Do Moral Communities Play a Role in Criminal Sentencing? Evidence From Pennsylvania

Abstract

Religion and social control have been a sociological concern since Durkheim and Weber, and the relationship between religion and punishment has long been the subject of speculation. However, surprisingly little empirical research exists on the role of religion or religious context in criminal justice, and almost no research on the role of religious context on actual sentencing practices. We conceptualize the potential relationships between religious context and sentencing severity by drawing from the focal concerns and court community perspectives in the sentencing literature and moral communities theory developed by Rodney Stark. We suspect that Christian moral communities might shape notions of perceived blameworthiness for court community actors. Such moral communities might also affect notions of community protection – affecting perceptions of dangerousness, or perhaps rehabilitation, and might influence practical constraints/consequences (e.g., local political ramifications of harsh or lenient sentences). We examine these questions using a set of hierarchical models using sentencing data from Pennsylvania county courts and data on the religious composition of Pennsylvania counties from the Associated Religion Data Archives. We find that county Christian religious homogeneity increases the likelihood of incarceration. In addition, Christian homogeneity as well as the prevalence of civically engaged denominations in a county condition the effects of important legally relevant determinants of incarceration. Furthermore, we find evidence that Christian homogeneity activates the effect of local Republican electoral dominance on incarceration. We argue that Christian homogeneity effects sentencing practices primarily through local political processes that shape the election of judges and prosecutors

Religion and the criminal justice system have been closely intertwined throughout the history of the United States. In the colonies it was the Quakers, led by William Penn, who first rejected the harsh criminal punishments favored in the English justice system. In the early 19^th century it was once again a group of prominent Quakers that lobbied for the development of penitentiaries; precursors to the modern U.S. prison. The Quakers believed that penitence and an understanding of one's “sins” would come only through the solitude and reflection offered by solitary confinement. Upon reflection it should not be surprising that religion and the criminal justice system are often connected. After all, both social institutions revolve around the concepts of social control and the maintenance of normative community. The modern criminal justice system seeks to maintain social control via the threat of legal sanctions and punishment. Religion has the capability to impose this-worldly and “otherworldly” sanctions upon believers to maintain the moral order.

By providing a unified system of morals, codes, beliefs and practices, Durkheim (1985, p. 121) argues, religion has the unique power to unite “into one single moral community...all those who adhere.” But while a wealth of studies have found that religion inhibits deviant behavior (e.g., Benda 1997; Evans et al. 1996; Grasmick, Kinsey, and Cochran 1991; Jang and Johnson 2001; Johnson et al. 2000a; Pearce and Haynie 2004), the relationship between religion and the justice system is severely understudied at the community or ecological level (see Savelsberg 2004). We only know of one published study of the effect of community religious contexts and sentencing outcomes (Fearn 2005).

In this study, we examine the impact of religious context on sentencing practices in Pennsylvania county courts. To do so, we use hierarchical models to determine if and how county religious contexts influence incarceration. We begin by discussing current theoretical conceptualizations of the sentencing process, and by presenting an overview of previous research relevant to the issue of religion and sentencing/punishment. Based on our review of existing research, we propose three different ways in which religion might impact the criminal sentencing process—through conservative religious ideologies, via religion's ability to serve as a “moral community,” and through the activities of civically engaged religious groups.

Contextual Influences on Criminal Sentencing: Courts as Communities

The focal concerns/court community perspective emphasizes particular kinds of substantive rationalities (Savelsberg, 1992) at work as court actors make sentencing decisions (see Steffensmeier et al. 1998; Steffensmeier and Demuth 2000; Kramer and Ulmer 2002; Engen et al. 2003; Ulmer and Johnson 2004 for more detailed explanations of this perspective). The focal concerns perspective argues that court actors orient toward three focal concerns when making sentencing decisions: perceived blameworthiness, protection of the community, and practical constraints and consequences of sentencing decisions (e.g., impact on victim or offender's family, offender health/age, correctional resources, political ramifications, interorganizational relationships, etc.). This focal concerns model parallels and partially incorporates Albonetti's (1987) uncertainty avoidance theory of criminal case processing and her causal attribution theory of judicial discretion (1991).

According to Albonetti (1991), sentencing reflects the use of bounded rationality (March and Simon, 1958), in which court actors make highly consequential decisions with insufficient information, which produces uncertainty. Even when more extensive information is available, the risk and seriousness of recidivism is never fully predictable, a defendant's moral character is never fully knowable, and human decision-making processes have built-in limitations to the amount and complexity of information that can be considered. In this context, judges and other court actors make situational imputations about defendants’ character and expected future behavior (Steffensmeier 1980; Farrell and Holmes 1991), and assess the implications of these imputed characteristics in terms of three focal concerns: defendant blameworthiness, community protection, and practical constraints and consequences connected to the punishment decision. These situational definitions of defendants vis-a-vis the focal concerns influence sentencing decisions.

The three focal concerns are in turn embedded in local court communities (Ulmer and Johnson 2004; Fearn 2005). This perspective views courts as communities (Eisenstein et al. 1988), or distinctive social worlds (Ulmer 1997; Kautt 2002), based on participants’ shared workplace, interdependent working relations between key sponsoring agencies (prosecutor's office, bench, defense bar), and local cultures. These distinctive local cultures shape formal and informal case processing and sentencing norms, and thus produce variation in case processing and sentencing outcomes (see Eisenstein et al. 1988; Ulmer and Kramer 1996, 1998; Ulmer 1997; Dixon 1995). Court communities are said to foster their own locally varying substantive rationalities (Savelsberg 1992; Ulmer and Kramer 1996) which shape sentencing outcomes and processes at least as much as formal policies and legal structures (see also Kautt 2002). Such local variations are especially likely in states (such as Pennsylvania) where judges and prosecutors are locally elected. Furthermore, court communities typically have locally distinctive, informal, and evolving case processing and sentencing norms, or “going rates” (Eisenstein et al. 1988; Ulmer 1997). These going rates often provide members of courtroom workgroups with “templates” for case processing strategies, typical plea bargaining terms, and sentences. In other words, this court community perspective embeds individual court decisions and decision makers in organizational and community contexts (see Ulmer and Johnson 2004; Kautt 2002).

Jenness and Grattet (2005) use the term “perviousness” to refer to the nexus between criminal justice organizations and their social environments. The court community perspective expects that courts will be “pervious” or permeable to the surrounding social environment. As Kautt (2002, p. 642) puts it, the embeddedness of sentencing law, policies, and processes in the social worlds of court communities “suggests that the impact of case-level factors should be conditioned by the characteristics of the court in which a case is adjudicated.” Thus, the court community perspective not only implies that sentence severity will vary across courts, but also that the effects of key predictors will also vary. This is because sentencing processes and workgroup members’ interpretations of sentencing criteria, such as focal concerns of sentencing, are embedded in court community culture, which is in turn influenced by the culture, politics, and social structure of the surrounding community (Ulmer and Kramer 1998; Ulmer and Johnson 2004). Thus, as we explain below, a county's religious context might influence the severity of sentencing, and condition the effect of key determinants of sentencing, such as offense severity and prior criminal record.

The lines of influence from community to court actors can be fairly direct. For example, in Pennsylvania, judges and prosecutors are locally elected in partisan campaigns, and are usually locals themselves (Ulmer 1997; Levin 1977). They are sensitive to local media scrutiny and the need to maintain legitimacy with the local public (Eisenstein et al. 1988). Defense attorneys also typically come from the surrounding local community (Flemming et al. 1992). Court community actors know and relate to one another (with varying degrees of cooperation or conflict), and fairly high attitudinal/normative homogeneity among court community participants, while not necessarily the rule, is not uncommon (see Ulmer 1997; Flemming et al. 1992). Thus, it is not far-fetched to suspect that religious contexts are one community feature that influences court communities and the severity of their punishments (Fearn 2005).

What Do We Know About Religion and Criminal Punishment?

