we discuss the practical implications of this result for anticorruption policy.

Introduction

The ethical qualities of public servants have attracted public and scholarly attention since the earliest known discussions of public administration (Frederickson 2002; Plato, The Republic; Hegel 2005 [1820]; Wilson 1892; Weber 1978 [1921]). One of the reasons for the importance of ethical behavior in the public sector is its implications for the quality of governance and the level of corruption. Drawing on field evidence of honest behavior of public- and private-sector workers in 40 countries, this article offers the most comprehensive answers thus far for the fundamental and enduring question of sector differences in honest behavior and provides empirical support for the role of public honesty in determining corruption level.

The Becker–Stigler model of corruption (Becker and Stigler 1974) specifies two main determinants of corruption—the incentive structure that public-sector workers face (e.g., anticorruption policies) and their intrinsic honesty. Drawing on this model, we propose a theory of public-sector honesty and its relationship with corruption. In contrast to most studies of this topic, we posit that public-sector honesty is primarily determined by societal culture—the set of ideas, attitudes, beliefs, and norms in the general society (Guiso, Sapienza, and Zingales 2015, Simpser 2020). However, the level of honest behavior among public-sector workers is also affect by public-sector culture—a collective norm regarding the role of the public sector in a given society—which is reflected by systematic differences in honesty behavior between the public and private sectors. Arguably, public-sector culture also affects the level of corruption through processes other than public-sector honesty, such as incentive structures, monitoring, and other anticorruption policies.

We utilize and extend a data set originally collected by Cohn et al. (2019) in a large-scale field experiment on “civic honesty”¹ to test this model in 40 countries. We find support for our hypotheses that (1) public-sector honesty is a strong predictor of corruption; (2) public-sector honesty is influenced by societal culture; and (3) public-sector culture affects the level of corruption, beyond the effect of public honesty, through processes other than public-sector honesty, such as incentive structures, and other anticorruption policies. Our results offer the most comprehensive answers thus far for the enduring question of sector differences in honest behavior. We find no support for a global difference in honest behavior between public- and private-sector workers. More specifically, sector differences in honest behavior do exist, yet they vary across countries, suggesting that public-sector culture varies in its effect on honest behavior.

This article offers three main contributions. First, we explicate the relationships between honest behavior, anticorruption policy, and corruption in a theoretical model. Second, relying on a refined, cross-country data set with behavioral measures, we offer new evidence on the role of public-sector honesty for cross-country variation in corruption levels. Third, this study suggests that sector differences in honest behavior are not due to the inherent nature of the public sector, but rather to the normative role collectively ascribed to it locally. We conclude by discussing the implications of these findings for the study of corruption, publicness, and sector differences, as well as the limitations of the current research, and future challenges for the study of behavioral ethics in the public sector.

Theoretical Framework

Public-Sector Honesty and Corruption

One of the most desirable traits of public servants is a high standard of moral and ethical behavior. Confucian administrative ethics (sixth century, BCE) stipulates that “all bureaucratic relations are to be guided by moral conventions” and conceives the notion of public administration as “morality in action” (Frederickson 2002, 617). Similarly, the prescribed training of Athenian public officials in the fifth-century BCE promoted a simple communal lifestyle aimed at suppressing their egocentric aspirations (Plato, The Republic). More recent scholars similarly described civil servants as a “universal class” whose role is to protect the common interests of society by engaging in impersonal, impartial, and rule-based conduct (Hegel 2005 [1820]; Wilson 1892; Weber 1978 [1921]). Thus, the notion of an ethical public servant is prominent in very different strands of normative theories of public administration. An important implication of public officials’ moral behavior is their impact on the quality of governance (e.g., Miller 2005), and specifically on the level of corruption. Corruption is defined as deviant behavior that involves self-interested abuse of government power by public officeholders (Gardiner 2008; Shleifer and Vishny 1993). It is of central concern in many countries (Overby 2017; Pew Research Center 2014) and was shown to produce a range of politically, socially, and economically detrimental consequences (see, e.g., Solaz, De Vries, and De Geus 2019; Yair, Sulitzeanu-Kenan, and Dotan 2020).

The point of departure in developing a theoretical model linking public servants’ honest behavior and corruption is the Becker–Stigler model (Becker and Stigler 1974; Olken and Pande 2012; Rose-Ackerman 2013). This model posits that a bureaucrat will act corruptly depending on the difference between public and private wages (w and v, respectively), the probability of detecting the corrupt behavior (p), the corrupt gain (b), and the “cost of dishonesty,” or rather, the importance of honesty norms for the public official $(H_{pub})$²:

$${\displaystyle \begin{array}{l}{\displaystyle w-v<\frac{1-p}{p}(b-H_{pub})}\end{array}}$$

(1)

Drawing on this model, one may distinguish between anticorruption policy—represented by a vector (w, p)—and intrinsic public-sector honesty $(H_{pub}),$ as two determinants of corruption. Although the bulk of the empirical research on corruption has centered on the various processes of the former, the focus of this research is on the latter. We begin by positing that public sector honesty$(H_{pub})$—measured by the propensity for honest behavior of public sector workers in country i—is expected to influence the level of corruption in that country $(C),$ as depicted by the α arrow in figure 1. Olsen et al. (2019) provide preliminary support for this expectation by finding a correlation between corruption level and average behavioral dishonesty among prospective public employees across 10 countries. Thus, a hypothesis that is directly inferred from the Becker–Stigler model is:

Figure 1

The Theoretical Model of Public-Sector Honesty and Corruption.Note: The relationships between societal culture (SC), public-sector honesty $(H_{pub})$⁠, private-sector honesty $(H_{priv})$⁠, public-sector culture (PSC), and corruption (C). The dashed double arrow represents a correlation rather than a causal relationship (see Pearl 2009).

 

H1: Public-sector honesty $(H_{pub})$ at the country level reduces corruption level.

Public-Sector Honesty and Societal Culture

What are the determinants of public-sector honesty? Given that norms are context dependent (Kimbrough and Vostroknutov 2016; List 2007), we posit that public-sector honesty $(H_{pub})$ is shaped by two factors—“societal culture” (SC) and “public-sector culture” (PSC). Bureaucrats are primarily members of their society, and, like other citizens, they learn and internalize the set of ideas, attitudes, beliefs, and norms—referred to as “culture” (Guiso, Sapienza, and Zingales 2015, Simpser 2020)—that shape social interactions in their society. Indeed, a number of studies have found significant differences in honest behavior across cultures (e.g., Gächter and Schulz 2016; Hugh-Jones 2016, but see also Pascual-Ezama et al. 2015). Thus, societal culture is expected to affect honest behavior of both public- and nonpublic-sector workers within a society, as depicted by arrow β in figure 1 and summarized in our second hypothesis:

 

H2: Public-sector honesty $(H_{pub})$ is affected by country-level societal culture.

