Racial/Ethnic Discrimination and Prosecution

Abstract

Unwarranted disparity taking place at the stage of prosecution has long been an interest for sentencing researchers. Research exploring the effect of offender race on prosecutorial decisions, however, has produced conflicting and inconclusive results. Some studies concluded that minority offenders faced more unfavorable outcomes than White offenders, whereas others found no significant impact of race/ethnicity in the prosecution process. Still others found a minority advantage. Given these inconsistencies, this research uses meta-analytic methodology to assess empirical findings from a body of scholarship that examined the relationship between race/ethnicity and prosecutorial outcomes. Analyses of homogeneity and moderator variables are also conducted to explore whether there are factors accounting for variability in effect sizes across studies. The result suggests that minority offenders face greater odds of being charged or fully prosecuted than White offenders. Moreover, several moderators, primarily methodologically relevant, account for variability across effect sizes.

Keywords

discretion charging meta-analysis prosecution prosecutor race/ethnicity

For the majority of criminal cases, the pretrial phase is perhaps the most crucial one, as more than 80% of cases in state courts and around 90% in federal courts are resolved at stages prior to trial (Motivans, 2013; Perry & Banks, 2011). Prosecutors, serving as key courtroom players and gatekeepers, control the case flow in pretrial proceedings by selecting the arrestee to be charged, the type of charge against the arrestee, and the case to be dropped or negotiated (Hartley, Maddan, & Spohn, 2007; Spohn & Holleran, 2001). The implementation of federal and state sentencing guidelines as an effort to curb discretion exercised at the sentencing stage has further increased prosecutorial power (Miethe, 1987; Spohn, 2009). These examples illustrate how influential prosecutorial charging and full-prosecution decisions are in case processing. As such, prosecution in connection with the exercise of broad discretion during the pretrial phase must be scrutinized in the public eye; in fact, researchers in courts and sentencing have shown a keen interest in this direction.

There has been a considerable amount of research on criminal justice decision making in recent decades (Devine, 2012; Free, 2005; Spohn, 2009). Researchers have invested heavily in the investigation of the extent to which offender characteristics as extralegal factors play a role in criminal justice decision making. The ban on the use of such extralegal factors as race and ethnicity in prosecution and sentencing has been written into policy guidelines or law (e.g., American Bar Association, 1993; Minnesota Sentencing Guidelines Commission, 1981; Nebraska Court Rules of Professional Conduct, 2008; U.S. Sentencing Commission, 1987). Despite the categorical prohibition, evidence still reveals that race and ethnicity affect not only judicial but also prosecutorial decision making (Hartley etal., 2007; Henning & Feder, 2005; Leiber & Blowers, 2003; O. Mitchell, 2005; Spohn, 2000; Spohn & Fornango, 2009; Wooldredge & Thistlethwaite, 2004). The effect of race and ethnicity in the courtroom may come from explicit bias, implicit bias, or the implementation of laws and practices with racially disparate effects.

With the rich literature examining racial/ethnic disparity and discrimination in the criminal justice field, it is not surprising that inconsistent findings have emerged as a result of methodological differences across studies. Criminologists have recently started a comprehensive approach to exploring the existence of racial discrimination in the criminal justice system by synthesizing past research findings. The research synthesis of prior studies on racial disparity and discrimination in courtroom decision making has been conducted qualitatively (Free, 2001, 2002, 2005) or quantitatively (Mazzella & Feingold, 1994; O. Mitchell, 2005; T. L. Mitchell, Haw, Pfeifer, & Meissner, 2005; Pratt, 1998; Sweeney & Haney, 1992). Research has also extended this line of investigation to police decision making (Kochel, Wilson, & Mastrofski, 2011). These efforts have been intended to resolve conflicting results from a large body of studies and to seek a single, elementary answer for policy implications.

Although several meta-analyses have endeavored to answer the question about minorities’ unfair treatment in the criminal justice system, gaps still exist. The major gap is that meta-analysis as the systematically quantitative review has not been used to examine prosecution. The issue of whether minority offenders are more likely than White offenders to be charged or fully prosecuted has only been systematically reviewed in a qualitative way (Free, 2001, 2002, 2005). In light of the gap, the current study conducts a quantitative meta-analysis to synthesize existing research findings on race, ethnicity, and criminal prosecution. The main purpose of the study is to center on the extent to which an offender’s minority status has an effect on prosecutors’ charging decisions or full prosecution. This goal is achieved through a mean effect size analysis. The secondary purpose is to test whether variability across the effect sizes of different studies is due to sampling error alone, and if not, to further identify moderators that serve as the sources of variability in effect sizes. This goal is achieved through a homogeneity analysis.

Legal/Extralegal Factors and Prosecutorial Decisions

Similar to judges’ decision making, prosecutors’ decision making is influenced by a number of legal and extralegal factors (Beichner & Spohn, 2005; Hartley etal., 2007). Research has recently extended Steffensmeier, Ulmer, and Kramer’s (1998) focal concerns model from judicial to prosecutorial decisions to explain warranted disparities stemming from the effects of legal factors as well as unwarranted disparities stemming from the effects of extralegal factors in the courtroom (Beichner & Spohn, 2005; Franklin, 2010b; Hartley etal., 2007; Spohn, Beichner, & Davis-Frenzel, 2001; Spohn & Fornango, 2009; Ulmer, Kurlychek, & Kramer, 2007). The focal concerns model has consolidated several other explanations to form a potent perspective in courts and sentencing (Ulmer etal., 2007). However, the application of the focal concerns model to prosecution does have considerations different from those for bail or sentencing. According to the focal concerns model, courtroom decision makers base their decisions on three concerns: (a) the assessment of an offender’s “blameworthiness” or culpability, (b) the assessment of an offender’s dangerousness for “the protection of the community” to be free from the threat of recidivism, and (c) “practical constraints and consequences” based on organizational and individual perspectives (Steffensmeier etal., 1998, pp. 766-767).

Blameworthiness is often associated with offense severity, offense type, and accountability attributable to the victim, and an offender’s criminal history has been the key predictor of offender dangerousness, with repeat offenders viewed as having a high risk of recidivism and posing a threat to the general public (Steffensmeier etal., 1998). In actuality, offense severity and criminal history representing the first two focal concerns serve as legal factors expected to explain the substantial portion of disparity in case outcomes. Regarding the third concern, practical constraints and consequences, prosecutors are highly concerned about the possibility of conviction by winning a case through plea agreement or trial. Therefore, convictability that reflects efficiency, effectiveness, and certainty from the organizational aspect becomes an important consideration in the prosecution process (Beichner & Spohn, 2005; Hartley etal., 2007; Ulmer etal., 2007). The primary factor for the organizational aspect of practical considerations is the strength of evidence, which also serves as a dominant legal factor (Albonetti, 1987; McCoy, Salinas, Walker, & Hignite, 2012; but see Spohn & Spears, 1997). In addition, there are extralegal factors falling within the individual aspect of practical considerations, such as victim characteristics and victim–offender relationships (see Beichner & Spohn, 2005; Kingsnorth, MacIntosh, & Wentworth, 1999; Spears & Spohn, 1997; Spohn etal., 2001; Spohn & Holleran, 2001).

Race and ethnicity as extralegal factors are commonly tied to blameworthiness, offender dangerousness, and practical consequences due to stereotypes and biases. Racial minorities are often perceived as offenders who commit violent street crimes, lack remorse, and tend to recidivate (Everett & Nienstedt, 1999; Steffensmeier & Demuth, 2000; Steffensmeier etal., 1998). Together with high levels of uncertainty about conviction in the absence of complete case information (Albonetti, 1987), a perceptual shorthand often has been used to predict the outcome of a criminal charge by counting on such extralegal factors as the defendant’s/victim’s race, ethnicity, gender, and age (Beichner & Spohn, 2005; Hartley etal., 2007). Researchers have highlighted the importance of holding legal and other extralegal factors constant before the unwarranted race/ethnicity effect can be accurately detected (O. Mitchell, 2005; Wilbanks, 1987; Wooldredge, 1998). The moderator analysis in this meta-analysis will examine whether controlling for theoretical factors underlying the focal concerns perspective makes a difference.

