Religion and the Acceptability of White-Collar Crime: A Cross-National Analysis

This article examines whether shared religious beliefs and religious social relationships (Durkheim) and belief in a personal, moral God (Stark) negatively affect attitudes toward the acceptability of white-collar crime. In addition, using a large cross-national sample and estimating multilevel models, we test whether effects are conditional on modernization and religious contexts characterized by belief in an impersonal or amoral God. Shared religious beliefs and the importance of God in one’s life are negatively related to the acceptability of white-collar crime. These effects, however, weaken in religious contexts characterized by belief in an impersonal or amoral God as do the effects of religious social relationships and belonging to a religious organization; modernization, on the other hand, does not have a moderating effect. In short, religious belief is associated with lower acceptance of white-collar crime and certain types of religious contexts condition this relationship.


INTRODUCTION
Unlike murder, robbery, or assault, white-collar crime is often perceived as "victimless" (see Greer 2009). Yet, the costs of white-collar crime are by no means insignificant, often far exceeding the costs of theft and robbery (Brown, Esbensen, and Geis 2001;Rosoff, Pontell, and Tillman 2004). Recent evidence suggests a shift in public attitudes toward a greater concern for whitecollar crime victimization as well as a desire for more punitive measures for white-collar crime perpetrators (Holtfreter et al. 2008;Rebovich and Kane 2002). No doubt, the Enron, Worldcom, and Bernie Maddoff scandals, as well as broader criticism of corporations in the current global recession, have increased the salience of the issue and focused public attention on the harm that results from financial and corporate crime and corruption.
Public opinion is vital for the creation of social policy in democratic nations as policymakers consider public preferences when making political decisions (see Burstein 2003;Brooks and Manza 2006;Kikusawa, Olafsdottir, and Pescosolido 2008). When it comes to criminal justice policy, strong public support for a more "get tough" approach on crime generally leads to more punitive measures (Cullen, Link, and Polanzi 1982;Katz 1980;Podgor 2007;Simpson 2002). Given the apparent shifts in public attitudes toward white-collar crime, identifying predictors of such attitudes is especially relevant and may contribute to an understanding of cross-national variation in laws, policies, and enforcement.
While there is extensive research exploring the relationship between religion and drug use, juvenile delinquency, and general index crimes like robbery and assault, white-collar crimes and attitudes toward them are underexplored in the literature (Stack and Kposowa 2006; see also Baier and Wright 2001;Johnson et al. 2000;Johnson and Jang 2011). Inspired by two theories within the sociology of religion-Durkheim's general theory of social order ([1897] 1979, [1925] 1961; see also Thorlindsson and Bernburg 2004) and Stark's (2001) more recent moral order hypothesis-we theorize and test the specific elements of religion that might foster or undermine intolerance of white-collar crime. 1 Durkheimian theory and the moral order hypothesis (Stark 2001) propose that religion-via different mechanisms-leads to the acceptance of societal norms, laws, and morals that, in turn, increase behavioral conformity. Thus, while neither perspective specifically addresses white-collar crime, they both provide general theories predicting normative attitudes that are applicable to all types of crime, including white-collar. The former perspective highlights the importance of religious communities normatively integrating individuals through shared beliefs and social relationships, and the latter underscores belief in a personal moral God. Both perspectives also emphasize the conditioning effects of contextual country-level factors: for Durkheim, the integrating negative effect of religion on tolerance of crime weakens with modernization and in societies characterized by a belief in an impersonal or inactive God; for Stark, the negative effect of belief in God on tolerance of crime is stronger in modernized societies as well as societies characterized by belief in a personal moral God.
Although there has been a recent upsurge in research testing conditional effects, nearly all of these studies focus on religious contexts or "moral communities" broadly defined (Cochran and Akers 1989;Maimom and Kuhl 2008;Regnerus 2003;Stack and Kposowa 2006;Stark 1996;Stark, Kent, and Doyle 1982;Tittle and Welch 1983;Wallace et al. 2007;Welch, Tittle, and Petee 1991). Much of this research attempts to establish conditional effects of religious contexts on individual-level religiosity. Yet, several studies find that not all operationalizations of religious contexts produce the same results (Finke and Adamczyk 2008;Maimom and Kuhl 2008;Regnerus 2003;Stack and Kposowa 2006). More research is needed to understand what exactly it is about religious contexts that condition individual-level religion effects and what types of religious contexts do so. Moreover, while this research has emphasized religious contexts, it has mostly neglected nonreligious contexts. The current study extends this research by investigating a nonreligious context-modernization-and also by theorizing one characteristic of religious contexts-belief in a moral, personal God-that when present should strengthen individual-level effects of religion and when absent should weaken them.
In this study, we use multilevel cross-national data and present results from a series of hierarchical generalized linear models (HGLMs) testing Durkheimian theory and the moral order hypothesis. We find mixed support for our hypotheses: several of the individual-level religion variables have direct negative effects on the acceptability of white-collar crime; modernization, however, does not condition these effects, while religious contexts characterized by belief in a moral, personal God do.

