Research Evaluation Report (Statistics)

Journal of Criminal Justice 42 (2014) 384–398

Contents lists available at ScienceDirect

Journal of Criminal Justice

A Critical Examination of the “White Victim Effect” and Death Penalty Decision-Making from a Propensity Score Matching Approach: The North Carolina Experience

Wesley G. Jennings a,b,⁎, Tara N. Richards c, M. Dwayne Smith d, Beth Bjerregaard e, Sondra J. Fogel f

a Dept. of Criminology, College of Behavioral & Community Sciences, SOC 333, University of South Florida, Tampa, FL 33620 b Courtesy Appointment, Dept. of Mental Health Law and Policy, College of Behavioral & Community Sciences, SOC 333, University of South Florida, Tampa, FL 33620 c University of Baltimore, School of Criminal Justice d University of South Florida, Department of Criminology e University of North Carolina – Charlotte, Department of Criminal Justice and Criminology f University of South Florida, School of Social Work

⁎ Corresponding author at: University of South Florid 4202 E. Fowler Ave., SOC 333, Tampa, FL 33620. Tel.: + 974 2803.

E-mail address: [email protected] (W.G. Jennings)

http://dx.doi.org/10.1016/j.jcrimjus.2014.05.004 0047-2352/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Available online 13 June 2014

Purpose: Death penalty research has rather consistently demonstrated a statistically significant relationship be- tween defendant race and victim race in general, and for the Black defendant/White victim race dyad specifically. The bulk of this evidence has been derived from correlational studies and from cases over relatively condensed

time frames. Methods: The current study uses data from North Carolina (n = 1,113) over several decades (1977–2009) to evaluate the link between defendant/victim racial dyad and jury death penalty decision-making. Results: Results suggest that there is an apparent “White victim effect” that can be observed in death penalty decision-making in traditional logistic regression models. Yet, once cases are matched via propensity score matching on approximately 50 case characteristics/confounders including the type of aggravators and mitigators accepted by the jury in addition to the number of aggravators and mitigators accepted, the relationship is rendered insignificant. Furthermore, these results hold for a defendant of any race killing a White victim and for the “most disadvantaged” situation for Black defendants (e.g., cases with White victims). Conclusions: The “White victim effect” on capital punishment decision-making is better considered as a “case effect” rather than a “race effect.”

© 2014 Elsevier Ltd. All rights reserved.

Introduction

One of the most widely debated issues in the criminological literature is whether there is racial discrimination in the implementation of the death penalty. Extant studies have investigated biases regarding the race of the defendant, the race of the victim, and different defendant/ victim racial dyad compositions. These important inquiries have produced a near voluminous body of work suggesting that a racial dis- parity in capital sentencing does exist (for comprehensive reviews, see Baldus & Woodworth, 2003; Kavanaugh-Earl, Cochran, Smith, Fogel, & Bjerregaard, 2008). Despite this prevailing wisdom, data and methodo- logical limitations in the prior research leave questions unanswered as to the true nature of this unexplained racial disparity. In this vein, while prior studies have indicated that overt race of defendant biases

a, Department of Criminology, 1 813 974 8024; fax: +1 813

.

are generally unfounded, two key questions regarding death penalty decision-making persist: 1) Whether the killers of White victims are disad- vantaged compared to the killers of Non-White victims in general; and 2) Whether cases involving Black defendants/White victims are disproportion- ately disadvantaged?

To address these research questions, the current study uses data from the North Carolina Capital Sentencing Project (NCCSP) to inquire as to whether (1) being a defendant of any race and killing a White victim, and (2) being a Black defendant and killing a White victim affects jury death penalty decision-making. Furthermore, the current study ap- proaches these research questions using a range of well-known and pre- viously reported significant covariates for influencing death penalty decision-making (see, for example, Baldus, Pulaski, & Woodworth, 1983; Paternoster & Brame, 2003, 2008; Williams, Demuth, & Holcomb, 2007). In addition, and more importantly, the current study also incorpo- rates a range of covariates that have heretofore not been used or avail- able in this literature or assessed simultaneously in the same analysis (in both the capital and non-capital context), particularly the novelty of incorporating the type of aggravators and mitigators accepted by the

385W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

jury alongside the total number of aggravators and mitigators accepted. Finally, the current study aims to contribute to this larger literature by assessing these linkages in a traditional logistic regression approach and then evaluating these relationships using propensity score matching, which is designed to be a more methodologically rigorous statistical ap- proach that can yield more precise estimates.

Theoretical Orientations Regarding Race and the Death Penalty

Examinations of a “White victim effect” or a “Black defendant/White victim effect” in capital sentencing decision-making are often grounded in conflict theory’s contention that Blacks are devalued relative to Whites (Hawkins, 1987) as well as focal concerns stemming from stereotypes regarding crime and violence among Blacks as “normative” and Whites as atypical (e.g., Baumer, Messner, & Felson, 2000). According to conflict theory, the social group with the power and authority (e.g., Whites and the wealthy) create and enforce laws to protect their own interests and to control the subordinate group (e.g., minorities and the poor) (Quinney, 1970). While conflict theorists maintain that the punishment of violent crimes such as homicide does not solely benefit the powerful, they argue that the socially disadvantaged are those most likely to be apprehended and punished for the commission of violent crimes.

Further, in his extension of conflict theory, Hawkins (1987) sug- gested that criminal sentencing and punishment is also affected by per- ceptions of crime seriousness that stem from differential social value placed on crime victims according to race and perceptions of threat to the normative social order. Specifically, Hawkins (1987) argued that Black crime victims are viewed as less worthy than White crime victims, that Black victimization is seen as less serious than White victimization, and that those who harm Blacks are punished less severely than those who harm Whites. Likewise, Black on White crime, especially violent crime, will be viewed as the most serious type of harm and punished the most severely because it violates both the social norms designating White power and privilege over Blacks and the prevailing legal norms against crime and violence more generally.

While conflict theory outlines the social conditions that orient stereotypes about Black criminality and White victimization, it does not provide insight into how such perceptions are translated into the individual decision-making process of criminal justice actors (Curry, 2010). Instead, Curry (2010) suggests that the focal concerns perspective may be informative in linking the social conditions outlined in conflict theory to racial biases in sentencing and punishment. Focal concerns the- ory stems from Albonetti’s (1991) contention that criminal justice actors make sentencing decisions using only limited information regarding a criminal case and the defendant involved and thus must operate under “bounded rationality” (March & Simon, 1958). Ulmer and Bradley (2006, p. 634–5) further indicate that even when extensive information regarding a defendant is available, as is usual in criminal trials, “the risk and seriousness of recidivism is never fully predictable, and a defendant’s moral character is never fully knowable”. Given this uncertainty, criminal justice actors make “situational imputations” (Steffensmeier & Demuth, 2000) regarding a defendant’s character, morality, and projected future behavior (Kramer & Ulmer, 2002, 2009; Ulmer & Bradley, 2006).

Focal concerns theory contends that criminal justice actors assess these situational imputations for three related focal concerns – the blame- worthiness of a defendant, their perceived dangerousness to the commu- nity, and the practical implications of sentencing (Steffensmeier & Demuth, 2000; Steffensmeier, Ulmer, & Kramer, 1998) – using both legal factors as well as extralegal factors such as the defendant’s and vic- tim’s age, race, and sex or in other words, variables that may be colored by stereotypes and perceptions regarding defendant/victim characteristics.

In addition, more recent discussions of focal concerns theory also highlight the importance of local court community context in actor’s in- terpretations of focal concerns (Kramer & Ulmer, 2002; Ulmer & Bradley, 2006; Ulmer, Kurlychek, & Kramer, 2007). Court community con- text may impact focal concerns because courtroom actors, which in the

previous research most often refer to judges and prosecutors, share “local- ized social worlds” stemming from their shared organizational con- straints, local politics, and norms. In this regard, we propose that the potential impact of court community context on perceptions of focal concerns may also be extended to discussions of capital juror decision-making given that they receive a unified set of instructions regarding their jury service (i.e., how to weigh aggravators and mitiga- tors) along with their shared knowledge regarding serious violent crime in their communities, fear of crime, and any community norms/expecta- tions regarding the sentencing of offenders who have been found guilty of first degree murder.

The focal concerns most relevant in capital juror decision-making are those of blameworthiness and protection of the community. Blame- worthiness refers to the idea that a defendant’s culpability directly influ- ences the severity of his/her sentence (Steffensmeier et al., 1998). Blameworthiness is usually associated with the notion of retribution – that the punishment should fit the crime, and in the capital context, that the most blameworthy defendants merit the death penalty versus life without parole. Relatedly, protection of the community centers upon the court actors’ responsibility to protect the community from re- cidivism by incapacitating and/or deterring offenders who pose a threat of future criminality. In regard to capital juror decision-making, where sentencing decisions consist of the choice between life without parole or the death penalty, the concept of general deterrence, or the idea that punishment meted out to one offender may deter other would-be offenders, may be especially germane.

Focal concerns theory maintains that the focal concerns of blamewor- thiness and dangerousness are mostly established by legal characteristics such as a defendant’s role in the crime (i.e., main offender or accomplice), criminal record, or the seriousness of the offense (Steffensmeier et al., 1998) in addition to the level of aggravation or mitigation. However, it is also expected that extralegal characteristics regarding defendant social status play a role in the attribution of blame and dangerousness. Specifi- cally, Steffensmeier et al. (1998, p. 768) contend that those who are young, non-White, and male are perceived as more blameworthy and dangerous than defendants of other age/race/sex dyads because of ste- reotypes or imagery identifying individuals in those social groups as “dangerous and crime prone”. For example, previous research demon- strates that probation officers and judges qualitatively score young, Black males’ prior records as more serious and view them as more dan- gerous and less remorseful/reformable than offenders with other age/ race/sex demographic makeups (Daly, 1994; Steffensmeier et al., 1998).

Further research examining the use of focal concerns in patterns of sentencing for serious offenders finds that young, Non-White males re- ceive more severe sentences (Auerhahn, 2007; Spohn & Holleran, 2000) and are less likely to receive downward sentencing departures (Kramer & Ulmer, 2002) than defendants in other age/race/sex dyads. In addi- tion, research on victim attributes indicates that Non-White victims are perceived as partially responsible for their victimization given per- ceptions that crime and violence, in the form of victimization and offending, is normative in the lives of minorities (Baumer et al., 2000). Likewise, female victims are more likely to be seen as innocent or sympathetic victims compared to male victims given their decreased in- volvement in crime (Baumer et al., 2000). As such, those who victimize Whites and/or women are perceived as more blameworthy than those who harm Non-Whites and/or males.

Prior Research on Race and the Death Penalty

Extensive research has documented a “race of victim effect” where- by cases involving one or more White victims are significantly more likely to result in the death penalty compared with similar cases in- volving Non-White victims (e.g., Baldus, Woodworth, & Pulaski, 1990; Keil & Vito, 1990; Paternoster & Brame, 2003; Pierce & Radelet, 2002; Radelet & Pierce, 1991; Williams & Holcomb, 2001). In fact, in a seminal study, Baldus et al. (1983) offered evidence that offenders who killed

386 W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

White victims were 4.3 times more likely to receive the death penalty, even after controlling for a myriad of important factors. Additional stud- ies from the 1980s across multiple states including California, Florida, Georgia, Illinois, Louisiana, Mississippi, New Jersey, North Carolina, Oklahoma, South Carolina, and Virginia found that the presence of a White victim was a significant predictor of a death sentence (U.S. General Accounting Office, 1990).

More recently, Baldus and Woodworth (2003) reviewed 18 studies from 1990 to 2003 and again found substantial support for the “race of victim effect” in that cases involving White victims were significantly more likely to result in the death penalty compared with cases involving Non-White victims. Most recently, analyses in Illinois (Pierce & Radelet, 2002), Nebraska (Baldus, Woodworth, & Grosso, 2002), California (Pierce & Radelet, 2005), Maryland (Pasternoster, Brame, Bacon, & Ditchfield, 2004), Colorado (Hindson, Potter, & Radelet, 2006), Louisiana (Radelet & Pierce, 2011a), and North Carolina (Radelet & Pierce, 2011b; Unah, 2011), as well as the United States military (Baldus, Grosso, Woodworth, & Newell, 2012), have provided evidence that homicide cases with White victims are more likely to receive the death penalty than comparable cases with Non-White victims.

