Petitioner's Supplemental Memorandum of Law and Proposed Finding of Fact

    IN THE UNITED STATES DISTRICT COURT 

FOR THE NORTHERN DISTRICT OF GEORGIA 
ATLANTA DIVISION 

  

WARREN McCLESKEY, 

Petitioner, 

-against- CIVIL ACTION 
: : 2 NO. C8l-2434A 

WALTER D. ZANT, Superintendent, 
Georgia Diagnostic & Classification : 
Center, 

LX
] 

Respondent. 

  

PETITIONER'S SUPPLEMENTAL MEMORANDUM OF LAW 
  

ROBERT H. STROUP 
1515 Healey Building 
Atlanta, Georgia 30303 

JOHN CHARLES BOGER 
10. Columbus Circle 

New York, New York 10019 

TIMOTHY K. FORD 

600 Pioneer Building 
Seattle, Washington 94305 

ANTHONY G. AMSTERDAM 

P New York University Law School 

: 40 Washington Square South 
New York, New York 10012 

ATTORNEYS FOR PETITIONER 

 



  

TABLE OF CONTENTS 
  

I. Petitioner's Burden of Proof On His Claim Of Racial 

DISC IME nA Or vy vi ie aie aii lle te eee a aie a 2 

II. The Methods Employed By Petitioner To Meet His Burden 
OF Proof. vi viieE ay var see ed ae Tee a ei ei wi 

ITI. Petitioner's Proof Of Discrimination. + sv % vis. vw a ov. +18 

Conclusion. LJ ° Ld LJ LJ ® . ° LJ Ld LJ LJ LJ LJ LJ Ld ® © eo Ld LJ LJ 2d 

 



  

IN THE UNITED STATES DISTRICT COURT 
FOR THE NORTHERN DISTRICT OF GEORGIA 

ATLANTA DIVISION 

  

WARREN McCLESKEY, 

Petitioner, 

-against- CIVIL ACTION 
NO. C81-2434A 

WALTER D. ZANT, Superintendent, 

Georgia Diagnostic & Classification 

Center, 

Respondent. 

  

PETITIONER'S SUPPLEMENTAL MEMORANDUM OF LAW 
  

Petitioner Warren McCleskey ("petitioner") submits this 

supplemental memorandum of law at the invitation of the Court, 

following the hearing of October 17, 1983, to address certain 

questions arising from the evidence presented by the parties. 

This memorandum is designed, not as a comprehensive statement of 

petitioner's case, but rather to supplement petitioner's previous 

memorandum of September 26, 1983. That initial memorandum provided 

the Court with an overview of petitioner's case and addressed at 

length the constitutional foundations of petitioner's arbitrariness 

and racial discrimination claims. 

In this brief, petitioner will not retrace that ground; 

instead, having already demonstrated that proof of persistent and 

intentional disparities by race in the treatment of capital cases 

in Georgia would suffice to make out a violation of the Equal 

 



  

Protection Clause of the Fourteenth Amendment, requiring petitioner's 

death sentence to be vacated, petitioner will now turn to the question 

of how such disparate racial treatment must be proven. Specifically, 

petitioner will address: (i) the burden of proof petitioner must 

shoulder to establish his evidentiary claims; (ii) the methods of 

proof petitioner has adopted to meet this burden; and (iii) the 

facts petitioner has established, measured by the prevailing legal 

standards. 

I. 

Petitioner's Burden of Proof 

On His Claim of Racial Discrimination 
  

Petitioner has shown in his initial brief (see Pet. Mem., 

86-92)1" that intentional discrimination sufficient to establish an 

Equal Protection Clause violation under the Fourteenth Amendment 

can be proven by statistical evidence alone: "In some instances, 

circumstantial or statistical evidence of racially disproportionate 

impact may be so strong that the results permit no other inference 

but that they are the product of a racially discriminatory intent 

or: purpose.” Smith v. Ballicom, 671 F.2d 3858, 859 (5th Cir. Unit. B 
  

1982) (on rehearing); accord, Spencer v. Zant, 715 F.2d 1562, 1581 
  

{11th Cir. 1083); cf. Adams v. Wainwright, 709 F.24 1443, 1449 
  

(11th Cir. 1983).%/ 

  

1l/ Each reference to Petitioner's Post-Hearing Memorandum of Law, 
dated September 26, 1983, will be indicated by the abbreviation 
"Pet. Mem." 

2/ By denying as irrelevant petitioner's prehearing request for 
discovery on other actions that might demonstrate a pattern of 

feont'd.] 

