Racial Discrimination in the Administration of the Death Penalty: The Experience of the United States Armed Forces (1984-2005)

Beginning of article

I. INTRODUCTION

This Article presents the results of an empirical study of racial discrimination in the administration of the death penalty in the United States Armed Forces between 1984, the year military law was brought into conformity with the requirements of Furman v. Georgia (1972), and 2005. (1) A main theme of this paper is the difference between the military system and the civilian systems that have been similarly studied in the operation, outcomes, racial disparities, and the sources of those disparities.

Our military evidence takes two forms. The first is evidence of systemic racial disparities in the charging and sentencing decisions of convening authorities and court-martial members that non-racial case characteristics do not explain. (2) The second is quantitative and qualitative assessments of the risk of racial discrimination in each case in which a minority accused was sentenced to death.

Our database includes military prosecutions for all "potentially death-eligible" murder cases known to us (n = 105), including all "factually death-eligible" murder cases that resulted in a capital murder conviction (by plea or at trial) with one or more statutory aggravating factors found or present (n = 97). (3) The sentencing dates of these cases range from July 16, 1984, to October 13, 2005. Fifteen of these cases resulted in a death sentence.

Part I of the Article identifies death-eligible offenses under military law, including premeditated and felony murder, which are the focus of this study. Part II describes the military capital charging and sentencing process for death-eligible murder cases. Part III explains our methodology. Part IV presents evidence of systemic racial disparities. Part V assesses the risk of racial discrimination in ten cases in which a minority accused received the death sentence. Part VI contrasts the findings of racial disparities in death-eligible capital cases with the evidence of racial disparities in non-capital sentencing outcomes among the sixty-six death-eligible murder cases that did not advance to a capital sentencing hearing.

Part VII presents our conclusions and policy recommendations. We found compelling evidence that the race of the accused and of the victim has influenced charging and sentencing decisions in the processing of death-eligible murder cases in the system since 1984. There is, however, an important distinction between the decisions made in the processing of these cases. The risk of racial prejudice is confined entirely to the decisions of convening authorities and members that lead up to and include the members' death-sentencing outcomes. Among the sixty-seven cases that did not advance to a capital sentencing hearing, the first sentencing issue was whether a life sentence or a term of years would be given. In the cases where an individual was sentenced to a term of years, the second issue was the duration of that sentence. There is no evidence of systemic racial effects in either of these decisions. (4)

Among the 105 cases in our study that potentially implicate the death penalty, there is evidence of a substantial risk of three forms of racial prejudice: white-victim discrimination, minority-accused/white-victim discrimination, and independent minority-accused discrimination. There is a risk of all three forms of prejudice in the imposition of death sentences among all death-eligible cases. (5) Closer scrutiny reveals that the source of the white-victim and minority-accused/white-victim effects in the imposition of death sentences among all death-eligible cases is convening authority decisions seeking death sentences (6) and the guilt trial decisions of court-martial members that advance cases to a capital sentencing hearing by a unanimous verdict on the accused's liability for capital murder. (7) The combined effects of these two decisions produce a substantial and statistically significant white-victim disparity in the rates that cases advance to a capital sentencing hearing. (8) The evidence further suggests that the principal source of the independent minority-accused racial disparity documented in the imposition of death sentences among all death-eligible cases is members' life/death decisions in capital sentencing hearings. Specifically, in white-victim cases, which constitute 97% of capital sentencing hearing cases, minority accused face a significantly higher risk of a death sentence than do similarly situated white accused. (9)

Our evidence also supports the hypothesis propounded by Justice Byron White that the risk of racial prejudice is greatest in highly aggravated minority-accused/white-victim cases, which are illustrated in this study by sixteen multiple-victim cases. (10)

Finally, our findings suggest that the risk of racial prejudice in the administration of the military death penalty for death-eligible murder would be greatly reduced if death sentencing in such cases were limited to cases with significant military implications in which the risk of the imposition of a death sentence has been low to non-existent since 1984.