THEORETICAL CONSIDERATIONS
Although multiple theoretical explanations exist for why religion affects crime, fewer theories can account for why religion would affect attitudes toward crime. Because our interest is in predicting attitudes toward white-collar crime, we focus on two theoretical perspectives that with societal density, challenging socialization into shared norms that restrict social atomism (i.e., normative integration). This, in turn, leads to societies based on individualism (e.g., the "cult of the individual") and personal liberty (Durkheim [1897(Durkheim [ ] 1979. Instead of normative integration through religious (or other) communities, functional integration (Durkheim [1897(Durkheim [ ] 1979Olsen 1965;Webb 1972) is achieved through social and economic exchanges between diverse people that generate trust and some shared values, which support and underlie these exchanges (Messner 1982;Neumayer 2003). Thus, Durkheim predicts that the effect of religion on the internalization of social norms and crime will decline as societies become more socially differentiated (see Durkheim [1897Durkheim [ ] 1979. Although often overlooked, Durkheim clearly reiterates this point several times in Suicide: "[in modernized societies] religion has lost most of its power" ([1897] 1979:255, emphasis added), it "has been able to exert a beneficent influence under given conditions [i.e., low societal differentiation]; but the necessary conditions are no longer given" ([1897] 1979:374), "unless humanity returns to its starting-point, and religions will no longer be able to exert very deep or wide sway on consciences" ([1897] 1979:375). Given this, we predict that (H2a) the negative effect of shared religious beliefs on acceptance of white-collar crime will be weaker in modernized societies, and (H2b) the negative effect of participation in religious social relationships on acceptance of white-collar crime will be weaker in modernized societies.
The second perspective, which we refer to as the moral order hypothesis (Stark 2001), proposes that belief in a powerful, moral, conscious God, who cares about the moral behavior of humans, reinforces the acceptance of societal norms and laws, which, in turn, serves as a deterrent for immoral behavior. In essence, if individuals believe in a God who requires moral conformity, they will be more likely to share the moral precepts of their God and behave accordingly. Stark's (2001) conception of the moral order is widespread acceptance and conformity to societal norms and laws (see also Finke and Stark 1992:35), such that his theory is applicable to all behaviors "subject to moral interpretation" (Stark 2001:623), including white-collar crime. Thus, we hypothesize that (H3) individuals who believe in a moral, personal God will have a lower acceptance of white-collar crime. Based on this, Stark (2001) argues that the negative relationship between religion and morality is not universal, but will be more prevalent in societies where religious beliefs entail conscious moral gods. He proposes that this tends to be the case in more complex cultures: as cultures develop, images of God(s) tend to shift away from many gods of little scope (or power) to few gods of great scope who have the power and the desire to enforce moral conformity. Stark predicts a weaker (or nonexistent) negative effect of individual-level belief in God, which he operationalizes as the importance of God in one's life, on morality in less complex or nonmodernized societies. However, he does not test this. Thus, we hypothesize that (H4) importance of God should have a greater negative effect on acceptance of white-collar crime in modernized societies compared to nonmodernized societies. Stark (2003:373) also argues that the morality of individuals will "be influenced by their religious commitments only in societies where the dominant religious organizations give clear and consistent expression to divine moral imperatives." Stark (2001Stark ( , 2003 describes how Protestantism, Catholicism, Judaism, Islam, and Hinduism tend to have conceptions of God as conscious, active, and moral, whereas Eastern Orthodoxy and Buddhism tend not to have such conceptions and instead focus more on rites and rituals. Consequently, he predicts a weaker (or nonexistent) negative effect of the importance of God in one's life on morality in countries where the majority religion is Eastern Orthodoxy or Buddhism (i.e., characterized by an impersonal, amoral, or nonexistent God). We therefore hypothesize that (H5) importance of God should have a weaker negative effect on acceptance of white-collar crime in societies where the dominant religion has more impersonal or amoral conceptions of God.
Although Durkheim ([1897Durkheim ([ ] 1979 asserts that belief in God is unnecessary for religion to have a protective effect against deviance (i.e., because integration, not the content of beliefs, is what matters for Durkheim), he describes certain beliefs in God that can serve to reinforce or undermine the authority of shared belief systems and social relationships. When the rules provided by the shared religious belief system "supposedly emanate from superhuman authority, human reflection has no right to bring itself to bear on them. It would be actual contradiction to attribute such an origin to them and permit free criticism of them" (Durkheim [1897(Durkheim [ ] 1979. In this way, the unquestionable nature of shared belief systems necessary for the regulation of behavior is supported by a belief that the regulations emanate from a God. However, Durkheim notes that belief in a God that is "outside of the universe and everything temporal cannot serve as a goal for our temporal activity, which is thus left without an objective. . . . Abandoning the world to us, as unworthy of himself, he simultaneously abandons us to ourselves in everything respecting the world's life" (Durkheim [1897(Durkheim [ ] 1979. Here, Durkheim indicates that the extent to which belief in God will matter depends on whether the God is viewed as more active in temporal affairs and as caring about the moral behavior of individuals, such that he or she provides behavioral regulations that must be followed unquestionably. As such, religious contexts where the rules are not believed to emanate from unquestionable divine authority should weaken the effects of shared religious belief systems and the social relationships that support them. This allows for the specification of a Durkheimian corollary hypothesis to H4: (H6a) the acceptance of shared religious belief systems should have a weaker negative effect on acceptance of white-collar crime in societies where the dominant religion has an impersonal or amoral conception of God and (H6b) participation in religious social relationships should have a weaker negative effect on acceptance of white-collar crime in societies where the dominant religion has an impersonal or amoral conception of God.