The great majority of the myriad studies of criminal sentencing focuses on investigating various forms of extralegal disparity at the individual defendant level (for examples of reviews, see Hagan and Bumiller 1983; Spohn 2000; Zatz 2000; Steffensmeier and Demuth 2000; Engen et al. 2003; Steen et al. 2005). This literature generally finds that offense severity/type, prior criminal record, and other legally relevant variables are almost always the strongest predictors of sentencing outcomes. In addition, case processing variables such as mode of conviction are often important, with defendants convicted by trial receiving more severe sentences than those who plead guilty, controlling for legally relevant factors (see Ulmer and Bradley 2006; King et al. 2005). The literature presents a mixture of findings about the degree to which extralegal defendant characteristics, such as socioeconomic status, race/ethnicity, gender, or age, affect sentencing outcomes directly or in combination with one another. In particular, some studies find that blacks and/or Hispanics, males, and defendants in their 20s are sentenced more harshly (see Zatz 2000; Spohn 2000; Steen et al. 2005).

However, a small but quickly growing literature on social contexts and sentencing has recently developed. Just as criminology has recently focused further attention on community and ecological contexts and crime, recent studies on sentencing have focused on how the community and ecological contexts of courts affect sentencing outcomes, and have done so with methods appropriate to multilevel data (such as hierarchical modeling) (e.g., Britt 2000; Kautt 2002; Ulmer and Johnson 2004; Wooldredge and Thistlethwaite 2004; Weidner et al. 2005; Fearn 2005; Johnson 2005; Ulmer and Bradley 2006).

In all these studies, individual case level factors explain most of the variation in sentencing outcomes, with between 5-11% of the variation in sentencing reported as existing at the jurisdictional level (depending on the study). This is not surprising--if contextual factors outweighed individual defendant and case factors like offense characteristics, criminal history, and mode of conviction (guilty plea vs. trial) in explanatory power, it would be truly remarkable and would have alarming implications for the nature of American criminal justice. However, some jurisdictional characteristics have been shown to significantly affect sentencing, such as racial or ethic composition, court size, court caseload, and crime rates. In addition, studies have found that the effects of individual case- or defendant-level sentencing predictors vary significantly between jurisdictions, and have found meaningful interaction effects between individual level and contextual factors (e.g., Kautt 2002; Ulmer and Johnson 2004; Ulmer and Bradley 2006).

Largely neglected in the literature on sentencing decisions, however, are the potential effects of religion. Consequently, researchers have limited guidance in developing relevant hypotheses. We propose three theoretical arguments for why we should expect a relationship between conservative religion and harsher attitudes regarding punishment: 1) conservative theology and punitiveness, 2) moral communities, and 3) civically engaged denominations.

Conservative Religion and Punitive Attitudes

As noted previously, we know of only one macro-level study that investigates the extent to which local religious contexts might affect sentencing (Fearn 2005). Fearn (2005) did not primarily focus on the effects of religious contexts on sentencing, but did include a county percent conservative Protestant measure as a predictor of sentencing outcomes in models using state court processing data from the 75 most populous counties in the US. She found no significant relationship between the local presence of conservative Protestants and the likelihood of offenders being sentenced to non-incarceration or to county jails. However, the likelihood of being sentenced to state prison rather than local jail increased with the local proportion of conservative Protestants.

Although they provide only limited guidance for understanding the role of religious context in sentencing decisions, there have been several studies of religion and attitudes towards punishment using survey data. Such research has found little or no difference between Catholics, Protestants and sometimes Jews in attitudes towards capital punishment (Hindelang 1974; Tyler and Weber 1982), preferred sentence severity (Blumstein and Cohen 1980; Hindelang 1974) and belief that courts are too lenient (Flanagan and Jamieson 1988).¹ Research that has specifically focused on conservative Protestant denominations has found a religion effect on attitudes towards punishment. For example, Myers (1988, 1989) finds that judges in Georgia affiliated with conservative Protestant denominations were less likely to grant probation. Grasmick et al. (1992) find a significant positive relationship between membership in a “fundamentalist” denomination and preference for retributive punishment for criminals (cf. Young 1992).

Some argue that “fundamentalists” are more punitive because they model the wrathful God portrayed in the Old Testament (see for example: Cook and Powell 2003; Garland 1990).² Those with a literal view of the Bible are said to accept passages about God's harsh judgment literally, applying that same judgment to contemporary deviance. Such beliefs, Lupfer, Hopkinson and Kelley (1988) argue, lead religious conservatives to place more blame on the individual for deviant behavior (cf: Lupfer and Wald 1985). This focus on the individual may lead religious conservatives to be more punitive in attitudes towards punishment. At the aggregate level this line of research would predict that higher rates of conservative Protestants would lead to a more punitive mentality and harsher sentences in the local justice system (see Fearn 2005).³

Such research suggests that it is the type of religion prevalent in an area (more specifically, the presence of conservative Protestants), that should impact criminal sentencing. This leads to hypothesis 1:

Hypothesis 1: The conservative Protestant adherence rate will have a significant impact on sentencing at the county level.

Moral Communities

While some research suggests that the type of religion prevalent in an area will impact sentencing decisions, a Durkheimian perspective would argue that majority religion, regardless of type, will produce a set of shared values and motivate adherents to punish those who stray from them. Drawing upon Durkheim's observations about religion's ability to produce conformity, Stark, Doyle and Kent (1980) developed the moral communities hypothesis. As originally formulated, the moral communities hypothesis argues that the effect of an individual's religious commitment on delinquent behavior depends upon the degree to which religion permeates the culture surrounding that individual (see Stark, 1984; 1996; Stark and Bainbridge 1996; Stark, Doyle, and Kent 1980). ⁴

Since “moral communities” are ones in which this religion permeates social life, this implies that attempts to capture these moral communities should go beyond measuring the proportion of a community that belongs to this or that Christian group. It seems to us that the moral communities concept implies a significant degree of religious consensus and homogeneity in a community, and it would be this local religious homogeneity that sets the context for social control decisions like sentencing. For example, Unnever and Cullen (2006) speculate that fundamentalist support for the death penalty might be stronger in areas exhibiting more homogeneous “moral” communities compared to communities with more religious diversity. In other words, they suspect that homogenous moral communities might mobilize stronger death penalty support among fundamentalists. Extending this logic, to the extent that members of a moral community share and agree upon religious “values,” we would expect the community to be more punitive towards those who reject those values through criminal acts, regardless of the denominational composition of the community. This leads to hypothesis 2:

Hypothesis 2: The level of religious homogeneity (regardless of the specific type of religion) will have a significant impact on sentencing at the county level.

Civically Engaged Denominations

Lee and Bartkowski (2004) draw upon the moral communities literature in their study of juvenile homicide rates. However, they modify the hypothesis by arguing that it is the presence of civically-engaged denominations that should impact crime rates. Per Tolbert, Lyson and Irwin (1998) a civically engaged denomination is a religious group actively involved in community programs that do not involve recruitment/conversion that help strengthen bonds between community members and build trust. Lee and Bartkowski (2004) find the prevalence of civically engaged denominations to be negatively related to juvenile homicide in rural areas. They argue that civic engagement is the mechanism by which religion creates a moral community. When a large proportion of a community is involved in denominations actively engaged in community programs, “horizontal social networks may be strengthened, normative consensus on acceptable and unacceptable behaviors may be elevated...and the community's ability to express and pursue collective goals may be bolstered” (Lee and Bartkowski 2004, p. 1003, our emphasis).

Although there have been no studies that examine the relationship between civically engaged denominations and criminal justice outcomes, we follow Lee and Bartkowski's (2004) logic and test whether deviants will be treated more harshly in areas with high levels of adherence to such denominations. Hypothesis 3 follows:

Hypothesis 3: The adherence rate of civically engaged denominations will have a significant impact on sentencing at the county level.