A growing body of literature has further considered the interactive effects of defendant and victim race and demonstrated that Black defendant/White victim dyads experience the greatest likelihood of re- ceiving the death penalty compared to other defendant/victim racial dyads (e.g., Bowers & Pierce, 1980; Gross & Mauro, 1989; Keil & Vito, 1995; Paternoster & Brame, 2003, 2008; Radelet & Pierce, 1991; Smith, 1987; Sorensen & Wallace, 1995). For example, Radelet and Pierce (1991) examined all first-degree murder cases in Florida from 1976 to 1987. Their results demonstrated that even after controlling for aggra- vating factors, Blacks who killed Whites were four times more likely to receive a death sentence compared to Whites who killed Blacks.

In addition, Keil and Vito (1995) looked at a sample of death eligible cases in Kentucky from 1976 to 1991. Logistic regression analyses esti- mated with six legal and extralegal variables demonstrated that cases in- cluding a Black defendant and White victim were 1.41 times more likely to receive the death penalty compared with cases with other defendant/ victim racial dyads. Similarly, Sorensen and Wallace (1995) examined Supplemental Homicide Report (SHR) data for Missouri homicides from 1977–1991, and their analysis indicated that after controlling for both legal and extralegal factors, Blacks who killed Whites were significantly more likely to receive the death penalty than Whites who killed Blacks or Whites who killed Whites (6.7% and 20.5%, respectively).

More recently, Paternoster and Brame (2003) examined all death eligible homicides in Maryland from 1978 to 1999 to investigate the in- fluence of race and geographic region on death penalty sentencing. Mul- tivariate models were estimated including a number of control variables that were significantly correlated with race or geographic region. Their findings revealed that Black defendants who killed White victims were about 2½ times more likely to receive the death penalty than Whites who killed Whites (.043 versus .017), 3½ times more likely than Blacks who killed Blacks (.043 versus .012), and almost eleven times more like- ly than Whites who killed Blacks (.043 versus .004). In a re-analysis of the same dataset in response to comments and analysis offered by Berk, Li, and Hickman (2005), Paternoster and Brame (2008) further confirmed these findings using propensity score matching to compare cases with Black defendants and White victims with cases including other defendant/victim racial dyads. Their results again demonstrated that cases including Black defendants and White victims were disad- vantaged in regards to their likelihood of receiving the death penalty compared to cases including defendants/victims of other racial combi- nations. Finally, Unah (2011) examined a sample of death eligible homi- cides in North Carolina from 1993 to 1997 and discovered that Black defendants who killed White victims were 8% more likely to receive a death sentence than Whites who killed Whites.

It is important to note that prior research has not been unequivo- cal in finding that the race of the victim and/or the defendant/victim

racial dyad affects death penalty sentencing (McAdams, 1998). For in- stance, Klein and Rolph’s (1991) investigation of death eligible cases in California from 1977–1983 revealed null findings regarding the ef- fect of race of the defendant and race of the victim in capital sentenc- ing decisions. Additionally, Baldus et al.’s (1990) seminal Charging and Sentencing Study suggested racial disparities in capital sentencing for cases with White victims regardless of the race of the defendant. Finally, Baldus, Woodworth, Young, and Christ’s (2001) evaluation of all death eligible cases in Nebraska from 1973 to 1999 found no interaction effect for Black defendant/White victim cases relative to other defendant/vic- tim racial dyads on death penalty decision-making.

Such inconsistencies in findings regarding the impact of Black defendant/White victim dyads on capital punishment may be explained, at least in part, by the myriad of potential confounding variables that are often unavailable to researchers and therefore often omitted from prior empirical assessments of capital sentencing. For example, prior research has indicated that case characteristics such as the number of aggravating factors and/or the presence of rape prior to the homi- cide (Stauffer, Smith, Cochran, Fogel, & Bjerregaard, 2006), whether the case is prosecuted in an urban jurisdiction (Holcomb, Williams, & Demuth, 2004), and the geographic region where the case is prosecuted (Baldus et al., 2001) may affect the role of race in capital sentencing decision-making. In addition, other victim characteristics including victim illegal activity (Gillespie, Loughran, Smith, Fogel, & Bjerregaard, in press; Stauffer et al., 2006) and victim sex (Holcomb et al., 2004; Williams et al., 2007) have been shown to moderate the relationship be- tween race and the death penalty.

Research by Holcomb et al. (2004) in Ohio and Williams et al.'s (2007) reanalysis of the “Baldus” data from Georgia demonstrate that much of the apparent main effect of victim race is driven by a “White fe- male victim effect” such that those defendants who kill White females are disproportionately more likely to receive the death penalty regard- less of defendant race (but see Stauffer et al., 2006). However, few stud- ies examining the relationship between race and capital sentencing have been able to include these covariates simultaneously in their anal- ysis. For example, in Paternoster and Brame’s (2003) rigorous and well- cited study on race and capital punishment in Maryland researchers were unable to account for victim sex in their analysis (also see Berk et al., 2005; Paternoster & Brame, 2008). Similarly, Holcomb et al. (2004) were unable to include information regarding aggravating fac- tors in their investigation. Moreover, factors that have been shown to impact capital sentencing decision-making more broadly, such as miti- gating factors, have rarely been included in investigations of race and the death penalty (for an important exception, see Paternoster & Brame, 2003). Furthermore, variables demonstrated to be important in studies on race and sentencing in the non-capital context such as de- fendant age (Steffensmeier et al., 1998) have also often not been a focus of prior race and the death penalty research.

Finally, data from many of the most rigorous prior studies in this literature often present other limitations such as small sample size (e.g., Baldus et al., 2001) or the inclusion of cases from a single urban ju- risdiction (Baldus, Woodworth, Zuckerman, Weiner, & Broffitt, 1998). In addition, studies on race and the death penalty that utilize the compre- hensive and well-cited dataset from the Charging and Sentencing Study (e.g., Baldus et al., 1990; Williams et al., 2007), include cases from 1973–1980 – data from more than thirty years ago, much of which was gathered during a time when capital punishment was under a mor- atorium stemming from the U.S. Supreme Court decision in Furman v. Georgia (1972).

Current Study

Recognizing the limitations of prior research on the relationship be- tween race of the victim and the defendant/victim racial dyad composi- tion and death penalty decision-making, the present study continues this line of important inquiry by examining this potential linkage

387W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

using a propensity score matching approach. As noted earlier, numerous investigations into race and capital sentencing have not had access to im- portant variables that may impact the relationship between race and death penalty decision-making. The present research is the first examina- tion of racial dyads and capital sentencing to investigate simultaneously the potential influence of several factors that previous research on race and sentencing (in both the capital and non-capital context) has demon- strated to be important independently; such factors include urbanicity, geographical region, the number of accepted aggravators, the number of accepted mitigators, victim marital status, and, perhaps most important in light of prior research, victim sex. Furthermore, the current study pro- vides another important and novel contribution by incorporating the type of aggravators and mitigators accepted by the jury (in addition to the number of aggravators and mitigators accepted). Finally, few prior studies have utilized data that include cases from contemporary time periods and/or cases that cover a broad time frame. As such, the current study seeks to provide a methodologically rigorous evaluation of the “White vic- tim effect” and jury decisions to assess the death penalty using a large sample of capital murder cases tried in North Carolina from 1977–2009 and including up to 50 legal and extralegal covariates.

Methods

Data and Sample

The data for this research were provided by the North Carolina Capital Sentencing Project (NCCSP; see Kavanaugh-Earl et al., 2008 for a full description). The NCCSP consists of a population of jury decisions in capital murder trials in North Carolina during the period 1977–2009. We emphasize here that the dataset focuses on jury decisions, the final step, excluding appeals, in which a sentence of death or life in prison is determined. More specifically, the dataset is comprised of cases in which (1) the state secured a first-degree murder conviction; (2) sought the death penalty; and (3) the trial advanced to the sen- tencing phase whereby the jury is provided a form entitled “Issues and Recommendation as to Punishment” and instructed to record their responses regarding aggravating factors submitted by the prosecu- tion, mitigating factors submitted on behalf of the defendant, and recommendation for a death or life (currently without parole) sentence. As a note, although termed a “recommendation,” the jury’s sentencing decision is binding unless negated by detection of judicial error.

Overall, the NCCSP dataset contains all cases from North Carolina meeting these criteria for the period June 1977–December 2009 (N = 1,356). The initial date marks the return to capital punishment in North Carolina following the Furman v. Georgia (1972) decision that suspended its use and the Gregg v. Georgia (1976) decision that allowed its resumption. The latter date is the last year for which a full contingen- cy of information is available. Following a series of methodological con- siderations, a final sample of 1,113 cases was utilized for this analysis.1

Of these, juries recommended a death sentence in 540 (48.5%) of the cases.2

Measures

Dependent Variable In North Carolina, capital jurors are afforded only two sentencing

options, life in prison (currently without the possibility of parole) or the death penalty. Therefore, the dependent variable, jury recom- mendation, is expressed dichotomously (1 = death penalty, 0 = life without parole).

Independent Variables Independent variables for the analysis included up to 50 legal and

extralegal characteristics, of which some have been used as control var- iables in the death penalty literature (Baldus et al., 1990; Bjerregaard, Smith, Fogel, & Palacios, 2010) and some represent novel relevant

factors that have not been assessed simultaneously in the same analysis as of current date. In addition, several case characteristics were included that were necessary for analysis given the structure of the NCCSP data. Specifically, defendant variables included sex (Male = 1, Female = 0), age (measured continuously); whether the defendant identified as White (Yes = 1, No = 0), Black (Yes = 1, No = 0), or Other Race (Yes = 1, No = 0)3; whether the defendant had a private attorney (Yes = 1, No = 0); the number of accomplices (measured continuously); and whether there was physical evidence that linked the defendant to the crime (Yes = 1, No = 0). Victim variables included sex (Male = 1, Female = 0), age (measured continuously), whether the victim was identified as White (Yes = 1, No = 0), Black (Yes = 1, No = 0), or Other Race (Yes = 1, No = 0),4 victim’s marital status (married/ widowed = 1, single/divorced = 0), whether the victim was involved in illegal activity at the time of the homicide (Yes = 1, No = 0), whether the defendant/victim were strangers (Yes = 1, No = 0), and the number of victims (measured continuously). In addition, variables concerning the aggravators accepted included the total number of aggravators accepted by the jury and the individual acceptance of 11 statutory aggravating factors listed under North Carolina law (North Carolina General Statutes, n.d.). Measures concerning the mitigators accepted included the total number of mitigators accepted by the jury and the independent acceptance of nine statutory mitigating circum- stances listed under North Carolina law (North Carolina General Statutes, n.d.) and 11 non-statutory mitigating factors.5 Finally, four case characteristics were included that represented whether the homi- cide was prosecuted in an urban area (Yes = 1, No = 0),6 whether the sentence was after the McKoy v. North Carolina (1990) U.S. Supreme Court decision (Yes = 1, No = 0),7 the year that the sentence was imposed, and, following Unah (2011) and O'Brien and Grosso (2011), we explicitly take into account county and geographic variation with three dichotomous (Yes = 1, No = 0) variables representing the three regions of North Carolina (Mountain Region, Piedmont Region, and the Coastal Region).