Taking this hypothesis to the extreme would suggest that public-sector workers are drawn from the general population within a country, with no (self or imposed) selection or any process that influences their intrinsic honesty, resulting in a state where public-sector honesty is no different than private-sector honesty. A core assumption in public choice theory indeed posits that bureaucrats are essentially no different from other economic actors in being self-interest maximizers (Mueller 2003, 1) of discretionary budgets (Migué, Bélanger, and Niskanen 1974; Niskanen 1975) and future employment prospects (Stigler 1971).³

Public-Sector Culture and Relative Public-Sector Honesty

However, public-sector institutions may promote a unique culture (i.e., a set of norms and informal rules), which can either emphasize or understate the honesty norm, causing the level of honest behavior among its workers to deviate from that of the general society (see, e.g., Cohn, Fehr, and Maréchal 2014, who identify the effect of banking culture on honest behavior). Institutional culture can affect the honesty norm by a range of processes including selection criteria into public employment, and work socialization (Bellé and Cantarelli 2017; Cohn, Fehr, and Maréchal 2014; Gino, Ayal, and Ariely 2009, O’Reilly and Chatman 1996). For example, if the public-sector culture places a premium on high ethical standards, we can expect that such considerations would affect recruitment, training, monitoring, and promotion processes.

Inversely, if the public-sector culture envisages the role of the public sector as a facilitator of rent extraction and a vehicle for obtaining corrupt rewards for its members, it will motivate similar processes, in this case, promoting less-honest entrants, and socialization of cooperative values that facilitate continued corruption (see Weisel and Shalvi 2015). Moreover, assuming that the public-sector culture is publicly known, it may also affect corresponding honesty-based self-selection processes (Brassiolo et al. 2020; Olken and Pande 2012, 496). Thus, public-sector culture is expected to affect public-sector honesty—as depicted by the causal path γ in figure 1. Assuming that private-sector workers are not subject to public-sector culture, the latter effect on public-sector workers may account for systematic sector differences in honest behavior. We refer to such differences as “relative public honesty”: $RPH=H_{Pub}-H_{Priv}$⁠. Relative public honesty may differ across countries, and its global mean value $(\overline{RPH})$ addresses the longstanding question of whether public-sector employees are globally more or less honest than their private-sector counterparts.

Indeed, a range of studies suggests that the level of honest behavior among public-sector workers may deviate from that of the general society. One such strand of studies relates to honesty-based selection into the public sector. People aspiring to enter the public sector in India exhibited less honesty than those aspiring to private-sector jobs, measured by cheating rates in a “corruption game” and a dice-roll game (Banerjee, Baul, and Rosenblat 2015; Hanna and Wang 2017, respectively). In contrast, a similar study in Denmark using a dice-roll game showed that students who aspire to public-sector jobs were more honest than students who aspired to private-sector jobs (Barfort et al. 2019). Similarly, police applicants in Germany were found to be more trustworthy than nonapplicants, based on an experimental trust game (Friebel, Kosfeld, and Thielmann 2019), and Russian students who prefer a public-sector career display less willingness to cheat or bribe in experimental games than those opting for the private sector (Gans-Morse et al. 2021).

Further indirect evidence for the possibility that public-sector workers differ in their level of moral behavior compared with their private-sector counterparts is based on studies of proximate values such as prosocial and public- sector values. Some experimental studies show more prosocial behavior among public-sector workers compared with private-sector workers in the Netherlands (Buurman et al. 2012) and among German students of public administration compared with students of law and business (Prokop and Tepe 2020; Tepe 2016; Tepe and Vanhuysse 2017). Sector-specific values and work orientation, known as “public service ethos” (Pratchett and Wingfield 1996) or “public service motivation” (PSM; see Bozeman and Su 2015; Perry, Hondeghem, and Wise 2010), may restrain unethical behavior among civil servants (Boyne 2002; Villadsen and Wulff 2018).⁴ Some studies have shown that PSM is positively associated with public-sector employment (e.g., Lewis and Frank 2002; Wright and Christensen 2010), and some studies suggest that PSM is positively associated with intentions to act in an honest and ethical manner (Brewer and Selden 1998; Meyer-Sahling, Mikkelsen, and Schuster 2019; Wright, Hassan, and Park 2016), yet not all studies provide support for this association (e.g., Christensen and Wright 2018).

Studies that estimated the correlation between PSM and actual ethical behavior provide mixed results. Olsen et al. (2019) found evidence in Denmark for a positive correlation between PSM and honest behavior in a dice-roll game, yet Sulitzeanu-Kenan et al. (2019) did not find support for such a relationship in Germany and Israel. A more fundamental limitation of PSM relates to its self-report-based measurement. The notion of “ethical blindspot” (Banaji and Greenwald 2016)—a process that hinders ethical violations from being consciously registered—may limit the validity of such measurements in the context of moral behavior (Zamir and Sulitzeanu-Kenan 2018). In sum, quite a number of studies suggest that public-sector honesty may differ from the honesty norm in the private sector. A systematic measure of relative public-sector honesty (RPH) thus enables us to extend our understanding of the predictors of corruption.

As suggested above, to the extent that public-sector culture (PSC) influences honest behavior, it is expected to produce systematic differences in honest behavior between public and private-sector workers $(\mathrm{R}\mathrm{P}\mathrm{H}\ne 0)$⁠. Yet public-sector culture is also expected to affect the level of corruption, through processes other than public-sector honesty, such as incentive structures—as represented by causal path δ in figure 1. The effect of public-sector honesty and the independent effect of public-sector culture on corruption correspond to the distinction between intrinsic honesty $(H_{pub})$ and anticorruption policy (the probability of detection, and public wages) in the Becker–Stigler model, respectively. The later type of processes, include, for example, monitoring and public wage differentials. For example, in some countries, civil servants operate in an environment of extensive external (e.g., politicians, media, and civil society) and internal monitoring, and public organizations generally have more formal procedures and red tape, and less autonomy in decision-making than private-sector organizations (Boyne 2002, 101). Due to these internal constraints, unethical behaviors are more likely to be detected in public-sector settings. Building on the relative wage hypothesis of Becker and Stigler (1974), Cornell and Sundell (2020) find a negative relationship between self-reported corruption and higher public-sector wages relative to other sectors. This theory yields our third hypothesis:

 

H3: Public-sector culture (PSC) affects the level of corruption (C), independently from the effect of public-sector honesty $(H_{pub})$⁠.