Offender Race, Prosecution, and Research Synthesis

A systematic review or synthesis of previous studies may be qualitative or quantitative. Several research syntheses have reviewed the literature on the extent to which an offender’s race affects prosecutorial decisions. By distinguishing the discrimination thesis from the nondiscrimination thesis, Marvin Free has conducted three qualitative syntheses of presentencing decision making (Free, 2001, 2002, 2005). Reviewing 26 studies of prosecutorial decisions on charging and dismissal, Free (2001) found that findings from nine studies (35%) supported the discrimination thesis whereas findings from 17 studies (65%) were consistent with the nondiscrimination thesis. He concluded with some support for the racially discriminatory argument in prosecutorial charging/dismissal decisions and further pointed out several methodological flaws for the nondiscriminatory argument in previous studies. Free (2002) further reviewed seven studies on prosecutorial decisions to seek the death penalty. He found that 71% of findings appeared to support discrimination against African American capital defendants. The most recent review by Free (2005) further revealed evidence that racial discrimination at prosecutorial charging/dismissal stages stemmed from prosecutors’ fear of minority group threat. However, he underscored the contradictory findings from previous studies.

In brief, having shed some light on the influence of an offender’s race on prosecution, previous research syntheses have been based solely on qualitative methods (e.g., Free, 2001, 2002, 2005). Quantitative meta-analysis has been employed only to synthesize research findings on jury and judicial decisions. The current meta-analysis seeks to fill the research gap by quantitatively synthesizing the extant findings of the relationship between offender race/ethnicity and prosecution. This meta-analysis will determine (a) whether race or ethnicity is a significant factor in charging or full-prosecution decisions, (b) what the magnitude of the race effect is, (c) whether moderators are present to account for variability in effect sizes across studies, and (d) how each moderator has an impact on the variability. Analyses are grounded in the assumption of random/mixed-effects models.

Method

Literature Search

To conduct a meta-analysis of the race effect on charging or full-prosecution decisions, three stages of the search strategy were created to screen previous studies that appeared to be qualified for the current meta-analysis. The search at the first stage was focused on the criteria taken to identify the literature that might be relevant to the subject matter of the study. A variety of combinations of keywords related to prosecutor, prosecution, charging, dismissal, discretion, decision making, disparity, discrimination, race, and ethnicity were utilized to conduct the search. The search targeted both federal and state adult prosecutions in the United States. Due to a distinct court process and special considerations in the courtroom for capital cases, the search is limited to noncapital cases (O. Mitchell, 2005). The search of the literature, which was confined to scholarly works published or produced between 1960 and 2012, was through multiple means. The first approach was to search primary online databases, such as APA PsycNET, Criminal Justice Abstracts, Google Scholar, HeinOnline, National Criminal Justice, ProQuest Dissertations and Theses, Reference Service, SocINDEX, and Sociological Abstracts.¹ To locate newly accepted articles before their appearance in print, the second approach was to examine several journal publishers’ websites that have frequently published criminal justice research, including Elsevier, Emerald, SAGE, Springer, Taylor and Francis, the University of Chicago Press, and Wiley. The last approach was to review the bibliographies of identified studies in the scholarship of courts and sentencing.

At the second stage, the search was focused on sorting out studies to identify those eligible for the later meta-analysis. The first stage of the search strategy might have generated studies without statistical information to calculate an effect size. Therefore, this meta-analysis essentially excluded studies with only literature summaries, argumentative components, qualitative methodology, and descriptive statistics. The analysis also omitted legal studies or law review articles that were not grounded in social science quantitative methodology. With the particular interest of the current study, only studies that had examined the main, direct effect of race or ethnicity on prosecution or controlled for this variable in the model were included in the meta-analysis. Furthermore, articles published online first before print by the end of 2012 were eligible for the meta-analysis (e.g., Romain & Freiburger, 2013). Last, there has been a debate over the inclusion of studies with varying methodological quality in a meta-analysis. Some meta-analysts prefer the inclusion of only rigorous research with “the best evidence,” whereas others argue for the inclusion of studies that present available evidence (for a discussion, see Lipsey & Wilson, 2001, pp. 9-10). Recognizing that no research was perfect, but that some research relied on outdated, seriously flawed analytic techniques, this meta-analysis followed the best evidence principle to include only empirical studies using multivariate analysis to control for crime severity or criminal history. Given a lack of their reliable findings, a sensitivity analysis by including seriously flawed studies was not performed (for a sensitivity analysis in meta-analytic research, see Kochel etal., 2011).

At the third stage, there was an additional screening procedure to ensure statistical independence for eligible studies, as identified at the first two stages (Kochel etal., 2011; O. Mitchell, 2005; Wu & Spohn, 2009). Incorporating all studies with overlapping data sources (e.g., the same or very similar data years or jurisdictions examined) would easily produce a biased estimate for the average effect size, either weighted or unweighted (O. Mitchell, 2005). Specifically, when studies had a large sample size or effect size, the sign (or direction) of the average effect size in the meta-analysis might be changed due mainly to immense influence from these studies. As a result, meta-analysts would likely reach a misleading conclusion about the actual effect of the primary variable of interest on the outcome. To tackle the issue of statistical independence for the coding purpose, priority was given to studies based on the following order: type of analytic model (partitioned over combined, see the “Classification of Effect Size” column in the appendix), information sufficiency (more sufficient over less sufficient), statistical methodology (multivariate over univariate or bivariate), number of variables (greater over smaller), sample size (larger over smaller), and year of publication (newer over older) (see Kochel etal., 2011; O. Mitchell, 2005; Wu & Spohn, 2009).

As shown in Table 1, the keywords used in the first and second search stages identified 271 studies.² Of the studies identified, 245 studies were excluded because they did not meet the criteria for inclusion (e.g., non-U.S. data, qualitative methods, insufficient statistical information, lack of focus on variables of interest, failure to examine the main effect, univariate/bivariate analyses, and similar data). In total, 37% of studies excluded lacked quantitative methods or statistical information for the analysis of effect sizes, and 45% of them did not examine the race effect or charging/full-prosecution decisions. Following the three stages of searches, only 26 studies were eligible for the later meta-analysis, thus generating 36 effect sizes and representing 86,877 cases (see the appendix).³

Table 1:

Summary of Sample of Studies

Status of Study	n
Number of studies retrieved^a	271
Number of studies excluded	245
Non-U.S. data	17
No quantitative analysis or no sufficient statistical information	90
No race and/or no investigation of whether to charge or fully prosecute a suspect	111
No main effect of offender race; only interaction effects	8
Use of the same data set	14
No control for primary legal factors (i.e., crime severity, criminal history, and strength of evidence)	5
Number of studies included in the analysis	26
Number of effect sizes included in the analysis	36

Studies retrieved do not include cases that focus on juvenile cases, plea bargaining, bail, and capital punishment.

Effect Size Coding and Adjustments

The dependent variable in the current meta-analysis was charging or full-prosecution decisions. This variable in previous studies was measured in varying ways, such as charge severity (e.g., Crew, 1991; Croyle, 1983), charge reductions (e.g., Albonetti, 1992; Bernstein, Kick, Leung, & Schulz, 1977; Bishop & Frazier, 1984; Bjerk, 2005; Farnworth & Teske, 1995; Langan, 1996; Miethe & Moore, 1985, 1986; Shermer & Johnson, 2010), or the likelihood of an offender being charged (e.g., Adams & Cutshall, 1987; Albonetti & Hepburn, 1996; Baumer, Messner, & Felson, 2000; Beichner & Spohn, 2005; Kingsnorth & MacIntosh, 2007; Miethe, 1987; Spohn & Holleran, 2001; Wooldredge & Thistlethwaite, 2004). This meta-analysis focused on the likelihood of a prosecutorial charge being filed, which was a dichotomy measuring whether a criminal charge was filed against an arrestee.⁴ However, the prosecution process involves different phases (Bernstein, Cardascia, & Ross, 1979). Attrition in prosecutors’ cases may result from an initial decision not to file a charge; it also may be the result of an initial decision to file a charge that is dismissed or dropped later (i.e., nolle prosequi). To have the appropriate number of studies for assessment, this meta-analysis merged full-prosecution decisions into charging decisions in the mean effect size analysis, but differentiated one stage from the other in the later moderator analysis.