In addition to testing these hypotheses, we explore two sets of cross-level interaction effects involving religious context variables and two key legally relevant individual level predictors. First, it is plausible that the local religious context might condition the impact of the legal culpability of offenders, or of likely indicators of perceived dangerousness or rehabilitative potential, on sentencing. Thus, we also determine if offense seriousness has a greater or lesser impact on sentencing decisions across our religious contexts. In Pennsylvania, definitions of legal blameworthiness are structured by the Pennsylvania sentencing guidelines, which rank all offenses on a 14 point severity scale,⁵ yet the guidelines allow significant local court discretion in the decision to incarcerate and, to a lesser extent, in deciding sentence lengths. It is plausible that court community actors in counties with greater proportions of conservative Protestants might be especially tough on more severe offenders, seeing them as more blameworthy and more deserving them of moral condemnation. Or perhaps the shared values fostered by religious homogeneity or civically engaged denominations will lead to harsher treatment of those who transgress serious social norms by committing serious crimes.

Second, criminal history has long been used formally and informally by criminal justice actors as a predictor of offenders’ risk of future crime and/or rehabilitative potential (see Albonetti 1991; Steffensmeier et al. 1993). In addition, there is a moral dimension to a criminal record (von Hirsch 1976; Kramer and Scirica 1986). Those who repeatedly commit crimes are often seen as “incorrigible,” as less deserving of leniency in a moral sense. Pennsylvania's sentencing guidelines rank the criminal histories of offenders on a weighted prior record score, enabling us to examine how the effect of prior criminal record varies between county courts.

We find it plausible that religious context could interact with offenders’ criminal records in affecting incarceration. Court actors in more conservative religious counties might see offenders with more extensive criminal histories as more blameworthy, in addition to being a threat of future crime and as undeserving of a more lenient sentence. Alternately, Applegate et al. (2000) and Unnever et al. (2005) found that conservative Christians’ criminal justice attitudes were more punitive, but at the same time also emphasized forgiveness and compassion. It is therefore possible that criminal history could have less of an impact on incarceration in counties with a preponderance of conservative Protestants. In addition, the shared values created by religious homogeneity or religious civic engagement might condition the effect of criminal history on sentencing in similar ways.

To review, court actors’ interpretations of focal concerns regarding sentencing are seen as embedded in local court communities, which are characterized by distinct local organizational relations, political environment, and culture. We are interested in the permeability or perviousness (Jenness and Grattet 2005) of court communities and their sentencing decisions to the religious context of their surrounding local environments.

Data and Methods

We test these expectations using individual-level sentencing data from county criminal trial courts⁶ in Pennsylvania, along with aggregate county characteristics and county religious context variables. The criminal sentencing data we employ span three recent years (1997-2000) and come from the Pennsylvania Commission on Sentencing (PCS, see http://pcs.la.psu.edu/). The PCS is required to collect information on all misdemeanors and felonies sentenced in the state (except for first and second degree murder, which are not covered by the guidelines since they are punishable by life imprisonment or death). Cases in this analysis were limited to the most serious offense per transaction and to those sentenced under the 1997 version of the Pennsylvania sentencing guidelines.

Pennsylvania has operated under a sentencing guideline system⁷ since 1982, so it presents a potentially strict test for the presence and strength of contextual variation in sentencing relative to non-guideline jurisdictions. Sentencing guidelines quantify and standardize sentencing decision criteria (e.g. offense severity and prior record), mandate court consideration of these criteria, and recommend a uniform matrix of sentence ranges. The presence of these sentencing guidelines, and the fact that they have been in place for nearly twenty years, may therefore foster sentencing uniformity. However, Pennsylvania is also a very diverse state, characterized by wide variations in local contextual characteristics. For example, Pennsylvania is home to two of the largest metropolitan areas in the U.S. (Philadelphia and Pittsburgh), but is also home to numerous medium-sized cities (e.g., Harrisburg, Erie, Reading, Scranton, Allentown) and a large number of small rural counties. The state is politically diverse as well, with the eastern and western ends of the state having a stronger organized labor presence, more racial, ethnic, religious, and cultural diversity, and a higher percent of voters registered as Democratic. Conversely, central Pennsylvania's counties tend to be more conservative, less diverse, and dominated by Republican voters. Furthermore, they differ substantially in their religious contexts (as our descriptive statistics later show), such as their rates of conservative Protestant adherents, adherents to civically engaged congregations, and Christian heterogeneity.

Dependent Variables

To gauge the influence of our three different measures of religious context, we examine the decision to incarcerate. Pennsylvania courts are given wide discretion regarding the incarceration decision under state sentencing guidelines. Courts may decide to sentence an offender to a term in a county jail or state prison or choose an alternative such as probation, intermediate sanctions, restitution or a diversionary program. If a certain type of religious context results in the harsher treatment of criminals, we should expect courts within such contexts to be more likely to incarcerate offenders.

Incarceration was coded 1 if the offender was sentenced to any length of confinement in a county jail or state prison and coded 0 if they were sentenced to any combination of non-incarceration options (e.g. probation, restitution, intermediate sanctions, etc.). We also explored other dependent variables, replicating the analyses we present below with sentence length and state imprisonment (as opposed to county jail) as sentencing outcomes. Our religious context variables of interest failed to exert any significant direct effects on sentence length (although the relationships were in the same directions as those reported below in the incarceration analysis), and no meaningful interaction effects. Essentially, then, our analyses produced null findings with regard to any effect of religious context on sentence length. Results are available from the first author on request.

There could be several reasons for the lack of a sizeable effect of religious context on sentence length. First, in the sentencing literature extralegal variables tend to affect the incarceration decision to a greater extent than sentence lengths (Spohn 2000; Zatz 2000). Second, Pennsylvania's sentencing guidelines arguably constrain the sentence length decision more than the decision to incarcerate. That is, courts have the discretion to incarcerate offenders or sentence them to some non-incarceration alternative (e.g., probation, intermediate punishments) throughout most of the guidelines grid. However, if incarceration is selected, sentence lengths are bounded. Third, the most qualitatively salient decision of punishment may be to send someone to jail or prison at all, rather than some non-incarceration option like probation or intermediate punishments (Holleran and Spohn 2004; Steffensmeier et al. 1993). Thus, incarceration might be a punishment threshold that is particularly meaningful and indicative of punitiveness. Religious context might affect the incarceration decision, while the length decision is primarily driven by other factors.

Previous research suggests that is it useful to further differentiate between state prison and county jail in modeling the incarceration decision (Ulmer 1997; Holleran and Spohn 2004), especially since Fearn (2005) found differences in the effects of percent conservative Protestant on the jail versus the state prison decision. In separate analyses, we examined hierarchical logistic regression models identical to the ones presented later except the dependent variable was state prison incarceration (1) vs. county jail and non-incarceration (0). Our religious context variables's effects were not statistically significant, although their signs were all in the same directions as reported below, and the effect for the Christian heterogeneity measures bordered on statistical significance (p<.10). Results are available from the first author on request.

In order to determine if there were notable differences in the effect of our measures on not only incarceration as a dichotomous outcome, but also on broken out as jail versus prison sentencing, we ran HLM multinomial regression models with a sentencing outcome variable containing three categories: no incarceration, jail, and prison (as suggested by Holleran and Spohn 2004). However, the prison sentences were too sparsely distributed across the counties. A check of crosstabulations showed that in three small counties, there were less than twenty state prison sentences. In 19 of the 60 counties, fewer than 50 people were sentenced to prison, while many of these same counties sentenced hundreds to jail time. This simply spread the data too thin (i.e., nine to eleven level 2 predictors, 60 counties, and many small counties with few prison sentences), and coefficients were very unstable.