Analytic Strategy

The availability and application of advanced statistical methods in criminology (propensity score matching in particular) now permits the ability for researchers to statistically approximate a quasi-experimental research design in an effort to better rule out confounding factors and in- crease the precision of estimates (for examples, see Gibson, Miller, Jennings, Swatt, & Gover, 2009; Gibson, Swatt, Miller, Jennings, & Gover, 2012; Jennings, Fox, & Farrington, 2014; Loughran et al., 2009; Richards, Jennings, et al., in press; Richards, Smith, Jennings, Bjerregaard, & Fogel, in press; Sampson, Laub, & Wimer, 2006). In its most basic form, propen- sity score matching (PSM) involves the estimation of propensity scores or the conditional probabilities of receiving a treatment on a vector of covariates/confounders (Rosenbaum & Rubin, 1983). Furthermore, these propensity scores can be considered balancing scores in the sense that the distribution of the observed covariates/confounders are independent of the treatment assignment and the resultant propensi- ty scores remove imbalances or systematic differences between the treat- ed and control groups prior to evaluating the outcome of interest (e.g., death penalty recommendation by the jury in this case). The propensity scores given the observed covariates/confounders are estimated via a logit model.

As applied in the current study, the race of the defendant and race of the victim (or race defendant/victim dyads) is considered the “treatment” for assessing the presence or absence of a “White victim effect.” Or in other words, the likelihood of receiving the death penalty in cases where a defendant of any race kills a White victim and cases where a Black defendant kills a White victim are compared to cases where a defendant of any race kills a Non-White victim and cases where a Non-Black defendant kills a Non-White victim, sepa- rately. These results would reveal whether or not there is a “White

388 W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

victim effect”, at least in terms of a statistically significant association. In order to more robustly evaluate the “White victim effect” and gen- erate more precise estimates, the estimation of the propensity score matching technique would then statistically approximate a quasi-ex- perimental research design where “treated” cases (cases that involve a defendant of any race killing a White victim and cases that involve a Black defendant killing a White victim) are matched to “control cases” (cases that involve a defendant of any race killing a Non-White victim and cases that involve a Non-Black defendant killing a Non- White victim) in an effort to rule out the potential confounding influ- ence of up to 50 legal and extralegal factors.

Thus, the analysis in the current study proceeds in six stages: First, a traditional logistic regression model is estimated in order to determine the presence/absence of a “White victim effect” alongside a host of legal and extralegal factors used as control variables. The presence/absence of a “White victim effect” for the Black Defendant/White victim dyad specifically is also assessed in a similar logistic regression model with the relevant control variables. Second, the “treated” and “control” cases are compared on up to 50 legal and extralegal factors to determine the nature and magnitude of standardized mean difference effect sizes (Cohen’s d; see Cohen, 1988) for the groups prior to propensity score matching. Third, propensity score matching methods are estimated using an R program (Hansen, 2004; Hansen & Bowers, 2008; Ho, Imai, King, & Stuart, 2007, 2011) in an effort to statistically match or balance the “treated” and “control” cases across the host of legal and extralegal factors.8 Fourth, the precision of the matching technique is assessed and the “treated” and “control” cases are compared across the legal and extralegal factors to evaluate whether any imbalances or notewor- thy standardized mean difference effect sizes are still present post- matching. Fifth, we examine the nature of the unexplained racial disparity between defendant race and victim race, or the presence of a “White victim effect”, and death penalty decision-making by comparing the death penalty decisions pre- and post-matching separately for cases: 1) that involve a defendant of any race killing a White victim versus cases that involve a defendant of any race killing a Non-White victim; and 2) that involve a Black defendant killing a White victim versus cases that involve a Non-Black defendant killing a Non-White victim. Last, in a final sensitivity analysis we estimate a propensity score weighted regression model in order to assess whether or not the effects reported in the propensity score matching analysis with the matched and reduced sample sizes of ‘treated’ and ‘control’ cases hold when estimated on the full samples of cases but adjusted by their re- spective propensity scores.

Results

Table 1 presents the results from the traditional logistic regression model evaluating the “White victim effect” alongside up to 50 legal and extralegal factors included as control variables.9 These results illustrate the presence of a statistically significant “White victim effect” (b = .331, se = .192, Odds Ratio = 1.393, p b .05, one-tailed) after con- trolling for a host of factors that vary in terms of their relevance and magnitude for affecting death penalty decision-making. Similarly, Table 2 reveals a statistically significant Black Defendant/White Victim effect on the likelihood of receiving a death penalty decision alongside controlling for the relevant legal and extralegal factors (b = 1.042, se = .420, Odds Ratio = 2.834, p b .01, one-tailed).10

Table 3 (Columns 1–3) presents the means and the standardized mean difference effect size comparisons of the propensity score con- founders prior to and post-matching. There were 22 significant mean differences in the confounders between the cases that involved a defen- dant of any race killing a White victim relative to cases that involved a defendant of any race killing a Non-White victim. Furthermore, the range in terms of the standardized mean difference effect sizes for the significantly different confounders were d = −.960 to d = .369 with a number of moderate (≥+/−.300 - ≤+/−.600: victim age, victim

married, victim involved in illegal activity, stranger victim, defendant suffered from father absence/abandonment, and case tried in Mountain Region) and large (≥+/−.600: Black defendant) effect sizes.

It is apparent from the statistical comparisons of up to 50 legal and extralegal factors that there are a number of statistically significant dif- ferences and substantive standardized mean differences between death penalty cases that involve a defendant of any race killing a White victim and a defendant of any race killing a Non-White victim. Recognizing these differences, we performed a propensity score matching analysis utilizing all of the defendant variables, victim variables, aggravating and mitigating factors, and case characteristics as potential confounders to generate propensity scores. Relying on statistical convention in the propensity score matching literature (Austin, 2009; Rosenbaum, 2002), we applied a nearest neighbor 1:1 matching algorithm with a strict caliper of .20 of the standard deviation of the logit of the propen- sity in order to exclude poor matches and increase the precision of the matches for the ‘treated’ and ‘control’ cases.11 After reviewing a series of diagnostics tests and plots (Bowers, Fredrickson, & Hansen, 2010; Hansen & Bowers, 2008), no imbalances were detected post-matching.

Turning back toward Table 3, Columns 4–6 present the means and the standardized mean difference effect size comparisons for up to 50 legal and extralegal confounders for the ‘treated’ and ‘control’ cases post-matching. As illustrated, once the cases that involve a defendant of any race killing a White victim (treated cases) are matched with sim- ilarly situated cases that involve a defendant of any race killing a Non- White victim (control cases), none of the 22 previously statistically sig- nificant mean differences between the treated and control cases are ob- served. In addition, all of the standardized mean differences between the confounders range from being small/negligible (d = −.148 to .158) to null (d = .000) effect sizes.

Fig. 1 graphically displays the relationship between being a defen- dant of any race who kills a White victim and a jury’s death penalty decision-making. As can be observed, the proportion of cases where a death penalty decision is reached is considerably greater for cases where a defendant of any race kills a White victim as opposed to a Non-White victim prior to matching {Before Matching (χ2 = 14.83, p b .001)}. However, after matching across up to 50 legal and extrale- gal factors, the “White victim effect” is no longer statistically significant or substantive {After Matching (χ2 = 0.59, p = .44)}. Or in other words, the odds of a defendant of any race receiving the death penalty for killing a White victim is 1.61 (95% CI = 1.26-2.06) and is statistical- ly significant prior to matching, but this increased odds is essentially negated and rendered insignificant after matching (OR = 1.16; 95% CI = 0.80-1.69).

The results from the previous analysis suggested that the “White vic- tim effect” was no longer apparent after applying propensity score matching. The argument could still be posed that perhaps there is a “White victim effect”, but it is being masked in the aggregate when con- sidering a defendant of any race killing a White victim. As such, the next stage of the analysis proceeded by applying the same analytic strategy but only for cases where Black defendants specifically can be considered the ‘most disadvantaged’ (e.g., the cases where a Black defendant kills a White victim).12

Table 4 (Columns 1–3) provides the means and the standardized mean difference effect size comparisons for the cases where a Black de- fendant kills a White victim compared with cases where a Non-Black defendant kills a Non-White victim before applying propensity score matching. Prior to matching, there were 17 statistically significant mean differences between the two defendant/victim dyads. Further- more, the standardized mean differences ranged from d = −.523 to d = .521 with a considerable number of standardized mean differences with moderate effect sizes (≥+/−.300 - ≤+/−.600: defendant age, defendant had a private attorney, physical evidence linked to crime, vic- tim age, victim married, stranger victim, total number of victims, total number of aggravators accepted, homicide committed in the course of another felony, homicide committed for monetary gain, total number

Table 1 Logistic Regression Results: The “White Victim Effect” in Death Penalty Decision-Making (n = 1,113)

Variables b (se) OR

Defendant Variables Defendant sex –.415 (.425) .661 Defendant age .019 (.010) 1.019 Black defendant –.207 (.194) .813 Other race defendant –.639 (.331) .528 Defendant had private attorney –.775* (.332) .460 Total number of accomplices –.025 (.075) .975 Physical evidence linked defendant to crime .409* (.171) 1.506

Victim Variables Victim sex –.061 (.171) .941 Victim age –.006 (.004) .994 White victim .331* (.192) 1.393 Victim married –.198 (.167) .820 Victim involved in illegal activity –.542* (.232) .582 Victim was a stranger –.119 (.178) .888 Total number of victims .087 (.109) 1.091

Aggravators Accepted Homicide committed in the course of another felony .539** (.180) 1.714 Homicide involved a rape .881* (.363) 2.414 Homicide committed for monetary gain .522** (.200) 1.685 Homicide was heinous and cruel 1.246*** (.179) 3.476 Homicide involved other crimes of violence .754*** (.189) 2.126 Defendant created a risk of death to more than one person .136 (.365) 1.146 Defendant previously convicted of a violent felony .539** (.193) 1.714 Homicide was committed while defendant was incarcerated .011 (.928) 1.011 Defendant previously convicted of a capital offense 1.780*** (.553) 5.932 Homicide committed in an effort to avoid arrest/escape .789** (.261) 2.202 Homicide committed to disrupt/hinder governmental function –.355 (.537) .701 Victim was law enforcement/criminal justice official .628 (.484) 1.874

Mitigators Accepted Statutory mitigators accepted Defendant had no significant history of criminal activity –.803*** (.183) .448 Defendant had mental/emotional disturbance –.387* (.179) .679 Defendant age −1.312*** (.292) .269 Defendant had impaired ability to appreciate his/her conduct −1.159*** (.185) .314 Defendant was a minor accomplice or played minor role −2.542*** (.693) .079 Defendant was under duress −1.170** (.371) .310 Defendant aided in apprehension/prosecution of another felon –.642 (.331) .526 Defendant military service –.186 (.323) .831

Non–Statutory mitigators accepted Defendant suffered from alcohol abuse –.565* (.255) .568 Defendant suffered from drug abuse .449 (.247) 1.567 Defendant was physically abused as a child/teenager .053 (.259) 1.055 Defendant was sexually abused as a child/teenager –.559 (.412) .572 Defendant was from a broken home .080 (.282) 1.084 Defendant suffered from father absence/abandonment –.357 (.240) .700 Defendant suffered from mother absence/abandonment –.137 (.338) .872 Defendant was in foster care .977 (.631) 2.656 Defendant witnessed parental misconduct .170 (.230) 1.186 Defendant had low intelligence (IQ) .364 (.261) 1.438 Defendant suffered from mental illness –.495* (.229) .610

Case Characteristics Homicide took place in urban area –.295 (.180) .745 Post McKoy v. North Carolina case 1.001* (.280) 2.721 Year sentence was imposed –.041* (.019) .960 Tried in piedmont region .369 (.250) 1.447 Tried in coastal region .278 (.244) 1.321

Model Diagnostics χ2 = 431.577 −2 log likelihood = 1,110.390 Nagelkerke R2 = .429

ABBREVIATIONS: b = unstandardized coefficient; se = standard error; OR = Odds Ratio. *p b .05 **p b .01 ***p b .001 (one-tailed). [The number of asterisks relates to the values as shown above.]