Race or ethnicity was the main independent variable of interest. Comparisons between two races or ethnicities could be modeled as non-White versus White, Black versus White, or Hispanic versus White. To capture all these possibilities, this meta-analysis referred to race or ethnicity as “minority.” The effect size drawn from each study was computed based on the odds ratio. There were several adjustments for effect size coding. First, when a study reported more than one odds ratio for the same outcome, an adjustment was made by averaging these effect sizes (Lipsey & Wilson, 2001; Wilson, Mitchell, & MacKenzie, 2006). The adjustment of taking an average value was mostly for a study with two or more racial/ethnic dummies in the analytic model simultaneously. Without this adjustment, the meta-analysis would inevitably violate the principle of statistical independence, as previously discussed. Second, when a study analyzed data using probit regression rather than logistic regression, a transformation of the probit unstandardized coefficient was conducted by multiplying it by π / √3 (Hasselblad & Hedges, 1995; Kochel etal., 2011) and exponentiating the result. Similarly, the probit standard error was adjusted by multiplying its variance by π² / 3 and taking the square root (Kochel etal., 2011; Lipsey & Wilson, 2001). Last, this meta-analysis treated standardized coefficients (i.e., βs) directly as correlation coefficients (i.e., rs) and employed the equation, $r = ({OR}^{3 / 4} - 1) / ({OR}^{3 / 4} + 1)$ , to obtain odds ratios (Bonett, 2007; Digby, 1983). Because of a direct transformation from unstandardized coefficients (i.e., bs) to βs and rs via the same formula (Knoke, Bohrnstedt, & Mee, 2002), researchers have suggested the imputation of beta coefficients in the absence of correlation coefficients (Baier & Wright, 2001; Peterson & Brown, 2005). This approach supplemented effect size coding to remedy the absence of odds ratios obtained directly from logistic regression models.

Missing statistical information was another issue that needed to be addressed. Several studies provided b coefficients or βs but did not report standard errors. Meta-analytic researchers have proposed several solutions (for a discussion, see Lipsey & Wilson, 2001). One of the solutions for the inclusion of studies with insufficient information in the meta-analysis was to impute an estimated value (see Lipsey & Wilson, 2001). Based on the equation $b / S E = t$ , this meta-analysis used the reported significance level (or the p value) and b coefficient to estimate the standard error of the race variable. In a two-tailed test with a sample size greater than 120, the p value of .05 equated to the t-value of approximately 1.96, the p value of .01 equated to the t-value of approximately 2.58, and the p value of .001 equated to the t-value of approximately 3.29 (Fisher & Yates, 1963). To avoid an overestimate or underestimate, the two-tailed t-values of 1.60, 2.27, 2.94, and 3.60 were used to acquire standard errors that reflected the p values greater than .05, between .05 and .01, between .01 and .001, and smaller than .001, respectively.

Moderator Variables

The moderator variable analysis in this study tested for two sets of moderators, either methodologically or theoretically relevant to variability in effect sizes (see Table 4). Regarding methodologically relevant moderator variables, whether a publication was refereed might influence the effect size. Journals tended to publish work with statistically significant results, which generally produced greater effect sizes than work with insignificant results (Lipsey & Wilson, 2001). Therefore, refereed publications were expected to have a greater race effect than nonrefereed publications. The region moderator had three categories (i.e., non-South, South, and multiple/unknown). Compared with the non-South, the South typically has been more punitive toward offenders and had more pronounced unwarranted racial disparity (Chiricos & Crawford, 1995; Frost, 2006; Kleck, 1981; O. Mitchell, 2005; Wu & Spohn, 2009).⁵ Type of jurisdiction distinguished the single jurisdiction from multiple jurisdictions. Based on the assumption of the offsetting effect or aggregation bias, studies pooling multiple jurisdictions together have displayed a race effect smaller than those analyzing a single jurisdiction (O. Mitchell, 2005; Wu & Spohn, 2009). Although several meta-analyses did not find the influence of data years (Kochel etal., 2011; O. Mitchell, 2005; Wu & Spohn, 2009), the current meta-analysis nonetheless incorporated this factor into the moderator variable analysis to explore potential variations in the methodological rigor of research design.

Type of standard error differentiated estimated standard errors from reported standard errors because some studies failed to provide information. Estimated standard errors likely overestimated or underestimated the effect of race and ethnicity on prosecutorial outcomes. Statistical methods from which effect sizes were derived referred to logistic regression, probit regression, and hierarchical linear modeling. This meta-analysis included studies with an interest in not only race but also ethnicity. As such, two types of coding for race and ethnicity were created to capture a single racial or ethnic group, such as Blacks versus Whites or Hispanics versus Whites, as well as the combined group, such as minorities versus Whites (cf. O. Mitchell, 2005). Last, to include studies investigating prosecutorial charging as well as full prosecution (i.e., nondismissal), a moderator to capture two points in time was created.

With respect to theoretically relevant moderator variables, researchers have pointed out an important distinction between studies with and without controls for legal and extralegal factors (see Hagan, 1974; Hawkins, 1987; O. Mitchell, 2005; Wilbanks, 1987; Wooldredge, 1998; Zatz, 1987). The focal concerns perspective has referred to the severity of crime, an offender’s criminal history, and evidentiary strength as three key legal factors to be held constant in the relationship between race/ethnicity and prosecution (Beichner & Spohn, 2005; Shermer & Johnson, 2010; Spohn etal., 2001). Failure to utilize multivariate analysis to control for the severity of crime and criminal history has been considered a serious flaw in methodology and has often resulted in an overestimation of the race effect (Wilbanks, 1987; see also O. Mitchell, 2005). This has also become a source of variability in effect sizes across studies (Wu & Spohn, 2009). The same is true of evidentiary strength. Not only has this factor helped prosecutors determine whether to pursue a case, but it has also affected case dismissal prior to trial (Albonetti, 1987; Free, 2005; McCoy etal., 2012). Although this meta-analysis had precluded studies without multivariate analysis to control for crime severity or criminal history, two effect sizes were still associated with the control, not for both, but only for one of these two legal factors. Moreover, a large number of effect sizes were obtained from studies without controlling for the strength of evidence. Given nearly a constant for crime severity or criminal history and clearly a variable for evidentiary strength in the present meta-analytic data, a moderator variable was created to capture the influence of whether an effect size was associated with controls for all three primary legal factors. An additional moderator variable was used to measure whether controlling for evidentiary strength alone would make a difference.

From the focal concerns perspective, extralegal factors to be held constant in the prosecution process have included victim characteristics, the victim–offender relationship, and other offender characteristics. Victims have played a key role in the prosecutorial decision to file a charge or to continue the prosecution (Beichner & Spohn, 2005; Kingsnorth etal., 1999; Spears & Spohn, 1997; Spohn etal., 2001; Spohn & Holleran, 2001). Holding victim characteristics and the victim–offender relationship constant might attenuate the effect of an offender’s race and ethnicity. Last, controlling for other offender characteristics would not only reduce the race/ethnicity effect but also differentiate racial discrimination from other types of discrimination, such as gender and age. No moderator variable was created, however, because only one effect size was not associated with the control for any of other offender characteristics. The moderator, if created, would be nearly a constant.

Statistical Analysis

The current meta-analysis operationalized both the independent (i.e., race and ethnicity) and dependent (i.e., charging and full-prosecution decisions) variables in a dichotomous form; thus, the inverse-variance weight method as performed on b coefficients (i.e., logged odds ratios) was appropriate (Borenstein, Hedges, Higgins, & Rothstein, 2009; Cooper, 1998; Hedges & Olkin, 1985; Kochel etal., 2011; Lipsey & Wilson, 2001; O. Mitchell, 2005). The mean effect size analysis was performed to examine the relationship. Findings then were reported through a conversion from b coefficients to odds ratios for the purpose of interpretation and for the determination of the magnitude of a significant race effect, if any. Furthermore, a Q test was performed for the homogeneity analysis. A significant Q value suggests that moderators related to the characteristics of studies or research methods exist to account for the variability across effect sizes obtained from different studies. It also suggests that meta-analysts should further investigate which and how moderators impact the variability. Lipsey and Wilson (2001) have offered an analog to the ANOVA and meta-regression as two equivalently effective options for the moderator variable analysis. The very small number of effect sizes (i.e., 36) hindered the effective use of meta-regression. The ANOVA procedure, therefore, served as the major method for the homogeneity analysis.