Although Durkheim ([1897Durkheim ([ ] 1979 and Stark (2001) use different terms-normative integration and the moral order respectively-both approaches predict that religion should decrease the acceptability of white-collar crime and this effect will depend on contextual factors. However, they disagree on what exactly it is about religion that generates such a relationship-unquestionable shared religious beliefs and religious social relationships in the case of Durkheim, or belief in a personal, moral God in the case of Stark. Figure 1 presents these two theoretical frameworks and their corresponding hypotheses.

Religion and Attitudes Toward the Acceptability of White-Collar Crime
There are few studies examining the effect of religion on attitudes toward white-collar crime (Stack and Kposowa 2006; see also Baier and Wright 2001). Stack and Kposowa's (2006) and Finke and Adamczyk's (2008) cross-national studies of the effect of religion on attitudes toward tax fraud and giving false information to the government, respectively, were a step toward filling this gap in the literature. Both studies find that church attendance significantly decreases one's tolerance of tax fraud and giving false information to the government, respectively. Stack and Kposowa (2006) suggest that future research should examine the effects of religiosity on attitudes toward other types of white-collar crimes and criminal behaviors. We extend this research by testing the effect of a variety of religion measures, derived from Durkheim and Stark's theories, on attitudes toward white-collar crimes (i.e., claiming government benefits to which one is not entitled, accepting a bribe, and cheating on taxes). Stack and Kposowa's (2006) and Finke and Adamczyk's (2008) studies exemplify a new and growing area of research examining the conditions under which religion reduces crime or the acceptance of crime. Most of this research tests the moral community hypothesis, which predicts "that the deterrent effect of religion on crime [and attitudes toward crime] is greatest in areas characterized by high rates of aggregate religiosity" (Baier and Wright 2001:6). According to this perspective, religion promotes belief in and conformity to norms through interaction with the larger community, but only to the extent that religiosity is integrated into the wider normative order and is accepted by the larger community that enforces this order (Stark 1996;Stark, Kent, and Doyle 1982). Recent studies have provided support for the moral community hypothesis,  finding that religious contexts do tend to amplify the negative effect of religion on crime and deviance and attitudes toward crime and deviance (Finke and Adamczyk 2008;Maimom and Kuhl 2008;Regnerus 2003;Scheepers, te Grotenhuis, and van der Slik 2002;Stack and Kposowa 2006;Wallace et al. 2007). This hypothesis has also been extended to other outcome variables of interest such as criminal sentencing (Ulmer, Bader, and Gault 2008), substance abuse treatment (Richard, Bell, and Carlson 2000;Shields et al. 2007), and religious participation (Hill 2009).
Taking all this research together, it is clear that religious contexts matter, yet not all measures of religious contexts produce the same results (see, for example, Finke and Adamczyk 2008;Maimom and Kuhl 2008;Regnerus 2003;Stack and Kpowosa 2006). Drawing on Stark (2001) and the Durkheimian tradition, we theorize that religious contexts characterized by belief in a personal, active, moral God will tend to amplify the individual-level religiosity effects, whereas religious contexts characterized by an impersonal, inactive, amoral, or nonexistent God will tend to weaken them. In this way, we suggest that not all religious contexts will support the moral community hypothesis. Moreover, although research on religion has increasingly emphasized the importance of context, it greatly neglects social and ecological environments that are not religious, such as macrostructural and secular conditions. Yet, both Durkheimian theory and Stark's moral order hypothesis emphasize the importance of nonreligious contexts. Modernization is a key societal context for both perspectives, but they offer competing hypotheses regarding how it conditions the effect of religion on tolerance of crime, with Durkheim arguing that it weakens the effect and Stark proposing that it amplifies it. Thus, our study extends the moral community literature by drawing on Durkheimian theory and Stark's moral order hypothesis (2001) to propose modernization as a secular contextual factor that might condition the individual-level effects of religion on acceptance of white-collar crime.

METHODS Data
We use data from the fourth wave (2000) of the World Values Survey (WVS) for 51 countries. The WVS is a nationally representative survey of randomly selected adults and contains rich biographical information on individual attitudes, values, and religious views. While the WVS was designed to compare individual attitudes across countries, the data are biased toward developed nations. 2 As a result, we include as many non-Western nations as missing data permit. Countrylevel measures come from a variety of sources: CIA World Factbook (2000), Economic Freedom of the World Project (Gwartney, Lawson, and Samida 2000), Freedom House (2001), and the World Bank. All country-level values are for the same year as the country's WVS sampling year.

Dependent Variable
We summed the following three items (WVS variable IDs: F114, F116, and F117) to create an attitude toward the acceptability of white-collar crime dimension (0 = never justifiable, 27 = always justifiable). Respondents were asked, "Please tell me for each of the following statements whether you think it can always be justified, never be justified, or something in between . . . ," and then were allowed to choose on a 10-point Likert scale from "Claiming government benefits to which you are not entitled," "Cheating on taxes if you have a chance," and "Someone accepting a bribe in the course of their duties." These items are typically considered victimless white-collar or occupational crimes and are strictly related to fraud and corruption, which are understudied in the literature (Stack and Kposowa 2006). The three items are highly correlated and have a Cronbach's alpha (i.e., α) of .65. See Table 1 for mean and median values of acceptability of white-collar crime by country.
