In April 1987, the United States Supreme Court dismissed statistical arguments in McCleskey v. Kemp that the Georgia capital sentencing system was racially biased, and generally rejected the use of any complex statistical model as evidence of discrimination in capital sentencing decisions. Several members of Congress have attempted to override the Supreme Court decision in McCleskey by proposing legislation, titled the Racial Justice Act, to prohibit the statewide application of the death penalty, if the penalty is imposed in a racially disproportionate pattern. This article discusses the general limitations of statistics to infer racial bias in capital cases by considering the statistical evidence presented in McCleskey and examines the impact that the Racial Justice Act would have on fairness in the application of the death penalty.

* I am grateful to Mary Beth Westmoreland (Georgia Attorney General's Office), George M. Palmer (Appellate Division for Los Angeles County), and William Miller (Past President of the Southeastern Legal Foundation) for their advice and suggestions on earlier versions of this paper. I would also like to thank the four anonymous reviewers for their recommendations for improving this article.

1. INTRODUCTION

    On Saturday May 13, 1978, about 2:15 in the afternoon, Warren McCleskey, Ben Wright, David Burney, and Bernard Dupree, four black men, robbed the Dixie Furniture Store on Marietta Street in Atlanta, Georgia. All four men were armed with either a gun or a shotgun. McCleskey entered through the front of the store, while the other three came in through the loading dock in the back of the store. The robbers tied up the employees and customers with hospital tape, forced them to lie on the floor, and threatened to kill anyone who moved.

    Answering a silent alarm, triggered by one of the furniture store employees that signaled a robbery in progress, Frank Schlatt, a white police officer, drove his patrol car to the front of the furniture store. McCleskey saw the patrol car arrive and hid behind a couch. Officer Schlatt entered the furniture store and proceeded down the center aisle to the middle of the store. McCleskey jumped up from behind the couch and fired two shots at the police officer. One bullet hit Officer Schlatt in the eye; the fatal shot. The second bullet ricocheted off a cigarette lighter that was in Officer Schlatt's left shirt pocket. Officer Schlatt fell backward to the floor. McCleskey escaped through the front of the store. Officer Schlatt was found in a pool of blood and died several hours later. In October, 1978, Warren McCleskey was tried, convicted, and sentenced to death for the murder of Officer Schlatt.

    One of the issues raised by McCleskey's lawyers on appeal in Federal Court was that McCleskey's death sentence was unfair because the Georgia capital sentencing system is racially biased against black murderers and killers of white victims. This claim was based on a statistical study of Georgia defendants conducted by Professors David C. Baldus, George Woodworth, and Charles Pulaski.

    An evidentiary hearing on McCleskey's claim of racial discrimination in death sentencing in Georgia was held in August 1983. Federal District Court Judge J. Owen Forrester ruled that the conclusion of the study, of racial bias in the Georgia system, was without merit. In 1985, the full 11th Circuit Court of Appeals, by a 9 to 3 vote, agreed with the District Court on this issue concluding (753 F.2d at 899) "Viewed broadly, it would seem that the statistical evidence presented here, assuming its validity, confirms rather than condemns the system." Finally, in April 1987, the United States Supreme Court, by a 5 to 4 count affirmed the decision of the 11th Circuit Court of Appeals. The court ruled that the statistical study, conducted by Professors Baldus, Woodworth and Pulaski, did not establish that the Georgia capital sentencing system discriminates against black defendants or killers of white victims. Furthermore, the Supreme Court flatly rejected the use of any complex statistical model as evidence to infer discrimination in capital sentencing decisions. The controversy did not end with the United States Supreme Court decision.

    In 1988, Michigan Representative John Conyers sponsored Bill H.R. 4442, titled the "Racial Justice Act of 1988," which sought to prohibit the application of the death penalty if the penalty is imposed in a racially disproportionate pattern. The Racial Justice Act of 1988 provided that a prima facie case of racial bias would be established when ordinary methods of statistical proof show that "death sentences are being imposed or executed-

(1) upon persons of one race with a frequency that is disproportionate to their representation among the total numbers of persons arrested for, charged with, or convicted of, death eligible crimes; or

(2) as punishment for crimes against persons of one race with a frequency that is disproportionate to their representation among the number of persons against whom death eligible crimes have been the subject of arrests, charges, or convictions.

To rebut a prima facie case, the Racial Justice Act of 1988 required that "a government must establish by clear and convincing evidence that identifiable and pertinent nondiscriminatory factors persuasively explain the observable racial disparities comprising the disproportion." Senator Kennedy offered the Racial Justice Act to the Senate as an Amendment to the Omnibus Drug Bill. The Amendment was soundly defeated. In 1990, a modified version of the Racial Justice Act was offered as an Amendment to the Crime Bill, and again defeated in the Senate. However, the House of Representatives narrowly approved an Amendment to the Racial Justice Act offered by Representative Hughes, which strengthened the defendant's evidentiary requirements for establishing a prima facie case, and lowered the standard of proof required from the prosecution to rebut a prima facie case. According to the Hughes Amendment, an inference that race was the basis of the death sentence is established if, at the time the death sentence was imposed, that death sentences were "imposed significantly more frequently in the jurisdiction in question--

(1) upon persons of one race than upon persons of another race;  or

(2) as punishment for capital offenses against persons of one race than as punishment for capital offenses against persons of another race.

    If a significant racial disparity in the imposition of the death sentence is found, the Hughes Amendment required the court to determine the validity of the inference that race was the basis of the death sentence by considering evidence of the statutory aggravating factors of the crimes involved and by comparing similar cases involving persons of different races. The Hughes Amendment lowered the standard of proof needed to rebut the inference that race was the basis of the death sentence, to a preponderance of the evidence. The Hughes Amendment did not survive the final compromise Crime bill passed by Congress in 1990. In 1991, the Hughes Amendment was again offered as part of the Crime bill, with a new name "Fairness in Death Sentencing Act." This time, the bill was defeated by both the Senate and the House of Representatives.

    Each version of the Racial Justice Act begins with the premise that a fair capital sentencing system should produce equal death sentencing rates by race, and racially disproportionate death sentencing rates are probably due to racial bias. This underlying premise is flawed because it ignores fundamental differences in racial homicide patterns.

    I was hired by the Georgia Attorney General to analyze the Baldus et al. databases, and testified as to my findings at McCleskey's evidentiary hearing. In my opinion, the Supreme Court ruled properly that statistical models are inappropriate and ineffective to measure the extent (if any) of discrimination in capital sentencing decisions. This article discusses the general limitations of statistics to infer racial bias in capital cases by considering the statistical evidence presented in McCleskey and examines the effect that the Racial Justice Act, if enacted, would have on the equitable application of the death penalty.

2. THE GEORGIA CAPITAL SENTENCING PROCESS

    In 1972, the United States Supreme Court, in Furman v. Georgia, voided all death penalty statutes nationwide, because they permitted the imposition of death sentences on defendants, who kill under extremely mitigated circumstances. The court concluded that the lack of safeguards against freakish and capricious death sentences violates the Eighth Amendment prohibition against "cruel and unusual punishment." In 1973, the Georgia Legislature enacted revised death penalty statutes to conform with the guidelines demanded by Furman, which the Supreme Court approved in 1976 in Gregg v. Georgia.

    The revised Georgia statutes divide murder trials into two parts. First, the jury hears evidence concerning the guilt or innocence of the defendant and renders a decision. If the defendant is found guilty of murder, then a second trial, or penalty trial, is held to determine the sentence, either life imprisonment or death. As a matter of practice, a defendant convicted of murder is automatically given a life sentence unless the prosecutor asks for the death penalty.

    During the penalty trial, the prosecution and defense are permitted to offer the jury additional aggravating and mitigating evidence unrelated to the murder. A death sentence may be imposed on the defendant only if the jury finds, beyond a reasonable doubt, that the murder was accompanied by one or more of the following 10 statutory aggravating circumstances:

(1) The offense of murder was committed by a person with a prior record of convictions for murder, armed robbery, kidnaping or rape, or the offense of murder was committed by a person with a substantial history of serious assaultive criminal convictions.

(2) The offense of murder was committed while the offender was engaged in the commission of rape, armed robbery, or kidnaping, or aggravated battery, or the offense of murder was committed while the offender was engaged in the commission of burglary or arson in the first degree.

(3) The offender by his act of murder, knowingly created a great risk of death to more than one person in a public place by means of a weapon or device which would normally be hazardous to the lives of more than one person.

(4) The offender committed the offense of murder for himself or another, for the purpose of receiving money or any other thing of monetary value.

(5) The murder of a judicial officer, former judicial officer, district attorney or solicitor or former district attorney or solicitor during or because of the exercise of his official duty.

(6) The offender caused or directed another to commit murder or committed murder as an agent or employee of another person.

(7) The offense of murder was outrageously or wantonly vile, horrible, or inhuman in that it involved torture, depravity of mind, or aggravated battery to the victim.

(8) The offense of murder was committed against any peace officer, corrections employee or fireman while engaged in the performance of his official duties.

(9) The offense of murder was committed by a person in, or who has escaped from, the lawful custody of a peace officer or place of lawful confinement.

(10) The murder was committed for the purpose of avoiding, interfering with, or preventing a lawful arrest or custody in a place of lawful confinement, of himself or another.

    If the jury does not find one or more of these 10 statutory aggravating factors present, beyond a reasonable doubt, a life sentence is automatically imposed. Furthermore, the jury has full discretion to decline to sentence a convicted killer to death, even if the defendant has violated one or more of the 10 statutory aggravating factors.

    The revised Georgia statutes also mandate that a death-sentenced defendant be granted an automatic appeal of his sentence to the Georgia Supreme Court. Among other issues to consider, the Georgia Supreme Court must conduct a proportionality review to determine whether the defendant's "sentence of death is excessive or disproportionate to the penalty imposed in similar cases, considering both the crime and the defendant." Thus, the Georgia Supreme Court is given the authority to vacate a death sentence if one is wantonly or freakishly imposed.