389W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

of statutory and non-statutory mitigators accepted, defendant had mental/emotional disturbance, defendant had impaired ability to ap- preciate his/her conduct, defendant suffered from alcohol abuse, homi- cide took place in an urban area, case was tried in Piedmont Region, and case was tried in Coastal Region).

Similar to the analytical strategy that was followed in the aggregate defendant race/White victim dyad and following statistical convention in the propensity score matching literature (Austin, 2009; Rosenbaum,

2002), we applied a nearest neighbor 1:1 matching algorithm with a strict caliper of .20 of the standard deviation of the logit of the propen- sity to identify the matched samples of ‘treated’ and ‘control’ cases balanced on the up to 50 legal and extralegal factors. Once again, the diagnostic tests and plots revealed that no imbalances were detected post-matching (Bowers et al., 2010; Hansen & Bowers, 2008), and that covariate balance was noticeably improved in the matched sample of “treated” and “control” cases. Furthermore, an examination of the

Table 2 Logistic Regression Results: The “Most Disadvantaged” Situation for Black Defendants (When They Kill Whites) Relative to When Non-Blacks Kill Non-Whites (n = 320)

Variables b (se) OR

Defendant Variables Defendant age .058 (.021) 1.060 Defendant had private attorney .017 (.788) 1.017 Total number of accomplices .055 (.161) 1.056 Physical evidence linked defendant to crime .837* (.376) 2.309

Victim Variables Victim sex –.143 (.369) .867 Victim age –.028 (.008) .973 Victim married –.843* (.364) .430 Victim involved in illegal activity –.177 (.502) .838 Victim was a stranger .084 (.357) 1.088 Total number of victims .065 (.311) 1.068 Black defendant/White victim 1.042* (.420) 2.834

Aggravators Accepted Homicide committed in the course of another felony 1.024** (.376) 2.784 Homicide involved a rape .613 (.758) 1.847 Homicide committed for monetary gain .535 (.384) 1.707 Homicide was heinous and cruel 1.006** (.373) 2.735 Homicide involved other crimes of violence .831 (.448) 2.295 Defendant created a risk of death to more than one person .408 (.766) 1.504 Defendant previously convicted of a violent felony .402 (.399) 1.517

Mitigators Accepted Statutory mitigators accepted Defendant had no significant history of criminal activity −1.056** (.380) .348 Defendant had mental/emotional disturbance –.243 (.389) .784 Defendant age −1.518** (.553) .219 Defendant had impaired ability to appreciate his/her conduct −1.757*** (.797) .172 Defendant was under duress −1.571* (.797) .208 Defendant aided in apprehension/prosecution of another felon .558 (.676) 1.746 Defendant military service –.828 (.770) .437

Non–Statutory mitigators accepted Defendant suffered from alcohol abuse .195 (.630) 1.216 Defendant suffered from drug abuse .869 (.519) 2.384 Defendant was physically abused as a child/teenager –.131 (.636) .877 Defendant was from a broken home .646 (.689) 1.909 Defendant suffered from father absence/abandonment –.049 (.475) .953 Defendant suffered from mother absence/abandonment –.652 (.747) .521 Defendant witnessed parental misconduct –.734 (.537) .480 Defendant had low intelligence (IQ) .046 (.590) 1.047 Defendant suffered from mental illness –.038 (.552) .962

Case Characteristics Homicide took place in urban area .212 (.410) 1.236 Post McKoy v. North Carolina case .958 (.651) 2.606 Year sentence was imposed –.099* (.043) .905 Tried in piedmont region .850 (.605) 2.339 Tried in coastal region .935 (.567) 2.546

Model Diagnostics χ2 = 156.540 −2 log likelihood = 286.624 Nagelkerke R2 = .516

ABBREVIATIONS: b = unstandardized coefficient; se = standard error; OR = Odds Ratio. *p b .05 **p b .01 ***p b .001 (one-tailed). [The number of asterisks relates to the values as shown above.]

390 W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

post-matching means and the standardized mean difference effect size comparisons demonstrated that none of the 17 previously statistically significant differences between the cases where a Black defendant killed a White victim as opposed to the cases where a Non-Black defendant killed a Non-White victim were detected after the matching procedure was applied (Table 2, Columns 4–6). In addition, all of the standardized mean differences between the confounders range from being small/ negligible (d = −.219 to .159) to null (d = .000) effect sizes.

Fig. 2 presents a graphical display of the proportional differences that exist in juror death penalty decision-making for cases involving a Black defendant killing a White victim and cases in which a Non-Black defendant kills a Non-White victim. These results suggest that the “White victim effect” that was initially observed in the aggregate defen- dant race/White victim dyad is also present and considerably robust in the “most disadvantaged” situation for Black defendants. Specifically, Black defendants who kill White victims are given the death penalty

at a greater proportion compared to Non-Black defendants who kill a Non-White victim {Before Matching (χ2 = 9.21, p b .01)}. Yet, once these cases are matched on up to 50 legal and extralegal factors, the “White victim effect” observed in the most disadvantaged situation for Black defendants is no longer significant {After Matching (χ2 = 0.14, p = .71)}. Or, in other words, the odds of a Black defendant receiving the death penalty for killing a White victim is 2.16 (95% CI = 1.31- 3.56) and statistically significant prior to matching, but the effect is re- duced to nearly zero and insignificant after matching (OR = 0.87; 95% CI = 0.41-1.82).

In the final stage of the analysis, we estimate a propensity score weighted regression model utilizing the full sample of cases where a de- fendant of any race kills a White victim and cases where a defendant of any race kills a Non-White victim (n = 1,113) and a propensity score weighted regression model based on the full sample of cases where a Black defendant kills a White victim and where a Non-Black defendant

Table 3 Statistical Comparisons of Factors Before (n = 1,113) and After (n = 436) Matching Cases Involving a Defendant of Any Race and a White Victim (treatment group) with Cases Involving a Defendant of Any Race and a Non-White Victim (control group)

Variables Before Matching n = 688 Any Race Defendant/ White Victim

n = 425 Any Race Defendant/ Non-White Victim

Standardized Mean Difference

After Matching n = 218 Any Race Defendant/ White Victim

n = 218 Any Race Defendant/ Non-White Victim

Standardized Mean Difference

Defendant Variables Defendant sex .04 .04 .000 .02 .02 .000 Defendant age 28.68 28.65 .002 27.44 28.39 –.107 Black defendant .33⁎ .78 –.960 .60 .66 –.117 Other race defendant .05⁎ .09 –.158 .09 .11 –.081 Defendant had private attorney .07⁎ .04 .121 .05 .05 .000 Total number of accomplices .78 .80 –.015 .99 .86 .098 Physical evidence linked defendant to crime .72⁎ .65 .158 .68 .68 .000

Victim Variables Victim sex .41 .45 –.068 .39 .43 –.075 Victim age 41.90⁎ 34.00 .369 38.58 38.08 .024 Victim married .46⁎ .28 .356 .41 .36 .110 Victim involved in illegal activity .12⁎ .22 –.301 .18 .20 –.042 Victim was a stranger .41⁎ .24 .351 .42 .34 .158 Total number of victims 1.47⁎ 1.67 –.244 1.55 1.50 .068

Aggravators Accepted Total number of aggravators accepted 2.13 2.00 .114 2.05 2.13 –.073 Homicide committed in the course of another felony .40⁎ .34 .124 .39 .38 .009 Homicide involved a rape .08 .06 .074 .06 .07 –.050 Homicide committed for monetary gain .30⁎ .19 .240 .26 .29 –.060 Homicide was heinous and cruel .40⁎ .30 .191 .33 .33 .000 Homicide involved other crimes of violence .43 .44 –.026 .46 .43 .056 Defendant created a risk of death to more than one person .04⁎ .06 –.156 .04 .05 –.025 Defendant previously convicted of a violent felony .25⁎ .33 –.192 .24 .30 –.148 Homicide was committed while defendant was incarcerated .01 .01 –.047 .01 .01 .000 Defendant previously convicted of a capital offense .03 .03 –.013 .02 .03 –.086 Homicide committed in an effort to avoid arrest/escape .11 .10 .045 .11 .11 .000 Homicide committed to disrupt/hinder governmental function

.02 .01 .068 .02 .02 .000

Victim was law enforcement/criminal justice official .05⁎ .01 .176 .04 .01 .109 Mitigators Accepted Total number of statutory and non–statutory mitigators accepted

9.87 10.93 –.099 10.25 9.73 .048

Statutory mitigators accepted Defendant had no significant history of criminal activity

.34 .34 –.008 .35 .29 .136

Defendant had mental/emotional disturbance .46 .50 –.077 .48 .42 .120 Defendant age .11⁎ .15 –.133 .13 .11 .059 Defendant had impaired ability to appreciate his/her conduct

.31 .35 –.069 .30 .28 .059

Defendant was a minor accomplice or played minor role .03 .05 –.096 .03 .03 .000 Defendant was under duress .07 .08 –.062 .09 .07 .073 Defendant aided in apprehension/prosecution of another felon

.08 .06 .078 .08 .08 .000

Defendant military service .07 .07 .004 .08 .06 .055 Non–Statutory mitigators accepted Defendant suffered from alcohol abuse .17 .15 .069 .16 .14 .036 Defendant suffered from drug abuse .20 .17 .065 .22 .17 .104 Defendant was physically abused as a child/teenager .15 .16 –.044 .14 .15 –.013 Defendant was sexually abused as a child/teenager .04 .06 –.074 .03 .06 –.135 Defendant was from a broken home .08 .12 –.124 .09 .11 –.067 Defendant suffered from father absence/abandonment .12⁎ .23 –.318 .19 .17 .056 Defendant suffered from mother absence/abandonment .07 .06 .067 .07 .06 .035 Defendant was in foster care .02⁎ .01 .114 .02 .01 .059 Defendant witnessed parental misconduct .17 .21 –.092 .17 .17 .000 Defendant had low intelligence (IQ) .09 .12 –.118 .10 .09 .032 Defendant suffered from mental illness .15 .16 –.027 .14 .12 .051

Case Characteristics Homicide took place in urban area .44 .49 –.089 .38 .47 –.185 Post McKoy v. North Carolina case .69⁎ .81 –.254 .71 .74 –.060 Year sentence was imposed 1992.98⁎ 1994.98 –.287 1993.24 1993.91 –.096 Tried in mountain region .19⁎ .07 .296 .15 .11 .090 Tried in piedmont region .44 .44 .002 .38 .45 –.139 Tried in coastal region .38⁎ .49 –.240 .45 .45 .000

⁎ p b .05.

391W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

Fig. 1. Death Penalty Decision Differences Before and After Matching Cases Involving a De- fendant of Any Race and a White Victim (treatment group) with Cases Involving a Defen- dant of Any Race and a Non-White Victim (control group). NOTES: Death Penalty Decision: Before Matching (χ2 = 14.83, p b .001) / After Matching (χ2 = 0.59, p = .44).

392 W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

kills a Non-White victim (n = 320). This type of analysis is beneficial as it allows for the estimation of the “White victim effect” without the related sample size loss when relying on the propensity score matched samples' analysis previously reported, but still with the benefit of adjusting the cases by their propensity to receive a jury recommendation for the death penalty derived from accounting for the up to 50 legal and extrale- gal confounders. The results of these two propensity score weighted re- gression models are displayed in Column 4 of Table 5, and they both suggest that the “White victim effect” is no longer statistically significant after the propensity score adjustment is taken into account.

Also, Table 5 provides the estimate of the “White victim effect” for all of the stages of analysis that were conducted in this study for both case comparisons (e.g., comparing cases where a defendant of any race kills a White victim and cases where a defendant of any race kills a non-White victim and comparing cases where a Black defendant kills a White victim and where a Non-Black defendant kills a Non-White victim). What is apparent from these comparisons is that the “White victim ef- fect” is statistically significant as a stand-alone effect (Column 1, Rows 1 and 2), and its effect is still observable and significant after controlling for an array of legal and extralegal factors in a traditional logistic regres- sion model (Column 2, Rows 1 and 2). However, the results presented in Columns 3 and 4 illustrate that the “White victim effect” is noticeably reduced in magnitude and rendered statistically non-significant after matching on confounders via propensity score matching (Column 3, Rows 1 and 2) or when utilizing propensity score weighted regression (Column 4).