A formidable challenge for any attempt to empirically test this theoretical model and its empirical implications is the need for a comparable measure of honest behavior across sectors and countries. To address this empirical challenge, we utilize data collected in a large-scale field experiment on civic honesty conducted by Cohn et al. (2019), extend it in order to attain detailed sector data, and merge it with corruption, and additional macro-level indicators. Based on these data we are able to test our hypotheses on a global scale. The following section describes our data set.

Data and Measures

Measuring Civic Honesty

Between 2013 and 2016, Cohn et al. (2019) conducted a series of field experiments in 40 countries to examine acts of civic honesty, “where people voluntarily refrain from opportunistic behavior” (70). For this purpose, 17,303 ostensibly lost wallets were handed to workers in public buildings; for each wallet, the researchers recorded whether an attempt was made to contact the owner (by email) in order to return it. The rate of attempts to return the wallets serves as a behavioral measure of honest behavior, as subjects in the experiment could save time and gain additional money by not returning the wallets.

The amount of money in the wallets was randomly varied; roughly half of them contained money (~13.5 USD) and the other half did not.⁵ The results showed that wallets containing money were about 10% more likely to be reported than wallets with no money. In line with previous findings in behavioral ethics, the researchers suggested that the results are explained by “a combination of altruistic concerns and an aversion to viewing oneself as a thief, which increase with the material benefits of dishonesty” (70).

Two important attributes of this experiment render it particularly beneficial for the study of public-sector honesty and corruption. First, wallets were “lost” in five types of institutions that are common in many countries: (1) banks; (2) theaters, museums, or other cultural establishments; (3) post offices; (4) hotels; and (5) public institutions (70). A research assistant walked into each institution with the wallet, approached an employee at the public reception counter, reported having found the wallet “on the street around the corner,” placed it on the counter, and asked the employee to take care of it (70–1). Although Cohn et al. (2019) aggregate honest behavior across the five different institutions, we take advantage of this variance in order to separately estimate public- and private-sector honesty. Since employees working in public institutions are public-sector workers, the data⁶ allow us to estimate the rate of attempts to return wallets among public- and private-sector employees, and thus to estimate public- and private-sector honesty, respectively, based on a common behavioral measure in 40 countries. Second, the behavior recorded in this research (returning/not a lost wallet) is peripheral to the task of the workers involved and unlikely to be influenced by institutional incentive structures; therefore, it offers a measure of intrinsic honest behavior.

Categorizing Public-Sector versus Private-Sector Organizations

Observations in public institutions are identified by the dummy variable Public sector (included in the original study)—our main independent variable. In order to assess the validity of this identifier, we conducted additional coding (beyond the original data of Cohn et al. 2019) of all 3,656 public institutions included in this category, which provided four subcategories: government departments/ministries; local administration; law enforcement, and public safety; and a residual “other” category, as shown in table 1. The coding procedure is detailed in supplementary section A of the online appendix.

Table 1

Proportions of Returned Wallets (Honesty Levels) across Sectors and Subsector Categories

	Private Sector			Public Sector
	Banks	Culture	Hotels	Government Departments/Ministries	Local Administration	Law Enforcement and Public Safety	Other Public	Post
N (% of sector)	4,056 (35.9%)	3,471 (30.7%)	3,764 (33.3%)	600 (16.4%)	1,654 (45.2%)	1,067 (29.2%)	335 (9.2%)	2,356 (13.6%)
Honesty level	48.9%	55.4%	45.4%	43.0%	48.0%	44.9%	65.4%	31.0%
N (% of total)	11,291 (65.3%)			3,656 (21.1%)
Honesty level	49.8%			47.9%

Note: The original categories—banks, culture, hotels, public sector, and post—rely on the original coding of Cohn et al. (2019). Subcategories within the public sector—government departments/ministries, local authorities, law enforcement and public safety, and “other public”—were coded as part of the present study (see supplementary section A of the online appendix for more details).

Observations from banks, cultural establishments, and hotels represent the behavior of private-sector employees.⁷ Observations from post offices are considered a third, distinct category, indicated by the dummy variable Post office. In numerous countries, formerly state-owned postal services have been partially or fully privatized. In the absence of an established cross-country measure of the privatization of postal services, we have chosen a conservative estimation approach and included a separate dummy to account for post offices in each regression model.⁸ This exclusion also conforms to the view that postal workers provide different types of services than our designated public-sector workers (see Crewson 1997).

Estimation Approach

We begin by estimating the difference in honest behavior between sectors globally (⁠$\overline{RPH}$⁠), and in each country. This analysis addresses the question of whether public-sector employees are globally more or less honest than private-sector employees, and it utilizes the individual-level data.

Our main dependent variable is the dummy variable for the original response, which takes the value 1 if an email was sent to the purported owner of the wallet, and zero otherwise. The main independent variable is the dummy variable Public that indicates an observation in a public institution. Since employees in some institutions were likely busier than employees in other institutions, we control for the original busyness measure, based on research assistants’ impressions of how busy the specific recipient of the wallet was on a seven-point scale from “not at all” (0) to “very busy” (6).⁹ Additionally, we control for the experimental conditions in the original Cohn et al.’s (2019) study, and country fixed-effects. These data enable us to estimate relative public-sector honesty as follows:

$${\displaystyle \begin{array}{ll}{\displaystyle Respons{e}_{ic}=}& \alpha +\beta \cdot Publi{c}_{ic}+\gamma \cdot Pos{t}_{ic}\\ & +\mu \cdot Z_{ic}+{\delta }_c+C_{ic}\end{array}}$$

(2)

where $Respons{e}_{ic}$ is the probability of an attempt to return wallet i in country c; $Publi{c}_{ic}$ and $Pos{t}_{ic}$ are indicators of observations (“lost” wallets) at public institution and post offices, respectively (private institutions as reference); $Z_{ic}$ is a matrix of the set of experimental conditions in the original study and busyness,¹⁰ and ${\delta }_c$ is a matrix of country fixed-effects. Parameter β captures our estimate of interest: the mean difference in the probability of returning a wallet between public- and private-sector organizations. To test whether public honesty is affected by societal culture (H2), we utilize the hierarchical nature of the data to estimate the variance fraction in individual honest behavior that is accounted for by country differences (intraclass correlation [ICC]).