Descriptive statistics and variable codes for individual-and country-level measures are provided in Table 2.

Country Measures
Economic development, which we use to measure modernization, is an often discussed correlate of crime (Antonaccio and Tittle 2007;LaFree and Kick 1986). Building on Lipset's (1959) scale, we use a dimension composed of three standardized items: the gross domestic product per capita (constant 2,000 U.S. dollars), the life expectancy at birth for the total population, and the percent of the total population that is urban dwelling (α = .82).
To test hypotheses 4 and 6 (a and b), we follow Stark's (2001) identification of dominant religions with an impersonal, inactive, amoral, or nonexistent God and dominant religions where this is not the case. Stark (2001) provides detailed descriptions of how conceptions of a personal, moral God vary across religions with Protestants, Catholics, Jews, Muslims, and Hindus tending to have more active, moral conceptions of God(s) and Eastern Orthodox Christians and Buddhists tending not to have such conceptions (see Stark 2001Stark , 2003 for further description). This is not to say that all members of a given religion hold the same conception of God. Stark does not deny that there is variation within religions in how members conceive of God; rather, his point is that, on average, members of these religions tend to have certain conceptions of God, which allows  (0) us to use the dominant religion of a particular country as a proxy for these conceptions. Given this, we create a binary measure where "1" represents a country in which Eastern Orthodoxy or Buddhism is the majority religion and "0" otherwise. There are nine countries in our sample that receive a "1": Belarus, Bulgaria, Greece, Japan, Macedonia, Moldova, Romania, Russia, and Vietnam. Following Nivette (2011) and Pridemore and Trent (2010), we also control for several country-level variables that have been shown to affect crime in previous studies: (1) percent female obtained from the World Bank (see Messner and Rosenfeld 1997;Pratt and Godsey 2003), (2) an index of ethnic and linguistic homogeneity obtained from the CIA World Factbook and the Encyclopedia Britannica (α = .75) (Hansmann and Quigley 1982;LaFree and Kick 1986), and the gini coefficient obtained from the World Bank to measure economic inequality (e.g., Chamlin and Cochran 2006;Lederman, Loayza, and Menendez 2002;Pratt and Godsey 2003;Robbins and Pettinicchio 2012). 3 We also estimated models with country-level corruption 4

31,022
. and moral community 5 as control variables (results not shown). However, because neither of these variables had statistically significant effects, we exclude these two variables to conserve degrees of freedom. 6

Individual Measures
Unfortunately, direct measures of conceptions of God are unavailable. Thus, following Stark (2001), we use religious affiliation (F025) to test H3 and measure belief in a moral, personal, active God (e.g., Protestant, Catholic, Jew, Hindu, and so on) and belief in an impersonal, inactive, amoral, or nonexistent conception of God (i.e., Orthodoxy and Buddhism). We expect individuals who are affiliated with religions that harbor moral, personal, and active conceptions of God to be less accepting of white-collar crimes than Buddhists (the referent category in our models). We also expect Orthodox Christians to be statistically similar to Buddhists with respect to the acceptance of white-collar crime. Because we have no theoretical predictions regarding them and due to their small sample size, other religious affiliations (e.g., Sikhs, Ancestral, and so on) were grouped together into "other religious affiliation." To test H4 and H5, we use the exact variable employed by Stark (2001): importance of God in one's life (F063). Respondents were asked "How important is God in your life? Please use this scale to indicate 10 means 'very important' and 1 means 'not at all important'." The variable was not recoded and left as is.
To test the Durkheimian hypotheses, we created two different latent constructs representing the two key factors affecting normative integration-shared beliefs and social relationships (or interactions). The first factor, an extensive shared belief system, can be operationalized as an individual's acceptance of a shared religious belief system. We combine the following three questions into a latent construct (α = .69) to measure this concept: how much confidence respondents have in churches (E069) and whether they think that the churches in their country "are giving adequate answers to: the moral problems (F035) and needs of the individual [and] the social problems facing our country today (F038)." Importantly, these questions do not capture the actual content of beliefs, which Durkheim viewed as mostly inconsequential, but instead they measure larger religious belief structures. According to Durkheim, shared belief systems provide rules and guidelines for how individuals and groups should behave and are effective to the extent that they are extensive and unquestionable. The first question in the construct captures the unquestionable quality of beliefs by measuring one's confidence in the religious institution providing them. The other two questions encompass the extensive rules or guidelines that provide development scale (r ≈ .70), we use democracy (i.e., political rights from Freedom House 2001) to measure corruption and democratic politics (see Grim and Finke 2007 for a study that also uses this measure). Note that our measure of democracy is highly correlated with the Governance Matters corruption measure (r ≈ 0.70) and only slightly correlated with our economic development scale (r ≈ −.30). 5 While the operationalization of moral communities varies from study to study, two common measures include (1) the proportion (or percentage) of the largest religious group in a geographic area (e.g., Moore and Vanneman 2003;Regnerus 2003;Stack and Kposowa 2006;Stark 1996), and (2) average church attendance (i.e., aggregated from the individual level) (e.g., Finke and Adamczyk 2008;Maimom and Kuhl 2008;Regnerus 2003;Scheepers, te Grotenhuis, and van der Slik 2002;Wallace et al. 2007). In line with the former, we operationalize moral community using a measure of religious homogeneity (or percent largest religious group in a country) obtained from the Encyclopedia Britannica (for similar measures, see Regnerus 2003;Stack and Kposowa 2006;Stark 1996). We also operationalized moral communities using a country's average level of religious service attendance. 