Discussion

The current study sought to provide a methodologically rigorous ex- amination of the “White victim effect” and death penalty decision- making. Utilizing data from a large sample (n = 1,113) of death penalty cases that have occurred in North Carolina over nearly four decades (1977–2009) and including covariates yet to have been estimated simultaneously in the same analysis or were previously unavailable (particularly the novelty of incorporating the type of aggravators and mitigators accepted by the jury in addition to the number of aggravators and mitigators accepted), our investigation represents one of the largest and most extensive studies in the literature looking at the “White victim effect” as it relates to juror’s decisions to recommend the death penalty. Furthermore, the application of propensity score matching methods places this study in a rare situation to assess the “White victim effect" alongside traditional logistic regression models in an effort to yield more precise estimates for comparison (see also Berk et al., 2005; Paternoster & Brame, 2008). A summary of the key findings that emerged from this effort is highlighted below.

Our initial analysis revealed that the “White victim effect” that is observable in isolation (as an independent effect) is still apparent after controlling for an array of legal and extralegal confounders esti- mated via traditional logistic regression models. Yet, after rigorously ac- counting for up to 50 legal and extralegal confounders through the application of propensity score matching along with the ability to ac- count for the type of aggravators and mitigators accepted by the jury in addition to the number of aggravators and mitigators accepted, we found no significant or substantive differences in the likelihood of re- ceiving the death penalty for cases with White victims (and defendants of any race) compared to cases with Non-White victims (and, again, de- fendants of any race). However, in line with conflict theory (Hawkins, 1987; Quinney, 1970) vis-à-vis focal concerns (Steffensmeier & Demuth, 2000; Steffensmeier et al., 1998), a race of victim effect may be most pronounced in cases that include a Black defendant and a White victim, and thus, we estimated a second model comparing cases with Black defendants/White victims versus cases with Non-Black defendants/ Non-White victims. Again, after balancing cases on a myriad of poten- tially confounding legal and extralegal factors and accounting for the type of aggravators and mitigators accepted by the jury in addition to the number of aggravators and mitigators accepted, we found no sta- tistical or substantive evidence that cases with White victims were more likely to result in the death penalty compared with cases involving similarly situated Non-White victims, even when the defendant is Black. Both of these findings regarding the absence of a “White victim effect” hold in propensity score weighted regression models as well, which make use of the full samples of comparison cases. This robust evidence is considerably in line with an emerging body of literature that has evaluated the role of race in a host of related outcomes such as speeding (Tillyer & Engel, 2012), school suspensions (Wright, Morgan, Coyne, Beaver, & Barnes, 2014), and criminal justice processing (Beaver et al., 2013) and have reported that race effects are either not observed or considerably reduced once important and relevant be- havioral and other germane covariates are considered.

Despite these results, it is not our contention that race plays no role in juror’s capital sentencing decision-making, but instead, we join other scholars (see for example, Berk et al., 2005; Paternoster & Brame, 2003, 2008) in their arguments that the relationship between race and capital punishment is more complex and nuanced than much of the extant lit- erature has suggested.13 For instance, in the current study, analyses es- timated prior to matching reveal significant and substantive differences on more than 20 potentially confounding characteristics of cases. There- fore, we suggest that the “race of victim effect” demonstrated in prior studies regarding the implementation of the death penalty may actually be better described as a “case effect” that represents the real and cumu- lative differences in many cases involving White victims and/or Black defendants that have largely not been captured using traditional logistic regression techniques and relying on death penalty cases from relative- ly condensed time periods.

For example, in cases with White victims, a significantly greater pro- portion of homicides were committed for monetary gain and/or in- volved a victim who was a law enforcement officer or criminal justice official. Furthermore, a significantly greater proportion of cases that have been considered the “most disadvantaged” – cases with Black de- fendants and White victims – were committed for monetary gain and during the course of another felony, in addition to these cases having a greater number of accepted aggravators and a fewer number of ac- cepted mitigators. As such, these cases were the most disadvantaged in the eyes of jurors, but this disadvantage cannot simply be attributed to the race of the victim and/or the race of the defendant alone. Such findings are consistent with research on focal concerns theory that con- siders assessments of defendant blameworthiness and dangerousness as being influenced by both legal and extralegal factors (Steffensmeier & Demuth, 2000; Steffensmeier et al., 1998).

In the present findings, the cumulative effects of both extralegal disadvantage – being a Black defendant and murdering a White victim

Table 4 Statistical Comparisons of Factors Before (n = 320) and After (n = 114) Matching Cases Involving a Black Defendant and a White Victim (treatment group) with Cases Involving a Non-Black Defendant and a Non-White Victim (control group)

Variables Before Matching n = 228 Black Defendant/ White Victim

n = 92 Non-Black Defendant/ Non-White Victim

Standardized Mean Difference

After Matching n = 57 Black Defendant/ White Victim

n = 57 Non-Black Defendant/ Non-White Victim

Standardized Mean Difference

Defendant Variables Defendant age 26.09⁎ 29.82 –.476 29.52 29.98 .117 Defendant had private attorney .03 .10 –.446 .05 .07 –.109 Total number of accomplices 1.04 .87 .128 .83 .79 .026 Physical evidence linked defendant to crime .77⁎ .61 .388 .63 .65 –.042

Victim Variables Victim sex .38 .34 .092 .37 .39 –.036 Victim age 45.63⁎ 35.57 .447 39.75 39.47 .012 Victim married .47⁎ .26 .425 .35 .33 .035 Victim involved in illegal activity .14 .19 –.128 .19 .23 –.101 Victim was a stranger .62⁎ .37 .521 .42 .42 .000 Total number of victims 1.39⁎ 1.55 –.235 1.40 1.47 –.101

Aggravators Accepted Total number of aggravators accepted 2.40⁎ 1.78 .469 1.81 1.88 –.054 Homicide committed in the course of another felony .45⁎ .27 .352 .26 .32 –.106 Homicide involved a rape .10 .05 .142 .07 .09 –.059 Homicide committed for monetary gain .41⁎ .19 .461 .26 .25 .036 Homicide was heinous and cruel .35 .27 .165 .33 .30 .073 Homicide involved other crimes of violence .41 .47 –.121 .40 .40 .000 Defendant created a risk of death to more than one person

.02 .07 –.295 .04 .04 .000

Defendant previously convicted of a violent felony .27 .27 .000 .32 .26 .118 Defendant previously convicted of a capital offense .03 .03 .000 .02 .05 –.219 Homicide committed in an effort to avoid arrest/escape .13 .10 .088 .05 .09 –.105

Mitigators Accepted Total number of statutory and non–statutory mitigators accepted

8.15⁎ 11.61 –.498 9.86 9.84 .003

Statutory mitigators accepted Defendant had no significant history of criminal activity .32 .40 –.175 .30 .32 –.038 Defendant had mental/emotional disturbance .35⁎ .52 –.367 .44 .46 –.037 Defendant age .15 .11 .113 .11 .09 .049 Defendant had impaired ability to appreciate his/her conduct

.23⁎ .39 –.388 .30 .32 –.042

Defendant was a minor accomplice or played minor role .04 .04 .000 .11 .07 .171 Defendant was under duress .06 .13 –.260 .09 .11 –.071 Defendant aided in apprehension/prosecution of another felon

.11 .07 .130 .05 .09 –.114

Defendant military service .04 .08 –.157 .07 .05 .085 Non–Statutory mitigators accepted Defendant suffered from alcohol abuse .08⁎ .21 –.445 .14 .16 –.063 Defendant suffered from drug abuse .20 .20 .000 .19 .23 –.087 Defendant was physically abused as a child/teenager .09 .15 –.227 .07 .11 –.124 Defendant was sexually abused as a child/teenager .02 .04 –.197 .05 .05 .000 Defendant was from a broken home .06 .11 –.197 .07 .09 –.073 Defendant suffered from father absence/abandonment .18 .21 –.388 .21 .23 –.046 Defendant suffered from mother absence/abandonment .07 .02 .189 .05 .04 .069 Defendant witnessed parental misconduct .12 .17 –.171 .12 .12 .000 Defendant had low intelligence (IQ) .10 .08 .082 .07 .09 –.058 Defendant suffered from mental illness .13 .14 –.042 .11 .11 .000

Case Characteristics Homicide took place in urban area .41⁎ .23 .356 .29 .26 .071 Post McKoy v. North Carolina case .73⁎ .75 –.049 .72 .74 –.039 Year sentence was imposed 1993.36 1994.15 –.114 1993.23 1993.70 –.068 Tried in mountain region .11 .08 .119 .12 .09 .110 Tried in piedmont region .43⁎ .21 .450 .30 .26 .071 Tried in coastal region .46⁎ .72 –.523 .58 .65 –.141

⁎ p b .05.

393W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

as well as legal factors – greater numbers of aggravators and/or fewer numbers of mitigators – may heighten jurors’ perceptions of defendant blameworthiness. In addition, cases involving Black defendants and White victims that were also committed for monetary gain (i.e., during a robbery) may invoke stereotypical imagery regarding the “dangerous Black offender”. Nevertheless, our results suggest that the “White victim effect” that is generally apparent in most of the prior research is still

apparent alongside legal and extralegal factors as control variables in traditional logistic regression models, but the effect does not hold once propensity score methods are used in either type of application, e.g., applying the propensity score estimates for matching purposes or utilizing the propensity score estimates in a propensity score weighted regression. These results lead us to echo the important suggestions made by Paternoster and Brame (2008, p. 994), that “critical thinking”

Fig. 2. Death Penalty Decision Differences Before and After Matching Cases Involving a Black Defendant and a White Victim (treatment group) with Cases Involving a Non- Black Defendant and a Non-White Victim (control group). NOTES: Death Penalty Decision: Before Matching (χ2 = 9.21, p b .01) / After Matching (χ2 = 0.14, p = .71).

394 W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

regarding the specific “stories that link race and decisions” is imperative for unpacking any potential complex relationships between race and capital punishment decision-making.

On another front, these results have direct implications for the victim- offender overlap literature (Jennings & Reingle, 2012; Maldonado-Molina, Jennings, Tobler, Piquero, & Canino, 2010) as it seems that when offenders kill similarly situated victims then the death penalty is not invoked at the same rate compared to when an offender kills a stranger (who is a non-offender). For instance, when a young black male robs a store and kills a white clerk (who is a non-offender) then the death penalty decision is certainly more likely in contrast to when a young black male who has a prior criminal history kills another similarly situated young black male with a prior criminal history in the course of commit- ting another felony. And, this same scenario applies to cases where the homicide is white-on-white and both the offender and the victim are similarly situated and evince analogous risky lifestyles and mirror a parallel profile of case characteristics and circumstances.

In light of these findings, further research is necessary to investigate why cases involving White victims that are tried capitally, at least in North Carolina, differ so greatly with regard to their legal and extralegal case characteristics. Specifically, it may be that there are qualitative dif- ferences in offending behavior that leads cases with defendants of any race and White victims to be more worthy of the death penalty in the eyes of jurors. In this regard, it may be that prosecutorial decisions re- garding first degree murder and death penalty notice leads to a body of death eligible cases that are widely different dependent on the race of the victim and/or race of the defendant. In fact, Paternoster and Brame’s (2008) results demonstrating that race was an important factor in determining prosecutors’ decisions to seek the death penalty in

Table 5 Comparison of Methods for Estimating the “White Victim Effect” in Death Penalty Decision-Ma

Unweighted Effecta

Effect Adjusted for Control Variables in a Traditional Logistic Regression Modelb

b (se) OR b (se) OR

White Victim .479*** (.125) 1.614 .373* (.190) 1.45 Black Defendant/ White Victim

.769** (.256) 2.158 1.042* (.420) 2.83

ABBREVIATIONS: b = unstandardized coefficient; se = standard error; OR = Odds Ratio. *p b .05 **p b .01 ***p b .001 (one-tailed). [The number of asterisks relates to the values as sh

a Total sample size for White Victim model (n = 1,113). Total sample size for Black defenda b Total sample size for White Victim model (n = 1,113). Total sample size for Black defend c Total sample size for White Victim model (n = 436). Total sample size for Black defendan d Total sample size for White Victim model (n = 1,113). Total sample size for Black defend

Maryland provide evidence supporting this latter assumption (see also Baldus & Woodworth, 2003; Paternoster, 1984).14 Therefore, we pro- pose that a critical area for further research involves the strategic collection of data and empirical analysis that links the reasons why prosecutors report choosing to seek the death penalty in each case and the cases’ documented legal and extralegal case characteristics, including racial composition that may or may not play a direct role in this decision-making process (Pokorak, 1998).