A. DEATH-ELIGIBLE OFFENSES UNDER THE UNIFORM CODE OF MILITARY JUSTICE

There is a long tradition of the use of capital punishment in the United States Armed Forces. There are currently fifteen death-eligible offenses in the Uniform Code of Military Justice (UCMJ). (11) All but two of them currently relate to crimes with important national security or military implications that have no counterparts in civilian death-penalty systems. Mutiny, sedition, and espionage are in the first, national security category. (12) There are also eight death-eligible offenses with serious military implications that apply only "in time of war" or during combat operations against a foreign power. (13) In addition, two other offenses have important military implications but no "time of war" requirement. (14) These are long-standing offenses that to our knowledge have not been applied since the Korean War. (15)

The fourteenth and fifteenth death-eligible offenses are murder (premeditated and felony murder) (16) (Section 118) and rape (17) (Section 120(a)) committed by U.S. military personnel during peacetime anywhere in the world. (18) A murder conviction is the basis of all of the military death sentences imposed since 1960. With one exception, murder and rape are the most recently established death-eligible military offenses, having been enacted by the UCMJ in 1950. (19)

Death eligibility for murder requires no connection between the murder and military interests or functions. Military status alone makes the statute applicable to military personnel and gives courts-martial jurisdiction. (20) In terms of the definition of capital murder, therefore, the UCMJ mirrored the provisions of typical 1950s civilian death-penalty statutes that defined first-degree and felony murder as capital offenses. Like the civilian systems of that era, the UCMJ also vested in the sentencing authority complete and untrammeled discretion to decide whether a sentence for capital murder should be death or life imprisonment.

This second feature of the military death-penalty system became important after Furman v. Georgia (1972), which held that the unguided discretion of sentencing authorities in civilian jurisdictions violated the cruel and unusual punishments provision of the Eighth Amendment to the United States Constitution. (21) Furman facially applied only to state death-penalty systems. Nevertheless, the military procedures in place in 1972 were identical to the civilian death-penalty procedures condemned in Furman. More than a decade passed, however, before a military court acknowledged the relevance of Furman to the military system.

In the meantime, the United States Supreme Court ruled in Gregg v. Georgia (1976) and Proffitt v. Florida (1976) (22) that the adoption of statutory lists of aggravating circumstances comparable to those found in the Model Penal Code (23) and the use of bifurcated guilt and penalty trials satisfied the requirements of the Eighth Amendment because they materially reduced the breadth of capital charging discretion. In the Court's view, these reforms limited death sentences to the most aggravated cases, thereby eliminating the risk of arbitrariness and discrimination in the administration of capital punishment. (24)

In 1983, eleven years after Furman, military courts ruled that Furman v. Georgia did apply to the military death-penalty system and that existing military procedures did not meet the requirements of Furman and Gregg. (25) To cure this defect, President Reagan issued a 1984 executive order that limited death eligibility to capital cases in which the fact-finder determined that one or more statutory aggravating circumstances were present in the case and "any extenuating or mitigating circumstances are substantially outweighed by any aggravating circumstances." (26) The aggravating circumstances, known as aggravating "factors" in current military parlance, embrace a number of situations with distinct military and national security implications (27) that facially apply to all death-eligible offenses, but in practice have no applicability to "civilian-style" premeditated and felony murders, which constitute the vast majority of death-eligible murders committed by military personnel in peacetime. Another distinctly military aggravating circumstance applies to murder and rape "committed in time of war"--a condition that has not existed since the Korean War. (28)

The executive order does, however, specifically exclude from murder offenses two omnibus aggravators with significant military implications. (29) In fact, only one of the omnibus aggravators applicable to all death-eligible military offenses (grave risk to non-decedent victims) has frequent relevance to murder cases. (30)

The executive order also defines, for premeditated murder cases under Article 118(1) of the UCMJ, an extensive list of distinctly civilian-style aggravating circumstances. (31) These aggravating circumstances were inspired by the typical state death-penalty statute whose aggravating circumstances are modeled after those found in the Model Penal Code. (32) The focus of the executive order on civilian-style aggravating circumstances for premeditated murder was understandable, given that six of the seven murder cases in which a death sentence had been imposed in a military court between 1979 and 1984 involved a distinctly civilian-style murder with no special military implications. (33)

In fact, only one part of one of the premeditated murder aggravating circumstances in the executive order is uniquely tailored to military circumstances. This aggravator (7G) classifies as death-eligible the premeditated murder of a "commissioned, warrant, noncommissioned, or petty officer of the armed services of the United States" killed "in the execution of office" when the accused had knowledge of the victim's status. (34) The balance of the 7G aggravating factor reflects an effort to provide special protection for law enforcement and corrections officers that is found in most civilian jurisdictions. (35)