6 We exclude the democracy and moral community variables in favor of inequality and economic development since (1) democracy is consistently unrelated to criminological outcomes in the cross-national literature (see Nivette 2011), (2) inequality and other measures of resource deprivation are consistently related to criminological outcomes in the crossnational literature (see Pridemore and Trent 2010), (3) economic development is a key predictor in the present article, (4) neither democracy nor the moral community variables yield statistically significant cross-level interaction effects with our level-1 religion variables of interest (results not shown), (5) including democracy or the moral community variables either alone or in concert to the remaining models does not alter the results with respect to the key religion variables, and (6) excluding these variables saves on degrees of freedom. individuals with moral and social answers to how they should live and solve problems. Following neo-Durkheimian research (for similar operationalizations, see Bjarnason et al. 2005;Pescosolido and Georgianna 1989;Pescosolido 1990Pescosolido , 1994Thorlindsson and Bernburg 2004;van Tubergen, te Grotenhuis, and Ultee 2005), we measure the religious social relationships (or interaction) factor with a latent construct that includes standardized values for frequency of participation in religious services (F028) and how much time a respondent spends with members of his or her church, mosque, or synagogue (A060) (α = .76). Neo-Durkheimian research often uses belonging to a religious organization (A065) as a measure of religious social relationships (or interaction), but this measure was not highly correlated with the other measures included in the latent construct. Because of its common use, we included it in the analysis as a discrete variable.
We also control for age (X003), sex (X001), marital status (X007), age when completed education (X023r), employment status (X028), income (X047r), a respondent's number of children (X011), and political views of the respondent (E033). 7 In terms of income, respondents were asked to place their household income into one of a range of values listed in their country's currency. These ranges were then categorized into values of "low," "medium," and "high." Therefore, this measure of income represents where an individual's household income ranks relative to others in his/her country. Measuring income in this way allows for comparability across countries. Finally, we include a count of how many types of secular organizations an individual participates in as a measure of secular communities that may also serve to integrate individuals (A064-A076) (Corcoran, Pettinicchio, and Young 2011;Kitts 1999;Klandermans, van der Toorn, and van Stekelenburg 2008).

Analytic Strategy
We use multilevel models, specifically HGLMs, to adjudicate between the two theoretical models. This analytic strategy allows us to assess the degree to which key individual-level predictors for the Durkheimian and moral order hypotheses impact individual attitudes toward white-collar crime while considering contextual-level factors and cross-level interaction effects. Overall, HGLM accurately estimates standard errors of clustered cases within larger units and permits the estimation of higher level factors (Raudenbush and Bryk 2002). Since our dependent variable is positively skewed (skewness = 1.76) and overdispersed (the variance-23.1-is roughly seven times greater than the mean-3.5), we use a two-level Poisson model with an overdispersion parameter (i.e., hierarchical negative binomial regression), which significantly improves model fit over a classic Poisson regression model, to simultaneously estimate individualand country-level predictors of acceptance of crime.
We estimate our models using restricted maximum likelihood with the default settings in HLM 6.02 (Raudenbush et al. 2004) except for setting the microiterations and the macroiterations to 1,000, which aids with convergence. Following Inglehart et al. (2004), all calculations are frequency weighted at the individual level (weight WVS s017). Finally, we estimate each unique covariance in the marginal covariance matrix (i.e., unstructured covariances) and grand-mean center all variables. 8 7 We used "age completed education" instead of the highest degree attained because it is less correlated with income (polychoric correlation = .29) than highest degree attained (polychoric correlation = .39) but still captures education well (the polychoric correlation between year completed education and highest degree attained = .75). 8 Due to missing data in the WVS, our fully saturated models yield 31,022 level-1 units and 51 level-2 units with an average of 608 level-1 units per level-2 unit. In other words, biased inferences due to statistical power are likely absent from the present article. As Moineddin, Matheson, and Glazier (2007) and Scherbaum and Ferreter (2009) reveal, 50 level-2 units with at least 50 level-1 units per level-2 unit are necessary to reach a satisfactory power level (i.e., β) of .80 in HGLM. This suggests that type II errors related to statistical power are not an issue and that statistical inferences derived from the present article will be unbiased in this respect.

RESULTS
We start our analysis with the estimation of a null random-intercepts-only model (not shown). From the null model, we find that acceptance of white-collar crime significantly varies across countries (variance = .25) and that overdispersion is present (variance = 6.14) and statistically significant, which yields an intraclass correlation of .04. The intraclass correlation shows that 4 percent of the total variance in the outcome is between countries, while 96 percent of the total variance in the outcome is at the individual level (AIC = 291,218). Although the intraclass correlation is small, the random intercept in the null model is statistically significant (p < .001). This suggests that the use of HGLM is appropriate for the present data.