Next, we constructed a country-level data set (N = 40) that includes country scores for public- and private-sector honesty, by computing the proportion of returned wallets in the public- and private-sector samples, respectively, for each of the 40 countries. To this country-level data set, we added two widely accepted indices of corruption—Transparency International’s Corruption Perceptions Index (CPI) and the World Bank’s Control of Corruption Index (CoC)—as well as GDP per capita. The corruption and GDP values for each country were taken for the year (between 2013 and 2016) in which the experiment was fielded in that country.

Hypotheses 1 and 3 are tested using the following linear model:

$${\displaystyle \begin{array}{l}{\displaystyle Corruptio{n}_c=\gamma +{\gamma }_1{H}_{{}_{\rho u{b}_c}}+{\gamma }_{2}PSC+{\gamma }_3{W}_c+{\epsilon }_c}\end{array}}$$

(3)

where $Corruptio{n}_c$ is a measure of corruption level (either CPI or CoC) for country c, parameter ${\gamma }_1$ captures the change in a country’s corruption score associated with 1 percentage point increase in the public-sector rate of wallets returned (⁠$H_{{}_{\rho ub}}$⁠), parameter ${\gamma }_2$ captures the average change in corruption score associated with a 1-unit increase in the public-sector culture, and parameter ${\gamma }_3$ captures the differences in corruption associated with a country’s wealth, measured by logged GDP per capita, a well-known predictor of corruption (e.g., Cornell and Sundell 2020; Treisman 2000). Since $PSC$ is assumed to be a confounder of the relationship between $H_{{}_{\rho ub}}$ and corruption, estimating the effect of $H_{{}_{\rho ub}}$ on corruption requires controlling for the direct effect of $PSC$ on corruption, and vice versa (see Pearl 2009).

Although public-sector culture (⁠$PSC$⁠) is not directly observable in our data, a key assumption in our theory is that the honest behavior of public- and private-sector workers is affected by societal culture, but only public-sector workers are affected by the public-sector culture. We can therefore utilize sector differences in honest behavior $(H_{pub}-H_{priv})$ at the country level,¹¹ as a proxy for the role of honesty in local public-sector culture (⁠$PSC$⁠)¹²:

$${\displaystyle \begin{array}{l}{\displaystyle ({H_{pub}}_c-H_{pri{v}_c})=a^{\prime }+PS{C}_c+{{ϵ}^j}_c}\end{array}}$$

(4)

We therefore augment equation (3) by replacing the unobserved PSC with its proxy $(H_{pub}-H_{priv})$⁠:

$${\displaystyle \begin{array}{ll}{\displaystyle Corruptio{n}_c=}& \gamma +{\gamma }_1{H_{pub}}_{{}_c}+{{\gamma }^{\prime }}_2({H_{pub}}_{{}_c}-H_{pri{v}_c})\\ & +{\gamma }_3{W}_c+{ϵ}_c\end{array}}$$

(5)

Note that the coefficient of $({H_{pub}}_{{}_c}-H_{pri{v}_c})$ is marked by an apostrophe, to distinguish it from ${\gamma }_2$ in equation (3), since $({H_{pub}}_{{}_c}-H_{pri{v}_c})$ is not identical to PSC, but a proportional proxy for it.

Empirical Results

Descriptive Analyses

Table 1 presents the different rates of returned wallets across the eight private and public institutions in the data set. Return rates vary across the eight institution types, as indicated by a chi-square test (χ ²(7) = 403.47; p < .001, two-tailed tests throughout). One institution stands out: return rates among post office employees (31.0%) are much lower than among employees in any of the other institution types. As we can see, the return rate among all public-sector employees is 47.9%. This estimate is 2.5 percentage points higher than for hotel employees but 1 percentage point lower than for bank employees and 7.5 percentage points lower compared with cultural establishments’ workers. Overall, the return rate in the three types of private institutions is 49.8%—1.9 percentage points higher than among public-sector workers and statistically significant (χ ²(1) = 3.91; p = .048).

Relative Public-Sector Honesty

The simple comparisons of sector honesty levels provided by table 1 do not control for various covariates and potential differences at the country level. We therefore estimated the sector difference in the probability of returning a wallet (honest behavior) in a set of ordinary least squares regressions with country fixed-effects, as shown in table 2.¹³ All the models include clustered standard errors at the country level, and weights are employed to account for different sample sizes across countries.

Table 2

Estimating the Global Mean of Relative Public-Sector Honesty

	(1)	(2)	(3)	(4)
Variables	Honest Behavior	Honest Behavior	Honest Behavior	Honest Behavior
Public sector	−0.009 (0.021)		−0.011 (0.021)	0.002 (0.022)
	[0.661]		[0.610]	[0.924]
Public (local administration)		−0.007 (0.023)
		[0.769]
Public (Government)		−0.021 (0.032)
		[0.512]
Public (Public safety and law)		−0.039 (0.030)
		[0.194]
Public (Other)		0.099 (0.037)
		[0.011]
Post office	−0.165 (0.021)	−0.165 (0.021)	−0.166 (0.021)	−0.159 (0.030)
	[0.000]	[0.000]	[0.000]	[0.000]
Busyness	−0.012 (0.004)	−0.012 (0.004)	−0.012 (0.004)	−0.012 (0.004)
	[0.007]	[0.006]	[0.006]	[0.007]
Experimental conditions
Money			0.115 (0.015)	0.124 (0.015)
			[0.000]	[0.000]
Big money			0.231 (0.028)	0.220 (0.026)
			[0.000]	[0.000]
No key			0.050 (0.033)	0.054 (0.022)
			[0.140]	[0.018]
Public × Money				−0.036 (0.021)
				[0.096]
Public × Big money				0.013 (0.036)
				[0.709]
Public × No key				0.049 (0.088)
				[0.580]
Post × Money				−0.008 (0.032)
				[0.802]
Post × Big money				0.080 (0.032)
				[0.017]
Post × No key				−0.177 (0.088)
				[0.051]
Constant	0.537 (0.009)	0.537 (0.009)	0.479 (0.010)	0.475 (0.011)
	[0.000]	[0.000]	[0.000]	[0.000]
Observations	17,294	17,294	17,294	17,294
R  2	.152	.153	.169	.170
Country fixed-effects	Yes	Yes	Yes	Yes

Note: SE clustered at the country level in parentheses. Two-tailed p-values are in brackets. Results are from linear probability models. Weights are employed in all models to account for different sample sizes across countries.

In Model 1, the coefficient of interest is the dummy variable Public sector, indicating public-sector observations, and thus the model captures the global mean of relative public-sector honesty $(\overline{RPH})$⁠. We control for post office observations (thus, the reference category is private-sector observations) and for the busyness measure. Model 2 replaces the Public sector dummy with four indicator variables that identify the four types of public organizations: local administration, government ministries/departments, law enforcement and public safety, and a residual category (“other”). (Private-sector organizations again serve as the reference category.) Model 3 adds dummy variables to account for the experimental conditions, and Model 4 includes interaction terms between the sectors and the experimental treatments, as detailed below.