The 1984 executive order does not list specific mitigating circumstances, which means under the Eighth Amendment that defense counsel may present any mitigating factor beating on the circumstances of the offense or the character of the accused. (36)

The drafters of the executive order likely shared the opinion of the Supreme Court in Gregg v. Georgia (1976) that the limitations which the military aggravating factors place on the exercise of discretion by sentencing heating members would significantly reduce the risk of arbitrariness and racial discrimination by limiting death eligibility to the most highly aggravated murder cases. Additional protections in this regard are provided by the UCMJ's two important predicates for the imposition of a military death sentence: (a) a unanimous finding of liability for capital murder by the court-martial members, (37) and (b) a unanimous finding by court-martial members "beyond a reasonable doubt" that one or more "aggravating circumstances" exist and that "any extenuating or mitigating circumstances are substantially outweighed by any aggravating circumstances." (38)

However, in spite of these protections, there are good reasons for concern about the risk of arbitrariness and discrimination in the administration of the military death penalty. First is the breadth of discretion exercised by senior commanders ("convening authorities") who make capital charging decisions in the system. Second, the discretion of the officers and enlisted personnel (the "courts-martial members") who determine capital murder liability and impose life and death sentences in capital sentencing hearings is also very broad.

In addition, there are historical and cultural grounds for concern that the 1984 executive order may have fallen short of its intended effect. First, there is historical evidence of racial disparities in the administration of the military death penalty. Most striking are Professor Lilly's studies of military executions in Europe during World War II, suggesting that black soldiers accused of rape and murder of white victims were disproportionately executed for their crimes. (39) Race relations and attitudes within the Armed Forces and within U.S. society generally have changed dramatically since then. Nevertheless, a 1970s Pentagon-sponsored study of the military justice system documented continuing race disparities at a number of levels. (40)

Second, the literature from 1980 through 2008 documents race effects in the administration of comparable civilian death-sentencing systems. (41) In terms of the scope of the discretion of civilian prosecutors and sentencing authorities, the state systems are comparable to the current military system. There is a risk, therefore, that the racial attitudes that commanders and court-martial members bring to the military from civilian life may similarly affect their decisions in the military justice system. (42)

Another concern is that the military justice system lacks the transparency of civilian systems. The capital charging decisions of civilian prosecutors operating in multiracial communities are often subject to close scrutiny. (43) So also are the decisions of civilian juries in highly visible cases. (44) In contrast, the decisions of commanders and courts-martial members typically receive little scrutiny either within or outside the military. (45) The main exceptions are highly aggravated murders that implicate the authority and effectiveness of the military command. (46)

II. DECISIONMAKING IN DEATH-ELIGIBLE MURDER CASES

The following Section presents an overview of decisionmaking in the U.S. military capital punishment system by documenting how the cases in the study move through different decision points.

A. CAPITAL CHARGING AND SENTENCING DECISIONS

Under military law and practice, the death-penalty statute is applied in a three-stage process by two decisionmakers--the convening authority and the court-martial members.

[FIGURE 1 OMITTED]

A capital prosecution in a death-eligible case is commenced by the "convening authority," normally a general or admiral in the accused's command, who has total discretion whether or not to seek a death sentence in a death-eligible case. A decision to seek a death sentence in the case is known as a "capital referral," a decision that is heavily influenced by the "advice" letter of the commander's staff judge advocate (SJA), his chief legal advisor. (48)

At stage one, if a case is capitally charged and the capital referral is not withdrawn by the convening authority, the case advances to a capital court-martial with the government seeking a death sentence. (49) As shown in Figure 1, at stage one, the capital referral and plea bargaining decisions of the convening authority took death off the table in 58% of the death-eligible cases (Box 1B). The remaining 42% advanced to a capital court-martial with the government seeking a death sentence. This rate is comparable to the 39% rate in the eight civilian jurisdictions on which post-Furman data are available from the 1970s and 1980s, (50) but it is higher than the civilian rates in most jurisdictions since 1990. (51) However, since 1990, the military capital charging rate has declined sharply, (52) as it has in many civilian jurisdictions. (53)

Since November 18, 1997, the military life-sentence option has included a life sentence without possibility of parole (LWOP), and four such sentences have been imposed in non-capital sentencing hearings during the period of this study. (54) The availability of LWOP may also have contributed to an increase in the frequency with which convening authorities have waived the death penalty in death-eligible cases by declining to charge them capitally or withdrawing a capital charge, usually as part of a plea bargain.