It is worth noting that two models, fixed-effects and random-effects, have competed with each other to tackle the issue of variability within and between studies. The fixed-effects model assumes that variability is associated with random error within the study because all studies share the same effect size from a common population, whereas the random-effects model assumes that variability is associated with random error not only within the study but also between studies (for a discussion, see Lipsey & Wilson, 2001). When a Q value is significant, it is possible to (a) use the moderator analysis to identify factors (or moderators) that have led to systematic sampling error based on the fixed-effects model, (b) import the random variance component between studies into the random-effects model, or (c) combine fixed-effects with random-effects to create a mixed-effects model that takes into account the systematic sampling error and random variance component (Lipsey & Wilson, 2001). This meta-analysis relied on the assumption from the random-effects model and conducted analyses using the mixed-effects model. Running the mixed-effects model helps researchers explore the impact of moderators with such methods as ANOVA and meta-analytic regression (Lipsey & Wilson, 2001). Meta-analysts have preferred this approach (Borenstein etal., 2009; Kochel etal., 2011; Lipsey & Wilson, 2001; O. Mitchell, 2005).

Results

All odds ratios for effect sizes were coded on the basis of how minority offenders were treated as compared with White offenders during prosecution. Regardless of statistical significance, odds ratios greater than 1.00 would suggest the minority disadvantage, whereas odds ratios smaller than 1.00 would indicate the minority advantage. The stem-and-leaf plot in Figure 1 shows the relatively even distribution between the minority disadvantage and advantage. Quantitatively, approximately 56% of odds ratios fell within the range of the minority disadvantage. The distribution of effect sizes for the minority disadvantage was more dispersed than that for the minority advantage, however. Seven odds ratios representing the minority disadvantage were greater than 1.40, compared with only four odds ratios representing the minority advantage that fell below the corresponding size of 0.71 (1/1.40 = 0.71). In addition, the greatest effect size for the minority advantage was 0.30, which corresponded with the odds ratio of 3.33 for the minority disadvantage. This size was much smaller than 4.17, as demonstrated by the greatest effect size for the minority disadvantage. Table 2 shows a simple vote-counting strategy to further explore the significance of effect sizes in their original studies. Each effect size represents one of four groups: (a) significant minority disadvantage, (b) insignificant minority disadvantage, (c) significant minority advantage, and (d) insignificant minority advantage. Only six of effect sizes (17%) displayed the significant minority disadvantage, and two of effect sizes (6%) displayed the significant minority advantage. The majority of effect sizes had no race/ethnicity effect. All of the above results depicted only a partial picture of the relationship between race/ethnicity and prosecution. An advanced statistical technique, such as a meta-analysis, was employed below to understand the complex dynamics of the relationship.

Figure 1:

Stem-and-Leaf Diagram of Odds Ratios as Effect Sizes

Table 2:

Vote-Counting by Direction and Significance of Effect Sizes

Direction and Significance^a	Frequency	%
Significant minority disadvantage	6	16.7
Insignificant minority disadvantage	14	38.9
Significant minority advantage	2	5.5
Insignificant minority advantage	14	38.9
Total	36

Minority disadvantage: Odds ratios > 1; minority advantage: Odds ratios < 1. Five studies (Crutchfield, Weis, Engen, & Gainey, 1995; Franklin, 2010a; Kingsnorth & MacIntosh, 2007; McCoy, Salinas, Walker, & Hignite, 2012; Romain & Freiburger, 2013) controlled for more than one racial/ethnic dummy in the same regression model, two of which (Crutchfield et al., 1995; McCoy et al., 2012) reported a difference in statistical significance across racial/ethnic dummies. Significance following an adjustment by averaging effect sizes of racial/ethnic dummies was determined by a comparison between the ratio of the b coefficient to its standard error and 1.96 (i.e., p = .05).

Table 3 presents findings from the mean effect size analysis and the homogeneity analysis. The null hypothesis for the mean effect size analysis under the random-effects model was that race and ethnicity did not affect charging or full-prosecution decisions. The original Q value of 95.730 was statistically significant at p = .001, revealing that there was significant between-study variability across effect sizes.⁶ With the underlying assumption from the random-effects model that permitted unobserved random error to exist, a random-effects variance component (τ²) was introduced and added back to the variance of each effect size to recalculate the mean effect size.⁷ As a result, the random-effects odds ratio as the weighted mean effect size was 1.093 (i.e., logged odds ratio = 0.089), with the 95% confidence interval between 1.028 and 1.162. Because the confidence interval did not include e⁰ = 1 (or zero, as converted to the logged odds ratio), the null hypothesis was rejected (p < .001), indicating that race and ethnicity were a significant factor in the prosecution process, at least in regard to decisions to charge or fully prosecute offenders. Minority offenders had an approximately 9% higher likelihood of being charged or fully prosecuted than White offenders. The finding comported with the theoretical hypothesis that minority offenders were more disadvantaged than White offenders in criminal courts in general and in prosecutorial decisions in particular.

Table 3:

The Effect Size Estimate and Q Statistic

Moderator Variable	Mean Effect Size^a	SE	95% CI		Q	k
Moderator Variable	Mean Effect Size^a	SE	Lowera	Uppera	Q	k
Race and ethnicity	1.093 (0.089)**	0.031	1.028 (0.028)	1.162 (0.150)	95.730***	36

Note. SE = standard error; CI = confidence interval; Q = between-study Q value; k = number of effect sizes.

Logged odds ratios (i.e., b coefficients) are in parentheses.

p < .01. ***p < .001.

Regarding the homogeneity analysis, the null hypothesis that variability in effect sizes across studies was due to sampling error alone was tested (Cooper, 1998; Lipsey & Wilson, 2001). Rejecting the null hypothesis would indicate significant differences across studies beyond what sampling error alone could account for. As mentioned, the statistically significant Q value of 95.730 (df = 35, p < .001) revealed substantial variability in the relationship between race/ethnicity and prosecution across studies. With the performance of statistical analyses under the mixed-effects model, sampling error and unobserved random error could not explain away the variability in effect sizes across studies. There was a need to perform the moderator variable analysis to identify methodological or theoretical characteristics contributing to systematic variations.

Table 4 reports the results of the moderator variable analysis.⁸ Of the sample and analytic characteristics in Panel A, type of publication, Q(34) = 0.271, p = .063, data years, Q(34) = 0.013, p = .908, statistical methods for effect sizes, Q(33) = 2.567, p = .277, and coding schemes for race and ethnicity, Q(34) = 0.870, p = .351, were not statistically significant. Moreover, except for controls for all three primary legal factors together, Q(34) = 3.912, p = .048, theoretical characteristics in Panel B, such as controls only for evidentiary strength, Q(34) = 3.076, p = .079, controls for victim characteristics, Q(34) = 0.000, p = .991, and controls for the victim–offender relationship, Q(34) = 0.461, p = .497, failed to reach the significance level at .05 under the mixed-effects model. The findings from Panel B, in general, were partially contrary to the expectation that analytic models controlling for theoretical characteristics would generate smaller effect sizes in racial or ethnic disparity. The findings regarding partial support were nonetheless congruent with those from Kochel etal.’s (2011) examination of the race effect on police arrest decisions.