The Public sector coefficient in Models 1 and 3 ranges from −0.9 to −1.1 percentage points and is not statistically significant across model specifications. These results suggest that public-sector employees are, in general, no more honest than their private-sector counterparts across the 40 countries. In Model 2, we see no significant difference in honest behavior between the three major public-sector organization types (jointly covering over 90% of the public-sector observations) and private-sector organizations. More honest behavior (higher than any other public or private subcategory) was found for the residual (other) subcategory of public organizations, which constitutes a mere 9.2% of the public-sector observations. These results provide support for the construct validity of the public-sector category.

As Cohn et al. (2019) show, the presence of money ($13.45) in a wallet raised the likelihood that it would be returned, and increasingly so when a large sum ($94.15) was in the wallet. Relying on further analyses, they suggest that this effect reflects “theft aversion”—“the cost associated with negatively updating one’s self-image as a thief” (2019, 71). Thus, the presence of money in the wallet implicitly enhances moral concerns among participants, resulting in an increased rate of returned wallets. If public-sector workers were characterized by greater honesty than private-sector employees, we would expect the presence of money in the wallet to result in a stronger effect on their propensity to return it. By interacting the Public sector dummy (as well as the Post office dummy) with the experimental conditions in Model 4, we estimate the difference between the effect of money on the behavior of workers in the private and public sectors. The lack of any significant interaction between the Public sector dummy and any of the experimental conditions suggests that public-sector workers did not react to the presence of money in the wallet differently from private-sector employees.¹⁴

In further analyses, we estimated the moderating effects of individual characteristics (age and gender), as well as situational characteristics (busyness, the presence of bystanders, and the presence of coworkers) on relative public- sector honesty by including interaction terms between these covariates and the Public sector variable. Overall, no significant interactions were found.

In all models, post office workers are much less likely to return wallets than either private- or public-sector employees. Post office employees were busier (M = 2.9) than other employees (M = 1.8, p < .001), yet even after controlling for busyness, post office workers are 16.5 percentage points less likely to return wallets than private-sector workers.

We conducted several robustness tests for the estimate of global relative public-sector honesty (full details are provided in supplementary section C of the online appendix). Across all these additional analyses, the results are consistent: no significant difference between the global mean honesty of public- and private-sector workers was found.

Country-Level Variance in Public-Sector Honesty

Given the lack of support for a global difference between public- and private-sector honesty, we utilize the hierarchical nature of the data to explore country-level difference in public-sector honesty and relative public-sector honesty. Figure 2 presents the distribution of country-level proportions of returned wallets in the public and private sectors. It is evident that honest behavior levels in the public and private sectors within each country are more similar than honest behavior across countries. This pattern conforms to the hypothesis that public (and private) sector honesty is shaped by a common country-level societal culture (H2). ICC estimated using a null multilevel model suggests that 15.0% [confidence intervals (CI): 11.2%–19.6%; N_total = 17,303; N_countries = 40] of the total variance in individual honest behavior is accounted for by country-level differences. These results providing support for the influence of local culture on honest behavior in both the public and private sector—in line with H2.¹⁵

Figure 2

Country Levels of Public- and Private-Sector Honesty.Note: The black and gray dots represent the proportions of returned wallets in the public and private sector, respectively, for each country.

The lack of support for global sector difference in honest behavior, coupled with evidence for the importance of local culture, points to the possibility that relative public honesty varies across countries. ICC estimates using a three-level null multilevel model (individual choices within sector, within countries) suggest that 16.3% [CI: 12.0%–21.6%] of the total variance in honest behavior is accounted for by country differences, and 18.2% [14.1%–23.2%] by sector-within-country differences (N_total = 14,947, N_countries = 40, N_{sectors-within-countries} = 80). These results suggest that a sizable share of individual honest behavior is shaped by local societal culture and that public-sector culture—represented by relative public-sector honesty—varies across countries.

To examine public-private honesty differences in each of the 40 countries, we fitted a model based on Model 3 in table 2 (without country fixed-effects and clustered standard errors) to the data from each country.¹⁶Figure 3 shows the distribution of the Public sector coefficient estimates with 95% confidence intervals across countries. The x-axis presents the difference (in percentage points) in the rate of returned wallets from public versus private institutions. Positive values indicate higher levels of honest behavior among public-sector workers than among private-sector workers, whereas negative values indicate the opposite.

Figure 3

Relative Public-Sector Honesty across 40 Countries.Note: The x-axis shows the difference in the likelihood of returning wallets in public institutions compared with private institutions (RPSH); the upper part presents a lowess smoothed histogram of RPSH (red line) against a normal distribution plot (black line); the lower part shows country estimates of relative public sector honesty with 95% confidence intervals (controlling for the experimental conditions, and busyness).

Overall, the public-sector coefficient does not reach conventional levels of statistical significance in 30 of the 40 countries. Among the remaining 10 countries, where a statistically significant difference (p < .05) was found, this difference is positive in three countries and negative in the remaining seven (top and bottom of figure 3, respectively).

Figure 3 reflects substantive variance in relative public-sector honesty across countries. The results range from Sweden, where the rate of returned wallets among public-sector workers is 18.5% higher (p < .001), to Serbia, where it is 34.1% lower than among private-sector workers (p < .001)—a substantial cross-national difference of 52.6 percentage points in relative public-sector honesty. Under the assumptions of our theory, these differences between public- and private-sector honesty imply cross-country differences in public-sector culture. Next, we test our hypotheses regarding the relationship between public-sector honesty and corruption (H1 and H3).

Estimating Corruption

The following analyses utilize a country-level data set (N = 40), as described in the “Estimation Approach” section. Table 3 provides descriptive statistics for this data set. Notably, the two measures of corruption are highly correlated. Public and privates sector honesty are strongly correlated (r = 0.90), in line with the individual-level analyses of country-level ICCs and with H2 (see also figure 2 above). Given this association, it is not surprising that both public- and private-sector honesty are correlated with corruption, yet the former correlation is stronger.