At stage two, a unanimous finding of liability for capital murder by the court-martial members will advance the case to a capital sentencing hearing. (55) Death came off the table for 27% of those found guilty of premeditated or felony murder by virtue of a non-unanimous finding of guilt (Box 2B). The remaining thirty cases with a unanimous guilty verdict advanced to a capital sentencing hearing.

At stage three, court-martial members consider the aggravating factors and mitigating circumstances and make a life or death determination. At this stage, members sentenced 50% of the accused to death (Box 3A). This rate is also comparable to the 53% penalty trial death-sentencing rate in the twelve civilian jurisdictions on which data are available from the 1970s and 1980s, (56) but it is higher than the average civilian penalty trial rate since 1990.57 In that regard, since 1984, panel members' death-sentencing rates have steadily risen, although the number of capital sentencing hearings during that time has been small. (58)

Overall, the death-sentencing rate among all death-eligible cases since 1984 is 15% (15/97). This rate is higher than comparable figures in most states on which we have data both before (59) and after 1990. (60) However, because of the increasing selectivity of convening authorities in pursuing capital punishment, the overall death-sentencing rate in all death-eligible cases has remained quite stable in recent years (despite the members' rising death-sentencing rate), declining from 20% (7/35) during the first period (1984-1989) and holding thereafter at 14% (4/28) (1990-1994), 10% (2/18) (1995-1999), and 14% (2/14) (2000-2005). (61)

B. THE DEATH-SENTENCED ACCUSED

Table 1 lists the fifteen cases that resulted in the imposition of a death sentence between 1984 and 2005, classified by the year of the sentence and the status of the murder as a "civilian-style" or "militarily implicated" murder (Column D). This "crime type" distinction is based on compelling evidence that in terms of criminal culpability and deathworthiness, military decisionmakers perceive murders that threaten important military interests to be more aggravated than civilian-style murders that do not. (62)

Militarily implicated cases involve lethal attacks on United States troops while on duty, any cases where the victim is a commissioned officer, (63) and other murders that directly threaten the authority or effectiveness of the military mission. (64) The following research note provides an example of a militarily implicated case involving a lethal attack on U.S. troops on duty and one or more officer victims:

   At night in Kuwait four days before the 2003 United States led
   invasion of Iraq, Hassan Akbar, a 31-year-old Army E5 with 4
   years of military service, feigned an attack on his unit by
   rolling live hand grenades into three tents with sleeping officers
   and opened fire on the occupants as they fled their tents. He
   killed one officer with a shot in the back. A second officer died
   of 87 shrapnel wounds. He also injured 14 other military victims,
   many seriously. (65)

The following note exemplifies a case with fewer significant military implications:

   Motivated by a perceived racial attack on a black marine by
   white marines, Kenneth Parker, a 21-year-old Marine Corps
   E3 with 3 years of military service, and five military
   co-perpetrators kidnapped, robbed and killed with a shot to the
   heart the first white person they encountered who happened to
   be a fellow marine. Parker was the shooter. On another occasion
   he killed the male spouse of a fellow marine's paramour with
   a shotgun blast to the chest at request of the follow marine.

In contrast, civilian-style murders involving family and acquaintance victims (67) or stranger victims in felony murders (68) pose comparatively little threat to military discipline and control and the effectiveness of the military mission. The following research notes provide good examples of strictly civilian-style murders:

   In 1987, to collect a life insurance payment, Joseph Thomas, a
   28-year-old field wireman (Marine Corps E5) with 10 years of
   military service, along with an accomplice, killed his white
   wife with a tire iron. They moved the body into Thomas' car
   and drove the car off a cliff and set it on fire.

   Todd Dock, a 19-year-old armor crewman (Army E3) with 3 years
   of military service, stabbed a German taxi driver in the neck
   until he lost consciousness. After the cab came to a stop, Dock
   dragged the driver's body from the car. At that point, the driver
   regained consciousness and attempted to grab Dock's arm. Dock
   then repeatedly stabbed the victim in the abdomen and
   chest until he died. Dock stole the victim's wallet and fled.