Model 1 in Table 3 includes the individual-level control variables. This is done for purposes of model comparison and to test whether or not the acceptance of white-collar crime varies by sociodemographic characteristics. Women and married, older, and more educated respondents are less accepting of white-collar crime. Interestingly, political beliefs, income, parenthood, employment status, and organizational involvement are statistically unrelated to the acceptance of white-collar crime. 9 Model 2 in Table 3 includes the individual-level religion predictors. First, the level-2 controls do not change in their directions of effect or statistical significance (except for the "Child" coefficient). Second, although the religious social relationships and the religious organization variables are statistically insignificant, the index measure of shared religious beliefs is statistically significant and negatively related to acceptance of white-collar crime, which supports H1a. Third, there is noticeable variation in the acceptance of white-collar crime across religious affiliations. Except for those who did not provide an affiliation, all other affiliations, including Protestant, Catholic, Orthodox Christian, Muslim, Jew, 10 and Hindu, are significantly less likely to accept white-collar crime than Buddhists. This provides some support for H3 as Buddhists are predicted to have a higher acceptance of white-collar crime than Protestants, Catholics, Jews, Muslims, and Hindus. However, contrary to expectation, Orthodox Christians also have significantly lower levels of acceptance of white-collar crime than Buddhists. Fourth, the importance of God variable yields a statistically significant negative relationship with acceptance of white-collar crime, which is contrary to Stark's (2001) conditional and situational prediction for importance of God (H4 and H5). Finally, as shown in model 2, including the individual-level predictors greatly reduces the AIC compared to model 1 ( AIC = 1,647).
The country-level control variables and the country-level impersonal God measure are included in models 3 and 4, respectively. Model 3 shows statistically significant country-level effects that parallel other cross-national criminological studies using behavioral outcomes (e.g., homicide): (1) countries that have a greater percentage of women have significantly higher acceptance of white-collar crime (Messner and Rosenfeld 1997;Pratt and Godsey 2003), and (2) greater ethnolinguistic homogeneity in a country is significantly associated with lower acceptance of white-collar crime (Hansmann and Quigley 1982). Interestingly, the inequality and economic development measures are unrelated to the acceptance of white-collar crime. Finally, model 4 illustrates how our country-level impersonal God measure does not significantly affect the acceptability of white-collar crime nor does it alter the findings from previous models. As expected, the level-2 country variance (i.e., between-country) and the overdispersion variance (i.e., withincountry) decrease with the inclusion of the level-1 and level-2 controls and predictors (see the random-intercepts-only variances above for comparison).  In model 5, the slopes are allowed to vary across countries for both the Durkheimian and moral order variables. In other words, the model estimates a set of nested random-coefficient Poisson regression models where the fixed effect coefficients for the Durkheimian and moral order variables represent weighted averages of slopes across the 51 countries. Model 5 shows that all of the coefficients for the Durkheimian variables significantly vary across countries (providing preliminary support for H2ab and H6ab) and that the weighted average of slopes for shared religious beliefs remains statistically significant. Moreover, since the variance of the random intercept decreases with the inclusion of the random coefficients (model 4 vs. model 5), this suggests that some of the between-country variation in acceptance of white-collar crime is accounted for by the Durkheimian variables. Moreover, when we allow the importance of God variable's slope to vary randomly, its fixed effect remains statistically significant. In addition, the random coefficient for importance of God is statistically significant (providing preliminary support for H4 and H5) and decreases variation in the random intercept. This suggests that some of the between-country variation in acceptance of white-collar crime is accounted for by crosscountry differences in the importance of God variable. And finally, of the five models presented in Table 3, model 5 yields the lowest AIC.
The models in Table 4 test whether the Durkeimian and moral order terms vary with random slopes and with economic development. 11 That is, we explore cross-level interaction effects and allow the individual-level coefficients to vary across countries. This produces an intercepts-and slopes-as-outcomes Poisson regression model. Contrary to Durkheimian theory (see H2a and H2b), none of the effects of the Durkheimian religion variables are significantly attenuated as economic development increases (models 2-4). Moreover, the results also fail to support Stark (2001) (see H4), as the interaction between importance of God and economic development is not statistically significant (model 1). Interestingly, the variances of the random coefficients for the Durkheimian and moral order variables remain largely unchanged from Table 3 to Table 4. This suggests that some other contextual factor(s) account for the residual variation in the coefficients for the Durkheimian and moral order variables across countries. Table 5 presents models allowing the Durkheimian and moral order variables to vary with random slopes and with the country-level measure of an impersonal or amoral God (i.e., countries where the dominant religion is Eastern Orthodoxy or Buddhism). The negative effects of all the Durkheimian and moral order terms are significantly lower in societies characterized by majoritarian beliefs in an impersonal God compared to societies where this is not the case. This provides strong support for H5, H6a, and H6b. The significant interaction effects also account for nearly half of the cross-national variation in the slopes for all of the Durkheimian and moral order variables (without cross-level interaction effects, models 1-4 in Table 5 generate random slope variances of .002, .004, .07, and .009, respectively, for the key religion variables). In spite of this, there remains residual variation in the slopes that is left unexplained. Finally, it is important to note that model 5 (the fully saturated model) yields the lowest AIC of any model presented in Tables 3, 4, or 5. Figure 2 provides graphical illustrations of the relationships between acceptance of whitecollar crime and the key religion variables. Figure 2a presents the predicted probability of not accepting white-collar crime (i.e., zero-count acceptance of white-collar crime) by importance of God and societies characterized by majoritarian beliefs in an impersonal God. Moving to the right along the horizontal (or x) axis, the results indicate that white-collar crime becomes much more acceptable for those who find God important in countries characterized by belief in an impersonal God. Indeed, in countries characterized by belief in an impersonal God, the predicted probability of not accepting white-collar crime (i.e., zero-count acceptance of white-collar crime) for a person   Figure 2 Predicted probability of not accepting white-collar crime as a function of importance of God, religious social relationships, religious organization, shared religious beliefs, and beliefs in an impersonal God: (a) importance of God and impersonal God, (b) religious social relationships and impersonal God, (c) religious organization and impersonal God, and (d)

shared religious beliefs and impersonal God
Note: Predicted probabilities were calculated based on results presented in models 1-4 in Table 4, holding all other terms constant. All calculations are based on uncentered values.