3. THE BALDUS, WOODWORTH, AND PULASKI STUDIES

    David C. Baldus was interested in studying whether Georgia's revised death sentencing statutes truly changed the way in which death sentences were imposed. With the help of George Woodworth, a statistician, and another legal colleague, Charles Pulaski, data was collected on Georgia murder cases to compare the sentencing pattern of defendants convicted of murder before the Furman ruling with defendants convicted of murder under the revised Georgia death penalty statutes. This effort was called the Procedural Reform Study (PRS) (see Baldus et al. (1984), (1985) and (1986) for the findings from this study). The NAACP Legal Defense Fund commissioned the Baldus team to conduct a second more detailed statistical study, called the Georgia Charging and Sentencing Study (GCSS), to focus on sentencing practices in Georgia after Furman (see Baldus et al. 1990). The purpose of the second study was to empirically test whether or not the revised Georgia death sentencing system is operating fairly, and formed the basis for the statistical analyses presented to the District Court.

    The second study attempted to collect information on over 500 crime related factors for a sample of 1,082 defendants convicted of voluntary manslaughter or murder in Georgia between 1973 and 1978. Drawing on information from data files at the Georgia Department of Pardons and Paroles, five 2nd year law students were paid $5 an hour to fill out a 42 page questionnaire for each defendant sampled. The bulk of the data on the circumstances of the offense was taken from a one or two page police or criminal investigation report. Most of the questionnaire items are categorical, indicating whether or not the coder believed a particular aggravating, mitigating or evidentiary factor occurred in the case.

4. LIMITATIONS OF THE GEORGIA CHARGING AND SENTENCING STUDY DATABASE

    Despite the persistent data-gathering effort, the final database has serious problems. Over 100 questionnaire items associated with the crime were unknown for a significant percentage of the 1,082 cases (see 580 F.Supp 358). For instance, in 41% of the cases, it was unknown if the homicide was motivated by racial hatred. In 46% of the cases, it was unknown if the homicide had been planned for more than five minutes. In 74% of the cases, it was unknown whether the victim pleaded for his life. In 82% of the cases it was unknown if the defendant expressed pleasure with the homicide. Even the race of the victim was unknown for 62 cases. The consistency and accuracy of the data was a further problem. When identical aggravating or mitigating factors were compared in the same case between the earlier Procedural Reform Study and the later Georgia Charging and Sentencing Study, differences were often found (see 580 F.Supp at 357-358).

    There are further structural problem with the questionnaire and the coding of items. Most crime related factors are collected as categorical items, which ignores qualitative differences in the factors between cases (for example torture, mental torture, victim beaten before killing, defendant bragged about killing, or victim had a criminal record).

    Other conventions confound the aggravation or mitigation effect of certain circumstances of the homicide. For example, one question lists 19 offenses, ranging from misdemeanors to felonies such as armed robbery, rape, and kidnaping, and asks the coder to indicate whether the homicide occurred while the defendant was engaged in the commission of each offense whether or not the defendant was charged and convicted of the offense.

    In the case of homicides with multiple offenders, the study adopted the protocol of attributing the aggravating circumstances of the crime to each offender without adjustment for his/her particular level of involvement. This muddling of the aggravating variables makes it more difficult for statistical models to predict or explain the sentencing outcomes for each offender. Multiple offenders are involved in 40.3% of the life sentence cases and 61.6% of death penalty cases (see Table I).

    To assess the reliability and face validity of the data from the two studies, I presented the coded data on the death penalty cases to the State attorneys, who litigate all death penalty appeals for Georgia. After examining the data, the consensus agreed that the two studies missed important aggravating and evidentiary facts in the death penalty cases. In fact, important aggravating factors were missing from McCleskey's own case. Consider the following comparison of several items from Warren McCleskey's questionnaire to the record from his trial transcript.

(1) In 1970, Warren McCleskey was charged with 25 to 30 counts of armed robbery. A jury convicted him on 3 counts of armed robbery and imposed 3 life sentences. McCleskey also pled guilty to 5 other counts of armed robbery. In 1971, McCleskey was resentenced to 18 years for 3 counts of armed robbery. The questionnaire records that McCleskey was sentenced to 18 years for 3 counts of armed robbery.

(2) McCleskey spent 7 years and 4 months in prison from 1970 through 1977. McCleskey was released into a work release program in September 1977 and paroled several months later. The questionnaire did not record any of these facts and the foil designed to record the amount of time McCleskey spent in prison was left blank, inferring that McCleskey spent no time in prison.

(3) McCleskey was implicated in the armed robberies of the Red Dot Fruit Stand in Powder Springs in March 1978, and the Red Dot Grocery Store on Edgewood Ave. in April 1978. The manager of the Red Dot Fruit Stand became hysterical during his testimony. He recalled being tied up with tape, a gun pointed at his head, thinking that he was about to die. A victim of the Red Dot Grocery Store robbery testified that she was held at gunpoint while the robbers stole her purse. The questionnaire did not ask for, nor does it record the brutal psychological abuse McCleskey and his coperpetrators inflicted upon their robbery victims.

(4) McCleskey visited the Dixie Furniture Store on the morning before the robbery and talked with a saleslady about buying a bed. Before returning in the afternoon to rob the store, McCleskey changed shirts, shaved his moustache, and used makeup to make his face appear bumpy with a scar. The questionnaire indicates that the contemporaneous offense was planned for more than 5 minutes.

(5) As to the actual sequence of events leading to the murder of Officer Schlatt, as described in the introduction section of this article, the questionnaire records simply that between 6 and 10 people were exposed to a risk of death during the course of an armed robbery by McCleskey and 3 coperpetrators.

(6) Concerning Officer Schlatt's murder, the questionnaire records that the homicide was bloody, the defendant intended to use deadly force, one of the two shots hit the victim in the head, and the defendant left the scene of the crime.

(7) After his arrest, McCleskey gave a written statement to the police in which he admitted that he had robbed the Dixie Furniture Store on May 13, 1978, but denied that he was responsible for killing Officer Schlatt. The questionnaire records that McCleskey gave an oral statement rather than a written statement.

(8) McCleskey testified at his trial that he was not at the Dixie Furniture Store on May 13th. Instead, he was at his sister's house, his mother's house and he played cards with his friends. No one testified at McCleskey`s trial to corroborate his story. The questionnaire records that McCleskey's defense at trial was "mistaken identity."

(9) McCleskey concluded his testimony with "I am not a killer. I am not a murderer. I am not a violent person at all. I never assaulted anyone or cut anyone or shot anyone in my life." The questionnaire records that it was unknown to the coder whether McCleskey was remorseful for the killing.

(10) McCleskey confessed to the murder of Officer Schlatt to Offie Evans, his cellmate in the Fulton County Jail. McCleskey bragged to Evans that he would "shoot his way out if it had been a dozen of them." The questionnaire records that it was unknown to the coder whether or not McCleskey expressed pleasure with the killing.

    Murder cases are stories, not a collection of categorical aggravating, mitigating, and evidentiary factors. Even if the foils available in McCleskey's questionnaire were completely and accurately filled out, there would still be insufficient detail to enable anyone, unfamiliar with his case, to recreate the crime story.

    The District Court ruled on the sufficiency of the database and found (580 F.Supp at 360) "the data base has substantial flaws and that petitioner has failed to establish by a preponderance of the evidence that it is essentially trustworthy," and "there are errors in coding the questionnaire for the case sub judice. This fact alone will invalidate several important premises of petitioner's experts."

5. ANALYSIS OF THE GEORGIA CHARGING AND SENTENCING STUDY IS THE GEORGIA SENTENCING SYSTEM ARBITRARY?

    If a defendant is convicted of murder and the prosecution asks for the death penalty, then the trial moves to the penalty phase. The jury is obligated to consider the combination of all relevant aggravating, mitigating and evidentiary facts of the case before imposing a sentence. Thus, it is impossible to judge the appropriateness of a defendant's sentence based upon the knowledge of whether or not one or several particular factors occurred. However, if the revised statutes are working as designed, then death-sentenced defendants should have a relatively high frequency of incidents that tend to exacerbate homicides such as the mutilation or rape of the victim. Furthermore, death-sentenced defendants should also have a relatively low frequency of incidents that tend to excuse homicides such as the killing resulting from a family quarrel or other dispute. This is precisely the pattern that is depicted by the Georgia Charging and Sentencing Study database. Table I below compares the incidence of important aggravating and mitigating factors of the crimes according to the offender's conviction and sentence.
 