Table 4:

Effect Size Analyses of Random-Effects Mean Odds Ratios by Moderators

		95% CI
Moderator Variable	M ES	Lower	Upper	z	p	Q	τ²	k
Panel A: Methodological (sample or analytic) moderators
Type of publication					.603	0.271
Nonrefereed publication	1.150	0.958	1.380	1.503	.133		.016	4
Refereed journal article	1.091	1.012	1.177	2.269	.023		.019	32
Region					.024	7.456*
Non-South	1.061	0.997	1.130	1.852	.064		.009	25
South	1.411	1.108	1.799	2.786	.005		.068	8
Multiple/not reported	0.983	0.888	1.087	−0.339	.735		.000	3
Type of jurisdiction					.005	7.918**
Single	1.159	1.054	1.275	3.048	.002		.027	26
Multiple	0.999	0.959	1.042	−0.042	.966		.000	10
Year of data					.908	0.013
Prior to 1991	1.089	1.022	1.162	2.611	.009		.008	19
1991 or later	1.100	0.940	1.288	1.192	.233		.049	17
Type of standard error					.012	6.240*
Provided by study	1.029	0.946	1.120	0.676	.499		.017	25
Estimated	1.224	1.100	1.361	3.717	.000		.014	11
Statistical method for effect size					.277	2.567
Logistic	1.116	1.042	1.195	3.125	.002		.013	32
Probit	1.108	0.988	1.242	1.755	.079		.000	2
Hierarchical linear modeling	0.875	0.654	1.169	−0.905	.365		.032	2
Coding for race and ethnicity					.351	0.870
Black or Hispanic vs. White (single)	1.063	0.965	1.172	1.239	.215		.020	16
Minority vs. White (combined)	1.132	1.036	1.237	2.745	.006		.012	20
Prosecutorial decision point					.015	5.947**
Screening	1.205	1.071	1.356	3.105	.002		.029	21
Prosecution	1.021	0.958	1.087	0.634	.526		.006	15
Panel B: Theoretical moderators
Controls for all three primary legal factors (i.e., crime severity, criminal history, and strength of evidence)					.048	3.912*
Yes	1.275	1.053	1.542	2.494	.013		.046	14
No	1.043	0.985	1.104	1.445	.149		.007	22
Controls for evidentiary strength					.079	3.076
Yes	1.249	1.031	1.515	2.268	.023		.049	16
No	1.044	0.986	1.105	1.490	.136		.007	20
Controls for victim characteristics					.991	0.000
Yes	1.092	0.916	1.302	0.983	.325		.014	13
No	1.093	1.024	1.168	2.658	.008		.013	23
Controls for victim–offender relationship					.497	0.461
Yes	1.158	0.931	1.441	1.322	.186		.067	12
No	1.071	1.008	1.139	2.207	.027		.009	24

Note. ES = effect size; CI = confidence interval; z = z score; p = critical value; Q = between-study Q value; τ² = random-effects variance component; k = number of effect sizes.

p < .05. **p < .01.

As in Table 4, the set of effect sizes divided by region was not homogeneous, Q(33) = 7.456, p = .024. Studies with data from Southern jurisdictions (odds ratio = 1.411, p = .005) reported more pronounced disparity than those with data from non-Southern jurisdictions (odds ratio = 1.061, p = .064). The same finding was also for the comparison between Southern jurisdictions (odds ratio = 1.411, p = .005) and multiple or unknown jurisdictions (odds ratio = 0.983, p = .735). In the South, minority offenders faced significantly greater odds of being charged or fully prosecuted than White offenders. Type of jurisdiction also played a significant role in accounting for systematic variations across studies, Q(34) = 7.918, p = .005. For the data based on the single city or county, which produced a more pronounced race effect, minority offenders had significantly greater odds of being charged or fully prosecuted than White offenders (odds ratio = 1.159, p = .002). The availability of a standard error from studies similarly affected homogeneity, Q(34) = 6.240, p = .012. Compared with those clearly reporting standard errors (odds ratio = 1.029, p = .499), studies failing to provide this information revealed significantly larger racial or ethnic disparity (odds ratio = 1.224, p = .000). In other words, the minority disadvantage concerning prosecution came mainly from studies without reporting the standard error. Significant variations across prosecutorial decision points took place as well, Q(34) = 5.947, p = .015. Studies with a focus on initial charging (odds ratio = 1.205, p = .002) reported greater disparity than those with a focus on full prosecution (odds ratio = 1.021, p = .526). The significant minority disadvantage occurred primarily at the charging stage rather than at the full-prosecution stage. Finally, studies controlling for all three primary legal factors (odds ratio = 1.275, p = .013), including crime severity, criminal history, and evidentiary strength, had more pronounced disparity than those without controlling for these variables (odds ratio = 1.043, p = .149). The former was also the significant source of racial/ethnic discrimination.

Discussion and Conclusion

The purpose of the current study was to examine the effect of race and ethnicity on disparity in charging or prosecuting decisions through a meta-analysis. Previous meta-analytic research has examined the extent to which an offender’s race affected decision making by the police (Kochel etal., 2011), jurors (Mazzella & Feingold, 1994; T. L. Mitchell etal., 2005; Sweeney & Haney, 1992), and judges (O. Mitchell, 2005; Pratt, 1998). Despite several qualitative syntheses of prosecutorial decision making (Free, 2001, 2002, 2005), no quantitative meta-analysis has been performed. This meta-analytic review sought to bridge the research gap by comprehensively evaluating whether an offender’s race and ethnicity are a major determinant of charging and full-prosecution decisions. Furthermore, to systematically and quantitatively synthesize research findings from a large number of previous studies, it is necessary for meta-analysts to address the issue of variability across these studies and identify moderators accounting for differences in effect sizes. This meta-analysis was based on the best evidence rather than the available evidence from previous studies. That is, studies without employing multivariate statistics to control for primary legal factors were excluded from analysis.

The finding from the mean effect size analysis revealed that an offender’s race and ethnicity did play a significant role in prosecutors’ decisions to file a charge or pursue a full prosecution. That is, the combination of a large amount of empirical evidence on the issue of race, ethnicity, and prosecution has pointed to the conclusion that minority offenders faced greater odds of being charged or fully prosecuted than White offenders. The finding from the mean effect size analysis was congruent with meta-analytic findings on police arrest and judicial sentencing decisions in support of the racial discrimination thesis—harsher treatment for racial minorities than for Whites (Kochel etal., 2011; O. Mitchell, 2005). The conclusion, however, was at odds with the current vote-counting and previous qualitative results, both showing no disparate treatment between minority and White offenders in 65% to 75% of studies (see Free, 2001, 2002, 2005). It is worth noting that the qualitative synthesis of the literature is often tied to and restricted to vote-counting, without the capability to account for the influence resulting from such factors as sample size and standard error. Moreover, due to its incapability to differentiate the magnitude of an effect size from that of another effect size, the qualitative approach inevitably overestimates the contribution of a study with a small race effect and underestimates the contribution of a study with a large race effect. The present quantitative review has utilized a more rigorous methodology than a qualitative review to address the issue of previous disparate findings based on varying research methods. It should not be surprising that the present conclusion did not confirm other qualitative findings.

Building on the random-effects assumption as opposed to the fixed-effects assumption, the moderator variable analysis in this study examined whether there were variations across effect sizes. The Q statistic for the homogeneity test showed a large amount of variation across effect sizes attributable to the existence of moderator variables. As indicated by further investigations, several methodological characteristics accounted for the observed variation, whereas only one theoretical characteristic had an impact. Among significant moderators were region (South, non-South, or unknown), type of jurisdiction (single or multiple), type of standard error (provided by studies or estimated), prosecutorial decision points (screening or full prosecution), and controls for all three primary legal factors—that is, crime severity, criminal history, and evidentiary strength (yes or no). One initial expectation (see Chiricos & Crawford, 1995; Frost, 2006; Kleck, 1981), although not supported by prior meta-analytic research on posttrial proceedings (e.g., O. Mitchell, 2005; Sweeney & Haney, 1992), was that data based on non-Southern jurisdictions should produce a smaller effect size than data based on Southern jurisdictions. This expectation has not been violated in the current meta-analysis that examined the pretrial phase. In addition, racial or ethnic discrimination took place primarily in the South, in the single jurisdiction, and in the prosecutors’ initial screening phase. All of the above evidence suggests that the prosecutorial culture varies not only in terms of geographical areas but also between the pretrial and posttrial processes or between separate decision-making points. Although the advantage of the mean effect size analysis in a meta-analysis is to allow for an overall, single conclusion from among conflicting research findings, it is worth cautioning that interpretation of findings should also pay attention to the influence of specific moderator variables pertaining to where and when the effect takes place.