Fifty-three percent (8/15) of the cases in Table 1 involve civilian-style murders that are reminiscent of the civilian-style death-sentenced cases that immediately antedated the adoption of the 1984 executive order. (69) However, since 1990, 86% (6/7) of the death sentences have been imposed in cases with military implications. These data suggest that since 1990 there has been a de facto presumption against the use of capital punishment in civilian-style murders. The only death sentence imposed in a civilianstyle murder since 1990 is Witt, (70) a brutal two-victim case with five aggravating factors, which plausibly could overcome this presumption. The nearly complete absence of death sentencing for civilian-style murder since 1990 is not explained by the absence of highly aggravated civilian murders since then. Rather, what explains the decline is a substantial shift by convening authorities and members away from death sentencing in civilianstyle murders. (71)

C. THE APPELLATE REVIEW PROCESS

The appellate review process following the imposition of a military death sentence commences with the accused's request for clemency by the convening authority, who has complete discretion to reduce both the crime of conviction and its punishment. (72) No comparable authority exists in civilian courts. Convening authorities disallowed the death sentence in two of the fifteen death-sentenced cases in this study.

For death sentences approved by the convening authority, appeals lie to the branch-specific courts of military review, the Court of Appeals of the Armed Forces (CAAF), and the United States Supreme Court. (73) Table 2 classifies appellate outcomes in the fifteen death sentences imposed since 1984 in terms of their risk of execution.

Rows 1 and 2 identify the cases with no continuing risk of execution. Melvin Turner and Curtis Gibbs in Row 1 are the two cases in which the convening authority exercised its clemency authority to overturn the death sentence. In Row 2, death was taken off the table for seven accused by an appellate military court decision, a decision on remand by the convening authority, or court-martial members in a second capital sentencing hearing.

Row 3 identifies five cases with a continuing risk of execution because military judicial appeals have not been exhausted, the case is in federal habeas corpus proceedings, or the death sentence was vacated subject to reimposition. Row 4 identifies one accused with a high risk of execution because he has exhausted his military and direct appeals to the United States Supreme Court. (74)

III. METHODOLOGY

This Part presents the methodology supporting the substantive racial findings presented in Parts IV through VI of this Article.

A. THE UNIVERSE AND SAMPLE

The universe of death-eligible cases between 1984 and 2005 is indeterminate because no branch of the military maintains a record of the death-eligible cases processed through its courts, and no military authority is responsible for tracking such information throughout the system. (75) To identify death-eligible cases we conducted an exhaustive search of every identified data source with information on military homicides.

1. Identifying Potentially Death-Eligible Cases

The first step was to search for the names of all accused who were prosecuted in the military criminal justice system and convicted of a homicide whose cases may have been factually death-eligible. For this purpose, we consulted all reported military court homicide decisions, appellate briefs of counsel in capital military appeals since 1985, a list of all offenders incarcerated for homicide at Fort Leavenworth as of June 2004, all Army homicide convictions from 1986 to 2004 maintained by the Army, and all Army unreported homicide decisions from 1984 to 2004. We also screened the Army Times, Navy Times, Marine Corps Times, and Air Force Times since 1984 for the names of accused in reported homicides. Finally, the judge advocates with whom we have had contact during the course of this research brought to our attention several cases based on their personal experiences. This search produced a master list of 440 homicide cases.

2. Collecting Data to Assess the Death Eligibility of Accused in Homicide Cases

The second step in our search for death-eligible cases was to obtain sufficient information on each homicide on our master list to determine the likelihood that the accused was death-eligible. We sought access to the complete military file for each case we deemed likely death-eligible and consulted judicial opinions concerning the accused and their coperpetrators. We also obtained the full files of co-perpetrators (where appropriate) and media reports of the accused's crimes through extensive online searches.

To bring a case into the study, our standard was "factual" death eligibility. This required strong evidence that (a) a case that was not capitally charged would, if it had been so charged, have supported a conviction of capital murder and a finding of one or more aggravating factors in the case, and (b) both of these findings would have been sustained on appeal if the sufficiency of the evidence supporting the death eligibility of the case had been challenged. We limited the application of this rule by what we describe as the "controlling fact finding" rule. This rule holds that if a case advances to a court-martial on a capital murder charge but the members find the accused not guilty of the capital offense, the case will not be viewed as death-eligible, even though the evidence would have supported a capital murder conviction on appeal had the members returned a unanimous capital murder verdict. Unless the decision of acquittal on the capital charge appears to have been a clear case of member nullification (and we found no such case), we treated such a judgment as a factual finding that the elements of death-eligible murder were not present in the case.