In this same vein, the selection bias that is apparent when evaluating race and the death penalty relying solely on post-conviction data (as most prior research has done as well) does have additional implica- tions that warrant discussion to contextualize our results. Specifically, self-selection into a jury trial and sentencing rather than pleading guilty, prosecutorial decisions for which cases to take to a jury trial, and certain factors such as total number of aggravators accepted may in fact be mechanisms by which victim race affects death penalty decision- making. A similar argument could be made for total number of mitiga- tors accepted as well. For example, jurors may view defendants who kill White victims as being more deserving of the death penalty, and as such, the jurors may be inclined to accept a greater number of aggravators and/or a lesser number of mitigators in cases with White victims. Thus, in this scenario, victim race (being a White victim) is not directly related to the jury’s decision to recommend the death penalty, rather the role of victim race is mediated by or operates through the role of certain legal and extralegal factors such as the total number of aggravators accepted and/or the total number of mitiga- tors accepted (which are factors that occur “after” victim race). Ulti- mately, the legal and extralegal factors, especially the number of mitigators and aggravators accepted, and exposure to the death penalty is endogenous.

It is important to note several limitations in an effort to situate these results in the larger literature on death penalty decision-making. For ex- ample, although these data represent a population of jury decisions in capital murder trials from 1977–2009 where the jury carried out their specific instructions regarding aggravation and mitigation, and where at least one aggravator was found so that the case remained death pen- alty eligible, the data is from only one state. There is certainly likely to be variability across jurisdictions and states in how the implementation and application of the death penalty and how the complex and nuanced process of death penalty decision-making plays out in their respective settings that has implications for generalizability. In fact, this variability has been previously documented in Illinois (Pierce & Radelet, 2002), Nebraska (Baldus et al., 2002), California (Pierce & Radelet, 2005), Maryland (Pasternoster et al., 2004), Colorado (Hindson et al., 2006), and North Carolina (Radelet & Pierce, 2011b; Unah, 2011) as well as in the United States military (Baldus et al., 2012). Future research is en- couraged to collect data from a similar and lengthy time period and apply rigorous empirical methods such as propensity score matching in an effort to determine whether the findings presented here can be extended beyond North Carolina.

king

Effect After Propensity Score Matching on Confoundersc

Effect After Propensity Score Weighted Regressiond

b (se) OR b (se) OR

2 .148 (.192) 1.159 .226 (.163) 1.253 4 –.143 (.378) 0.867 .159 (.338) 1.173

own above.] nt/White victim model (n = 320). ant/White victim model (n = 320). t/White victim model (n = 114). ant/White victim model (n = 320).

395W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

Related to the discussion of limitations above, it is worthy to note that our results differ somewhat from those that have recently been published by others relying on North Carolina data. For example, Unah (2011; see also Radelet & Pierce, 2011b) relied on five years of first and second-degree murder cases (1993–1997) and reported a race of victim effect after controlling for legal factors, institutional factors related to politics, and structural factors. Unah also demonstrated that the racial disparity was observable at the jury stage, and that it was also affected by factors at the prosecutorial stage. Comparatively, we replicated these results from the jury stage (as we do not have data from the prosecutorial stage) in our traditional logistic regression analysis using nearly 35 years of North Carolina death penalty data (1977–2009). Yet, our propensity score analysis relying on up to 50 legal and extralegal covariates (including many of those used in Unah’s analysis) and with the novelty of accounting for the type of aggravators and mitigators accepted by the jury alongside the total number of aggravators and mitigators accepted suggested that the racial disparity was not apparent after applying propensity score matching.

Therefore, it is quite possible that our results differ from those produced by Unah (2011) and others in North Carolina for a variety of reasons such as: 1) sample differences (we rely on a near population of cases from 1977–2009 versus a near population of cases from 1993–1997); 2) time period/length of time differences (35 years of death penalty data from 1977–2009 versus 5 years of death penalty data from 1993–1997); 3) absence of data from the prosecutorial stage versus Unah (2011) having included measures observed at this stage; 4) inclusion of the type of aggravating and mitigating factors ac- cepted in addition to the total number of aggravators and mitigators ac- cepted versus not; and/or 5) differences in the precision of estimates derived from propensity score matching versus modeling the racial dis- parity in death penalty decision-making using traditional regression- based approaches and their related limitations (e.g., sample size con- cerns, multicollinearity, etc.). Having said this, we are not suggesting that propensity score methods are a panacea for investigations into ra- cial disparity and death penalty decision-making as these methods also have limitations as well (e.g., model misspecification as PSM only accounts for observed and observable covariates, potential for hidden bias, etc.) (see Pearl, 2000). As always, caution should be taken when interpreting the results from any empirical analysis recognizing the specific strengths and limitations of the statistical methods employed.

In addition, and related to the issue of generalizability, the evidence presented here suggests that in death eligible cases that do go to trial in North Carolina, the likelihood of a jury arriving at a death penalty deci- sion as opposed to a life (currently without parole) decision is about 50/ 50 in the aggregate. Or, in other words, a jury’s decision in favor of the death penalty is an outcome in about half of the death penalty eligible cases that go to trial, proceed to the penalty phase, and the jury finds at least one aggravating circumstance. Furthermore, as reported previ- ously, the North Carolina data suggest that more than 20 legal and extralegal case characteristics significantly vary when comparing cases involving White victims to cases involving Non-White victims. Thus, the findings from this particular study may not necessarily apply to ju- risdictions where a decision to send a death penalty eligible case to trial and to a jury for recommendation and/or where a jury’s recom- mendation for the death penalty are statistically rarer events. Similarly, these results may not extend to jurisdictions where there is significantly and/or substantively less variability in the legal and extralegal case characteristics observed in death penalty cases that go to trial and when a jury is commissioned to provide a recommendation as to punishment. Finally, another interesting point for discussion centers on the negative effect of time (year that sentence was imposed) on death penalty decision-making that was observed in the traditional logistic regression models. This is suggestive of a trend that has received a growing amount of attention in the popular press and media where the application of the death penalty, in general, is becoming less com- mon in more recent times and public perceptions have varied to a

degree in their support of the death penalty in the U.S. and abroad (Applegate, Cullen, & Fisher, 2000; Behnken, Caudill, Berg, Trulson, & DeLisi, 2011; Bohm & Vogel, 2004; Bohm, Vogel, & Maisto, 1993; Jiang, Lambert, & Nathan, 2009; Liang, 2005; Lu & Zhang, 2005; Mancini & Mears, 2010; Wozniak & Lewis, 2010). Thus, it is an empirical question as to whether or not the application of this sentence (perhaps now) be- coming reserved for only the most aggravated and least mitigated cases will have an influence on racial disparity going forward. Future research is encouraged to investigate research questions such as this as the data accumulates.

Nevertheless, the accumulated evidence gleaned from the larger body of literature on race and death penalty decision-making generally illustrate a “race of victim effect” that is correlational and observable in analyses of data from varying snapshots of time (Baldus et al., 1990; Baldus et al., 2001; Keil & Vito, 1995; Paternoster & Brame, 2003, 2008; Pierce & Radelet, 2002; Radelet & Pierce, 1991; Williams & Holcomb, 2001). Still, other research has identified alternative factors such as the number of aggravating factors, the presence of rape prior to the commission of the homicide (Stauffer et al., 2006), and victim in- volvement in illegal activity (Gillespie et al., in press; Stauffer et al., 2006), along with victim sex (Holcomb et al., 2004; Richards, Jennings, et al., in press; Richards, Smith, et al., in press; Williams et al., 2007) and the geographic region where the case was prosecuted (Baldus et al., 2001) that may condition the relationship between race and death penalty decision-making. Our investigation using data from a large sample (n = 1,113) of death penalty cases that have occurred in North Carolina over nearly four decades (1977–2009) seems to comple- ment and extend the important research that has been carried out thus far in the extant race and death penalty decision-making literature by illustrating that there is an observable “race of victim effect”, e.g., a “White victim effect” specifically. Yet, its role in death penalty decision- making is much more complex and nuanced, as others have suggested (Berk et al., 2005; Paternoster & Brame, 2003, 2008), and may be better described as a “case effect.”15

In conclusion, the contribution of the current study is that it demon- strates that legal and extralegal factors that have not been previously available or assessed simultaneously in prior race and death penalty re- search such as urbanicity, geographical region, the number of accepted aggravators, the number of accepted mitigators, victim marital status, and, perhaps most important in light of prior research, victim sex may play some role in influencing jury decision-making. The simultaneous use of these novel factors, alongside other more well-known and docu- mented legal and extralegal factors (e.g., victim and defendant age, victim/defendant relationship, defendant prior record) (Baldus et al., 1990; Paternoster & Brame, 2003, 2008; Williams et al., 2007), all likely to varying degrees affect jury decision-making in the capital context. Having said this, before now, this complex process with this range of factors at the death penalty phase, particularly with the ability to ac- count for the type of aggravators and mitigators accepted by the jury in addition to the number of aggravators and mitigators accepted, has not been evaluated generally or using propensity score matching tech- niques specifically.

In the end, what these results seem to point toward is that when modeling this process in a traditional logistic regression model frame- work within the constraints of recognizing the rarity of the event (cases that are considered for the death penalty and cases where the death penalty is recommended as the sentence), the “White victim effect” is illustrated as having an independent effect. Yet, when con- straints such as sample size concerns relative to the number of variables that can be estimated in a traditional logistic regression model and other issues including multicollinearity are no longer an influence, pro- pensity score methods seem to yield more precise estimates of the “White victim effect” by making use of the range of well-known and novel legal and extralegal case characteristics that contribute to the complex and varied pathway that ultimately lead jurors to recommend the death penalty in any one particular case.

396 W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

Notes

1. Reasons for the elimination of cases leading to the final sample used in the current analysis were as follows: (1) Seventy-four cases were not available for analysis because the jury did not find any aggravators. When this occurs, the defendant is no longer eligible for a death sentence and the deliberations end there. In this situation, the jury does not consider mitigating factors; (2) Sixty-nine cases were removed because the jury became deadlocked and, despite their instructions, did not complete the “Issues and Recommen- dation as to Punishment” form in which a death or life sentence is recorded, rendering those cases absent of any information concerning jury responses to aggravation or mitiga- tion. When a jury does not provide a sentencing recommendation, there is an automatic default to life in prison; (3) Ninety-two cases were removed because the earliest versions of the “Issues and Recommendation as to Punishment” form presented mitigators as a group; jurors were asked whether they accepted one of more of them and were not re- quired to make an independent determination of each mitigator; and (4) Eight cases were removed because victim marital status could not be determined. Despite the necessary re- moval of these cases, the resulting data set consists of all capital murder jury decisions in North Carolina during the specified time period with data sufficient to carry out the anal- yses reported here.

2. It is important to note that the percentage of cases where a jury recommended the death penalty in these 1,113 cases was 48.5%, and it is a somewhat more frequent occur- rence than in the total population of 1,356 cases during this time period, of which the death penalty was recommended in 42.5% of the cases. This situation is largely attributable to the fact that the outcome in the cases that were filtered out of the population of cases (as described in Endnote 1) were generally, by default, life sentences.