		TABLE 1
	COMPARISON OF AGGRAVATING AND MITIGATING CRIME FACTORS BY CONVICTION AND SENTENCE
	VOLUNTARY MANSLAUGHTER CONVICTION: 1 - 20 YEARS		MURDER CONVICTION: LIFE SENTENCE		MURDER CONVICTION: DEATH SENTENCE
AGGRAVATING FACTORS	COUNT**	PERCENT	COUNT**	PERCENT	COUNT**	PERCENT
VICTIM KILLED DURING ARMED ROBBERY	14/504	2.8%	118/422	28.0%	77/138	55.8%
VICTIM KIDNAPED	2/504	0.4%	27/422	6.4%	51/138	37.0%
VICTIM RAPED	1/504	0.2%	15/422	3.6%	28/138	20.3%
VICTIM BEATEN	21/502	4.2%	50/420	11.9%	28/140	20.0%
VICTIM WITNESS TO CRIME	3/501	0.6%	51/390	13.1%	56/117	47.9%
EXECUTION STYLE MURDER	24/468	5.1%	65/380	17.1%	67/125	53.6%
VICTIM MUTILATED	7/502	1.4%	43/420	10.2%	43/140	30.7%
KILLING UNNECESSARY TO FINISH CRIME	7/499	1.4%	45/395	11.4%	50/119	42.0%
VICTIM PLED FOR LIFE	11/118	9.3%	32/98	32.7%	44/67	65.7%
VICTIM TORTURED	2/502	0.4%	5/420	1.2%	17/140	12.1%
VICTIM FORCED TO DISROBE	2/502	0.4%	20/420	4.8%	44/140	31.4%
DEFENDANT RESISTED ARREST	31/474	6.5%	75/403	18.6%	53/138	38.4%
MITIGATING FACTORS
JEALOUSY MOTIVATED THE KILLING	67/421	15.9%	43/392	11.0%	1/139	0.7%
DEFENDANT ENRAGED	389/444	87.6%	166/348	47.7%	23/109	21.1%
FAMILY, LOVER, LIQUOR, BARROOM FIGHT	321/443	72.5%	150/390	38.5%	12/129	9.3%
KILLING RESULTED FROM DISPUTE	373/458	81.4%	178/380	46.8%	18/128	14.1%
DEFENDANT REMORSEFUL	32/56	57.1%	20/74	27.0%	2/39	5.1%
VICTIM PROVOKED PASSION IN DEFENDANT	92/397	23.2%	23/391	5.9%	0/131	0.0%
DEFENDANT/VICTIM STATUS
DEFENDANT POOR	389/503	77.3%	327/417	78.4%	38/100	38.0%
DEFENDANT HIGH STATUS	12/504	2.4%	12/416	2.9%	4/127	3.1%
VICTIM A STRANGER TO DEFENDANT	38/495	7.7%	122/413	29.5%	82/140	58.6%
VICTIM FAMILY, FRIEND, OR INTIMATE	272/495	54.9%	149/419	35.6%	20/140	14.3%
VICTIM A POLICE OFFICER	0/507	0.0%	13/421	3.1%	9/140	6.4%
MULTIPLE OFFENDERS	45/500	9.2%	170/422	40.3%	85/138	61.6%
NUMBER OF CASES IN DATABASE	516		426		140
* Cases in which the occurrence of a factor is unknown are excluded from the percentage calculation for that factor.
** Under the column "COUNT," the denominator indicates the number of defendants for which the factor was known and the numerator indicates the number of defendants for which the factor was indicated present in the case.

    The 1,082 defendants were divided into three distinct groups by crime of conviction (voluntary manslaughter or murder) and sentence: (1) defendants convicted of voluntary manslaughter and sentenced to prison for 1 to 20 years, (2) defendants convicted of murder and sentenced to life imprisonment and (3) defendants convicted of murder and sentenced to death. For several important aggravating and mitigating features of homicides, Table I displays the percentage of cases for which each of these factors occurred, broken down by crime of conviction and sentence. Table I is an abbreviated version of a more detailed exhibit with over 100 aggravating and mitigating factors that I presented to the court.

    According to Table I, the victim was killed during an armed robbery in 55.8% of the death sentence cases, 28.0% of the life sentence cases, and 2.8% of the voluntary manslaughter cases. On the other hand, the homicide arose as the result of a dispute in 81.4% percent of the voluntary manslaughter cases, 46.8% of the life sentence cases and only 14.1% of the death sentence cases.

    Cases in which the occurrence of a factor is unknown are excluded from the percentage calculation for that factor. Admittedly, the vast number of unknown and incorrect data in the Georgia Charging and Sentencing Study database limit the accuracy and reliability of any analysis of the data. Nevertheless, Table I shows a consistent pattern of higher percentages for aggravating factors and lower percentages for mitigating factors as the sentence increases in severity. Death sentence cases tend to exhibit the highest percentage of each aggravating factor and the lowest percentage of each mitigating factor. Unless the coders consistently misrepresented the general facts of most cases, it is unlikely that complete correct data would substantially change or reverse the general pattern of higher percentages for aggravating factors and lower percentages for mitigating factors as the sentence increases in severity.

    On viewing the more extensive exhibit I prepared for the court, the District Court was moved to find (580 F.Supp at 365) "These observations, other testimony by all of the experts and the court's own analysis of the data put to rest in the court's mind any notion that the imposition of the death penalty in Georgia is a random event unguided by rational thought."

    Of particular interest is the third section of Table I, where 77.3% of the defendants convicted of voluntary manslaughter and 78.4% of defendants convicted of murder and sentenced to life are classified as poor as compared to only 38.0% of the defendants sentenced to death. This is inconsistent with the theory that only the poor and the powerless are sentenced to death. Furthermore, high status defendants are not prone to commit homicides which explains the extremely low percentages within each sentence category. Yet, further analysis reveals that 14.3% (4/28) of high status defendants are sentenced to death to only 5.0% (38/754) of poor defendants.

RACE OF VICTIM ANALYSIS

    The Georgia Charging and Sentencing Study estimated that 7% of white defendants were sentenced to death as compared to only 4% of black defendants. Therefore, death sentences were imposed on a higher percentage of whites than blacks. However, when the cases were divided up by the race of the victim, 11% of the white victim killers were sentenced to death as compared to only 1% of the black victim killers. There are two plausible explanations for this 10 percent disparity.

    Baldus testified that racial factors play a role in determining who receives the death penalty in Georgia. He contends that at least part of the difference in death sentencing rates is due to institutionalized racism in the Georgia sentencing system, in that a black life is not held to be as valuable as a white life. An alternative explanation for the 10 percent disparity in death sentencing rates is that white victim homicides generally occur in more aggravated circumstances than black victim homicides. The Georgia Charging and Sentencing Study database supports this position. Table II displays the percentage of cases in which several of the more important aggravating and mitigating features are broken down by the victim's race.
 

			TABLE ll
	COMPARISON OF AGGRAVATING AND MITIGATING CRIME FACTORS BY THE RACE OF THE VICTIM
	WHITE VICTIM CASES		BLACK VICTIM CASES
AGGRAVATING FACTORS	COUNT	PERCENT	COUNT	PERCENT	PER CENT DIFFERENCE	Z SCORE
VICTIM KILLED DURING ARMED ROBBERY	170/510	33.3%	37/498	7.4%	25.9%	10.2
VICTIM KIDNAPED	68/510	13.3%	11/498	2.2%	11.1%	6.6
VICTIM RAPED	29/510	5.7%	15/498	3.0%	2.7%	2.1
VICTIM BEATEN	63/512	12.6%	31/493	6.3%	6.3%	3.3
VICTIM WITNESS TO CRIME	92/471	19.5%	17/481	3.5%	16.0%	7.8
EXECUTION STYLE MURDER	118/465	25.4%	34/456	7.5%	17.9%	7.3
VICTIM MUTILATED	76/512	14.8%	14/493	2.8%	12.0%	6.7
KILLING UNNECESSARY TO FINISH CRIME	81/474	17.1%	21/484	4.3%	12.8%	6.4
VICTIM PLED FOR LIFE	60/156	38.5%	26/117	22.2%	16.3%	2.9
VICTIM TORTURED	18/512	3.5%	6/493	3.0%	6.8%	4.3
VICTIM FORCED TO DISROBE	50/512	9.8%	15/493	3.0%	6.8%	4.3
DEFENDANT RESISTED ARREST	112/493	22.7%	42/469	9.0%	13.7%	5.8
MITIGATING FACTORS
JEALOUSY MOTIVATED THE KILLING	29/477	6.1%	70/427	16.4%	-10.3%	-5.0
DEFENDANT ENRAGED	200/423	47.3%	334/428	78.0%	-30.7%	-9.3
FAMILY, LOVER, LIQUOR, BARROOM FIGHT	165/468	35.3%	284/444	64.0%	-28.7%	-8.7
KILLING RESULTED FROM DISPUTE	205/464	44.2%	327/450	72.7%	-28.5%	-8.7
DEFENDANT REMORSEFUL	21/91	23.1%	29/71	40.8%	-17.7%	-2.4
VICTIM PROVOKED PASSION IN DEFENDANT	34/451	7.5%	73/417	17.5%	-10.0%	-4.5
DEFENDANT/VICTIM STATUS
DEFENDANT POOR	302/469	64.4%	407/488	83.4%	-19.0%	-6.7
DEFENDANT HIGH STATUS	20/493	4.1%	8/494	1.6%	2.5%	2.3
VICTIM A STRANGER TO DEFENDANT	184/509	35.8%	55/483	18.8%	17.0%	9.1
VICTIM FAMILY, FRIEND, OR INTIMATE	167/509	34.6%	242/483	39.6%	-5.0%	-5.5
VICTIM A POLICE OFFICER	18/512	3.5%	3/498	0.6%	2.9%	3.2
MULTIPLE OFFENDERS	227/507	44.8%	65/495	13.3%	31.5%	10.9
NUMBER OF CASES IN DATABASE**	516		504
DEATH SENTENCE PERCENTAGE		11%		1%
* Cases in which the occurrence of a factor is unknown are excluded from the percentage calculation for that factor.
** The percentage calculations omit the 62 cases in which the race of the victim was unknown.
*** The Z-value indicates the number of standard deviations the calculated percent difference is from 0.

    White victim homicides have a much higher incidence of armed robberies, kidnapings, and rapes. For example, in 33.3% of the white victim cases, the victim was killed during the course of an armed robbery as compared to only 7.4% of the black victim cases. The 25.9% difference in armed robbery rates between white victim and black victim cases is approximately 10.2 standard deviations above 0%. Furthermore, white victim homicides show a greater percentage of mutilations, execution style murders, tortures, and beaten victims, features which generally aggravate homicides and increase the likelihood of a death sentence if one or more statutory aggravating factors is also present.

    On the other hand, cases involving black victims are more likely to involve an enraged and remorseful defendant, who kills as the result of a fight or dispute. For instance, in 44.2% of the white victim cases, the homicide was precipitated by a dispute or fight as compared to 72.7% of the black victim homicides.

DEFENDANT-VICTIM RACIAL COMBINATION ANALYSIS

    The Georgia Charging and Sentencing Study projected that 22% of black killers of white victims, 8% of white killers of white victims, 1% of black killers of black victims, and 3% of white killers of black victims are sentenced to death. Table III shows the incidence of important aggravating and mitigating factors broken down by the four defendant-victim racial combinations.
 