We identified eight cases that appeared to be factually death-eligible in the eyes of the convening authority where the defendants were charged with capital murder, but the members acquitted on the capital charge and instead found liability for a lesser-included non-capital form of murder or manslaughter. (76) Because we saw no evidence of jury nullification in those cases, they were excluded from the portion of the study that focused on the death-sentencing outcomes. However, we included these cases in our analysis of convening authorities' charging decisions. (77) For those analyses, therefore, our sample of cases included 105 cases.

We excluded one factually death-eligible case that was committed abroad in a country that would not surrender custody of the accused until the United States agreed that the case would not be referred capitally. (78) We also excluded approximately four cases that, on the basis of news reports, appeared likely to be factually death-eligible under military law, because the military authorities deferred to the jurisdiction of civilian authorities who sought to prosecute them. All of these crimes were committed off the accused's military facility, had no military implications, and normally were sufficiently aggravated to attract substantial attention in the civilian community in which they were prosecuted. (79) Because none of these cases resulted in a criminal prosecution by U.S. military officials, they were excluded from our sample of death-eligible cases. (80)

Appellate courts vacated three of the death sentences in our sample on appeal, and convening authorities then prosecuted the cases a second time. Because of the first prosecution and death sentence, there is legitimate concern about the independence of the second charging decision. (81) Due to our small sample size, we included these three cases in our primary analyses of charging decisions. We replicated the final analysis without these three cases, and their elimination had no impact on the results.

Additionally, because military sentencing is done by court-martial members, there is little or no risk of a lack of independence in their decisions. For this reason, we included the second prosecution in one case (82) in our analysis of the imposition of death sentences in capital sentencing hearings.

Our final sample consisted of ninety-seven death-eligible cases, thirty of which advanced to a capital sentencing hearing; of these, fifteen resulted in a death sentence. Eighty of the ninety-seven death-eligible cases involved one or more white victims. (83)

3. Data Collection and Entry

For every case identified as death-eligible and included in the study, we sought access to the accused's full file from the National Records Center in Suitland, Maryland. Army, Navy, and Marine Corps judge advocates facilitated this effort and permitted our interns to copy relevant portions of the files in their offices in Washington, D.C. Air Force judge advocates facilitated our access to their files through the Freedom of Information Act (FOIA). Without the support of these officers, for which we are grateful, we could not have conducted this study.

For each full case file that we accessed, law student interns in the District of Columbia copied the relevant portions of the case files. This copying process was directed by a protocol prepared by the investigators of this study. Moreover, the investigators trained the interns who did the copying in Washington, D.C., at the outset of their internships. They also received long-term on-the-job training from the Army, Navy, and Marine Corps judge advocates in whose offices they copied the cases. For the Air Force cases that were copied through our FOIA request, we submitted the same copying guidelines for the personnel copying the cases.

We could not locate full files for certain cases at the National Records Center or otherwise. For those cases, we relied on the information found in military judicial opinions and the information that we could find online in the media, including military newspapers, such as the Army Times, and local newspapers. Law students at the University of Iowa College of Law coded the data on each case into a fifty-three-page detailed data collection instrument (DCI) prepared by the investigators. (84) This process produced a machine-readable database for all of the cases in the study. With this information, we created recoded variables that are suitable for statistical analysis. Appendix A reports the distribution of our data for each of these recoded variables for each of the ninety-seven death-eligible cases in our sample. The coders also prepared a detailed narrative summary of each case.

B. ANALYSIS

This Section provides details on each method of analysis and each measure of culpability used in the study.

1. Overview

We estimated racial disparities in the imposition of death sentences among all death-eligible cases. We then focused separately on (1) convening authority decisions advancing cases to a capital court-martial, (2) members' unanimous capital murder convictions advancing cases to a capital sentencing hearing, and (3) members' death-sentencing decisions in capital sentencing hearings. All of these decisions combine to produce the racial disparities documented in the first step, the threshold analysis of death sentencing among all death-eligible cases. The threshold analysis considers only whether a case ultimately received a death sentence.