Independent Variables

Individual Level Predictors

We employ several individual case and contextual level factors as controls, as suggested by the prior sentencing literature. The legally relevant sentencing variables that we include in the analyses are the severity of the current offense, the offense type, the prior criminality of the offender, and the presumptive guideline sentence recommendation. The Pennsylvania sentencing guidelines provide direct measures of both the severity of the offense and the prior criminality of the offender, and these are supposed to be the primary determinants of sentencing decisions. Offense severity is measured by the Offense Gravity Score (OGS). The OGS itself ranges from 1 to 14 (14 is the most serious). The OGS rankings by the PCS incorporate such considerations of harm and loss to victims and society, victim vulnerability characteristics, drug amounts for drug trafficking cases, as well as the state's overall judgment of criminal wrongfulness and culpability attached to a crime. Offense type is measured with dummy variables, coded 1 if the offender was convicted of a particular offense, and 0 otherwise (theft serves as the reference category in our models). It is important to note that offense severity can vary within these offense types as well as between them, so offense type and severity are not redundant. The offender's prior record or criminal history is measured by the Prior Record Score (PRS), a weighted eight-category scale, with the last two categories reserved for repeat felons and repeat violent offenders. This scale represents an offender's past convictions for misdemeanors and felonies, as well as certain juvenile adjudications, and is weighted by the severity of prior convictions.

Albonetti (1998) and Engen and Gainey (2000) argue for the importance of including the presumptive guideline sentence recommendation as a further control when examining extra-legal effects, so we include it here as well. The presumptive guideline sentence recommendation variable provides a measure of what the sentencing guidelines indicate is an appropriate sentence. We coded the variable presumptive disposition as 1 if the applicable guideline range minimum recommends a period of state prison or county jail incarceration and 0 otherwise (Ulmer 2000).

We also included the race/ethnicity, gender, and age of the offender. We created dummy variables for our racial/ethnic and gender distinctions. “Black” was coded 1 if the offender was African American and 0 otherwise. Similarly, “Hispanic” was coded 1 if the offender was Hispanic and 0 otherwise. “Female” was coded 1 if the offender was a woman and 0 otherwise. Age was coded as the number of years of the offender at the time of sentencing. We coded mode of conviction with two dummy variables, coded 1 if the offender was convicted through a bench or jury trial, and coded 0 if the offender pled guilty.

Contextual Variables

We include several county and court characteristics as controls when examining the effect of religious context measures on sentencing. First, caseload is the mean of the annual number of cases processed by a given court from 1997-2000. Other contextual variables include percent black (coded from the 1998 Current Population Survey) and percent Republican voters. The latter is a particularly important control variable because we want to distinguish the effects of religious context per se from the effects of political ideological context. We also control for court size (as defined by number of judges and share of state caseload, for details see Ulmer 1997), with dummy variables for medium and large county courts. In particular, the state's two largest counties, Philadelphia and Allegheny (Pittsburgh), are quite different from the rest of the state in size, other social and cultural characteristics, and sentencing patterns (Kramer and Ulmer 2002). Together, these two counties also account for 28% of the state's criminal court caseload. We want to ensure that any contextual effects we find are not confounded by effects simply attributable to these two large and distinctive metropolitan counties. By controlling for court size, we can be more confident that county level effects we find are not simply driven by idiosyncratic Philadelphia or Pittsburgh sentencing patterns.

Other contextual controls we included in supplemental analyses are county poverty rates, percent Hispanic, county unemployment rates, and UCR crime rates. These exhibited high multicollinearity (with correlations in the .80 - .90 range) with each other and with court caseload and percent black. Because we were conceptually more interested in controlling for percent black (as a proxy attempt to address the county presence of African American conservative Protestant adherents, whose political attitudes are quite different from white conservative Protestants—see footnote 10) and percent Republican, we therefore omitted the other county level control variables in our final models. We retained court caseload and court size as controls, however, because these variables have been shown to significantly influence sentencing in previous research (Ulmer and Johnson 2004). While the omission of these contextual control variables deserves caution, we felt that the level of multicollinearity, plus the need to have parsimonious models that preserved statistical power clearly warranted removal of some variables that were either collinear with others, or did not have significant effects. In order to adequately deal with the potential presence of African American conservative Protestants, we needed to retain the percent black variable, thus creating the need to exclude the poverty and unemployment rates (which did not significantly predict sentencing outcomes in any case). Similarly, the severity of punishment may be quite dependent on the court caseload (Ulmer and Bradley 2006; Ulmer and Johnson 2004), therefore creating the need to omit the variable highly correlated with caseload, the UCR crime rates for the counties (which was also highly correlated with percent black). Importantly, the addition or omission of our extra level two predictors did not substantively alter the effects reported below for the religious context variables of interest.

Using the typology developed by Steensland et al. (2000), we created rates of adherents per 1,000 for mainline Christian denominations (e.g., Episcopalian, Lutheran, Methodist, Presbyterian), conservative Protestant denominations (e.g., various Baptist groups, Assembly of God, Church of Christ), Roman Catholics, and a combined measure of total adherents to all denominations. Conservative Protestant denominations (labeled “Evangelical” in the Steensland et al. typology) share certain characteristics, including a tendency to view the Bible in more literal terms (i.e., fundamentalism) and a focus upon “born again” experiences and personal responsibility for salvation (Emerson and Smith 2000). Therefore, we use the conservative Protestant adherence variable to test that the prevalence of conservative/fundamentalist Christians will result in harsher sentencing practices.

The religious context variables were developed using data from the 2000 Religious Congregations and Membership Study (RCMS) available from The Associated Religion Data Archives (www.thearda.com). Collected by the Association of Statisticians of American Religious Bodies (ASARB) and funded by the Lilly Endowment, the RCMS study gathered data on 268,254 congregations that included 139 Christian denominations and six Eastern religions. Estimates for Jewish and Islamic groups are also included in the data (Jones et. al, 2002). Since most Pennsylvania counties have very small percentages of non-Christian religious adherents, however, we focus only on the Christian groups. In supplemental analyses, we included percent Jewish, percent Islamic, and percent other religions, but none of these exhibited statistically or substantively significant effects, and did not alter the effects of our other religious context predictors of interest. We also included a measure of total Christian adherents in models, both separately and together with the other religious context measures, but it never attained substantive or statistical significance.

Per Durkheim, a moral community could be composed of any type of majority religion. In other words, there is no theoretical reason why a Muslim moral community should not exist. However, given that the United States, and Pennsylvania in particular, is predominately Christian, our moral communities variable is an index of dissimilarity for county Christian homogeneity/heterogeneity, calculated from the rates of membership for three groups: mainline Christian, conservative Protestant, and Roman Catholic. The index is calculated as follows, where H represents the index of religious heterogeneity/homogeneity:

$$\mathrm{H}=1-({(\text{conservative Protestant rate}∕\text{total adherents rate})}^2+{(\text{mainline Protestant rate}∕\text{total adherents rate})}^2+{(\text{Catholic rate}∕\text{total adherents rate})}^2)$$

This index varies between 0 and 1, and in our data, high scores indicate counties with more heterogeneous populations with respect to the three Christian groups, and low scores indicate more religious homogeneity.⁸ In other words, this measure indicates the degree to which the Christians in a county are members of the same or different groups among the three Christian groupings.

We created our measure of adherence to civically engaged denominations by following Tolbert, Lyson and Irwin (1998). Using the General Social Survey, Tolbert, Lyson and Irwin (1998) determined the number of voluntary associations to which respondents claimed affiliation. They then compared mean levels of membership by denomination. They considered those denominations whose adherents reported above average involvement in voluntary associations to be civically engaged. With this method, Tolbert, Lyson and Irwin (1998) identify twelve civically engaged denominations.⁹ Using this list we calculated the number of adherents per 1,000 to civically engaged denominations.¹⁰

Some limitations of our data are that they lack measures of defendants’ socioeconomic status, type of defense attorney, pretrial release status (bail), or detailed information on the status of victims. Unfortunately, these are limitations shared with the majority of sentencing studies (see review by Zatz 2000). These variables are not collected by the PCS. More importantly for this study, we have no information on the sentencing judges’ (or prosecutors’) religious beliefs or ideologies of punishment (for an analysis of the importance of judge characteristics in sentencing, see Johnson 2006). This means we cannot explore the relative importance of key court decision makers’ religious view or ideologies, versus the religious context of the surrounding county, or the possible interrelationships between these two sets of factors.