3. Among the defendants who were of Other races (n = 75), 79% were American Indian, 16% were Hispanic, and 5% were Asian.

4. Among the victims who were of Other races (n = 66), 62% were American Indian, 27% were Hispanic, 6% were Asian, and 5% were of other races.

5. The additional 11 mitigators utilized here were developed from thematically cate- gorizing a range of commonly occurring factors that were submitted. These were that the defendant suffered from: drug use, alcohol use, physical abuse, sexual abuse, broken home, father absence/abandonment, mother absence/abandonment, being placed in fos- ter care, parental misconduct, low IQ, and mental illness/disorder.

6. The 15 North Carolina counties classified by the N.C. Rural Economic Development Center, Inc (2012) as “urban” (more than 250 people per square mile in density) include: Alamance, Buncombe, Cabarrus, Catawba, Cumberland, Davidson, Durham, Forsyth, Gaston, Guilford, Mecklenburg, New Hanover, Orange, Rowan, and Wake County. This def- inition and its rural counterpart are incorporated into North Carolina legislation.

7. The U.S. Supreme Court decision McKoy v. North Carolina (1990) altered the manner in which the presentation of mitigating circumstances, a crucial legal feature of the sen- tencing phase of capital murder trials in North Carolina, is responded to by the jury. Con- sequently, the variables measuring the impact of mitigation in cases before and after the McKoy decision must be accounted for in the analysis (Kremling, Smith, Cochran, Bjerregaard, & Fogel, 2007). We do so in our analysis by controlling for whether the case was pre- or post-McKoy to determine if there is a unique period effect associated with the cases.

8. The R packages that were used for the propensity score matching techniques ap- plied in the current study were MatchIt, Rtools, and cem (Hansen, 2004; Hansen & Bowers, 2008; Ho et al., 2007, 2011). These statistical packages will run on any platform where R is installed (Windows, Unix, or Mac OS X), and these tools are available from the Compre- hensive R Archive Network http://CRAN.R-project.org/package=MatchIt). Specifically, MatchIt was designed for causal inference with a dichotomous dependent variable and any number of pretreatment control variables/confounders. Furthermore, while the statis- tical package does work with experimental data, it is generally designed for observational data where random assignment was not assigned or when it was not plausible to use ran- dom assignment (the latter of which is the case in the current study). For a more detailed description on MatchIt, see Ho et al. (2007, 2011).

9. Parameters not estimated for total number of aggravators or total number of miti- gators due to multicollinearity. Specifically, the VIFs for these two variables were above 5 indicating collinearity concerns (O'Brien, 2007; Stine, 1995).

10. Similar to the previous traditional logistic regression model examining the “White victim effect”, parameters were not estimated in the Black defendant/White victim model for total number of aggravators or total number of mitigators due to collinearity concerns as the VIFs for these two variables were above 5 (O'Brien, 2007; Stine, 1995).

11. There are a number of different criteria/modeling strategies that can be applied when using propensity score matching. As such, it is important for us to briefly review the pros and cons of our strategy/criteria in order to better contextualize our results. Gen- erally, Ho et al. (2011, p. 4) suggest that “Because the outcome variable is not used in the matching procedure, any number of matching methods can be tried and evaluated, and the one matching procedure that leads to the best balance can be chosen.” Furthermore, propensity score matching permits the inclusion of any number of covariates/confounders and is not limited by the number of variables that can be used under certain sample size constraints or multicollinearity concerns like traditional regression methods are subject to. In this sense, more covariates/confounders are typically better, and more covariates/ confounders increase the precision of the obtained propensity score estimates. More spe- cifically, propensity scores are most often utilized for direct matching which enables ex- actness in matching, and the most common and suggested form is nearest-neighbor or 1:1 matching (Ho et al., 2011; Rosenbaum, 2002). Although it is true that direct 1:1 matching will result in data loss for the subjects that are the outliers and where no

common degree of support/or overlap in the propensity scores are observed, the ‘benefit’ of the data loss is the increase in the precision of the matching in that the matched treated and control cases represent the most comparable comparison groups (Ho et al., 2011; Rosenbaum, 2002). In fact, Paternoster and Brame (2008) provide an explicit example of this when they report their unweighted and weighted analyses where the sample sizes for the defendant sample is n = 1,041 (n = 214 Black defendant/White victim cases; n = 827 Other Race cases) in the unweighted analysis and the sample size is n = 522 (n = 214 Black defendant/White victim cases; n = 308 Other Race cases) in their propen- sity score weighted analysis. Thus, their sample size loss is nearly 50% after matching (weighting), yet Paternoster and Brame (2008, pp. 978–988) argue that their weighted sample size “…implies that many non-BD–WV cases were relatively incomparable with the BD–WV cases. These incomparable cases can be viewed as the source of the many un- weighted imbalances reported…After weighting, the influence of these cases is reduced, but the price for that benefit is reduced effective sample size and reduced statistical power for the between-group comparisons.” Also, when applying propensity score matching an- alysts have to make a determination of the ‘caliper’ that they select as the predetermined range around the estimated propensity score that can be used to match treated and con- trol cases. The most commonly used caliper and one that is recommended by propensity score experts and statistical convention in the literature is .20 (Austin, 2009; Rosenbaum, 2002). As such, this caliper was applied in the current study. Finally, estimated propensity scores can be used in a propensity score weighted regression model, rather than strictly used as estimates for matching treated and control cases and discarding unmatched treated and control cases. In this regard, all matched and unmatched cases are available for the analysis. As such, we performed our final analysis in this fashion as a sensitivity analysis, which replicated the results from our propensity score matched analyses but with the full sample of cases.

12. It could also be argued that the “White victim effect” or this “disadvantage gap” may be observed when comparing cases with Black defendants/White victims and cases with White defendants/White victims. Therefore, in an effort to investigate this further, we estimated bivariate comparisons of death penalty decision-making using these defen- dant/victim race dyads. The results of this analysis failed to demonstrate a significant asso- ciation between being a Black defendant and killing a White victim compared with being a White defendant and killing a White victim and receipt of the death penalty (M = 0.54 versus M = 0.52, respectively; t = −0.42, p = .67). This analysis suggests that the “White victim effect” “disadvantages” both Black and White defendants relatively equally.

13. In fact, Berk et al. (2005, p. 386) report that standard “regression models did not fare especially well…and that when stronger adjustments for confounding were employed, the racial effects were substantially diminished.” Similarly, Paternoster and Brame (2008, p. 975) argue that, “the examination of Black Defendant–White Victim racial disparity, then, requires an explicit consideration of the counterfactual—‘What would have happened had this case involved some other racial combination of offender and victim?’… The propensity score approach, therefore, emphasizes the counterfactual approach to causal thinking and inference.”

14. In this vein, Paternoster and Brame’s (2008) re-analysis of the Maryland data did suggest that cases where there was a Black defendant and a White victim were treated more adversely at the death notice decision point (e.g., the point at which prosecutorial discretion can be observed).

15. On this score, in one of the classic death penalty studies, Radelet (1981) assessed 637 homicides in 20 Florida counties in 1976 and 1977 and demonstrated/re-confirmed that most crime was intraracial rather than interracial. Furthermore, he also noted that when murder was interracial these events were disproportionately black-on-white homi- cide not white-on-black homicide and that the offenders and victims were strangers. In fact, stranger-related homicides were 700% more likely to be black-on-white than white-on-black. Nevertheless, once the defendants were indicted for first-degree murder, Radelet reported that neither defendant nor victim race nor any dyad effect was observed on the likelihood of being sentenced to death. In this same vein, Kleck (1981) also offered evidence to the fact that the race effect may not be as pronounced as commonly reported and that legal factors play a larger role in capital sentencing.

References

Albonetti, C. A. (1991). An integration of theories to explain judicial discretion. Social Problems, 38, 247–266.

Applegate, B., Cullen, F., & Fisher, B. (2000). Public views toward crime and correctional policies: Is there a gender gap? Journal of Criminal Justice, 30, 89–100.

Auerhahn, K. (2007). Just another crime? Examining disparity in homicide sentencing. Sociological Quarterly, 48, 277–313.

Austin, P. C. (2009). Some methods of propensity-score matching had superior perfor- mance to others: Results of an empirical investigation and monte carlo simulations. Biomedical Journal, 51, 171–184.

Baldus, D. C., Grosso, C. M., Woodworth, G., & Newell, R. (2012). Racial discrimination in the administration of the death penalty: The experience of the United States Armed Forces (1984–2005). Journal of Criminal Law and Criminology, 101, 1227–1335.

Baldus, D. C., Pulaski, C. A., Jr., & Woodworth, G. (1983). Comparative review of death sentences: An empirical study of the Georgia experience. Journal of Criminal Law and Criminology, 74, 661–753.

Baldus, D. C., & Woodworth, G. (2003). Race discrimination in the administration of the death penalty: An overview of the empirical evidence with special emphasis on the post-1990 Research. Criminal Law Bulletin, 39, 194–226.

397W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

Baldus, D. C., Woodworth, G., & Grosso, C. M. (2002). Arbitrariness and discrimination in the administration of the death penalty: A legal and empirical analysis of the Nebraska experience (1973–1999). Nebraska Law Review, 81, 486–753.

Baldus, D. C., Woodworth, G., & Pulaski, C. A., Jr. (1990). Equal justice and the death penalty: A legal and empirical analysis. Boston, MA: Northeastern University Press.

Baldus, D. C., Woodworth, G., Young, G. L., & Christ, A. M. (2001). The disposition of Nebraska capital and non-capital homicide cases (1973–1999): A legal and empirical analysis. Lincoln, NE: Nebraska Commission on Law Enforcement and Criminal Justice.

Baldus, D. C., Woodworth, G., Zuckerman, D., Weiner, N., & Broffitt, B. (1998). Racial dis- crimination in the post-Furman era: An empirical and legal overview, with recent findings from Philadelphia. Cornell Law Review, 83, 1643–1770.

Baumer, E. P., Messner, S. F., & Felson, R. B. (2000). The role of victim characteristics in the disposition of murder cases. Justice Quarterly, 17, 281–307.

Beaver, K. M., DeLisi, M., Wright, J. P., Boutwell, B. B., Barnes, J. C., & Vaughn, M. G. (2013). No evidence of racial discrimination in criminal justice processing: Results from the National Longitudinal Study of Adolescent Health. Personality and Individual Differences, 55, 29–34.

Behnken, M. P., Caudill, J. W., Berg, M. T., Trulson, C. R., & DeLisi, M. (2011). Marked for death: An empirical criminal careers analysis of death sentences in a sample of convicted male homicide offenders. Journal of Criminal Justice, 39, 471–478.

Berk, R., Li, A., & Hickman, L. J. (2005). Statistical difficulties in determining the role of race in capital cases: A re-analysis of data from the state of Maryland. Journal of Quantitative Criminology, 21, 365–390.

Bjerregaard, B., Smith, M. D., Fogel, S. J., & Palacios, W. (2010). Alcohol and drug mitigation in capital murder trials: Implications for sentencing decisions. Justice Quarterly, 27, 517–537.

Bohm, R. M., & Vogel, B. L. (2004). More than ten years after: The long-term stability of informed death penalty opinions. Journal of Criminal Justice, 32, 307–327.

Bohm, R. M., Vogel, R. E., & Maisto, A. A. (1993). Knowledge and death penalty opinion: A panel study. Journal of Criminal Justice, 21, 29–45.

Bowers, J., Fredrickson, M., & Hansen, B. (2010). RItools: Randomization Inference Tools. R package version 2.15.3.

Bowers, W. J., & Pierce, G. L. (1980). Arbitrariness and discrimination under post-Furman capital statutes. Crime and Delinquency, 26, 563–635.

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Earlbaum Associates.

Curry, T. R. (2010). The conditional effects of victim and offender ethnicity on sentencing for non-capital cases. Punishment and Society, 12, 438–462.