2. Unadjusted Disparities

We commenced each phase of the analysis with estimates of unadjusted racial disparities that take no account of the differential culpability levels of the accused. These analyses focus on minority-accused disparities, white-victim disparities, and minority-accused/white-victim disparities. (85)

3. Adjusted Disparities

We principally used three different methods to control for the culpability of cases: logistic regression, a regression-based scale, and a salient factors scale. This section explains these three alternative measures of offender culpability.

i. Logistic Regression Analysis

For each stage in the analysis, our first set of controls was a logistic regression model based on legitimate non-racial variables and race variables. (86) At each stage, we considered a large number of candidate variables for inclusion in the model. In screening variables for inclusion, our goal was to produce a legally acceptable model to verify whether any unadjusted racial disparities persisted, declined, or disappeared upon adjustment for measures of criminal culpability. (87)

These models document the linear effect of each variable, meaning the enhanced or diminished probability that, on average, the applicable charging or sentencing outcome will occur when a variable is present in the case, after controlling for all other variables in the model. (88) For example, our core analysis of death sentences imposed among all death-eligible cases indicates that, on average, accused in multiple-victim cases had an adjusted 18-percentage-point higher probability of being sentenced to death than accused in single-victim cases. (89)

Logistic regression analysis produces another measure of impact estimating the extent to which the presence of individual case characteristics enhance or diminish the odds rather than the probability that a given outcome will occur. For example, in the multiple-victim analysis noted above, the adjusted 8.3 "odds ratio" or "odds multiplier" estimated that, on average, the odds of a death sentence being imposed were 8.3 times higher in multiple-victim cases than they were in single-victim cases. We also report the odds ratios. (90)

ii. Regression-Based Scale

On the basis of these logistic regression results, we created our second culpability measure--three five-level scales. (91) Although these scales are substantially based on regression results, the process by which race effects are estimated with these scales is more transparent than the regression analysis. Also, the validity of the estimated disparities is more verifiable through close analysis of the facts of the cases that we define as similar in terms of accused culpability. Each of these scales discriminates well in predicting charging and sentencing outcomes. (92)

iii. Salient Factors Scale

Our third measure of accused culpability is a six-level scale based on three salient case factors that bear on offender culpability. (93) This culpability measure is completely independent of the regression results. It also lends itself to validation through qualitative assessments of the cases deemed to be comparable by the procedure. (94) The measure also discriminates very well in explaining charging and sentencing outcomes. (95)

Our assessment of the risk of systemic race effects overall and at each stage of the process depends on the magnitude, statistical significance, and consistency of the estimated racial disparities. Particularly important in this regard is the consistency of the disparities estimated with controls for the salient factors scale and the disparities estimated with controls for the two regression-based measures of criminal culpability. (96)

4. Other Approaches

We conducted a comparable analysis of death-sentencing disparities among the sixteen multiple-victim cases in the database, which have an unusually high death-sentencing rate (50%) and account for 53% (8/15) of the death sentences imposed since 1984. (97) In this analysis, we applied an additional measure of offender culpability based on the qualitative assessments of five law students who analyzed the comparative culpability of the sixteen multiple-victim cases. (98)

We also conducted case-specific quantitative and qualitative analyses, in which we assessed the risk of racial prejudice in the ten minority-accused cases that resulted in a death sentence. This analysis is based on an analysis of charging and sentencing outcomes in cases with levels of criminality that are comparable to each death-sentenced minority defendant. (99) Finally, we estimated race effects in sentencing outcomes among the sixty-six death-eligible cases that did not advance to a capital sentencing hearing as a basis for comparing those racial effects with the racial effects documented among the cases that did advance to a capital sentencing hearing. (100)

We conducted generally accepted diagnostics for the six logistic multiple-regression models that are the centerpiece of our analysis of racial disparities. If a regression diagnostic suggested a possible problem, we took corrective measures to determine if the suggested modification affected the magnitude and statistical significance of our core findings. (101)

5. Methodological Issues

In this Section we consider methodological issues that may be of interest to social science methodologists and statisticians.

i. Adjustment and Modeling

The first issue concerns the importance of adjusting the unadjusted racial disparities that we estimate in this Article with multivariate statistical analyses that control for offender culpability. We agree with Professor John Tukey's argument in an influential paper (102) addressed to professional statisticians that in an observational study, which focused on the impact of an anesthetic (halothane) on surgical mortality rates, "adjusting for a variety of covariates was essential. Otherwise, the results of the analysis [of anesthetics] might have been dominated by the practices of anesthesiologists, rather than by the effects of different anesthetics. Here defense is not at all a casual matter: it is vital." (103) Similar concerns convinced us of the importance of adjusting all of our important racial disparities for the criminal culpability of the accused involved in each analysis.