Table 3

Country-Level Descriptive Statistics

Descriptive Statistics				Pairwise Correlations
	N	Mean	SD	(1)	(2)	(3)	(4)	(5)
(1) CPI score	40	55.60	20.74
(2) CoC score	40	0.55	1.05	0.99 [0.000]
(3) Private-sector honesty $(H_{priv})$	40	48.67	21.10	0.68 [0.000]	0.67 [0.000]
(4) Public-sector honesty $(H_{pub})$ Hpub	40	46.14	22.68	0.82 [0.000]	0.81 [0.000]	0.90 [0.000]
(5) Public-sector culture $(H_{pub}-H_{priv})$	40	−2.53	9.86	0.43 [0.005]	0.43 [0.005]	−0.07 [0.680]	0.37 [0.018]
(6) Logged GDP per capita	40	9.93	0.77	0.75 [0.000]	0.74 [0.000]	0.69 [0.000]	0.70 [0.000]	0.13 [0.436]

Note: Two-tailed p-values are in brackets.

Table 4 presents the results of the country-level analyses, in order to test H1 and H3. In Models 1–3, the CPI score is used as a measure of corruption level, and in Models 4–6, the CoC score is used (CPI range: 25–92; CoC range: −1.0 to 2.28). Models 1 and 4 provide baseline models of corruption scores predicted by logged GDP per capita, one of the most dominant predictors of corruption (e.g., Treisman 2000). In line with previous studies, logged GDP per capita is associated with less corruption and accounts for a medium share of the variance in corruption (.556 and .539). Models 2 and 5 add the proxy for public-sector culture—measured by the country difference between public and private-sector honesty $(H_{pub}-H_{priv})$⁠; Models 3 and 6 add public-sector honesty and estimate the model specified in equation (5).

Table 4

The Relationships between Public-Sector Honesty, Public-Sector Culture, and Corruption

	Corruption Perceptions Index (CPI)			Control of Corruption (CoC)
Variables	(1)	(2)	(3)	(4)	(5)	(6)
Public-sector honesty $(H_{pub})$			0.411 (0.111) [0.001]			0.021 (0.006) [0.001]
Public-sector culture $(H_{pub}-H_{priv})$		0.724 (0.196) [0.001]	0.452 (0.184) [0.019]		0.037 (0.010) [0.001]	0.023 (0.010) [0.024]
Logged GDP per capita (mean-centered) (w)	20.207 (2.862) [0.000]	19.036 (2.499) [0.000]	11.066 (3.041) [0.001]	1.013 (0.148) [0.000]	0.953 (0.130) [0.000]	0.537 (0.159) [0.002]
Constant	55.600 (2.185)	57.436 (1.957)	37.769 (5.559)	0.546 (0.113)	0.639 (0.102)	−0.388 (0.290)
	[0.000]	[0.000]	[0.000]	[0.000]	[0.000]	[0.189]
Observations	40	40	40	40	40	40
R  2	.567	.684	.771	.551	.668	.760

Note: SE in parentheses. Two-tailed p-values are in brackets. Results are from ordinary least squares models with two indices of corruption as dependent variable.

Given the very similar results for the two corruption indices, we describe the results of Models 2 and 3 (CPI). Model 2 accounts for 66.7% of the variance in CPI scores in the sample. An increase of 1 point in the public-sector culture (the difference in percentage points between public- and private-sector honesty) is associated with an increase of .724 in CPI score. Model 3 estimates the effect of public-sector honesty (H_pub) and the direct effect of public-sector culture on corruption (controlling for GDP per capita). This model accounts for 75.2% of the variance in CPI scores. An increase of 1 percentage point in public-sector honesty (H_pub) is associated with an increase of .411 in CPI score, and an increase of 1 point in public-sector culture (PSC) is associated with an increase of .452 in CPI score. These results provide support for H1 and H3: Public-sector honesty and public-sector culture are both negatively associated with corruption (positive relationships with CPI and CoC). This latter association offers empirical support for our theoretical assumption that country-level differences between public- and private-sector honesty provide a proxy for public-sector culture. Have these differences been merely random, they would not have been associated with corruption.

To assess the relative predictive power of the three independent variables included, we conducted a dominance analysis (Azen and Budescu 2006), using the domin command (Luchman 2021) in Stata 17. The dominance statistic represents the average net increase in model R² resulted by adding an independent variable to models constructed using all possible subsets of the other predictors. These analyses show that public-sector honesty is the dominant predictor, as it accounts for largest share of the country variance in corruption (see supplementary section D in the online appendix).

Drawing on existing studies of country-level predictors of corruption (see, e.g., Cornell and Sundell 2020; Letki 2006; Treisman 2000), we conducted several additional robustness tests, in which we control for (1) the public-sector wage premium compared with the private sector, in order to assuage concerns that the findings in table 3 are the results of wage differences across sectors in the given countries (see also Cornell and Sundell 2020); (2) a country’s level of democracy, which was found to negatively correlate with corruption (Cornell and Sundell 2020; but see Letki 2006); and (3) whether a country is a federal country, as federalism is positively correlated with corruption (Treisman 2000). As shown in supplementary table C9 in the online appendix, adding these variables does not substantively alter our results.

Our theory posits that the level of corruption is determined by public-sector honesty and public-sector culture. However, an alternative hypothesis is that private-sector honesty may affect on the level of corruption, alongside public-sector honesty. Under such a theory the “supply” side of corruption, provided by public employees, is (partially) captured by public-sector honesty, and the “demand” for it is represented by private-sector honesty. If this is the case, we should expect a negative relationship between private-sector honesty and the level of corruption (positive associations with CPI and CoC), controlling for public-sector honesty. Regressing the level of corruption on public- and private-sector honesty levels is mathematically equivalent to regressing it on public-sector honesty and relative public-sector honesty (H_pub − H_priv)(as detailed in supplementary section E in the online appendix).¹⁷ However, the estimated associations between private-sector honesty and both corruption indices are negative (as reported in supplementary table E1 in the online appendix). These results imply that in more honest societies one should expect more corruption—an implausible result, given the underlying theory, and likely the product of model misspecification. We therefore argue that the more plausible theoretical model is the one that identifies public-sector honesty and relative public-sector honesty (a proxy for the public-sector culture) as predictors of corruption.

Discussion

Our results offer the most comprehensive answers thus far for the fundamental and enduring question of sector differences in honest behavior, provide empirical support for the role of public honesty in determining corruption level, and extend our understanding of honest behavior in public organizations. First, we find no support for a global difference in honest behavior between public- and private-sector workers. Sector differences in honest behavior do exist, yet they vary across countries, suggesting that public-sector culture varies in its effect on honest behavior (sectors within countries account for 18.2% of the individual-level variance in honest behavior). Second, public-sector honesty is shaped by societal culture (H2) (county differences account for 16.3% of the individual-level variance in honest behavior, and public- and private-sector honesty at the country level are highly correlated). Third, public-sector honesty is a strong predictor of corruption (H1). Lastly, public-sector culture predicts corruption levels, beyond the effect of public honesty, suggesting that it influences the level of corruption through processes such as incentive structures (H3).