who finds God to be very important is approximately 1.37 times lower than a similar individual in a country where the dominant religion professes belief in a moral, personal God. The other plots- Figures 2b, 2c, and 2d-demonstrate comparable conditional relationships between the Durkheimian variables and the country-level belief in an impersonal God (1.59, 1.59, and 1.42 times lower, respectively). These results provide important confirmation of the moderating effect of country-level belief in an impersonal God on the relationship between religion and acceptance of white-collar crime.

DISCUSSION AND CONCLUSIONS
White-collar crime has become an important political topic for a variety of reasons: the highly publicized cases of corporate crime and corruption; the global economic recession and the juxtaposition of bailing out Wall Street and not main street; and the occupy or 99 percent movement, all of which have propelled the issue of white-collar crime and corruption into the public forum and forced the issue onto the political agenda. Although we may not have declared a "war on white-collar crime," the increasing salience of the issue signals that the public has thought about white-collar crime more seriously in recent times, which, in turn, means that public attitudes will become more central in determining political responses to white-collar crime and corruption. Unlike attitudes about drug use and violent crime, which have been both shaped and fueled by the war on drugs and crime campaigns, little is known about how individuals form their opinions about white-collar crime. We do know that there is a negative relationship between religion and criminal behaviors and attitudes toward them (Baier and Wright 2001;Johnson et al. 2000;Johnson and Jang 2011). Yet, crime in these studies generally refers to juvenile delinquency, drug use, or general index crimes, leaving a dearth of research on the relationship between religion and attitudes toward white-collar crime.
To address this gap in the literature, we use Word Values Survey public-opinion data and employ multilevel modeling to test the relationship between religion and acceptance of whitecollar crime. Religious beliefs, whether measured as importance of God in one's life or as the Durkheimian shared religious beliefs index, are negatively related to the acceptance of whitecollar crime. By contrast, religious social relationships and belonging to religious organizations are unrelated to the acceptability of white-collar crime. The results also reveal variation in the acceptance of white-collar crime by religious affiliation. As predicted, Buddhists are significantly more likely to accept white-collar crime compared to nearly all other religious affiliations, including Orthodox Christians. The difference between Buddhists and Orthodox Christians may be due to the former tending to have conceptions of Gods as amoral and limited in power, and the latter believing in a God that, while "remote and inactive" (Stark 2001:627), is still considered moral. The moral nature of God might be more important than inactiveness and remoteness for influencing normative conformity. In short, religious belief and affiliation are important for understanding variation in the acceptance of white-collar crime.
The results provide mixed support for both Durkheimian theory and Stark's moral order hypothesis. Contrary to Durkheimian theory, we find no evidence for the hypothesis that individuals are socialized into societal norms through religious social ties and interactions. As Stark (2001) predicts, the social dimension of religion seems unnecessary for sustaining the moral order, or in this case, intolerance of white-collar crime. However, these results also fail to support Stark's assertion that only belief in a moral, active God produces normative conformity. Regardless of how religious belief was measured (i.e., by the Durkheimian shared religious beliefs index, importance of God measure, or religious affiliation), it was significantly related to the acceptance of white-collar crime.
The discrepant findings between different religion measures is interesting in light of Baier and Wright's (2001) meta-analysis of religion and crime studies, which finds that behavioral and belief measures of religion have equivalent negative effects on crime. A key difference between the studies in their meta-analysis and the current study is the investigation of acceptance of white-collar crime. Religion and crime studies have consistently found a stronger deterrent effect of religion on victimless crimes (e.g., drug use and gambling) than victimful crimes (e.g., theft and murder) (see Burkett 1980;Baier and Wright 2001). Burkett (1980) argues that this is due to both religious and secular social institutions strongly condemning victimful crimes, whereas religious institutions generally act alone in their condemnation of victimless crimes. While there is little known about the relationship between religion and white-collar crime, it may be the case that white-collar crime receives less attention from religious institutions and religious individuals compared to secular institutions, such as places of employment. If this is the case, then religious social ties, participation in religious rituals, and religious organizational involvement, consistent with our results, would not serve to reduce tolerance of white-collar crime. On the other hand, religious beliefs can provide extensive guidelines for moral behavior, including general support for criminal laws (Bainbridge 1989;Durkheim [1897Durkheim [ ] 1979Ellis 1985;Hirschi and Stark 1969;Stark 2001). This may facilitate intolerance of white-collar crime, regardless of whether such beliefs are actively reinforced by religious social relationships and organizations.