			TABLE llI
	COMPARISON OF AGGRAVATING AND MITIGATING CRIME FACTORS BY THE DEFENDANT VICTIM RACIAL COMBINATION
	BLACK KILLS WHITE		WHITE KILLS WHITE		BLACK KILLS BLACK		WHITE KILLS BLACK
AGGRAVATING FACTORS	COUNT	PERCENT	COUNT	PERCENT	COUNT	PERCENT	COUNT	PERCENT
VICTIM KILLED DURING ARMED ROBBERY	96/143	67.1%	74/367	20.2%	31/471	6.6%	6/27	22.2%
VICTIM KIDNAPED	26/143	18.2%	74/367	20.2%	10/471	6.6%	6/27	22.2%
VICTIM RAPED	14/143	9.8%	42/367	11.4%	15/471	3.2%	0/27	0.0%
VICTIM BEATEN	16/143	11.2%	15/367	4.1%	30/467	6.4%	1/26	3.8%
VICTIM WITNESS TO CRIME	47/129	36.4%	47/369	12.7%	14/457	3.1%	3/24	12.5%
EXECUTION STYLE MURDER	42/127	33.1%	76/338	22.5%	131/432	7.2%	3/24	12.5%
VICTIM MUTILATED	25/143	17.5%	51/369	13.8%	13/467	2.8%	1/26	3.8%
KILLING UNNECESSARY TO FINISH CRIME	47/124	37.9%	34/350	9.7%	21/461	4.6%	0/23	0.0%
VICTIM PLED FOR LIFE	25/43	58.1%	35/113	31.0%	23/108	21.3%	3/9	33.3%
VICTIM TORTURED	3/143	2.1%	15/369	4.1%	6/467	1.3%	0/26	0.0%
VICTIM FORCED TO DISROBE	18/143	12.6%	32/369	8.7%	15/467	3.2%	0/26	0.0%
DEFENDANT RESISTED ARREST	37/141	26.2%	75/352	21.3%	37/444	8.3%	5/25	20.0%
MITIGATING FACTORS
JEALOUSY MOTIVATED THE KILLING	0/140	0.0%	29/337	8.6%	70/403	17.4%	0/24	0.0%
DEFENDANT ENRAGED	22/121	18.2%	178/302	58.9%	321/409	78.5%	13/19	68.4%
FAMILY, LOVER, LIQUOR, BARROOM FIGHT	10/133	7.5%	155/335	46.3%	273/420	65.0%	11/24	45.8%
KILLING RESULTED FROM DISPUTE	24/132	18.2%	181/332	54.5%	311/426	73.0%	16/24	66.7%
DEFENDANT REMORSEFUL	3/22	13.6%	18/69	26.1%	26/67	38.8%	3/4	75.0%
VICTIM PROVOKED PASSION IN DEFENDANT	4/132	3.0%	30/319	9.4%	67/394	17.0%	6/23	26.1%
DEFENDANT/VICTIM STATUS
DEFENDANT POOR	93/125	74.4%	209/344	60.8%	386/460	83.9%	21/28	75.0%
DEFENDANT HIGH STATUS	0/133	0.0%	20/360	5.6%	7/466	1.5%	1/28	3.6%
VICTIM A STRANGER TO DEFENDANT	101/143	70.6%	83/366	22.7%	44/457	9.6%	11/26	42.3%
VICTIM FAMILY, FRIEND, OR INTIMATE	7/143	4.9%	160/366	43.7%	235/457	51.4%	7/26	26.9%
VICTIM A POLICE OFFICER	15/142	10.6%	3/370	0.8%	3/472	0.5%	0/26	0.0%
MULTIPLE OFFENDERS	82/140	58.6%	145/367	39.5%	58/468	12.4%	8/27	29.6%
NUMBER OF CASES IN DATABASE**	145		371		477		28
DEATH SENTENCE PERCENTAGE		22%		8%		1%		3%
* Cases in which the occurrence of a factor is unknown are excluded from the percentage calculation for that factor.
** The percentage calculations omit the 62 cases in which the race of the victim was unknown.

    The four defendant-victim racial combinations provide the key that unlocks the race-of-victim sentencing puzzle. Each defendant-victim racial combination portrays a fundamentally different homicide pattern. The black defendant white victim cases are the most aggravated of all four defendant-victim racial combinations. In 67.1% of the cases, the homicide results from an armed robbery whereas only 18.2% of the time is the homicide precipitated by a dispute. The interracial nature of this kind of homicide minimizes the possibility that the killing arose due to a family dispute, a quarrel between lovers, or arguments between friends and relatives. The victim was a stranger 70.6% of the time, and a family member or friend in only 4.9% of the cases. Multiple offenders were involved in 58.6% of the cases. Black defendant white victim killings are invariably linked to felony circumstances which legally qualifies the defendant for a death sentence.

    Of all four defendant-victim racial combinations, the black defendant black victim homicides occur most frequently with 477 cases. Only 6.6% of the time did the homicide result from an armed robbery, whereas 73.0% of the time the killing was precipitated by a dispute. In this category of homicide, the victim is a stranger only 9.6% of the time as compared to a family member or friend 51.4% of the time. Multiple offenders are involved in only 12.4% of the cases. This kind of homicide is characterized by poor defendants (83.9%), acting alone, who kill family members, friends, or other acquaintances during a fight or argument. This type of killing tends to have mitigating circumstances which explains why few of these defendants are ever sentenced to death.

    The white defendant white victim cases reflect a mixture of two basic types of killings. Although the majority of these homicides are precipitated by disputes and fights, a significant percentage occur during the course of a felony or involve horrible brutal killings.

    Only 28 cases are categorized as white defendant black victim homicides. Four of these cases are misclassified as interracial killings due to the convention adopted by the study of classifying Hispanic defendants as white but Hispanic victims as black. The relatively few incidents of white defendant black victim homicides and the uncertain data make it difficult to ascertain any distinctive pattern. Yet, one important fact is clear. This type of homicide rarely occurs in Georgia. In fact, over 95% of the black victim cases have black defendants.

    The black victim homicide patterns revealed by the Georgia Charging and Sentencing Study are generally consistent with earlier studies of homicides in Atlanta, Georgia. For the 383 murder and nonnegligent homicides in Atlanta between 1958 and 1962, James Imes (1972) found 83.3% were black victim homicides. Over 99% of the black victim homicides had black offenders. Robbery precipitated 3.2% of the black victim homicides, as compared to 18.2% of the white victim homicides. A dispute, domestic quarrel, jealousy, or revenge caused 90.1% of the black victim homicides, compared to 77.3% of the white victim homicides. Rivers (1975) investigated socioeconomic factors related to 237 Atlanta homicides in 1972 of which 75.5% of them were black offender-black victim. Rivers concluded that "It can be said most emphatically that systemic socioeconomic factors are significant in explaining the large number of black on black homicides."

    The F.B.I. Uniform Crime Reports for murder and nonnegligent manslaughter, document the consistent intra-racial nature of homicides nationwide, and the disproportionate higher number of black offender black victim homicides. The Georgia Department of Human Resources reports that between 1976 and 1985, homicide was the leading cause of death for black males between 20 and 44. Many researchers, for example Pettigrew and Speir (1962), Wolfgang and Ferracuti (1967), Harvey (1986), Staples (1986), and Hawkins (1986), in attempting to explain the differential racial homicide rates, point generally to cultural differences in the black community and the greater frustration experienced by blacks in trying to escape from their low social and economic standing.

6. MULTIVARIATE ANALYSES OF THE GEORGIA CHARGING AND SENTENCING STUDY

    Although black victims homicides are more mitigated and less aggravated than white victims homicides, this fact, by itself, might not account for the complete 10% disparity in death sentencing rates between killers of whites and killers of blacks. Baldus contended that some slight, systemic bias, still existed against defendants who kill white victims after controlling for all possible factors that could conceivably affect the sentencing outcome. The District Court was presented with the results of numerous multivariate statistical analyses, which Baldus relied on to support his theory.

    Multivariate analyses require that the variable values be known for each case. Thus, Baldus was forced to either recode all the unknown values for the variables in his database or abandon the use of multivariate analysis. Baldus chose to recode the tens of thousands of unknown values for all variables to "0" (i.e. to indicate that the factor did not occur). For example, if it was unknown to the coder whether or not the defendant expressed pleasure with the killing, then that foil was treated as if the defendant did not express pleasure with the killing. Baldus reasoned that if his data collector was unable to determine the status of these factors, given the summary reports that were examined, then the occurrence of the factor must also be unknown to the prosecutors and juries. This assumption is not even true for McCleskey's own questionnaire, and serves to understate the number and degree of aggravating factors that are present in his case. Furthermore, cases in which the race of the victim was unknown were recoded so that the race of the victim matched the race of the defendant.

CROSSTABULATION ANALYSIS

    Initially, to control for the effect of nonracial variables, Baldus used crosstabulation tables to subdivide the cases into distinct groups based upon selected aggravating, mitigating and evidentiary variables. This method allows a comparison of death sentencing rates, by race, for cases whose joint occurrence of the selected aggravating, mitigating, and evidentiary factors are identical. But Baldus limited his crosstabulation analyses to at most three variables at a time.

    Controlling for n variables simultaneously produces 2ⁿ subgroups (assuming each control variable has two values). Thus, controlling for 10 variables results in 1,024 subgroups and controlling for 230 variables produces approximately 1.7 x 10⁶⁹ subgroups. Obviously, controlling for only 25 to 30 relevant factors could cause each of the 1,082 cases to be uniquely classified, which completely eliminates racial disparities in death sentencing rates and explains (although not necessarily correctly) death sentencing results through nonracial factors. There are numerous factors that can influence whether or not the defendant is sentenced to death and each case is a unique combination of these factors. Thus, numerous nonracial explanations of sentencing outcomes exist, even if the capital sentencing system is racially biased.

MULTIPLE REGRESSION AND LOGISTIC REGRESSION

    The bulk of the evidence purporting to show the existence of racial bias in the Georgia capital sentencing system was based upon weighted multiple regression and logistic regression equations. The dependent variable was defined to indicate decisions at various stages of the capital sentencing process, such as whether or not the prosecutor sought a penalty trial or whether or not the defendant was sentenced to death at the penalty trial.