The second issue concerns the behavioral accuracy of the models and measures of offender culpability on which we based our adjusted racial disparities. In this regard, it is important to recognize that we are not attempting to construct a behaviorally correct model of the military charging and sentencing system. Instead we are investigating the simpler question of whether unadjusted racial disparities can be explained away by the presence or absence of some combination(s) of legally relevant, legitimate case characteristics. (104)

With respect to adjustment methodology, Tukey urges restraint, but not abstention, in using regression methods for statistical adjustment. (105) He also strongly recommends using composite covariates as a complement to and a check on regression methods, advice we endorse and have followed in the analysis reported in this paper. Tukey proposed two broad methods for constructing such composites: weighted sums and "smear and sweep." (106) We used weighted sums. (107)

ii. The Identification of Non-Racial Control Variables to Include in the Logistic Regression Analysis

In addition to meeting potential statistical challenges, we believe that capital charging and sentencing studies must also meet standards of legal acceptability. (108) In our screening of candidate variables for inclusion in the logistic regression models and the creation of our salient factors scale, we did two things to meet this requirement. First, we based the foundations of the culpability measures on legally mandated case characteristics. We also constructed a composite that we call the "salient factors scale," based entirely on statutory relevance and expert knowledge of how the charging and sentencing system does and should operate. (109)

Second, experience indicates that exclusive reliance on tests of statistical significance as a basis for screening variables for inclusion in regression models can produce erroneous substantive results. (110) With this in mind, we assessed candidate variables not only on the basis of their impact on the goodness of fit of the models, but also on the basis of our a priori understanding of military law and how the system should operate and did operate in fact. So guided, we rejected candidate variables that were statistically significant in the model but had a "perverse" sign (e.g., a statutory aggravating circumstance with a negative sign suggesting that it had a mitigating effect in the system) and included some statistically insignificant variables whose effect was substantial and consistent with theory, (111)

iii. Sample Size Limitations on the Number of Control Variables Included in the Logistic Regression Analysis

Although our sample includes the entire universe of death-eligible cases prosecuted in the Armed Forces during the period of our study, the total sample (n = 97) is small compared to similar studies of state court capital charging and sentencing. Also, death sentences were imposed in only fifteen (15%) of the cases, and the estimated race-of-victim effects are based on a sample of only seventeen (17%) non-white-victim cases.

The small number of military death sentences imposes limits on the number of racial and non-racial variables (covariates) that may be included in a logistic regression analysis. Professor Tukey reports a broad spectrum of opinion among statisticians, from "optimist" to "pessimist" in terms of the number of covariates that may be included in a logistic regression model. He summarizes this opinion to the effect that each additional covariate added to a regression model in effect reduces the "equivalent number of rarer endpoints" by from "one or two" to "at least 10 rarer end points for each parameter fitted." (112) In our military study, the "rare endpoint" is the more punitive outcome at each decision point. (113) Thus, under the two "rare endpoints" test per covariate, the appropriate number of explanatory variables for our model of death sentences imposed among all death-eligible cases would be 7.5 (15/2). We share these concerns and, in the construction of our regression models, strived to follow the guidelines referred to by Tukey. In those models, the ratio of death sentences to covariates ranged from 1.9 to 5.2. (114) Because our models fall on the low end of the guidelines accepted by statisticians, we place particular weight on our composite culpability scales, one of which--the salient factors scale--is completely independent of multiple regression results.

The comparatively small number of non-white-victim cases (n = 17) also affects the number of non-racial explanatory variables that may be included in a multiple regression analysis with the white-victim variable included. The potential of the non-racial variables (covariates) to explain away the white-victim effect has the same implications for the total number of variables included in the model as the capacity of the model to predict an outcome variable with a small number of adverse outcomes, such as the imposition of capital punishment. The same concern applies to the application of our offender-culpability scales to the extent they could …