Contrary to the original expectation regarding the influence of theoretical moderators (see O. Mitchell, 2005; Wilbanks, 1987), studies with legal and other extralegal variables as controls were not necessarily more precise, and in turn did not produce a smaller race effect, than those without these variables as controls. Excluding statistical significance, comparisons in the magnitude of mean effect sizes between studies with and without legal/extralegal variables as controls also failed to form a clear pattern in this regard (cf. O. Mitchell, 2005). In fact, Kochel et al. (2011) in their meta-analysis of police decision making also reported a larger mean effect size for studies with a control for several legal/extralegal factors, despite no statistically significant difference. A close look revealed that the significant minority disadvantage occurred when studies controlled for only evidentiary strength or for all three primary legal factors at the same time (i.e., criminal history, offense severity, and evidence strength). These results have mirrored the finding that the minority disadvantage was greater for the best evidence method (i.e., 9%) than for the available evidence method (i.e., 7%, not shown) and that the two methods of analysis did not reach the same conclusion in terms of the significance test. Conversely, consistent with the expectation of the focal concerns perspective regarding the individual aspect of practical considerations, the mean effect size derived from studies without controlling for victim characteristics or the victim–offender relationship appeared to have resulted in minority disadvantage. That is, victim-related factors have been important in the prosecution process (Beichner & Spohn, 2005; Kingsnorth etal., 1999; Spears & Spohn, 1997; Spohn etal., 2001; Spohn & Holleran, 2001), and controlling for them has averted the overestimation of the difference in prosecution between minority and White defendants. In sum, these findings suggest that the argument about the relationship between methodological rigor and the magnitude of the race effect may appear to hold true for a single study but not for a research synthesis. Future research may further focus on the examination of this relationship.

The current meta-analytic research has its limitations and should guide future scholarship. First, prosecution includes a wide range of decisions. In addition to the decision to charge or fully prosecute an arrestee as reviewed in this meta-analysis, there has been research on the magnitude of charge reductions, motions for departures, mandatory minimum penalties, prosecution for capital cases, and plea negotiation deals (Albonetti, 1992; Bishop & Frazier, 1984; Bjerk, 2005; Farnworth & Teske, 1995; Hartley etal., 2007; Langan, 1996; Miethe & Moore, 1985, 1986; Shermer & Johnson, 2010; Songer & Unah, 2006; Sorensen & Wallace, 1999; Spohn & Fornango, 2009; Ulmer etal., 2007). With limited space, this research was unable to have a thorough review of all prosecutorial decisions. Despite the limitation, it should be noted that not all prosecutorial decisions merit a meta-analytic review at this time. For example, quantitative research on prosecutorial discretion and departure decisions has been conducted heavily with federal data and found similar outcomes. There is really no need to conduct an additional meta-analysis. The same is true of prosecutorial discretion in capital cases. Moreover, the fact that there is little quantitative research on discretion in mandatory minimum sentence cases guarantees no need for a meta-analysis at the present time. Meta-analysts in the future should consider the possibility of whether more studies are available or whether conflicting outcomes exist.

Second, it may be instrumental in separating charging decisions from full-prosecution decisions for analysis. However, this approach will necessarily make the sample size for each type of decision much smaller than that in the current meta-analysis. Given the moderate number of studies contained in this meta-analysis, charging and full-prosecution decisions were not modeled independently in the mean effect size analysis.⁹ To remedy the situation, this study created a moderator in the moderator variable analysis to allow for an observation about variability between charging and full-prosecution decisions. Meta-analysts should be attentive to future prosecutorial studies to determine whether it is feasible to differentiate between the two types of decisions when studies based on various data sets become sufficient.

Last, eight studies with 11 effect sizes contained in this meta-analysis did not provide information about the standard error. To retain the number of studies and effect sizes for analysis, the standard error for each missing effect size was imputed through an estimated value (see Lipsey & Wilson, 2001). Estimated standard errors may arguably be distorting the outcome for the mean effect size analysis. The 11 effect sizes, however, have represented both significant and insignificant race effects in their original studies. Given that the key purpose of a meta-analysis is to incorporate as many studies as possible into the mean effect size analysis rather than a purpose to exclude studies on the basis of a publication’s quality of writing, the imputation of estimated values should be viewed as an appropriate method.

From a policy perspective, meta-analysis has become an important tool in policy research because it has functioned to integrate a large number of prior research findings on the same topic into a single result through systematic analysis. In practice, the social world changes in terms of location, time, or culture, making methodology and results in social science research vary accordingly. It is not surprising to see why policy makers have been handicapped by lack of consistent research results to propose potential policy changes. The conclusion from the current meta-analysis concerning racial/ethnic minorities’ differential treatment in the courtroom process is in line with conclusions from prior meta-analytic research on jury and judicial decisions. Consistent meta-analytic results within the court system suggest that policy makers need to be highly concerned about racial/ethnic offenders’ treatment in adult criminal court. With the minority disadvantage commonly found in prosecutorial, jury, and judicial decisions, it is important for both decision makers and policy makers to seek reforms of the current practice of handling cases involving minority offenders in the court system. For instance, Bibas (2009) detailed the very minimal effect of legislative, judicial, and state bar efforts to keep prosecutors in compliance with laws and regulations. Instead, Bibas offered an alternative approach to the reform by focusing on external pressure from stakeholders (e.g., community members or defendants) as well as internal management (e.g., institutional culture, organizational structure, and hiring/promotion policies). It is also important for researchers to be attentive to the areas of prosecution that have not come under the survey of meta-analysis.

Footnotes

Appendix

Summary of 26 Studies Included in Meta-Analysis

Study	Data Source	Data Period	Classification of Effect Size	Significance (p < .05)	Sample Size	Effect Size (Odds Ratio)	Coding of Race
Adams and Cutshall (1987)	The United States (Washington, D.C.)	1974-1975		Yes	745	1.650	White/Black
Albonetti (1987)	The United States (Washington, D.C.)	1974		No	6,014	1.158	White/Non-White
Albonetti and Hepburn (1996)	Maricopa County, Arizona	1989-1990		No	5,554	1.127	White/Non-White
Baumer, Messner, and Felson (2000)	33 counties in the United States	Prior to 1988		No	1,990	1.306	White/Non-White
Beichner and Spohn (2005)	Miami, Florida	1997		No	127	0.486	White/Non-White
Bernstein, Cardascia, and Ross (1979)	One metro in New York	1974-1975		Yes	2,972	1.041	White/Non-White
Chiricos and Bales (1991)	Two counties in Florida	1982		No	1,970	0.861	White/Black
Crutchfield, Weis, Engen, and Gainey (1995)	King County, Washington	1994	Property	No	4,735	1.070	White/Non-White
Crutchfield, Weis, Engen, and Gainey (1995)			Personal	Yes	3,573	1.487
			Drug	No	3,967	1.551
Curran (1983)	Dade County, Florida	1965-1966	1965-1966	No	188	1.273	White/Black
		1971	1971	No	87	1.718
		1975-1976	1975-1976	No	181	1.342
Franklin (2010a)	33 counties in the United States	1998		No	15,557	0.987	White/Non-White
Franklin (2010b)	33 counties in the United States	1998		No	4,126	0.950	White/Black
Frazier and Haney (1996)	One Midwestern metro	1991		No	105	1.000	White/Non-White
Ghali and Chesney-Lind (1986)	Honolulu, Hawaii	1979-1980		No	5,226	1.103	White/Non-White
Henning and Feder (2005)	Shelby County, Tennessee	2000-2001		Yes	2,521	1.960	White/Non-White
Kingsnorth and MacIntosh (2007)	Sacramento County, California	1999-2001	Male	No	4,905	1.137	White/Non-White
			Female	No	643	1.262
Leiber and Blowers (2003)	One county in a Southeastern state	1990		Yes	1,727	1.879	White/Black
McCoy, Salinas, Walker, and Hignite (2012)	Harris County, Texas	1999		Yes	2,356	1.790	White/Non-White
Miethe (1987)	Minnesota	1978-1982	1978	No	1,273	0.994	White/Black
			1980	No	1,369	1.044
			1982	No	1,738	0.988
Romain and Freiburger (2013)	One county in a Midwestern state	2009		Yes	1,009	0.492	White/Non-White
Spears and Spohn (1997)	Detroit, Michigan	1989		No	318	0.830	White/Black
Spohn and Holleran (2001)	Kansas City, Missouri, Philadelphia, Pennsylvania	1996-1998	Sexual assault—Stranger	No	109	0.787	White/Non-White
			Sexual assault—Acquaintance	No	277	0.926
			Sexual assault—Partner	No	114	4.166
Spohn and Horney (1993)	Detroit, Michigan	1970-1984	Sexual assault—Prereform	No	243	0.441	White/Black
			Sexual assault—Postreform	No	505	0.298
Spohn and Spears (1997)	Detroit, Michigan	1976-1978		No	6,009	0.907	White/Black
Wooldredge (1998)	Doña Ana County, New Mexico	1983-1988		No	1,586	0.863	White/Mexican
Wooldredge and Thistlethwaite (2004)	Hamilton County, Ohio	1993, 1995-1996		Yes	2,948	0.730	White/Black
Worrall, Ross, and McCord (2006)	Southern California	2003		No	110	0.914	White/Non-White

Note. The appendix contains studies for the best evidence synthesis, which is the primary meta-analytic method in the present research.