    If the dependent and predictor variables are categorical, multiple regression, like the crosstabulation method, can control for background variables, if all interaction variables are specified in the equation. Controlling for n variables using the crosstabulation method produces 2ⁿ subgroups but the equivalent regression equation, that controls for the same n background variables, requires a total of 2ⁿ - 1 regular and interaction variables.

    The interaction variables are important because in many instances, the aggravating or mitigating effect of a factor can be deduced only within the context of the facts of the case (see also Barnett (1985)). For instance, the fact that the victim used alcohol before the homicide would mitigate the homicide if the intoxicated victim provoked a fight with the defendant, but would aggravate the homicide if the defendant robbed and killed a drowsy victim. Also, the fact that a victim is beaten before the homicide would aggravate the homicide if 4 men rob, attack, rape, and kill a woman abducted from a nearby parking lot, but is almost inconsequential if the victim started a fight with the defendant. As a rule, interaction variables were omitted from the regression models presented to the court.

    The court was given numerous examples of regression and logistic regression equations, which attempted to show the average amount of racial bias present at several stages of the capital sentencing process (see Baldus et al. (1990)). One model contains 236 factors. Invariably, the coefficients for the racial variables were statistically significant at the .05 level.

    Professor Baldus testified that if important background variables are not controlled for, then the coefficients of the racial variables would not necessarily provide the whole picture. For this reason, the legal opinions focus on the interpretation of the regression equation with the largest number of variables, the 230+ variable model (Baldus et al. (1990), Appendix L, pp. 620 - 629). The categorical dependent variable (DPMURIDT) assigns a 1 to cases in which the death penalty was imposed, and a 0 to cases in which the defendant was indicted for murder but the death penalty was not imposed. The 230+ variable model produced a coefficient of +.066 (p = .0145) for the defendant is black (BLACKD) variable and a coefficient of +.065 (p = .0153) for the recoded victim is white (WHVICRC) variable. But the 230+ variable model also has other coefficients that make no sense. Aggravating variables have negative or mitigated weights, such as the victim was a hostage (-.11, p = .2876), the victim was a witness to the crime (-.15, p = .0692), and the victim offered no provocation (-.13 p = .0095). Mitigating variables have positive or aggravated weights, such as defendant had no significant criminal history (+.06, p = .0981), coercion was used on the defendant (+.28, p = .0032), and family dispute other than spouse, or ex-spouse (+.06 p = .0984). Furthermore, the importance and effect of the aggravating, mitigating, and evidentiary factors change dramatically from model to model depending upon the specific combination of factors that are present in the regression or logistic regression equation. The District Court observed (580 F.Supp at 363) "In the Charging and Sentencing Study a very substantial proportion of the variables are correlated to the race of the victim and to the death sentencing result. All or a big proportion of the major nonstatutory aggravating factors show positive correlation with both the death sentencing result and the race of the victim. More than 100 variables show statistically significant relationships with both death sentencing results and the race of the victim." Obviously, the existence of multicollinearity makes the coefficients unreliable as the marginal impact of that variable to the probability of the imposition of a death sentence.

    To impose a death sentence, the prosecution must prove, beyond a reasonable doubt, that the defendant is also guilty of one or more of ten statutory aggravating circumstances. Accordingly, models of capital sentencing decisions must be limited to those offenders who were eligible for a death sentence. The 230+ variable regression model includes irrelevant defendants who were indicted for murder but had no statutory aggravating factors. These defendants had the same probability of being sentenced to death as parking violators. Furthermore, it is noteworthy that Baldus et al. report several multiple regression and logistic regression models for actual jury decisions at penalty trial with nonsignificant racial coefficients. (for example, see Baldus et al. (1990), pp. 644-645).

    The Georgia Charging and Sentencing Study is missing important aggravating and evidentiary factors in the death penalty cases. These omissions serve to make death penalty and life sentence cases appear more factually similar, which impedes the ability of any multivariate methodology to differentiate the effect of the aggravating, mitigating, and evidentiary variables on the likelihood of being sentenced to death. This data collection problem could explain the fact that a large number of models have statistically significant race of victim coefficients, but with absurd coefficients for the nonracial variables (signs in the wrong direction, and magnitudes out of proportion to their expected effect).

    Furthermore, using the same weighted multiple regression methodology but with different combinations of predictor variables, I have produced thousands of regression equations, of different sizes, in which the racial variables are not statistically significant at the .05 significance levels. Some of these models exceed 250 variables. One example, an equation with 34 predictor variables, is displayed in Table IV. Several others can be found in Katz (1989). These models with nonsignificant racial coefficients do not prove that the Georgia capital sentencing system is free from racial bias. The models with statistically significant racial coefficients do not prove that the Georgia capital sentencing system is racially biased. This exercise merely confirms the fact that different combinations of legitimate variables in the regression or logistic regression models produce different estimates of the racial and nonracial coefficients.
 

TABLE IV THE COEFFICIENTS AND P-VALUES FOR THE VARIABLES FOR AN EXAMPLE OF A REGRESSION MODEL WITH RACIAL VARIABLES THAT ARE NOT STATISTICALLY SIGNIFICANT
DEPENDENT VARIABLE = DEATH SENTENCE GIVEN INDICTMENT FOR MURDER. NUMBER OF PREDICTOR VARIABLES: 34 R-SQUARE: .3337 ADJUSTED R-SQUARE: .3106 F VALUE: 14.435 (DF. 34, 980) INTERCEPT: -.049
LIST OF VARIABLES	PARTIAL REGRESSION COEFFICIENT	2 TAIL P-VALUE
RACIAL VARIABLES
1) DEFENDANT WAS BLACK	-.007	.7551
2) ONE OR MORE WHITE VICTIMS (RECODE)	-.003	.9012
AGGRAVATING VARIABLES
3) COUNT OF STAT. AGG. FACTORS, B7 EXPANDED	.051	.0001
4) MALICE OR VIOLENCE	-.009	.6480
5) NUMBER OF MINOR AGGRAVATING FACTORS	.024	.0102
6) VICTIM SHOWED MONEY	-.063	.2566
7) NO VICTIM PROVOCATION	.032	.1553
8) MALICE OR DELIBERATION	-.006	.6974
9) DEFENDANT EXPRESSED PLEASURE WITH KILLING	.323	.0001
10) GREAT RISK OF DEATH TO OTHERS IN PUBLIC	-.047	.0172
11) MURDER WAS OUTRAGEOUS, WANTON, HORRIBLE	-.064	.0052
12) VICTIM WAS ON DUTY POLICE OFFICER	.041	.4798
13) MULTIPLE SHOTS	.022	.2632
14) NO VICTIM MITIGATING CIRCUMSTANCE	.010	.5823
15) NON-PROPERTY RELATED CRIME	.119	.0001
16) NUMBER OF OTHER VIOLENT PERSONAL CRIME CONVICTIONS, SIMPLE ROBBERY/AGGRAVATED ASSAULT OR BATTERY/ PERSONAL VIOLENT MISD	-.048	.1758
17) NUMBER OF VIOLENT PERSONAL CRIME CONVICTIONS BESIDE HOMICIDE/ARMED ROBBERY/RAPE/KIDNAP	.052	.1320
18) DEFENDANT LAY IN WAIT TO AMBUSH VICTIM	-.011	.6640
19) NUMBER OF CONVICTIONS FOR VIOLENT PERSONAL CRIMES, BURGLARY OR 1ST DEGREE ARSON	.007	.6640
20) DEFENDANT NOT KILLER BUT DID VIOLENCE	-.063	.1791
21) DEFENDANT FIRED 5 OR MORE SHOTS	.031	.2347
22) VICTIM WEAK	.018	.6740
23) NUMBER OF PRIOR ARRESTS	-.001	.4592
MITIGATING VARIABLES
24) NUMBER OF MITIGATING FACTORS	.005	.4225
25) DEFENDANT PANIC DURING CRIME	.041	.2566
26) DEFENDANT SURRENDERED WITHIN 24 HOURS	.003	.8796
27) DEFENDANT ACCOMPLICE IN KILLING	-.052	.1329
28) HISTORY OF DRUG OR ALCOHOL ABUSE	-.025	.1014
29) REVENGE MOTIVE	-.005	.7687
30) HATE MOTIVE	-.019	.3428
31) DEFENDANT HAD NO SIGNIFICANT CRIMINAL REC.	.021	.2758
32) DEFENDANT HURT BY POLICE	.249	.0958
EVIDENTIARY VARIABLES
33) SCIENTIFIC EVIDENCE OTHER THAN MED./WEAPON	.055	.0007
34) EYE WITNESS ID OF DEFENDANT	-.010	.4630

    The thousands of regression models with statistically nonsignificant racial coefficients provide thousands of nonracial explanations for the death sentencing outcomes. None of these models truly reflect the operation of the Georgia charging and sentencing system. The regression models generated by Baldus et al. with statistically significant racial coefficients explains death sentencing outcomes, in part, through racial factors. None of these models truly reflect the operation of the Georgia charging and sentencing system.

    In fact, it is possible to specify numerous multiple regression models based upon legitimate aggravating, mitigating, and evidentiary factors, that perfectly explain the death sentencing outcomes (R-square = 1) and assign the racial variables coefficient values of 0. For each death penalty case, identify legitimate aggravating, mitigating, and evidentiary factors that, in combination, uniquely distinguish that death penalty case from all other cases (both death penalty and non-death penalty) in the database. Define the interaction variable for each death penalty case to be the joint occurrence of those factors that uniquely identify it. Consider the multiple regression model with dependent variable indicating the sentencing outcome (death penalty or non-death penalty) and with the previously defined interaction variables and two racial variables as predictor variables. Each interaction variable will receive a coefficient value of 1. The coefficients of the racial variables will be 0. The constant term will be 0. The R-square value will be 1. Effectively, this model shows that the interaction variables, which were defined as a unique combination of legitimate factors, perfectly explain the death sentencing outcomes without any racial effects. As innumerable interaction variables can be defined for the death penalty cases, there exist innumerable multiple regression models of this type.