This research is the first, to our knowledge, to offer evidence for the role of intrinsic honesty among actual public-sector workers in shaping corruption levels across 40 countries. Using dominance analysis, we show that the predictive power of public-sector honesty exceeds that of GDP per-capita—an established key predictor of corruption (Treisman 2000). Although this relationship has been theoretically posited by Becker and Stigler (1974), little scholarly attention has been devoted to it, while institutional factors such as incentive structures, monitoring, and enforcement were extensively studied (Simpser 2020).

The results of this study fail to support the selection thesis; namely, that honesty-based selection results in different levels of honest behavior across sectors. Selection studies showed that people aspiring to enter the public sector in India exhibited less honesty than those aspiring to private-sector jobs (Banerjee, Baul, and Rosenblat 2015; Hanna and Wang 2017), whereas more honest behavior was found to characterize public-sector applicants and students who aspire to join the public sector in Denmark (Barfort et al. 2019), Germany (police applicants: Friebel, Kosfeld, and Thielmann 2019), and Russia (Gans-Morse et al. 2021). We do not find a significant difference between the honest behavior of public- and private-sector workers in India (b = −.002, p = .971, N = 400), Denmark (b = .010, p = .856, N = 300), or Germany (b = .031, p = .603, N = 400). In Russia, our results show a lower level of honest behavior among public-sector workers (b = −.165, p = .032, N = 302), in contrast to what would be expected based on the results of Gans-Morse et al. (2021).

Three potential explanations for these contradictory results may be considered. The first relates to the situations in which honest behavior is measured. Our study relies on field behavior of actual public- and private-sector workers, whereas selection studies examine the behavior of students and job market entrants. After applicants join the public sector, social processes such as attrition (selection out of the public sector) and socialization may act to temper the prior selection patterns. Second, the behavioral ethics literature delineates the psychological processes that lead well-intentioned people to violate moral norms (Feldman 2018; Zamir and Sulitzeanu-Kenan 2018). These social and psychological processes might be more influential than selection in determining subsequent behavior of public-sector workers, as Gans-Morse et al. (2021) also suggest. Third, the type of participants included in the selection studies mentioned above, typically college and university students, may find themselves in different positions and hierarchical levels than the workers whose behavior was recorded by Cohn et al. (2019).

This study is not without limitations. We have presented evidence regarding one measure of honest behavior, that is, the likelihood of contacting a person who ostensibly lost a wallet with the intention of returning it. Although this measure likely captures people’s inclination to act honestly despite the costs involved in doing so, it may not fully reflect other dimensions of ethical conduct, which may differ among public- and private-sector workers. Relatedly, one might question whether failing to contact the owner of a lost wallet (with or without money) should be considered a dishonest act. Some institutions may implement alternative procedures for dealing with lost items—such as placing them in a designated “lost and found” office—and recipients adhering to these procedures should not necessarily be considered dishonest.¹⁸ We certainly acknowledge this possibility and, again, accept that additional valid measures of (dis)honesty would surely help gauge public–private differences in honest behavior. That said, we believe it is reasonable to assume that within countries we are unlikely to see disparate procedures across private and public institutions, as related sociocultural norms are likely to apply to both sectors in a particular country. To date, this is the most globally comprehensive behavioral study of honest behavior across public and private workers.

In this study, we examine sector differences in honest behavior only among a certain subset of workers: those who work behind a reception desk in various institutions. It is therefore possible that in other professions that exist in both sectors, and at different hierarchical levels within these institutions, the results would be different. The limited research on public-sector motivations across hierarchical positions suggests similar levels of PSM among blue-collar workers and executive managers, and they differ with respect to PSM’s subdimensions (Desmarais and Gamassou 2014). Future results could determine the extent to which the results of this study are generalizable across ranks and professions.

Our method of identifying public and private observations in this study is based on the type of organization. However, in some public organizations, front-desk workers might be contracted private-sector workers.¹⁹ Notwithstanding this potential for misclassification of some observations, there are two reasons to suspect that this limitation is not likely to have biased our findings. First, the potential misclassification caused by this is restricted to “public-sector” observations and does not apply to private-sector observations.²⁰ Second, selective comparison of honest behavior in a subcategory of public organizations in which contracting of private-sector front-desk workers is rare—law enforcement and public safety—to the results in the two other public-sector subcategories (government ministries/departments and local administration) do not show significant differences between these public subcategories (see table 1, and Model 2 in table 2), suggesting no bias due to this limitation.

Conclusion

The honesty of public-sector workers has attracted both normative and empirical interest over the years, not least due to its potential implication for the level of corruption (Becker and Stigler 1974). Drawing on recent studies in economics (e.g., Graham et al. 2017, Guiso, Sapienza, and Zingales 2015) and political science (Simpser 2020) that demonstrate the importance of cultural values and norms in determining behavior, we develop and empirically test a theoretical model that incorporates both societal and public-sector culture as determinants of honest behavior and corruption. We empirically test this theory by utilizing a unique data set of honest behavior of public- and private-sector workers across 40 countries.

The findings of this research hold several theoretical and empirical implications for the study of sector differences in honest behavior and for sector differences in general. The substantive cross-country variation in sector differences in honest behavior, points to the limited external validity of single country studies of this attribute. These results beg the question of whether other recorded sector differences—for example, concerning risk taking (Bellante and Link 1981; Bozeman and Kingsley 1998; Chen and Bozeman 2012), perceptions of red tape (Bozeman and Loveless 1987; Bozeman et al. 1992; Feeney and Bozeman 2009), values (Lyons et al. 2006), importance of job security (Frank and Lewis 2004), and motivations and incentives (Crewson 1997; Jurkiewicz et al. 1998; Su et al. 2013)—systematically hold across societies.

Relatedly, despite the normative and practical reasons for placing a premium on the ethical standards of public-sector workers, our results fail to support the proposition that enhanced honest behavior is a fundamental characteristic of public-sectors workers. Rather, the emphasis assigned to honesty of public-sector workers within each country is locally determined by the prevailing public-sector culture. The findings may also contribute to the study of publicness (Bozeman 1987; Rainey, Backoff, and Levine 1976; Rainey and Bozeman 2000). The theory of dimensional publicness accounts for varying sector difference in honest behavior across countries, assuming that countries differ in range of organizations that are subject to political authority. However, this explanation is compatible with cross-national variation that ranges between a null difference (reflecting no political authority or its irrelevance for the specific attribute) and a particular difference (reflecting political authority). However, the variation in sector differences in honesty across countries ranges between significantly positive to significantly negative differences—as shown in figure 3. These results point to the possibility that the very content of publicness—the set of values and norms explicitly and implicitly promoted by political authority, which we refer to as public-sector culture—may vary across countries. Such a possibility presents a challenge to the publicness literature and further complicates the publicness puzzle.