Analytical Techniques

The logic of our analysis requires a multilevel analysis, and hierarchical modeling (HLM) is a widely used method for this throughout sociology, psychology, and other fields. The importance of inter-court variation in sentencing outcomes and the factors that affect them, and the desirability of using HLM for analyses of sentencing outcomes has by now been well established (see Britt 2000; Kautt 2002; Ulmer and Johnson 2004; Wooldredge and Thistlethwaite 2004; Fearn 2005). For our analyses, we used HLM 6 software (Raudenbush et al. 2004). We used hierarchical logistic regression (Bernoulli distribution) for incarceration decisions (for similar applications, see Ulmer and Johnson 2004; Wooldredge and Thistlethwaite 2004; Kautt 2002; Britt 2000). Our individual-level effects are centered around their grand means. Grand mean centering risks the introduction of estimation bias in the individual level effect because this coefficient is a weighted combination of the between and within county effects. While group mean centering provides an unbiased estimator for the individual level effect (Bryk and Raudenbush 1992: 117-121; Britt 2000), though, it artificially constrains county level compositional differences, thereby complicating assessment of sentencing variation across counties. Given the present focus on these county level differences, we elected to grand mean center our individual level predictor variables.

Findings

The descriptive statistics for our dependent variable, individual case level variables (level 1), and county contextual variables (level 2) are shown in Table 1. A correlation matrix is available upon request.¹¹

Table 1.

Descriptive Statistics

Dependent Variables	Freq. (%)	Mean (S.D)
Incarcerated	61,052 (49.3)
Level 1 Independent Variables	Freq. (%)	Mean (S.D)
Black	45,758 (36.9)
Hispanic	9,817 (7.9)
Bench Trial	4,264 (3.4)
Jury Trial	2,038 (1.6)
Female	20,391 (16.5)
Homicide	918 (0.7)
Aggravated Assault	3,851 (3.1)
Simple Assault	18,936 (15.3)
Arson	443 (0.4)
Burglary	6,340 (5.1)
Criminal Trespassing	3,704 (3.0)
Drug Felony	21,309 (17.2)
Drug Misdemeanor	18,569 (15.0)
Forgery	4,046 (3.3)
Weapons	3,959 (3.2)
Robbery	4,531 (3.7)
Theft	36,802 (29.7)
Rape/IDSI	599 (0.5)
Presumptive Disposition	50,532 (40.8)
Offense Gravity Score		4.04 (2.30)
Prior Record Score		1.50 (1.93)
Age		29.83 (9.70)
Guideline Minimum		5.12 (11.80)
Level 2 Dependent Variables	Freq. (%)	Mean (S.D)
Medium Court	14 (23.3)
Large Court	2 (3.3) *
Percent Black		3.38 (6.19)
Percent Republican		44.55 (8.09)
Mean Annual Caseload		2,271.07 (3,031.04)
Mainline Denominations – Rates of		208.62 (58.79)
Adherence per 1000 population
Evangelical Denominations – Rates of		71.04 (39.00)
Adherence per 1000 population
Catholic Church – Rates of Adherence per		228.16 (130.95)
1000 population
Civically Engaged Adherence Rate		376.02 (860.38)
Logged Civically Engaged Adherence
Rate		4.91 (1.46)
Index of Religious Heterogeneity		0.58 (0.06)
Heterogeneity, 1st quartile	15 (25.0)
Heterogeneity, 2nd quartile	15 (25.0)
Heterogeneity, 3rd quartile	15 (25.0)
Heterogeneity, 4th quartile	15 (25.0)

^*-- These two large counties also account for 28% of the state’s criminal court caseload.

Notably, Roman Catholics are the most populous Christian denomination in Pennsylvania counties, followed closely by mainline Protestant denominations. Conservative Protestant Christians are third, with a rate of about 71 per 1000 county residents. The mean for the index of Christian heterogeneity is .58, with a fairly narrow range of variation (the minimum is .41 and the maximum is .67, with two thirds of counties falling between .52 and .64). As we explain later, we also categorize the index of Christian heterogeneity into four quartiles. The first quartile is made up of the most homogeneous counties, while the fourth quartile is made up of the most heterogeneous counties. The civically engaged adherents variable is highly left-skewed, so we included the log (ln) of this variable in our models, rather than the raw rate variable.

Before estimating our main models, we estimated an unconditional model (available upon request) with no predictors. The level two variance component was .22, and was statistically significant a p < .001, indicating that incarceration varied significantly between counties and that HLM was appropriate.¹² The amount of level two variation is not the only consideration in deciding the appropriateness or relevance of multilevel modeling; rather, we are also conceptually interested in the degree to which county religious characteristics condition certain effects at level one (offense severity and prior record). Because we are primarily interested in county level effects here, we present fixed effects models that do not allow individual level effects to vary randomly between counties.

The results of our hierarchical logistic regression models of incarceration are shown in Table 2.

Table 2.

Hierarchical Models of Incarceration

Models	Incarceration
	(Odds Ratio)
	(1)	(2)
Intercept (B0, G00)	113.13***	14.32**
Level 1 Predictors
Offense Gravity Score	1 43 ***	1 43 ***
Prior Record Score	146***	146***
Age	99***	99***
Black	1.51***	1.51***
Hispanic	1 47***	1 47***
Bench Trial	1.55***	1.55***
Jury Trial	3.25***	3.25***
Female	64***	64***
Homicide (not including murder I or II)	181 ***	1.81 ***
Aggravated Assault	144***	144***
Simple Assault	.97	.97
Arson	1.13	1.13
Burglary	1 53 ***	1 53 ***
Criminal Trespassing	1.50***	1.50***
Drug Felony	.97	.97
Drug Misdemeanor	.56***	.56***
Forgery	1.22***	1.22***
Weapons Violation	81 ***	81 ***
Robbery	2 77 ***	2 77 ***
Rape/Involuntary Deviant Sexual Intercourse	1.45	1.45
Presumptive Disposition	1 49***	1 49***
Level 2 Predictors
Percent Black	.99	.99
Percent Republican	.99	.99
Medium Court	1.87*	2.19*
Large Court	1.90*	3.11*
Mean Caseload	999**	999**
Mainline Rate	.997	.997
Conservative Prot. Rate	.9995	.9999
Catholic Rate	.998	.999
Civically Engaged Adherents Rate (logged)	.971	.999
Index of Christian Heterogeneity	.054 *	-
Heterogeneity, 2nd quartile	-	.75
Heterogeneity, 3rd quartile	-	.75
Heterogeneity 4th quartile	-	.59 *
Level 1 N	122,096	122,096
Level 2 N	60	60

^*p<.05

^**p<.01

^***p<.001 Theft is the reference category for offense type.

As is typically found in the sentencing literature, the strongest determinants of incarceration are legally prescribed effects at the individual case and offender level. Each increase in offense severity increases the odds of incarceration by 42%, and each increase in prior record score increases incarceration odds by 44%. Furthermore, the presumptive disposition increases incarceration odds by 49%. In other words, if the sentencing guidelines recommend incarceration, this substantially increases an offender's chances of being incarcerated (note, however, that this relationship is not stronger because local courts have the discretion to depart from the guidelines). Additionally, incarceration chances vary by offense type as well. For example, robberies and homicides are substantially more likely to result in incarceration, while those convicted of drug misdemeanors are significantly less likely to be incarcerated.

Some interesting level-1 case processing and extralegal effects appear as well. For example, those convicted by trial, especially jury trial, are considerably more likely to be punished with incarceration. In fact, jury trials increase defendants’ incarceration odds 3.25 times, an effect stronger than the legally prescribed effect of the guideline presumptive disposition. Also, black and Hispanic defendants each have about 50% greater incarceration odds than whites. Females are significantly less likely to receive incarceration than males. Overall, these extralegal and case processing effects are not substantially different from the findings in other sentencing studies, particularly analyses of Pennsylvania data (e.g., Ulmer and Johnson 2004; Johnson 2006; Steffensmeier and Demuth 2001; Steffensmeier et al. 1998).