    Several researchers (for example Gross (1985), Kaye (1986), and Kennedy (1988)) fault the courts for not accepting the multiple regression and logistic regression evidence offered by Professor Baldus as sufficient proof of racial discrimination in death sentencing. Others (for example Barnett (1985), Wilbanks (1987), and Edmunson (1990), have recognized limitations to the interpretation of the racial coefficients in regression models for capital sentencing decisions.

    The District Court rejected these models as relevant to the measurement of racial effects in the context of capital sentencing, noting that it (580 F.Supp at 370, 372) "can produce endless series of self-fulfilling prophecies because it always attempts to explain actual outcomes based on whatever variables it is given," and "the presence of multi-colinearity substantially diminishes the weight to be accorded to the circumstantial statistical evidence of racial disparity."

    The 11th Circuit Court of Appeals found that the general approach of attempting to control for all possible legitimate aggravating and mitigating factors to see if some racial bias was left over inappropriate for the capital sentencing decision (753 F.2d at 899), "From a legal perspective, petitioner would argue that since the difference is not explained by facts which the social scientist thinks satisfactory to explain the differences, there is a prima facie case that the difference was based on unconstitutional factors, and the burden would shift to the state to prove the difference in results from constitutional considerations. This approach ignores the realities. It not only ignores quantitative differences in cases: looks, age, personality, education, profession, job, clothes, demeanor, and remorse, just to name a few, but it is incapable of measuring qualitative differences of such things as aggravating and mitigating factors. There are, in fact, no exact duplicates in capital crimes and capital defendants."

THE AGGRAVATION AND MITIGATION INDEX

    The aggravation and mitigation index approach develops a formula based on aggravating, mitigating, and evidentiary factors to classify each case according to its severity. Baldus selected the 500 defendants which he believed were at the greatest risk of a death sentence because they scored the highest on his aggravation and mitigation index. The 500 defendants were then categorized into 8 distinct groups. Baldus assumed that cases within each group were comparable in level of aggravation and mitigation and that higher numbered groups represented more aggravated cases. Baldus provided the court with a breakdown of the death sentencing rates, by defendant victim racial combination, for these cases within each of the 8 levels. This data appears in Table V.
 

TABLE V DEATH SENTENCING RATES BY DEFENDANT VICTIM RACIAL COMBINATION FOR CASES CATEGORIZED BY BALDUS INDEX*
CASE SEVERITY LEVEL	OVERALL	BLACK KILLS WHITE	BLACK KILLS BLACK	WHITE KILLS WHITE	WHITE KILLS BLACK
1	0/33	0/9	0/19	0/5	0/0
	0%	0%	0%	0%	---
2	0/55	0/8	0/27	0/19	0/1
	0%	0%	0%	0%	0%
3	6/76	3/10	2/18	1/39	0/9
	8%	30%	11%	3%	0%
4	4/57	3/13	0/15	1/29	0/0
	7%	23%	0%	4%	---
5	15/58	9/26	2/12	4/20	0/0
	27%	35%	17%	20%	---
6	11/64	3/8	1/20	5/32	2/4
	17%	38%	5%	16%	50%
7	29/71	9/14	5/13	15/39	0/5
	41%	64%	39%	39%	0%
8	51/58	20/22	6/8	25/28	0/0
	88%	91%	75%	89%	---
* The denominator indicates the number of cases that were classified in the case severity level and the numerator counts the number of cases in which the defendant was sentenced to death.

    Based on this index, Baldus observed that the overall death sentencing rates tend to increase as one moves from level 1 to level 8. Furthermore, according to the index, the death sentencing rates for white victim cases are generally higher than the death sentencing rates for black victim cases in levels 3 through 7. Baldus testified that racial bias in sentencing is concentrated in cases at the midrange level of aggravation and mitigation (levels 3 through 7).

    Baldus did not explain to the court exactly which variables were utilized in this index or how the formula worked. The list of factors comprising this index can be found in Baldus et al. (1990), although the point values for the factors are omitted. Analysis of the Georgia Charging and Sentencing Study database has enabled me to identify the particular index (called DPMRIDX3) that reproduces the summary statistical results in Table V. Table VI lists the factors used to produce this index and their associated point values.
 

TABLE VI THE AGGRAVATION AND MITIGATION INDEX
AGGRAVATING FACTORS	POINTS
A1) MOTIVE TO AVENGE ROLE OF JUDICIAL OFFICER, DISTRICT ATTORNEY, OR POLICE OFFICER	48
A2) VICTIM PHYSICALLY TORTURED	27
A3) DEFENDANT EXPRESSED PLEASURE WITH KILLING	26
A4) KIDNAPPING INVOLVED	16
A5) VICTIM KILLED SO DEFENDANT COULD COLLECT INSURANCE MONEY	14
A6) DEFENDANT'S DEFENSE WAS INSANITY OR DELUSIONAL COMPULSION	12
A7) KILLING MOTIVATED BY RACE HATRED	8
A8) MENTAL TORTURE	3
A9) BRUTAL CLUBBING OR OTHER UNNECESSARILY PAINFUL METHOD OF ATTACK	3
A10) MUTILATION OF BODY DURING HOMICIDE	3
A11) MUTILATION OF BODY AFTER DEATH	3
A12) SLASHED THROAT	3
A13) EXECUTION STYLE HOMICIDE	3
A14) VICTIM PLEADED FOR LIFE	3
A15) SEXUAL PERVERSION OR ABUSE OTHER THAN RAPE	3
A16) DEFENDANT ON ESCAPE FROM LAWFUL CUSTODY OF POLICE OFFICER OR LAWFUL CONFINEMENT	3
A17) BRUTAL STOMPING OR BEATING WITH HANDS OR FEET	1
A18) MULTIPLE GUNSHOTS TO HEAD OF VICTIM	1
A19) MULTIPLE STABBING	1
A20) BLOODY	1
A21) VICTIM FORCED TO DISROBE	1
A22) HOMICIDE PLANNED FOR MORE THAN 5 MINUTES	1
A23) CONTEMPORANEOUS OFFENSE PLANNED FOR MORE THAN 5 MINUTES	1
A24) HOMICIDE WAS UNNECESSARY TO COMPLETE CRIME	1
A25) DEFENDANT COMMITTED ADDITIONAL CRIMES AFTER HOMICIDE	1
GEORGIA STATUTORY AGGRAVATING FACTORS
B1) PRIOR CONVICTION OF MURDER, ARMED ROBBERY, RAPE, OR KIDNAPING 3	3
B2) KILLING OCCURRED WHILE DEFENDANT ENGAGED IN ARMED ROBBERY, RAPE KIDNAPING, BURGLARY, ARSON IN 1ST DEGREE, OR AGGRAVATED BATTERY	3
B3) GREAT RISK OF DEATH TO OTHERS IN PUBLIC PLACE	3
B4) DEFENDANT MURDERED TO OBTAIN MONEY	3
B5) DEFENDANT KILLED JUDICIAL OFFICER OR DISTRICT ATTORNEY FOR PERFORMING HIS OFFICIAL DUTIES	3
B6) DEFENDANT INVOLVED IN MURDER FOR HIRE	3
B7) THE MURDER WAS OUTRAGEOUSLY OR WANTONLY VILE, HORRIBLE OR INHUMAN:
a) THE VICTIM WAS MUTILATED BEFORE OR AFTER DEATH	3
b) BRUTAL STOMPING OR BEATING WITH HANDS OR FEET, OR MULTIPLE GUNSHOT WOUNDS, OR MULTIPLE STABS, OR OTHER MODE OF MULTIPLE LETHAL OR PAINFUL ATTACK	3
c) RAPE OR KIDNAPING OF WEAK OR DEFENSELESS VICTIM	3
d) DURING THE COURSE OF AN ARMED ROBBERY, RAPE, OR KIDNAPING, THE VICTIM IS KILLED TO OBTAIN MONEY OR TO PREVENT THE VICTIM FROM TESTIFYING AS A WITNESS TO THE CRIME. IN ADDITION THE VICTIM IS EITHER WEAK, DEFENSELESS, KILLED EXECUTION STYLE OR PLED FOR HIS LIFE	3
e) VICTIM BRUTALLY CLUBBED, STOMPED ON, BEATEN WITH HANDS OR FEET STRANGLED, DROWNED, POISONED, OR SHOT AND SLASHED TO DEATH	3
f) SEXUAL ABUSE, SEXUAL PERVERSION, OR VICTIM PLED FOR LIFE	3
B8) VICTIM WAS ON DUTY POLICE OFFICER	3
B9) DEFENDANT ON ESCAPE FROM LAWFUL CUSTODY OR LAWFUL CONFINEMENT	3
B10) DEFENDANT MURDERED TO AVOID ARREST OR SILENCE WITNESS TO CRIME	3
MITIGATING FACTORS
M1) DEFENDANT HAD IMPAIRED MENTAL CAPACITY	-46
M2) COPERPETRATOR RECEIVED A LESSER SENTENCE	-8
M3) DEFENDANT NOT TRIGGERMAN	-7
M4) DEFENDANT ADMITTED GUILT ASSERTING NO DEFENSE	-7
EVIDENTIARY FACTORS
E1) WITNESS IDENTIFIED DEFENDANT NEAR SCENE OF CRIME	1
E2) PRETRIAL IDENTIFICATION OF DEFENDANT	1
E3) DEFENDANT ADMITTED HOMICIDE WITH NO DEFENSE	1
E4) COPERPETRATOR IMPLICATED DEFENDANT IN HOMICIDE	1
E5) WITNESS HEARD INCRIMINATING REMARK BY DEFENDANT OR COPERPETRATOR	1
E6) SCIENTIFIC EVIDENCE OTHER THAN WEAPON OR MEDICAL	1
E7) MURDER WEAPON FOUND, LINKED DEFENDANT TO HOMICIDE	1
E8) MEDICAL REPORT LINKED DEFENDANT TO HOMICIDE	1
E9) WITNESS WITH INFORMATION ON PREPARATION OR PRECIPITATING EVENTS	1

    To calculate the aggravation and mitigation index score for a defendant convicted of murder or voluntary manslaughter, identify the factors listed in Table VI which apply to the particular defendant and the circumstances surrounding his crime. Then, add up the points associated with each of these factors. The total number of points is the index score for that defendant.