We also modeled religious belief at the contextual level by using a country's dominant religion-Eastern Orthodoxy and Buddhism-as a proxy variable for belief in an impersonal or amoral God (Stark 2001). Several studies include the proportion of the largest religious group in a geographic area as a measure of moral community or religious context and generally predict that the larger the market share of a religious group or the more religiously homogeneous a geographic area, the stronger the moral community (e.g., Moore and Vanneman 2003;Regnerus 2003;Stack and Kposowa 2006;Stark 1996). The country-level belief in an impersonal, amoral God variable, measured as countries where the largest religious group is Eastern Orthodoxy or Buddhism, is therefore a measure of moral community. Yet, this measure does not strengthen the negative effects of our individual-level religion variables, but instead weakens them. Moreover, the moral community control variable-religious homogeneity-did not produce significant direct or conditional effects (results not shown). This suggests that not all religious contexts are equal and the "moral" nature of religious communities-their ability to instill moral attitudes and ensure behavioral conformity-may depend on whether the religious community believes in a personal, moral God. Our findings contribute to an understanding of what types of religious contexts will tend to weaken the negative effects of individual-level religiosity contrary to the moral community hypothesis.
These contextual results further highlight the importance of religious beliefs for explaining the acceptance of white-collar crime. The country-level belief in an impersonal God measure conditions the negative effects of all the religion variables, including religious social relationships and belonging to a religious organization. Thus, the social dimension of religion only has a negative effect on the acceptance of white-collar crime in societies characterized by belief in a personal, moral God. This suggests that social integration into religious communities promotes normative conformity to the extent that the community supports belief in a moral, personal God. When investigating individual-level effects of religious social ties, memberships, and rituals on the acceptance of moral codes, it is vital for researchers to consider how these relationships may be entirely conditional on the types of beliefs held by the larger religious community.
While most research contextualizing religious effects on crime or attitudes toward crime has focused on aggregate religious contexts, we also contextualize religion's effect using a country's level of modernization and estimate interaction effects with our religion variables. We find, contrary to Durkheimian and Stark's (2001) predictions, that none of our religion predictors yield statistically significant interaction effects with modernization. The results suggest that religious contexts-as measured by country-level belief in an impersonal God-are more important for understanding the contextual effect of individual religiosity on acceptance of white-collar crime than secular contexts (i.e., modernization). Yet, even this measure of religious context does not fully account for all of the variation in individual-level religiosity slopes across countries. Future research should explore other unobserved contextual-level factors generating the variation in acceptance of white-collar crime across countries.
The present investigation is not without limitations. Our country-level sample is biased toward economically developed, Western Christian countries. Although a more economically and religiously diverse country sample is preferable, our results nevertheless offer insight into the generalizability of theories beyond what can be provided by one-country samples, which have dominated the literature (see Baier and Wright 2001). The sample is also cross-sectional, which does not allow us to examine how effects change over time as countries modernize.
Instead, consistent with analyses performed by Durkheim, Stark, and neo-Durkheimian studies, this study investigates the predicted relationships cross-sectionally with the expectation that if the hypotheses are correct, then a snapshot of countries at different stages of modernization should produce results that support the hypotheses. While different historical conditions may affect acceptance of crime rates in countries, as Stack and Kposowa (2006) note, one of the benefits of multilevel models is that they automatically account for unobserved heterogeneity in acceptance of crime that is due to between-country variation. Still, a disadvantage of this approach is that we cannot attend to historical influences outside of controlling for relevant structural factors (e.g., ethnic heterogeneity and inequality) and national residence through multilevel modeling. However, there are two important advantages associated with the use of a large crossnational and cross-sectional sample of individuals like the WVS. First, it allows for the testing of general models predicting attitudes toward white-collar crime to determine their cross-national generalizability, while at the same time providing a large enough sample to control for other important country-level factors. Although historical case studies are useful for providing rich detailed historical information and identifying underlying causal processes, these advantages make large-N analysis untenable. This makes it difficult to examine cases across a wide variety of structural conditions and limits the generalizability of the findings. Second, the use of a large cross-national and cross-sectional sample allows researchers to consider both individual and contextual effects across a variety of countries.
Another limitation of our study is that we cannot directly measure conceptions of God as such data are unavailable for a large cross-national sample. Thus, following Stark (2001), we use religious affiliation as a proxy measure at the individual and country level. At the country level, our measure captures societies where the dominant religion is Eastern Orthodoxy or Buddhism. Given the sample, this means that countries receiving a "1" for this variable are predominately postcommunist states (e.g., Belarus, Moldova, Romania, and Russia). The interaction effects may therefore represent the weakened effects of individual-level religiosity during communism (see Stack and Kposowa 2006;Stark 2001). However, as Stark (2001) notes, not all postcommunist states have Eastern Orthodoxy as the dominant religion. Because the "0" value for this variable also contains several postcommunist states (e.g., Albania, Hungary, Kyrgyzstan, Latvia, Lithuania, Poland, Slovania, and Slovakia), this strengthens the interpretation presented in this article, while encouraging further research in this area using more nuanced survey questions regarding how individuals view God (i.e., God as authoritarian, benevolent, critical, or distant) (see Bader and Froese 2005). While it has limitations, our analysis offers insights on how individuals come to understand and develop opinions about white-collar crime and corruption. This is important given that politicians, more often than not, enact policies congruent with public opinion (Burstein 2003;Monroe 1998;Stimson, Mackuen, and Erikson 1995). This study contributes to a broader understanding of how religion might indirectly shape punitive policy on white-collar crime through affecting individual attitudes.