Daly, K. (1994). Gender, crime, and punishment. New Haven, CT: Yale University Press. Furman v. Georgia, 408 U.S. 238 (1972). Gibson, C. L., Miller, J. M., Jennings, W. G., Swatt, M. W., & Gover, A. R. (2009). Using pro-

pensity score matching to understand the relationship between gang membership and violent victimization: A research note. Justice Quarterly, 26, 625–643.

Gibson, C., Swatt, M., Miller, J. M., Jennings, W. G., & Gover, A. (2012). The causal relation- ship between gang joining and violent victimization: A critical review and directions for future research. Journal of Criminal Justice, 40, 490–501.

Gillespie, L., Loughran, T. A., Smith, M. D., Fogel, S. J., & Bjerregaard, B. (2014). Exploring the role of victim sex, victim conduct, and victim-defendant relationship in capital punishment sentencing. Homicide Studies (in press).

Gregg v. Georgia, 428 U.S. 153 (1976). Gross, S., & Mauro, R. (1989). Death and discrimination: Racial disparities in capital sentencing.

Boston, MA: Northeastern University Press. Hansen, B. B. (2004). Full matching in an observational study of coaching for the SAT.

Journal of the American Statistical Association, 99, 609–618. Hansen, B. B., & Bowers, J. (2008). Covariate balance in simple, stratified and clustered

comparative studies. Statistical Science, 23, 219–236. Hawkins, D. F. (1987). Beyond anomalies: Rethinking the conflict perspective on race and

criminal punishment. Social Forces, 65, 719–745. Hindson, S., Potter, H., & Radelet, M. (2006). Race, gender, region and death sentencing in

Colorado, 1980–1999. University of Colorado Law Review, 77, 549–594. Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2007). Matching as nonparametric preprocessing

for reducing model dependence in parametric causal inference. Political Analysis, 15, 199–236.

Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2011). Matchit: Nonparametric preprocessing for parametric causal inference. Journal of Statistical Software, 8, 1–28.

Holcomb, J. E., Williams, M. R., & Demuth, S. (2004). White female victims and death penalty disparity research. Justice Quarterly, 21, 877–902.

Jennings, W. G., Fox, B., & Farrington, D. P. (2014). Inked into crime? An examination of the causal relationship between tattoos and life-course offending among males from the Cambridge Study in Delinquent Development. Journal of Criminal Justice, 42, 77–84.

Jennings, W. G., & Reingle, J. (2012). On the number and shape of developmental/life- course violence, aggression, and delinquency trajectories: A state-of-the-art review. Journal of Criminal Justice, 40, 472–489.

Jiang, S., Lambert, E., & Nathan, V. (2009). Reasons for death penalty attitudes among Chinese citizens: Retributive or instrumental? Journal of Criminal Justice, 37, 225–233.

Kavanaugh-Earl, J., Cochran, J. K., Smith, M. D., Fogel, S. J., & Bjerregaard, B. (2008). Racial bias and the death penalty. In M. J. Lynch, E. B. Patterson, & K. K. Childs (Eds.), Racial divide: Racial and ethnic biases in the criminal justice system. Monsey, NY: Criminal Justice Press.

Keil, T. J., & Vito, G. F. (1990). Race and the death penalty in Kentucky murder trials: An analysis of post-Gregg outcomes. Justice Quarterly, 7, 189–207.

Keil, T. J., & Vito, G. F. (1995). Race and the death Penalty in Kentucky murder trials: 1976–1991. American Journal of Criminal Justice, 20, 17–36.

Kleck, G. (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.

Klein, S., & Rolph, J. (1991). Relationship of offender and victim race to death penalty sentences in California. Jurimetrics, 32, 33–48.

Kramer, J. H., & Ulmer, J. T. (2002). Downward departures for serious violent offenders: Local court “corrections” to Pennsylvania’s sentencing guidelines. Criminology, 40, 897–932.

Kramer, J. H., & Ulmer, J. T. (2009). Sentencing guidelines: Lessons from Pennsylvania. Boulder, CO: Lynne Rienner Publishers.

Kremling, J., Smith, M. D., Cochran, J. K., Bjerregaard, B., & Fogel, S. J. (2007). The role of mitigating factors in capital sentencing before and after McKoy v. North Carolina. Justice Quarterly, 24, 357–381.

Liang, B. (2005). Severe strike campaign in transitional China. Journal of Criminal Justice, 33, 387–399.

Loughran, T. A., Mulvey, E. P., Schubert, C. A., Fagan, J., Piquero, A. R., & Losoya, S. H. (2009). Estimating a dose-response relationship between length of stay and future recidivism in serious juvenile offenders. Criminology, 47, 699–740.

Lu, H., & Zhang, L. (2005). Death Penalty in China: The law and the practice. Journal of Criminal Justice, 33, 367–376.

Maldonado-Molina, M., Jennings, W. G., Tobler, A., Piquero, A. R., & Canino, G. (2010). Assessing the victim-offender overlap among Puerto Rican youth. Journal of Criminal Justice, 38, 1191–1201.

Mancini, C., & Mears, D. P. (2010). To execute or not to execute? Examining public support for capital punishment of sex offenders. Journal of Criminal Justice, 38, 959–968.

March, J. G., & Simon, H. A. (1958). Organizations. New York, NY: Wiley. McAdams, J. C. (1998). Racial disparity and the death penalty. Law and Contemporary

Problems, 61, 153–170. McKoy v. North Carolina, 494 U.S. 433 (1990). N.C. Rural Economic Development Center, Inc (2012). Rural/Urban Counties in North

Carolina. http://www.ncruralcenter.org/index.php?option=com_content&view= article&id=75&Itemid=126 (accessed 12 March, 2014).

North Carolina General Statutes (n.d). Subchapter XV. Capital Punishment. Article 100. http://www.ncga.state.nc.us/enactedlegislation/statutes/html/bysection/chapter_ 15a/gs_15a-2000.html (accessed 12 March, 2014)

O'Brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality & Quantity, 41, 673–690.

O'Brien, B., & Grosso, C. M. (2011). Confronting race: How a confluence of social movements convinced North Carolina to go where the McCleskey Court wouldn’t. Michigan State Law Review, 2011, 463–504.

Pasternoster, R., Brame, R., Bacon, S., & Ditchfield, A. (2004). Justice by geography and race: The administration of the death penalty in Maryland, 1978–1999. Margins, 4, 1–98.

Paternoster, R. (1984). Prosecutorial discretion in requesting the death penalty: A case of victim-based racial discrimination. Law & Society Review, 18, 437–478.

Paternoster, R., & Brame, R. (2003). An empirical analysis of Maryland’s death sentencing system with respect to the influence of race and legal jurisdiction. http://www. sentencingproject.org/doc.publications/rd_sentencing_review.pdf (accessed 12 March, 2014).

Paternoster, R., & Brame, R. (2008). Reassessing race disparities in Maryland capital cases. Criminology, 46, 971–1008.

Pearl, J. (2000). Causality: Models, reasoning, and inference. New York: Cambridge Univer- sity Press.

Pierce, G. L., & Radelet, M. L. (2002). Race, region and death sentencing in Illinois, 1988–1997. Oregon Law Review, 81, 39–96.

Pierce, G. L., & Radelet, M. L. (2005). The impact of legally inappropriate factors on death sentencing for California homicides. Santa Clara Law Review, 46, 1–47.

Pokorak, J. J. (1998). Probing the capital prosecutor’s perspective: Race of the discretionary actors. Cornell Law Review, 83, 1811–1820.

Quinney, R. (1970). The problem of crime. New York: Dodd, Mead. Radelet, M. L. (1981). Racial characteristics and the imposition of the death penalty.

American Sociological Review, 46, 918–927. Radelet, M. L., & Pierce, G. L. (1991). Choosing those who will die: Race and the death

penalty in Florida. Florida Law Review, 43, 1–34. Radelet, M. L., & Pierce, G. L. (2011a). Death sentencing in East Baton Rouge Parish,

1990–2008. Louisiana Law Review, 71, 647–673. Radelet, M. L., & Pierce, G. L. (2011b). Race and the death penalty in North Carolina,

1980–2007. North Carolina Law Review, 89, 2119–2159. Richards, T., Jennings, W. G., Smith, M. D., Sellers, C. S., Fogel, S. J., & Bjerregaard, B. (2014).

Explaining the "female victim effect" in capital punishment: An examination of victim sex specific models of capital juror sentence decision-making. Crime & Delinquency (in press).

Richards, T., Smith, M. D., Jennings, W. G., Bjerregaard, B., & Fogel, S. (2014). An examination of the sex disparity in capital sentencing: A propensity score matching approach. American Journal of Criminal Justice (in press).

Rosenbaum, P. R. (2002). Observational studies (2nd ed.). New York: Springer-Verlag. Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in obser-

vation studies for casual effects. Biometrika, 70, 41–55. Sampson, R. J., Laub, J. H., & Wimer, C. (2006). Does marriage reduce crime? A counterfac-

tual approach to within-individual causal effects. Criminology, 44, 465–508. Smith, M. D. (1987). Patterns of discrimination in assessments of the death penalty: The

case of Louisiana. Journal of Criminal Justice, 15, 279–286. Sorensen, J. R., & Wallace, D. H. (1995). Capital punishment in Missouri: Examining the

issue of racial disparity. Behavior Sciences & the Law, 13, 61–80. Spohn, C., & Holleran, D. (2000). The imprisonment penalty paid by young, unemployed

black and Hispanic male offenders. Criminology, 38, 281–306.

398 W.G. Jennings et al. / Journal of Criminal Justice 42 (2014) 384–398

Stauffer, A. R., Smith, M. D., Cochran, J. K., Fogel, S. J., & Bjerregaard, B. (2006). The inter- action between victim race and gender on sentencing outcomes in capital murder trials. Homicide Studies, 10, 98–117.

Steffensmeier, D., & Demuth, S. (2000). Ethnicity and sentencing: Outcomes in U.S. federal courts: Who is punished more harshly?”. American Sociological Review, 65, 705–729.

Steffensmeier, D., Ulmer, J., & Kramer, J. (1998). The interaction of race, gender, and age in criminal sentencing: The punishment cost of being young, black, and male. Criminology, 36, 763–798.

Stine, R. A. (1995). The graphical interpretation of variance inflation factors. American Statistician, 49, 53–56.

Tillyer, R., & Engel, R. S. (2012). Racial differences in speeding patterns: Exploring the differential offending hypothesis. Journal of Criminal Justice, 40, 285–295.

U.S. General Accounting Office (1990). Death Penalty Sentencing: Research Indicates Pattern of Racial Disparities. GGD-90-57. Washington, DC: U.S. General Accounting Office.

Ulmer, J. T., & Bradley, M. S. (2006). Variations in trial penalties among serious violent offenses. Criminology, 44, 631–669.

Ulmer, J. T., Kurlychek, M. C., & Kramer, J. (2007). Prosecutorial discretion and imposition of mandatory minimum sentences. Journal of Research in Crime and Delinquency, 44, 427–458.

Unah, I. (2011). Empirical analysis of race and the process of capital punishment in North Carolina. Michigan State Law Review, 2011, 610–658.

Williams, M. R., Demuth, S., & Holcomb, J. E. (2007). Understanding the influence of victim gender in death penalty cases: The importance of victim race, sex-related victimization, and jury decision-making. Criminology, 45, 865–891.

Williams, M. R., & Holcomb, J. E. (2001). Racial disparity and death sentences in Ohio. Journal of Criminal Justice, 29, 207–218.

Wozniak, K. H., & Lewis, A. R. (2010). Reexamining the effect of Christian denominational affiliation on death penalty support. Journal of Criminal Justice, 38, 1082–1089.

Wright, J. P., Morgan, M. A., Coyne, M. A., Beaver, K. M., & Barnes, J. C. (2014). Prior prob- lem behavior accounts for the racial gap in school suspensions. Journal of Criminal Justice, 42, 257–266.