    The higher the index score, the more aggravated the homicide, and factors that are not included in Table VI are assigned no points by default. Baldus used the index score to classify all cases into three levels of aggravation and mitigation, low (less than 17 points, levels 1 and 2 in Table V), midrange (17 points to under 44 points, levels 3 through 7 in Table V), and high (44 points or more, level 8 in Table V). Baldus observed that death sentences are rarely imposed on defendants with low index scores but quite often on defendants with high index scores, and concluded that race has little effect on sentencing at these levels. However, Baldus argued that race of victim discrimination was most pronounced against black defendants in the midrange of cases where he believes that the room for prosecutor and jury discretion is the greatest.

    To be considered for a death sentence, the prosecution must prove, beyond a reasonable doubt, that the defendant is also guilty of one or more of ten statutory aggravating circumstances. If the prosecution cannot prove the existence of one or more statutory aggravating factors, then the defendant is not eligible for a death sentence, and there is no capital decision for the prosecutor or jury to make. Furthermore, homicides with statutory aggravating factors should be excluded if there are other legal reasons which bar the imposition of the death sentence. Accordingly, tests for racial bias in capital sentencing decisions, must compare only those offenders who were eligible for a death sentence. The inclusion of non-death eligible offenders improperly bloats the denominators in death sentencing rate comparisons. Baldus did not establish that each of the 500 cases contained one or more statutory aggravating factors. Therefore, it is possible that a significant percentage of the cases in the midrange might not even be eligible for a death sentence under Georgia Law.

    There are many unusual features of the aggravation and mitigation index, itself. A defendant who merely expresses pleasure with the homicide is assessed 26 points, which is 26 times worse than beating or stomping the victim to death. A defendant who claims insanity or delusional compulsion as his defense immediately aggravates his crime by 12 points. The aggravating factors with the highest point values (revenge killing of a judicial officer, district attorney, or police officer (48 points), physically torturing a victim (27 points), expressing pleasure with the murder (26 points), kidnaping (16 points), killing to collect insurance (14 points, have the strongest influence on which defendants will fall in the high aggravation range. Other aggravating factors such as armed robbery, rape, burglary, arson, aggravated battery, multiple killings, serial murders, etc. are assigned little importance by the index.

    Even accepting the index at face value for the sake of argument, it is still impossible to classify each case properly because of the vast number of items that are unknown or incorrectly coded. For instance, it was unknown in 82% of the cases whether the defendant expressed pleasure with the homicide and it was unknown in 41% of the cases whether the homicide was motivated by race hatred. Baldus assumed for the purpose of the midrange analysis (and all multivariate analyses) that items whose occurrence was unknown to the coder, did not happen. McCleskey is a case which exemplifies the problem with this assumption.

    According to the Georgia Charging and Sentencing Study database, McCleskey was assessed 1 point because the murder scene was bloody (A20), 1 point because the armed robbery was planned for more than 5 minutes (A23), 3 points for his prior armed robbery convictions (B1), 3 points for killing his victim during an armed robbery (B2), 3 points for murdering to obtain money (B4), 3 points for murdering a police office (B8), 3 points for killing to resist arrest (B10) and 1 point each for the seven evidentiary factors E1, E2, E4, E5, E6, E8, and E9. Therefore, McCleskey's index score is 24 which places his case in the midrange level of aggravation and mitigation. However, McCleskey should also receive an additional 26 points for bragging about the murder of Officer Schlatt to cellmate Offie Evans (A3). McCleskey's questionnaire indicates that it was unknown to the coder whether McCleskey had expressed pleasure with the killing. Furthermore, McCleskey should be granted 3 more points for holding 9 customers and employees of the Dixie Furniture Store as virtual hostages before Officer Schlatt arrived, putting them at great risk of death (B3). McCleskey's questionnaire correctly codes this feature but one item was incorrectly entered into the computer database. Finally, take away 8 points from McCleskey's score because all three of his coperpetrators, who were in the back of the furniture store at the time Officer Schlatt was shot in the front of the store, were sentenced to either 20 years or life (M2). This item was incorrectly coded in the database due to an error in the computer program that created this variable. Therefore, McCleskey's final corrected score is 45 points which bumps him out of the midrange into the highest aggravation category.

    The data in McCleskey's own case proves that the data coders were unable to identify important aggravating factors which were known to the prosecution and the jury. Generally, the understatement of the incidence of aggravating factors in death penalty cases causes death penalty and non-death penalty cases to be erroneously categorized together in the "mid-aggravation range." Furthermore, since the vast majority of the death penalty cases are white victim cases, the "mid-aggravation range" would exhibit a higher death sentencing rate for white victim cases.

    The District Court observed (580 F.Supp at 371) "The whole study rests on the presumption that cases with similar aggravation indexes are similarly situated. This presumption is not only rebuttable, it is rebutted, if by nothing else, then by common sense." He went on to explain that "there is no logical basis for the assumption that cases with similar aggravation indices are at all alike. Further, the aggravation index for any given case is a function of the variables that are included in the model. Any change in the variables included in the model will also change the aggravation index of most, if not all, cases."

THE BARNETT STUDY

    Arnold Barnett (1985) reexamined the data on Georgia homicide cases from the Procedural Reform Study and proposed a more generalized crosstabulation method for assessing racial bias in capital sentencing. Professor Barnett defined three dimensions for classifying homicide cases; (1) the certainty the defendant is a deliberate killer (DELIBERATE), (2) the status of the victim (STATUS), and (3) the heinousness of the murder (HEINOUS), along with elaborate rules for scoring each case.

    The DELIBERATE dimension allows three values, 0, 1 or 2. Generally, a score of 0 indicates a weak case as to the certainty that the defendant was the killer or the deliberateness of the killing, 2 represents strong evidence of certainty and deliberateness, and a score of 1 is assigned to cases that fall in between. The STATUS dimension can take on 0 or 1, which corresponds roughly to victim is a stranger (STATUS = 1) and victim is a nonstranger (STATUS = 0). The HEINOUS dimension allows 0, 1 or 2. Generally, a score of 0 indicates the killing was in "self-defense", a score of 2 is consistent with a "vile" murder, and all other cases are assigned 1.

    Each case was assigned a three-tuple (i,j,k) indicating their respective scores on the DELIBERATE (I), STATUS (j), and HEINOUS (k) dimensions. Barnett defined an overall measure of severity, or category, by summing the three scores for each case and compared death sentencing rates within each category by race of the victim. This chart is reproduced as Table VII.
 

TABLE VII DEATH SENTENCING RATES BY RACE OF VICTIM AND CATEGORY**
CASE CATEGORY	BLACK VICTIM**	WHITE VICTIM**
CATEGORIES 0-2:	.01(184)	.01(189)
CATEGORY 3	.11(45)	.32(89)
CATEGORY 4:
(1,1,2)	.75(8)	.84(64)
(2,0,2)	.57(7)	.55(11)
CATEGORY 5:	1.00(1)	.86(7)
* This is CHART A from Barnett (1985), pp. 1350.
** The numbers in parentheses represent the number of cases considered in the category.

    Table VII shows roughly equal death sentencing rates by race of victim for all categories except category 3. Category 3 shows a 21 point higher death sentencing rate for killers of white victims, which is statistically significant. Barnett attempted to control cases in category 3 further by considering the prior record of each defendant, but this factor did not eliminate the significant disparity.

    Barnett classification scheme does not require that each case contain one or more statutory aggravating factors. The disparity in death sentencing rates by race of victim in category 3 could be a direct result of Barnett's rules, which do not break out felony and nonfelony homicides. Every case in category 3 was assigned the score (1,1,1). Barnett characterizes homicides with a score of 1 on the DELIBERATE dimension as "Most 'common' slayings, such as killing during armed robberies or during barroom fights." But killing during the course of an armed robbery is a statutory aggravating factor. Killing in a barroom fight is not. As shown in Table II, whites are more likely to be killed during armed robberies (33.3% to 7.4%) and blacks are more likely to be killed due to disputes or during barroom fights (64.0% to 35.3% for FAMILY, LOVER, LIQUOR, BARROOM FIGHT, and 72.7% to 44.2% for KILLING RESULTED FROM DISPUTE).

    According to Barnett, STATUS = 1 indicates, generally, that the killer and victim were strangers. Killing a stranger is not a statutory aggravating factor. Killing a stranger who is a police officer, fireman, or judicial officer in the performance of his/her duties is a statutory aggravating factor. Table II shows that a higher percentage of white victims are police officers than black victims (3.5% to 0.6%). Furthermore, police officers are more likely to be killed due to a contemporaneous armed robbery. The third dimension, HEINOUS = 1 is a wide ranging default score assigned to homicides which were not motivated by self-defense and which were not terribly vile or gruesome.

    It is important to note that Barnett's analysis was based upon the data and narrative summaries of homicides from the Procedural Reform Study. The Procedural Reform Study was the predecessor to the Georgia Charging and Sentencing Study and contains much less detail of the aggravating, mitigating, and evidentiary factors pertaining to each case. As with the crosstabulation method, Barnett's results could change if homicides are generally misclassified due to incorrect or missing information. Barnett acknowledges this potential threat to the analysis, and considers various aspects of this problems such as data collection errors. Ultimately, he discounts the importance of these errors to the final results, conceding that "Brief case descriptions can only crudely portray the credibility of the various witnesses, the eloquence of the opposing attorneys, and the emphasis placed in court on the various elements of the case." Barnett does not consider the possibility that the case summaries omit, as in McCleskey's case, important aggravating facts that were known to the prosecution and the jury.