Two of the level two control variables significantly affect incarceration: court size and mean annual caseload. Defendants in large and medium courts (compared to small courts) and have notably greater odds of receiving incarceration, a finding that contrasts with some previous analyses of Pennsylvania sentencing, which found more leniency among large courts, but which did not include religious context predictors (e.g., Ulmer and Johnson 2004; Ulmer and Bradley 2006). Also, a one unit increase in mean annual caseload decreases the odds of incarceration by .001 %. The effect of caseload might seem trivial, but in cumulative terms, it would produce sizeable differences between counties with disparate caseloads (recall that the standard deviation for mean caseload is 3,031). Percent black and percent Republican voters both show negative associations with incarceration, but neither is significant.

Turning to the county religious context effects, in our first model, we enter the rates of adherents for the three Christian denominations (in a separate model not shown, we also entered a measure of total county religious adherents instead of the other religious context measures, but it failed to attain significance). Two of the three measures (mainline Protestant and Roman Catholic) are negatively related to the odds of incarceration while the conservative Protestant rate is positively related to incarceration. However, none of them attain significant statistical significance (the effect for mainline Protestant denominations comes the closest to significance, with a p. value of .11). Thus, we find no evidence that counties with more conservative Protestants punish defendants more severely. In addition, the log of the rate of civically engaged congregation adherents shows a non-significant and substantively trivial direct effect on the likelihood of incarceration. Thus, it also appears that the presence of adherents to civically engaged congregations does not directly effect incarceration.

We also enter the index of Christian heterogeneity into the first model, and it shows a significant and very strong negative relationship to incarceration odds. As Christian heterogeneity increases, incarceration odds greatly decrease, and vice versa. Thus, counties with more Christian homogeneity are more likely to incarcerate offenders. In other words, it is not the Christian group composition of a county that affects its court's willingness to incarcerate offenders, but rather the Christian homogeneity/heterogeneity of the county, regardless of composition.

The somewhat unrealistic size of the effect for the continuous variable index of Christian heterogeneity led us to suspect that there could be clusters of counties at the extremes of heterogeneity/homogeneity driving this effect. In other words, we suspected that the effect of heterogeneity is not monotonic and linear.¹³ To assess this, we broke the index into four quartiles from lowest to highest heterogeneity, and included these in our models as an alternative to the continuous heterogeneity index, with the most homogeneous quartile as the reference category. The results suggest that clusters of very heterogeneous and very homogeneous counties are the most dissimilar to each other in terms of their use of incarceration. The key contrast here is between the first and fourth quartiles – the most homogeneous and the most heterogeneous counties. The fourth quartile has a significant negative effect on odds of incarceration (compared to the first quartile reference category), while the second and third quartiles show decreased odds of incarceration compared to the first quartile, but their effects approach but do not attain significance. Defendants in counties with the most Christian heterogeneity have significantly decreased incarceration odds as compared to those with the least Christian homogeneity (or the most homogeneity). An offender being sentenced in one of the heterogeneous counties in the fourth category, then, would have about 42% lesser odds of incarceration than someone in one of the most homogeneous counties. Alternatively, in an identical model (not shown) with the fourth heterogeneity quartile as the reference category, the first (most homogeneous) quartile exhibits incarceration odds of 1.5—offenders in the most homogeneous counties have incarceration odds 50 percent greater than their counterparts in the most heterogeneous counties.

We also note that these findings are not just attributable to the presence of the two largest counties, Philadelphia and Pittsburgh. Our control for county size effectively controls out the influence of these two counties, since they comprise our “large” county category. Second, we obtain substantively similar findings when we remove these two counties from the data. Furthermore, Philadelphia and Pittsburgh are not the highest ranking counties on Christian heterogeneity.

Table 3 presents tests of our cross-level interaction effects of interest.

Table 3.

Cross Level Interaction Effects (From Separate Models)

	Incarceration
	Beta	S.E.	Odds
Offense Gravity Score	.318	.01	1.374***
Evangelical Rate	.0004	.0030	1.0004
Interaction Term	.001	.0001	1.001**
Prior Record Score	.382	.01	1.47***
Evangelical Rate	−.0004	.003	.9996
Interaction Term	−.0004	.0001	.9996**
Offense Gravity Score	.346	.008	1.413***
Heterogeneity 4th Quartile	−.5254	.2623	.591*
Interaction Term	.0301	.0116	1.031**
Prior Record Score	.3688	.0058	1.446***
Heterogeneity 4th Quartile	−.57	.2618	.566*
Interaction Term	−.054	.0115	.948***
Offense Gravity Score	.347	.008	1.414***
Civically Engaged Adherence Rate (logged)	.011	.07	1.0001
Interaction Term	.016	.002	1.016***
Prior Record Score	.49	.011	1.63***
Civically Engaged Adherence Rate (logged)	−.014	.07	.986
Interaction Term	−.029	.002	.972***

^*p < .05

^**p < .01

^***p < .001

We tested cross-level interaction effects for offense severity and percent conservative Protestant, and for prior record and percent conservative Protestant. The effect of offense severity increases very slightly with percent conservative Protestant, while the effect of prior record decreases very slightly with percent conservative Protestant. Although these two interaction terms are statistically significant, they are quite small in substantive terms. Their statistical significance is likely due to the large number of cases at level 1 and the strong effect of offense severity and prior record on incarceration. That is, with over 122,000 cases at level 1, even substantively quite small effects might be statistically significant. A one standard deviation (39 adherents per 1000) increase in the conservative Protestant rate would correspond to a roughly .03 increase in the effect of offense severity (increasing the offense severity odds ratio from 1.37 to 1.40). Likewise, a one standard deviation increase in the conservative Protestant rate would be associated with a .0156 decrease in the effect of prior record (changing the prior record odds ratio from 1.47 to about 1.45). Overall, these results suggest that the conservative Protestant adherence rate does little to color the relationship between offense severity, priors and sentencing decisions.

Another modest but significant interaction appears between offense severity and Christian heterogeneity, but in the opposite direction from that expected. In the most heterogeneous counties (the 4^th heterogeneity category), offense severity increases incarceration odds to a modestly greater extent. That is, in the counties with the most Christian heterogeneity, the effect of offense severity on the odds of incarceration is 3% greater than in the most homogeneous counties (changing the effect offense severity odds ratio from 1.41 to 1.44).

Also, Christian heterogeneity interacts with prior record in its effect on incarceration, but here the effect is positive, as we expected. The positive effect of prior record on incarceration odds is about 6% less in the most heterogeneous counties. In other words, a court's membership in the fourth heterogeneity quartile decreases its prior record effect on the odds of incarceration from 1.45 to about 1.39. Prior record increases the odds of incarceration in all counties, but does so to a moderately greater extent in the most homogeneous counties. That is, courts in counties characterized by more Christian homogeneity appear to weigh prior record more heavily in incarceration decisions. We also tested these cross-level interactions with the continuous variable Christian heterogeneity index, and the results were substantively the same as with the heterogeneity categories. The effect of offense severity significantly increased with Christian heterogeneity, while the effects of prior record significantly decreased with heterogeneity.

Finally, modest but significant cross-level interactions appear between the rate of civically engaged congregation adherents, offense severity, and prior record. A one unit increase in the logged civically engaged adherents rate (the mean for which is 4.9 and standard deviation is 1.46) is associated with a .016 (odds) increase in the effect of offense severity. Thus, a one standard deviation difference between counties in the log of civically engaged adherents rate would be associated with a .024 difference in the offense severity effect (increasing the offense severity effect from 1.41 to about 1.43). A one unit increase in the log of civically engaged adherents is associated with a .028 (odds) decrease in the effect of prior record. For example, a one standard deviation (1.46) increase in the logged civically engaged adherents rate would be associated with a .041 decrease in the effect of prior record (for example, decreasing it from 1.63 to about 1.59). Thus, courts in counties with more adherents to civically engaged congregations appear to emphasize offense severity to a slightly greater extent than those with fewer such adherents. However, such courts seem to slightly deemphasize prior record compared to counties with fewer civically engaged adherents.