Acknowledgements

The author wishes to thank the anonymous reviewers and the editor for their valuable comments. The author would also like to thank Scott W. Phillips for his assistance in editing and proofreading the article.

Notes

Jawjeong Wu is an associate professor in the Criminal Justice Department at SUNY Buffalo State. His research interests include comparative criminal justice, criminological theory, and sentencing disparity.

References

*Adams

Cutshall

C. R.

(1987). Refusing to prosecute minor offenses: The relative influence of legal and extralegal factors. Justice Quarterly, 4, 595-609.

*Albonetti

C. A.

(1987). Prosecutorial discretion: The effects of uncertainty. Law & Society Review, 21, 291-313.

Albonetti

C. A.

(1992). Charging reduction: An analysis of prosecutorial discretion in burglary and robbery cases. Journal of Quantitative Criminology, 8, 317-333.

*Albonetti

C. A.

Hepburn

J. R.

(1996). Prosecutorial discretion to defer criminalization: The effects of defendant’s ascribed and achieved status characteristics. Journal of Quantitative Criminology, 12, 63-81.

American Bar Association. (1993). ABA standards for criminal justice prosecution function and defense function (3rd ed.). Washington, DC: Author.

Baier

C. J.

Wright

B. R. E.

(2001). “If you love me, keep my commandments”: A meta-analysis of the effect of religion on crime. Journal of Research in Crime & Delinquency, 38, 3-21.

*Baumer

E. P.

Messner

S. F.

Felson

R. B.

(2000). The role of victim characteristics in the disposition of murder cases. Justice Quarterly, 17, 282-307.

*Beichner

Spohn

C. C.

(2005). Modeling the effects of victim behavior and moral character on prosecutors’ charging decisions in sexual assault cases. Violence and Victims, 27, 3-24.

*Bernstein

I. N.

Cardascia

Ross

C. E.

(1979). Defendant’s sex and criminal court decisions. In Aalvarez

Lutterman

K. G.

, & Associates (Eds.), Discrimination in organizations (pp. 329-354). San Francisco, CA: Jossey-Bass.

10.

Bernstein

I. N.

Kick

Leung

J. T.

Schulz

(1977). Charge reduction: An intermediary stage in the process of labelling criminal defendants. Social Forces, 56, 362-384.

11.

Bibas

(2009). Prosecutorial regulation versus prosecutorial accountability. University of Pennsylvania Law Review, 157, 959-1016.

12.

Bishop

D. M.

Frazier

C. E.

(1984). The effects of gender on charge reduction. The Sociological Quarterly, 25, 385-396.

13.

Bjerk

(2005). Making the crime fit the penalty: The role of prosecutorial discretion under mandatory minimum sentencing. Journal of Law & Economics, 48, 591-625.

14.

Bonett

D. G.

(2007). Transforming odds ratios into correlations for meta-analytic research. American Psychologist, 62, 254-255.

15.

Borenstein

Hedges

L. V.

Higgins

J. P. T.

Rothstein

H. R.

(2009). Introduction to meta-analysis. West Sussex, UK: John Wiley.

16.

*Chiricos

T. G.

Bales

W. D.

(1991). Unemployment and punishment: An empirical assessment. Criminology, 29, 701-724.

17.

Chiricos

T. G.

Crawford

(1995). Race and imprisonment: A contextual assessment of the evidence. In Hawkins

D. F.

(Ed.), Ethnicity, race, and crime: Perspectives across time and place (pp. 281-309). Albany: State University of New York Press.

18.

Cooper

H. M.

(1998). Synthesizing research: A guide for literature reviews. Thousand Oaks, CA: SAGE.

19.

Crew

K. B.

(1991). Racial differences in felony charging and sentencing: Toward an integration of decision-making and negotiation models. Journal of Crime and Justice, 14, 99-122.

20.

Croyle

J. L.

(1983). Measuring and explaining disparities in felony sentences: Courtroom work group factors and race, sex, and socioeconomic influences on sentence severity. Political Behavior, 5, 135-153.

21.

*Crutchfield

R. D.

Weis

J. G.

Engen

R. L.

Gainey

R. R.

(1995). Racial and ethnic disparities in the prosecution of felony cases in King County: Final report. Olympia: Washington State Minority and Justice Commission.

22.

*Curran

D. A.

(1983). Judicial discretion and defendant’s sex. Criminology, 21, 41-58.

23.

Devine

D. J.

(2012). Jury decision making: The state of the science. New York: New York University press.

24.

Digby

P. G. N.

(1983). Approximating the tetrachoric correlation coefficient. Biometrics, 39, 753-757.

25.

Everett

R. S.

Nienstedt

B. C.

(1999). Race, remorse, and sentence reduction: Is saying you’re sorry enough? Justice Quarterly, 16, 99-122.

26.

Farnworth

Teske

R. H. C.

Jr. (1995). Gender differences in felony court processing: Three hypotheses of disparity. Women & Criminal Justice, 6, 23-44.

27.

Fisher

R. A.

Yates

(1963). Statistical tables for biological, agricultural, and medical research (6th ed.). New York, NY: Hafner.

28.

*Franklin

T. W.

(2010a). Community influence on prosecutorial dismissals: A multilevel analysis of case- and county-level factors. Journal of Criminal Justice, 38, 683-701.

29.

*Franklin

T. W.

(2010b). The intersection of defendants’ race, gender, and age in prosecutorial decision making. Journal of Criminal Justice, 38, 185-192.

30.

*Frazier

P. A.

Haney

(1996). Sex assault cases in the legal system: Police, prosecutor, and victim perspectives. Law and Human Behavior, 20, 607-628.

31.

Free

M. D.

Jr. (2001). Racial bias and the American criminal justice system: Race and presentencing revisited. Critical Criminology, 10, 195-223.

32.

Free

M. D.

Jr. (2002). Race and presentencing decisions in the United States: A summary and critique of the research. Criminal Justice Review, 27, 203-232.

33.

Free

M. D.

Jr. (2005). Prosecutorial decision making and minority group-threat theory. Criminal Justice Studies, 18, 7-28.

34.

Frost

N. A.

(2006). The punitive state: Crime, punishment, and imprisonment across the United States. New York, NY: LFB Scholarly Publishing.

35.

*Ghali

Chesney-Lind

(1986). Gender bias and the criminal justice system: An empirical investigation. Sociology and Social Research, 70, 164-171.

36.

Hagan

(1974). Extra-legal attributes and criminal sentencing: An assessment of a sociological viewpoint. Law and Society Review, 8, 357-384.

37.

Hartley

R. D.

Maddan

Spohn

C. C.

(2007). Prosecutorial discretion: An examination of substantial assistance departures in federal crack-cocaine and powder-cocaine cases. Justice Quarterly, 24, 382-407.

38.

Hasselblad

Hedges

L. V.

(1995). Meta-analysis of screening and diagnostic tests. Psychological Bulletin, 117, 167-178.

39.

Hawkins

D. F.

(1987). Beyond anomalies: Rethinking the conflict perspective on race and criminal punishment. Social Forces, 65, 719-745.

40.

Hedges

L. V.

Olkin

(1985). Statistical methods for meta-analysis. Orlando, FL: Academic Press.

41.

*Henning

Feder

(2005). Criminal prosecution of domestic violence offenses: An investigation of factors predictive of court outcomes. Criminal Justice and Behavior, 32, 612-642.

42.

Higgins

J. P. T.

Thompson

S. G.

Deeks

J. J.

Altman

D. G.

(2003). Measuring inconsistency in meta-analyses. British Medical Journal, 327, 557-560.

43.

*Kingsnorth

R. F.

MacIntosh

R. C.

(2007). Intimate partner violence: The role of suspect gender in prosecutorial decision-making. Justice Quarterly, 40, 553-578.

44.