7. CONCLUSION

    The Racial Justice Act begins with the premise that overall racial disparities in death sentencing is a basis to infer racial bias in the sentencing system. This view ignores the existence of different racial homicide patterns. As a whole, black victim homicides in Georgia are less aggravated and more mitigated than white victim homicides. Accordingly, a higher percentage of killers of whites are sentenced to death than killers of blacks. The racial death sentencing patterns found in Georgia are not unique to Georgia. Similar aggregate racial death sentencing patterns have been identified by Bowers and Pierce (1977) in Florida, Texas, and Ohio, by Gross and Mauro (1984) in Mississippi, Oklahoma, Illinois, Arkansas, and Virginia, by Nakell and Hardy (1987) in North Carolina, and by Jacoby and Paternoster (1982) in South Carolina.

    The Racial Justice Act would protect prosecutors and jurisdictions with equal racial death sentencing rates from charges of racial bias. But in Georgia, equal death sentencing rates by race of the victim is more suggestive of racial bias than higher death sentencing rates for white victim cases. Furthermore, to rectify the imbalance in death sentencing rates by race, Georgia prosecutors and juries would be required to consider race along with the aggravating, mitigating, and evidentiary facts of the case before deciding to seek or impose the death penalty.

    Baldus et al. (1990), Kennedy (1988), Gross (1985), and others have argued that the low death sentencing rate for black victim cases is due, in part, to a lower value that the state of Georgia places on a black life. Following this line of reasoning, the equalization of death sentencing rates by race in Georgia would require that more death sentences be imposed on black defendants who kill black victims. But black on black homicides are the most mitigated of the four defendant-victim racial combinations.

    The Racial Justice Act permits the prosecution to rebut a prima facie case of significant racial death sentencing disparities by explaining the disparities through pertinent nonracial factors. The crosstabulation and regression methods emphasize the fact that this is not very difficult to accomplish. There is never a shortage of nonracial reasons one can offer for explaining why a defendant, convicted of murder, was or was not sentenced to death.

    Both the Charging and Sentencing Study and the Barnett Study attempt to categorize defendants that are "similarly situated," based upon the aggravating, mitigating, and evidentiary factors in their case. Death sentencing disparities, by race, of "similarly situated" cases are interpreted as evidence of racial discrimination. A fair test of capital decisions made by Georgia prosecutors, juries, and trial judges, requires that the analysis be limited to "death eligible" defendants and all relevant information be considered that was known to the decision maker at the point in time that the capital decision was made. Yet every murder case is a unique combination of aggravating, mitigating, and evidentiary facts. The determination of these facts for a particular case is not always straightforward, as witnesses can give conflicting testimony at trial, or the prosecuting attorney and defense attorney can interpret the evidence differently. Cataloguing the facts of the case into simple categorical variables loses critical information that can mean the difference between life and death.

    Justice Powell wrote in McCleskey (481 U.S. 279) "Most importantly, each particular decision to impose the death penalty is made by a petit jury selected from a properly constructed venire. Each jury is unique in its composition, and the constitution requires that its decision rest on consideration of innumerable factors that vary according to the characteristics of the individual defendant and the facts of the particular capital offense." Later, Justice Powell noted "Numerous legitimate factors may influence the outcome of a trial and a defendant's ultimate sentence, even though they may be irrelevant to his actual guilt. If sufficient evidence to link a suspect to a crime cannot be found, he will not be charged. The capability of the law enforcement agency can vary widely. Also, the strengths of the available evidence remains a variable throughout the criminal justice process and may influence a prosecutor's decision to offer a plea bargain or go to trial. Witness availability, credibility, and memory also influence the result of prosecutions. Finally, sentencing in state courts is generally discretionary, so a defendant's ultimate sentence necessarily will vary according to the judgement of the sentencing authority. The foregoing factors necessarily exist in varying degree throughout our criminal justice system."

    At the same time that the District Court denied McCleskey's claim of racial bias in the Georgia sentencing system, the court threw out McCleskey's conviction and sentence on the grounds that one of the witnesses, Offie Evans (McCleskey's cellmate), was promised leniency for his testimony against McCleskey and this promise was not divulged to the jury. The 11th Circuit Court of Appeals disagreed and reversed the District Court on this issue. Baldus did not collect any data concerning promises to witnesses nor concerning most of the myriad technical issues which death-sentenced litigants regularly raise on appeal. Ironically, at the same time as McCleskey's lawyers argued that Baldus had controlled for every possible factor in his 230+ variable model that could conceivably cause a death sentence, they also argued that other factors were present that served to deny McCleskey a fair trial and could have drastically changed the outcome of his trial.

    The criminal justice system affords each capital defendant an opportunity to a fair trial with impartial jurors. Two-hundred years of careful jurisprudence has attempted to preserve and improve the integrity of the trial process. The appellate courts rely upon voluminous records to carefully consider every aspect of a death penalty case to determine if the sentence is appropriate and if the trial conforms to the highest standards of fairness and equity. General statistics on homicide cases based on limited and incomplete data is a poor substitute for the judicial process.

BIBLIOGRAPHY

Baldus, David C., and James W. L. Cole (1980). "Statistical Proof of Discrimination," McGraw-Hill Book Company, New York.

Baldus, David, Pulaski, Charles, and George Woodworth (1984). Comparative Review of Death Sentences: An Empirical Study of the Georgia Experience," Journal of Criminal Law and Criminology, Vol. 74, pp. 661-753.

Baldus, David C., Woodworth, George and Pulaski Charles, (1985). "Monitoring and Evaluating Contemporary Death Penalty Systems: Lessons From Georgia," U. C. Davis Law Review, Vol. 18, pp. 1327-1349.

Baldus, David C., Woodworth, George and Pulaski Charles, (1986). "Arbitrariness and Discrimination in The Administration of The Death Penalty: A Challenge To State Supreme Courts," Stetson Law Review, Vol. 15, pp. 133-158.

Baldus, David C., Woodworth, George and Pulaski Charles, (1990). Equal Justice And The Death Penalty A Legal and Empirical Analysis, Boston: Northeastern University Press.

Bowers, William J. and Glenn L. Pierce (1977). "Arbitrariness and Discrimination Under Post-Furman Capital Statutes," Crime and Delinquency, Vol. 26, pp. 563-635.

Furman v. Georgia, 408 U. S. 238 (1972).

GA. CODE ANN 27-2534-1 (1983).

Georgia Department of Human Resources (1986). "1976-85 Top Ten Causes Of Death By Sex-Race For State Of Georgia."

Gregg v. Georgia, 428 U. S. 153 (1976).

Gross, Samuel, and Mauro (1984). "Patterns of Death: An Analysis of Racial Disparities in Capital Sentencing and Homicide Victimization," Stanford Law Review, Vol. 37, pp. 27-153.

Gross, Samuel (1985). "Race and Death: The Judicial Evaluation of Evidence of Discrimination in Capital Sentencing," U. C. Davis Law Review, Vol 18., pp. 1275-1325.

Harvey, William B. (1986). "Homicide Among Young Black Adults: Life In The Subculture Of Exasperation," in Homicide Among Black Americans, edited by Darnell F. Hawkins, Lanham: University Press of America.

Hawkins, Darnell F. (1986). "Black And White Homicide Differentials: Alternatives To An Inadequate Theory," in Homicide Among Black Americans, edited by Darnell F. Hawkins, Lanham: University Press of America.

H.R. 4442, (1988). The Racial Justice Act of 1988, U. S. House of Representatives, Documents.

H.R. 5269, (1990). The Racial Justice Act, U. S. House of Representatives, Documents.

H.R. 5269 Amendment, (1990). Amendment to The Racial Justice Act offered by Mr. Hughes of New Jersey, U. S. House of Representatives, Documents.

H.R. 3371, (1991). Fairness in Death Sentencing Act, U. S. House of Representatives, Documents.

Imes, James Douglas (1972). "Homicide And Social Conditions In The City Of Atlanta 1958-1962," Master's Thesis, Georgia State University Library, Atlanta, Georgia.

Jacoby, Joseph E. and Raymond Paternoster (1982). "Sentencing Disparity and Jury Packing: Further Challenges To The Death Penalty," The Journal of Criminal Law & Criminology, Vol. 73, No. 1, pp. 379-387.

Kaye, D. H. (1986). "The Place of Statistics in Establishing Unconstitutional Acts of Discrimination," in Statistical Methods In Discrimination Litigation, edited by D. H. Kaye and Mikel Aickin, Marcel Dekker, New York, New York.

Katz, Joseph L. (1989). "Statement of Dr. Joseph L. Katz to the Senate Committee of the Judiciary Concerning the Relationship Between Race and the Death Penalty," October 2, 1989.

Kennedy, Randall L. (1988). "McCleskey v. Kemp: Race, Capital Punishment, and the Supreme Court," Harvard Law Review, pp. 1388-1443.

McCleskey v. Zant, 580 F. Supp. 338 (N.D. Ga. 1984).

McCleskey v. Kemp, 753 F.2d 877 (11th Circuit en banc, 1985).

McCleskey v. Kemp, 481 U.S. 279(1987).

Nakell, Barry and Kenneth A. Hardy. (1987). The Arbitrariness of the Death Penalty, Philadelphia: Temple University Press.

Pettigrew, Thomas F. and Spier, Rosalind B. (1962). "The Ecological Structure of Negro Homicide," American Journal of Sociology, Vol. 67, May, pp. 621-629

Rivers, David Eugene (1975). "Homicide In Atlanta Incidence, Patterns And Policy Consequences," Master's Thesis, Georgia State University Library, Atlanta, Georgia.

Staples, Robert (1986). "The Masculine Way Of Violence" in Homicide Among Black Americans, edited by Darnell F. Hawkins, Lanham: University Press of America.

Wilbanks, William. (1987). The Myth Of A Racist Criminal Justice System, Monterey: Brooks/Cole Publishing Company.

Wolfgang, Marvin E. and Ferracuti, Franco. (1967). The Subculture Of Violence. London: Travistock Publications.