In sum, religious context--and other county and court characteristics--are inferior to individual case-level and offender-related predictors in determining the likelihood of incarceration. However, Christian homogeneity, but not the specific type of group predominant in an area (conservative Protestant or civically engaged) is associated with greater incarceration of criminal offenders. Further, county percent conservative Protestant, Christian heterogeneity, and the rate of civically engaged congregation adherents all intensified the effects of offense severity on incarceration, although the conservative Protestant effect was especially weak. Courts in counties characterized by the least Christian homogeneity (in other words, the most heterogeneity) appear to put modestly more emphasis on offense severity in incarceration decisions. Thus, Christian heterogeneity conditions the effect of offense severity on incarceration, but in the opposite direction than expected. Counties with the most Christian heterogeneity were slightly “tougher” on those who commit more severe offenses, even though these counties were less likely to incarcerate offenders overall.

Interestingly, county percent conservative Protestant and civically engaged adherents both dampened the effects of prior record on incarceration, while Christian heterogeneity also decreased prior record's effects. Counties with greater Christian homogeneity were “tougher” on offenders with more extensive criminal histories than counties with the least Christian heterogeneity.

Although we did not hypothesize further interaction effects, to check the robustness of our findings we examined whether our effects for Christian homogeneity were conditional on other county characteristics, or vice versa. In particular, we were concerned with whether homogeneity/heterogeneity interacted with the county political context, in this case percent republican. We thus added a multiplicative interaction term to a model containing all the other predictors in Model 2 of Table 2 for county percent Republican by our two dummy variables for the two heterogeneity quartile dummy variables that had the strongest contrast with each other—the first and fourth quartile, or the most and least heterogeneous Christian counties. Results are presented in Table 4.

Table 4.

Hierarchical Model of Incarceration with Interaction Between Christian Heterogeneity First Quartile and Percent Republican

Model	Incarceration (Odds Ratio)
Intercept (B0, G00)	14.32**
Level 1 Predictors
(Same as Table 2)
Level 2 Predictors
Percent Black	.99
Percent Republican	.98
Large Court	1.61
Mean Caseload	999**
Mainline Rate	.997
Evangelical Rate	.9999
Catholic Rate	.999
Civically Engaged Adherents Rate (logged)	.999
Index of Christian Heterogeneity (4th quartile reference):
Heterogeneity, 1st quartile	1.21 *
Heterogeneity, 2rd quartile	.73
Heterogeneity 3rd quartile	.58
Heterogeneity, 1st quartile * Percent Republican	1.08 *
Level 1 N	122,096
Level 2 N	60

^*p<.05

^**p<.01, ***p<.001

We found no significant interaction between the fourth (most heterogeneous) quartile and percent Republican, but did find such an interaction for the first (most homogeneous) quartile. The coefficient for the interaction term percent Republican * first Christian heterogeneity quartile was .08 (odds ratio 1.08, p<.02), with the fourth quartile as the reference category. We interpret this interaction as the difference between the effect of percent Republican in the counties with the most Christian homogeneity vs. the other counties. While percent Republican has no significant direct effect overall (.99 in Table 2 and .98 in Table 4), in the most homogeneous Christian counties, percent Republican's effect on the odds of incarceration is .08 stronger in a positive direction. Thus, in these counties, each increase in percent Republican is accompanied by a 1.06 increase in the odds of incarceration (.98 + .08 = 1.06). Christian homogeneity thus appears to activate the effect of percent Republican so that Republican electoral predominance effects incarceration only in homogeneous Christian counties. It also suggests that the homogeneous Christian counties differ from one another in their propensity to incarcerate, based on their political characteristics.

Conclusion

For Stark and Bainbridge (1987) a religious doctrine is a set of instructions. Every religion tells its believers what is required to reach their ultimate goal, whether it is Heaven, Nirvana or some form of enlightenment. In doing so the religion also tells the believer what they may not do – what behaviors are forbidden or unacceptable. Durkheim, of course, noted religion's power as a form of control. Religion, he argued, has the unique ability to provide shared rules and norms. In sum, both common sense and classical theory suggest that religion is an important part of normative culture, and should impact the treatment of deviants, and hence the criminal justice system (see Savelsberg 2004).

Since most research on religion and crime has focused on the prediction of individual deviant behavior or crime rates we have little guidance on exactly how religion should impact sentencing. Drawing upon the religion and deviance literature, we have determined three possible ways that religion might have a contextual effect upon the justice system. Myers (1988, 1999) and others find that conservative Protestants are more punitive. Therefore, it is possible that jurisdictions with a high proportion of conservative Protestants will be harsher on criminals. Lee and Bartkowski (2004) draw upon Tolbert, Lyson and Irwin (1998) and argue that civically engaged denominations will strengthen community ties and discourage deviance. Perhaps, strong community ties will also lead to a stronger rejection of deviants, resulting in harsher punishments. Finally, Stark, Kent and Doyle (1982) suggests that it is not the type of religion that predominates in an area but religious homogeneity that will deter deviance. By extension, perhaps religious homogeneity will lead to greater punishment of deviants, resulting in harsher treatment by the justice system.

Ultimately we do not find strong support that the type of Christianity predominant in a place will influence incarceration. Neither the conservative Protestant adherence rate, nor the rates of Roman Catholics or mainline Protestants, nor the rate of adherence to civically engaged denominations, had a significant impact on the odds of incarceration. Unnever and Cullen's (2006) research finding no direct effect of fundamentalist Christianity on death penalty support is instructive here. They argued that conservative Christians actually espouse countervailing beliefs (e.g., images of a harsh, punitive God vs. beliefs in compassion and forgiveness) that would seem to both foster and discourage punitiveness (Unnever and Cullen 2006; Unnever et al. 2005; Applegate et al. 2000). A major empirical goal of our paper was to examine the effects of three alternative measures of moral communities on incarceration (Christian homogeneity, particular Christian denominational families, and civically engaged congregational adherents). It does not trouble us that only one of these measures directly affected sentencing (modest cross level interaction effects notwithstanding): this suggests that Christian homogeneity is the most relevant measure when addressing moral communities and punishment, at least in our data.

Incarceration is more likely in places characterized by homogenous Christianity, no matter what specific forms that Christianity takes. In fact, the most substantial difference was found between counties at the lowest and highest levels of Christian homogeneity/heterogeneity. In their research on death penalty support, Unnever and Cullen (2006) speculate that fundamentalist punitiveness might be stronger in areas exhibiting more religious “moral” communities compared to communities with more religious diversity. In other words, they predict that homogenous moral communities might mobilize stronger death penalty support among fundamentalists. Extending this logic, to the extent that members of a moral community share normative values, we would expect them to be more punitive towards those that reject those values through acts of deviant behavior, regardless of the denominational composition of the community.

District Attorneys and judges face popular partisan election in Pennsylvania. Moral communities would most likely affect what courts do when Christian communities are homogeneous, and thus perhaps more able to mobilize as a unified voice in local politics. Our findings are consistent with a process in which homogeneous Christian communities can mobilize enough local political strength (particularly in local District Attorney or judicial elections) to affect local courts’ propensity to incarcerate offenders. Recall also that our models controlled for percent Republican voters, and thus our effects are not simply attributable to the partisanship of the local electorate. The effect of Christian homogeneity on incarceration apparently applies regardless of whether a county is predominantly Republican or Democratic.