The dominance of societal culture in determining the level of honest behavior of both private- and public-sector workers points to the importance of social parameters for the study of public administration. To the extent that our interest is in general human traits—such as intrinsic honesty—societal culture is expected to account for a sizable share of their variation across societies, regardless of sector. Second, incorporating both societal culture and public-sector culture theoretically and empirically is particularly important in comparative studies.

To demonstrate this point, let us consider the neglect of the societal benchmark in some recent honesty-based selection studies. Given a negative honesty-based selection into the public sector found in India (Banerjee, Baul, and Rosenblat 2015; Hanna and Wang 2017), and a positive selection found in Denmark, Barfort et al. (2019) propose that the latter finding may account for Denmark’s low level of corruption. Even if we accept that these selection patterns hold lasting effects on sector differences in honest behavior,²¹ the theory and findings of the current research show that sector differences in honesty provide a partial predictor of public-sector honesty, and ignoring the societal baseline may lead to misleading predictions of public-sector honesty, and consequently of corruption level.

Moreover, the lack of support for the selection thesis in all four countries for which data is available entails a practical implication for anticorruption policy. If honesty-based selection into the public sector does not provide lasting sector differences in honest behavior, anticorruption policies should prioritize other processes, such as socialization, training, and monitoring, in order to influence the level of honest behavior of public-sector workers, and consequently of corruption level.

To conclude, drawing on a behavioral measure of honesty in the field, across 40 countries, we show that public-sector honesty is primarily determined by societal culture and attribute the differences in public- and private-sector honesty to the public-sector culture. We find support for a sizable effect of public-sector honesty on the level of corruption and for the cumulative effects of public-sector culture on corruption both through public-sector honesty and through incentive structures. Lastly, we find no support for a global mean difference in the honest behavior of public- and private-sector workers. To the extent that such differences exist, they are due to the normative role collectively ascribed to the public sector in each society.

Footnotes

¹

Defined by Cohn et al. (2019, 70) as “the extent to which people voluntarily refrain from opportunistic behavior.”

²

This element is denoted in the original equation by the letter d, but for the purpose of developing our theory, we prefer the use of $H_{pub}$ (public-sector honesty).

³

Public choice theory may allow for prosocial public choices, however, only under circumstances that inhibit one’s ability to act in a self-interested manner. A notable example in the theory of constitutional government is by inducing uncertainty as to one’s future position (Buchanan and Tullock 1962), and other examples can be found in the theory of public goods, the prisoner’s dilemma, and insurable risks (Mueller 2003, 603).

⁴

Whether and how PSM might correspond with honest behavior remains a theoretically challenging question, as prosocially motivated civil servants might bend rules and act unethically for the benefit of their clients (e.g., DeHart-Davis 2007).

⁵

All the wallets also contained three identical business cards with the (fictitious) owner’s name and email address, a grocery list, and a key. In 3 of the 40 countries, there were two additional experimental conditions, in which the wallets either contained more money (~94 USD) or they contained the same amount of money as in the “Money” condition (13.5 USD) but no key. We include these two additional conditions in our analyses.

⁶

Downloaded from: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/YKBODN.

⁷

These institutions, at least in most countries, are not public. We conducted a set of robustness tests to assess the stability of our results to account for national variation in the public/private nature of these types of institutions, as detailed in the “Robustness Tests” section. Our results are substantively identical if we omit nondemocratic countries (a Polity IV score lower than 6) from our analyses (see supplementary section C of the online appendix).

⁸

See Son and Zohlnhöfer (2019) on the challenges of developing cross-country privatization measures.

⁹

As detailed in the Supplementary Materials of the original Cohn et al.’s (2019) article, on page 11: https://science.sciencemag.org/content/suppl/2019/06/19/science.aau8712.DC1.

¹⁰

“No money” (reference category), “Money,” “Big money,” “No key.” For details, see Cohn et al. (2019).

¹¹

Specifically, for each country we created a variable (proxy for PSC) by subtracting the proportion of returned wallets in the private sector, from the proportion of returned wallets in the public sector.

¹²

The relationship in equation 4 relies on the assumption that SC and PSC are unassociated. This assumption is supported by the lack of significant association between $H_{priv}$ and ${H_{\mathrm{p}\mathrm{u}\mathrm{b}}}_{{}_c}-H_{\mathrm{p}\mathrm{r}\mathrm{i}{\mathrm{v}}_c}$ (r = −.067, p = .680, N = 40).

¹³

Ordinary least squares estimates are reported for ease of interpretation. Logistic models provide substantively identical conclusions (see supplementary table C1 in the online appendix).

¹⁴

The significance of the joint Wald test of the three interactions of Public sector with the experimental conditions is p = .832.

¹⁵

Tabulated results are provided in supplementary table C2 in the online appendix.

¹⁶

Average country sample size is M = 432.6 (median = 400, min = 274, max = 1,132).

¹⁷

Note that the strong correlation between public- and private-sector honesty presents a risk of multicollinearity (VIF = 4.4), which does not exist when relative public honesty is used as predictor (VIF = 1.85). However, this technical challenge for estimation does not seem to bias results when using public- and private-sector honesty as predictors, since both coefficients and SEs seem perfectly consistent with their predicted values based on the mathematical prediction (see supplementary section E in the online appendix).

¹⁸

See some of the responses to the original Cohn et al.’s (2019) article: https://science.sciencemag.org/content/365/6448/70/tab-e-letters.

¹⁹

We thank an anonymous reviewer for pointing this out.

²⁰

The existence of private-sector contract workers in some public organizations is not mirrored by parallel public-sector front-desk workers in private organizations.

²¹

An assumption that is unsupported by our findings, as noted in the Discussion section.

Acknowledgments

We thank Yuval Feldman, Micha Mandel, Eyal Pe’er, seminar participant at Tel-Aviv University, the editor of JPART, and anonymous reviewers for their insightful comment. We are grateful for Michel André Maréchal for his invaluable advice regarding the extension of the original data. We thank Gian Fink for excellent research assistance. We acknowledge the generous support of the Niedersachsen Vorab (Research Cooperation Lower Saxony—Israel) grant.

Data Availability

Data of this research are available at: https://doi.org/10.7910/DVN/JECEH3.