Kingsnorth

R. F.

MacIntosh

R. C.

Wentworth

(1999). Sexual assault: The role of prior relationship and victim characteristics in case processing. Justice Quarterly, 16, 275-302.

45.

Kleck

(1981). Racial discrimination in criminal sentencing: A critical evaluation of the evidence with additional evidence on the death penalty. American Sociological Review, 46, 783-805.

46.

Knoke

Bohrnstedt

G. W.

Mee

A. P.

(2002). Statistics for social data analysis (4th ed.). Belmont, CA: Wadsworth.

47.

Kochel

T. R.

Wilson

D. B.

Mastrofski

S. D.

(2011). Effect of suspect race on officers’ arrest decisions. Criminology, 24, 460-495.

48.

Langan

P. A.

(1996). Sentence reductions for drug traffickers for assisting federal prosecutions. Unpublished manuscript. Retrieved from http://bjs.gov/content/pub/pdf/srdtafp.pdf

49.

*Leiber

M. J.

Blowers

A. N.

(2003). Race and misdemeanor sentencing. Criminal Justice Policy Review, 14, 464-485.

50.

Lipsey

M. W.

Wilson

D. B.

(1993). The efficacy of psychological, education, and behavioral treatment: Confirmation from meta-analysis. American Psychologist, 48, 1181-1209.

51.

Lipsey

M. W.

Wilson

D. B.

(2001). Practical meta-analysis. Thousand Oaks, CA: SAGE.

52.

Mazzella

Feingold

(1994). The effects of physical attractiveness, race, socioeconomic status, and gender of defendants and victims on judgments of mock jurors: A meta-analysis. Journal of Applied Social Psychology, 24, 1315-1344.

53.

*McCoy

Salinas

P. R.

Walker

J. T.

Hignite

(2012). An examination of the influence of strength of evidence variables in the prosecution’s decision to dismiss driving while intoxicated cases. American Journal of Criminal Justice, 37, 562-579.

54.

*Miethe

T. D.

(1987). Charging and plea bargaining practices under determinate sentencing: An investigation of the hydraulic displacement of discretion. Journal of Criminal Law and Criminology, 78, 155-176.

55.

Miethe

T. D.

Moore

C. A.

(1985). Socioeconomic disparities under determinate sentencing systems: A comparison of pre-guideline and post-guideline practices in Minnesota. Criminology, 23, 337-363.

56.

Miethe

T. D.

Moore

C. A.

(1986). Racial differences in criminal processing: The consequences of model selection on conclusions about differential treatment. The Sociological Quarterly, 27, 217-237.

57.

Minnesota Sentencing Guidelines Commission. (1981). Minnesota sentencing guidelines and commentary. Saint Paul: Author.

58.

Mitchell

(2005). A meta-analysis of race and sentencing research: Explaining the inconsistencies. Journal of Quantitative Research, 21, 439-466.

59.

Mitchell

T. L.

Haw

R. M.

Pfeifer

Meissner

C. A.

(2005). Racial bias in jury decision-making: A meta-analytic review of defendant treatment. Law and Human Behavior, 29, 621-637.

60.

Motivans

(2013). Federal justice statistics, 2010-statistical tables (BJS Publication No. NCJ-239913). Washington, DC: U.S. Department of Justice.

61.

Nebraska Court Rules of Professional Conduct, § 3-508.4 (2008).

62.

Perry

S. W.

Banks

(2011). Prosecutors in state courts, 2007-statistical tables (BJS Publication No. NCJ-234211). Washington, DC: U.S. Department of Justice.

63.

Peterson

R. A.

Brown

S. P.

(2005). On the use of beta coefficients in meta-analysis. Journal of Applied Psychology, 90, 175-181.

64.

Pratt

T. C.

(1998). Race and sentencing: A meta-analysis of conflicting empirical research results. Journal of Criminal Justice, 26, 513-523.

65.

*Romain

D. M.

Freiburger

T. L.

(2013). Prosecutorial discretion for domestic violence cases: An examination of the effects of offender race, ethnicity, gender, and age. Criminal Justice Studies, 26, 289-307.

66.

Shermer

L. O.

Johnson

B. D.

(2010). Criminal prosecutions: Examining prosecutorial discretion and charging reductions in U.S. federal district courts. Justice Quarterly, 27, 394-430.

67.

Songer

M. J.

Unah

(2006). The effect of race, gender, and location on prosecutorial decisions to seek the death penalty in South Carolina. South Carolina Law Review, 58, 161-210.

68.

Sorensen

Wallace

D. H.

(1999). Prosecutorial discretion in seeking death: An analysis of racial disparity in the pretrial stages of case processing in a Midwestern county. Justice Quarterly, 16, 559-578.

69.

*Spears

J. W.

Spohn

C. C.

(1997). The effect of evidence factors and victim characteristics on prosecutors’ charging decisions in sexual assault cases. Justice Quarterly, 14, 501-524.

70.

Spohn

C. C

. (2000). Thirty years of sentencing reform: The quest for a racially neutral sentencing process. In Horney

(Vol. Ed.), Criminal justice 2000: Vol. 3. Policies, processes, and decisions of the criminal justice system (pp. 427-501). Washington, DC: National Institution of Justice, Office of Justice Program, U.S. Department of Justice.

71.

Spohn

C. C.

(2009). How do judges decide? The search for fairness and justice in punishment (2nd ed.). Thousand Oaks, CA: SAGE.

72.

Spohn

C. C.

Beichner

Davis-Frenzel

(2001). Prosecutorial justifications for sexual assault case rejection: Guarding the “gateway to justice.” Social Problems, 48, 206-235.

73.

Spohn

C. C.

Fornango

(2009). U.S. attorneys and substantial assistance departures: Testing for interprosecutor disparity. Criminology, 47, 813-846.

74.

*Spohn

C. C.

Holleran

(2001). Prosecuting sexual assault: A comparison of charging decisions in sexual assault cases involving strangers, acquaintances, and intimate partners. Justice Quarterly, 18, 651-688.

75.

*Spohn

C. C.

Horney

(1993). Rape law reform and the effect of victim characteristics on case processing. Journal of Quantitative Criminology, 9, 383-409.

76.

*Spohn

C. C.

Spears

J. W.

(1997). Gender and case processing decisions: A comparison of case outcomes for male and female defendants charged with violent felonies. Women & Criminal Justice, 8, 29-59.

77.

Steffensmeier

Demuth

(2000). Ethnicity and sentencing outcomes in U.S. federal courts: Who is punished more harshly? American Sociological Review, 65, 705-729.

78.

Steffensmeier

Ulmer

J. T.

Kramer

J. H.

(1998). The interaction of race, gender, and age in criminal sentencing: The punishment cost of being young, black, and male. Criminology, 36, 763-798.

79.

Sweeney

L. T.

Haney

(1992). The influence of race on sentencing: A meta-analytic review of experimental studies. Behavioral Sciences & the Law, 10, 179-195.

80.

Ulmer

J. T.

Kurlychek

M. C.

Kramer

J. H.

(2007). Prosecutorial discretion and the imposition of mandatory minimum sentences. Journal of Research in Crime & Delinquency, 44, 427-458.

81.

U.S. Sentencing Commission. (1987). Guidelines manual. Washington, DC: Author.

82.

Wilbanks

(1987). The myth of a racist criminal justice system. Monterey, CA: Brooks/Cole.

83.

Wilson

D. B.

Mitchell

MacKenzie

D. L.

(2006). A system review of drug court effects on recidivism. Journal of Experimental Criminology, 2, 459-487.

84.

*Wooldredge

J. D.

(1998). Analytic rigor in studies of disparities in criminal case processing. Journal of Quantitative Criminology, 14, 155-179.

85.

*Wooldredge

J. D.

Thistlethwaite

(2004). Bilevel disparities in court dispositions for intimate assault. Criminology, 42, 417-456.

86.

*Worrall

J. L.

Ross

J. W.

McCord

E. S.

(2006). Modeling prosecutors’ charging decisions in domestic violence cases. Crime & Delinquency, 52, 472-503.

87.

Spohn

(2009). Does an offender’s age have an effect on sentence length? A meta-analytic review. Criminal Justice Policy Review, 20, 379-413.

88.

Zatz

M. S.

(1987). The changing forms of racial/ethnic biases in sentencing. Journal of Research in Crime & Delinquency, 24, 69-92.