At the same time, local Christian homogeneity appears to activate the effect of percent Republican on incarceration: percent Republican only had a meaningful effect on incarceration among the most homogeneous Christian counties. We believe this finding indirectly strengthens and clarifies our argument above that Christian homogeneity fosters a particular moral community, and this moral community effects sentencing practices through local electoral politics affecting judges and prosecutors. It also suggests that the homogeneous Christian counties differ from one another in their punitiveness— such counties not characterized by Republican electoral dominance would not be as likely to incarcerate offenders. One would expect especially pronounced use of incarceration in counties where a homogeneous Christian moral community exists, and local Republican-dominated electorates favor more punitive criminal sanctions and elect judges and prosecutors in line with those views.

This political process might also explain the lack of a direct effect for county percent conservative Protestant. As suggested by the descriptive statistics in Table 1, perhaps conservative Protestants simply do not have the raw numbers to be much of a political presence in county court politics. The mean conservative Protestant rate is 71 per 1,000 county residents with a standard deviation of 39, and the conservative Protestant rate never exceeds 189 per 1,000 residents (or about 19 percent).

The effect of homogeneity suggests that it is the prominent and homogeneous combined Christian presence that can influence official reaction to crime, more than a single denominational grouping could. As we mentioned earlier, homogeneity and/or the size of the presence of civically engaged congregations might be an indicator of local Christian political strength. Such homogeneous or engaged communities might affect the tenor of local justice more than any one grouping (such as conservative Protestants) can on its own. In other words, Christian homogeneity likely represents the collective cultural and political influence of local Christian communities on local justice. This, in turn, might explain why the effect for homogeneity is not linear and monotonic. Perhaps it is only when a community reaches a certain threshold level of religious homogeneity that it can mobilize and exert an influence on local court politics.

It is possible that the effects for Christian homogeneity we found are explained by broader social processes associated with other types of homogeneity (e.g., race, class, political, cultural). Since our effects persisted after controlling for county percent Republican and racial composition (as well as poverty, percent Hispanic, and other variables in supplemental analyses), we can say that the Christian homogeneity effects are not spurious in the face of these other types of heterogeneity/homogeneity. Still, an important direction for future research would be to distinguish the effects of religious homogeneity from the effects of other types of homogeneity. In particular, our finding that percent Republican affected incarceration specifically in homogeneous Christian counties suggests that different types of cultural and political homogeneity might combine to shape criminal sanctions.

We also found that religious context somewhat affected the relationships between key legally prescribed determinants of sentencing (prior record and offense severity) and incarceration. Most notably, counties with greater Christian homogeneity were modestly more punitive toward those with prior criminal records. That is, prior record carried more weight in determining who was incarcerated in such counties. In other words, more homogeneously Christian counties are tougher on offenders overall, and tougher on those with extensive criminal histories in particular. Perhaps homogeneous Christian counties preference for incarceration reflects a community sentiment of “tough love.” Or perhaps Christian counties draw upon the historical influence of the Quakers, hoping that incarceration will result in moral accountability, reform and the opportunity to “turn one's life around.” In such counties, there may be less tolerance of repeat wrongdoing, and a greater demand for formal intervention and accountability. This, in turn, may be coupled with a hope that incarceration might be a “wake up call” to moral reform. We want to reiterate, however, that this homogeneity*prior record effect was quite modest.

Further, counties with higher rates of adherents to civically engaged congregations weigh offense severity slightly greater, but are somewhat more lenient to those with prior records. The civically engaged denominations measure developed by Tolbert, Lyson and Irwin (1998) primarily consists of mainline Protestant groups (with the exception of the AME Zion, Jews, and Latter-Day Saints). Again, we believe the answer lies in local politics. Where prominent clusters of civically engaged congregations exist, they may be a presence in local politics and help to shape the court community culture. Future research should explore the conditions under which civically engaged congregations might affect local courts. Also, future research should investigate whether secular groups’ civic engagement affect the nature of local criminal punishment.

Most of all, our findings suggest that researchers should attempt to further develop the theoretical linkages between moral communities and the criminal justice system, as called for by Savelsberg (2004). We may speculate in this regard. Our findings are consistent with the notion that courts are pervious (Jenness and Grattet 2005) to their surrounding moral communities. Perhaps moral communities shape court actors’ definitions and perceptions of the focal concerns of sentencing (Steffensmeier et al. 1998; Steffensmeier and Demuth 2000): notions of perceived blameworthiness, community protection, and practical constraints/consequences. Moral communities could shape notions of perceived blameworthiness for court community actors by shaping local norms of conduct and views of appropriate punishment for transgressions. Such communities might also affect notions of community protection – affecting perceptions of dangerousness, or perhaps belief in rehabilitation. As suggested by the interaction between Christian homogeneity and percent Republican, this might be particularly true in a locality characterized by both a homogeneous moral community and a politically conservative electorate. Future research should examine this possibility, and also examine the extent to which homogeneous Christian localities enable conservative political interests to be especially influential.

Moral communities are especially likely to influence practical constraints/consequences attached to sentencing decisions in jurisdictions where key legal actors are locally elected. For example, the nature of any local political ramifications of judges’ and prosecutors’ decisions, such as sentences or plea bargains in publicly visible cases, could be shaped by local moral communities. Such moral communities might demand a certain level of severity of punishment from their elected prosecutors and judges. Future research should examine the actual mechanisms, including local politics, by which moral communities can affect court communities and their actors’ interpretations of sentencing's focal concerns. We also suspect that local religious homogeneity may not have similar effects in other countries, such as in Europe or the Middle East. An interesting set of empirical questions is whether local community religious homogeneity has a particularistic effect on punishment in the U.S. that does not generalize to other countries, or whether local religious homogeneity might have equally strong but different effects outside the U.S.

Future studies should consider the characteristics of the judges and prosecutors themselves (Johnson 2006). Previous research has noted the effect of patriarchal attitudes among judges and the disposition of cases involving domestic violence (Cassidy and Tafimow 2002), the impact of judge's gender on decision-making in state supreme courts (Songer and Crews-Meyer 2000), and judges’ race/ethnicity, gender, and career factors on sentencing (Johnson 2006). We suspect that judges who have retributive attitudes towards punishment will be more harsh in their decision-making, regardless of the surrounding context. Further, the judge's personal religiosity may prove an important predictor, as Myers (1988, 1989) found in a study of judges in Georgia. Perhaps the judge's personal religiosity will interact with the community religious contexts, with the harshest sentences occurring when religious judges reside in the most homogeneous Christian contexts.

Further, we suggest that research on religious context should take into account the religion of the defendant. Based on his studies of the Bay colony witch trials, Erikson (1966) notes the importance of “outsiders” (accused witches in this case) in maintaining a sense of community. From a Durkheimian perspective, the colony's identification and sanctioning of witches performed “a needed service to society by drawing people together in a common posture of anger and indignation” (p. 4). Such a perspective suggests that those outside the dominant religion of an area may be more likely to be sanctioned, or face stiffer penalties. Such data would not be easy to collect, but if possible, studies of religion and sentencing could include measures of the defendant's religiosity and an interaction term between individual and contextual religion measures. Unfortunately, our data did not include information on the religiosity or ideologies of judges, or the religiosity of defendants, so we could not test such hypotheses.

The intersection of religion, politics and the judicial system remains at the center of American national debate. At the 2005 confirmation hearing for Supreme Court nominee John Roberts, Senate committee members asked pointed questions about government endorsement of religion. At issue were the mention of God in the Pledge of Allegiance, government funding of faith-based organizations, and the posting of the Ten Commandments in courthouses. Behind such debate is the concern that religion will exert undue societal influence by pressuring students to accept religion, turning social programs into attempts at religious conversion, or by encouraging judges and jurors to allow religion to influence their verdicts. Key to such a debate, however, is a clear understanding of the extent to which and mechanisms by which religion has a contextual influence on justice outcomes. We hope that our research will inspire further studies of a nexus between religious contexts and the criminal justice system, as well as further explorations of the importance of Christian homogeneity in particular as a means by which religion impacts American culture.