 



  

In Castaneda v. Partida, 430 U.S. 482 (1977), the Supreme 
  

Court has held that to make out a prima facie statistical case, at 

least "in the context of grand jury selection," requires a 

petitioner to establish that 

"the [racial] group is one that is a recognizable, 
distinct class . . . Next the degree of underrepre- 
sentation must be proved, by comparing the propor- 
tion of the group in the total population to the 
proportion called to serve as grand jurors 

Finally, a selection procedure that is susceptible 

of abuse or is not racially neutral [must be shown 
which] supports the presumption of discrimination 
raised by the statistical showing." 

Castaneda v. Partida, supra, 430 U.S. at 494. At that point, the 
  

Court continued, petitioner "has made out a prima facie case of dis- 

criminatory purpose, and the burden then shifts to the State to 

rebut that case.”  1d., at 495. The Eleventh Circuit recently 

adopted a virtually identical procedure in analyzing a Poaroenioh 

Amendment . equal protection claim stemming from the detainment of 

Haitian immigrants: 

Although the standard of proof in Title VII cases 
differs from that in constitutional equal protection 
cases, the framework for proving a case, i.e. prima 
facie case, rebuttal, ultimate proof, is the same. 
See, e.g., Castaneda v. Partida, 430 U.S. at 495-96, 

. . . Because of the similar framework, and because 
there are few equal protection cases relying on 
statistics, when appropriate we draw upon Title VII 

cases.” 

  

  

2/ coni-'ad. 

racial discrimination in the criminal justice system in Fulton 

County and the State of Georgia, the Court necessarily limited 

petitioner's proof to statistical evidence, supplemented by 

reported decisions evidencing racial discrimination of which the 

Court might take judicial notice. (Pet. Mem. 101-02; see also 
  

Petitioner's First Interrogatories to Respondent, dated April 

13, 108%, 99 9-18; Order of June 3, 198%, at 2.) 

 



  

Jean v. Nelson, 711 F.2d 1455, 1486 n.30 (llth Cir.), vacated and 
  

  

pending on reh'g en banc, 714 F.2d 96 (llth Cir. 1983); cf. 
  

Eastland v. Tennessee Valley Authority, 704 F.2d 613, 618 (llth 
  

Cir. 1983). 

A proper analysis therefore requires, first, the 

determination of whether petitioner has established a prima facie 

case; second, the examination of respondent's rebuttal case, if 

any; and third, an assessment of whether, in light of petitioner's 

responsive evidence, he has ultimately met his burden of proof. To 

prevail, petitioner must demonstrate an Equal Protection violation 

"by a preponderance of the evidence." Jean v. Nelson, supra, 711 
  

F. 2d at 1487, citing Texas Dep't. of Community Affairg v. Burdine, 
  

450 U.S. 248, 252-53 (1981). 

Since petitioner asserts systemwide discrimination, the 
  

principal focus of analysis should be, not upon the evidence of 

discrimination in petitioner's individual case, as would be appro- 

. priate in analogous individual Title VII cases, see McDonnell 
  

Douglas Corp. v. Green, 411 U.S. 792 (1973); Mt. Healthy Board of 
  

  

Education v. Doyle, 429 U.8. 274 (1977), but rather upon systemwide 
  

(or perhaps judicial circuitwide, see Pet. Mem. 104-09) statistical 

evidence of disparities, as in analogous jury cases and other cases 

alleging classwide or systemwide discrimination. See, e.g., 
  

Castaneda v. Partida, supra; see also Washington v. Davis, 426 
  
  

U.S. 229, 241-42 (1976); Arlington Heights v. Metropolitan Housing 
  

Authority, 429 U.S. 252, 266 (1977); Hazelwood School District v. 
  

  

United States, 433 U.S. 299, 307-08 (1977). 
  

The precise evidentiary burden necessary to establish a 

prima facie capital sentencing case has never been definitively 

il a 

 



  

established. In Smith v. Balkcom, on rehearing, the former Fifth 
  

Circuit strongly suggested by negative implication what might 

suffice to establish a prima facie case: 

"No data is offered as to whether or not charges 

, or indictments grew out of reported incidents or 

as to whether charges were for murder under aggra- 

. vating circumstances, murder in which no aggra- 

vating circumstances were alleged, voluntary man- 

slaughter, involuntary manslaughter, or other 

offenses. The data are not refined to select 

incidents in which mitigating circumstances were 

advanced or found or those cases in which evidence 

of aggravating circumstances was sufficient to 
warrant submission of the death penalty vel non 

to a jury. No incidents resulting in not t guilty 

verdicts were removed from the data. The unsupported 

assumption is that all such variables were equally 

distributed « . + 

  

smith v. Balkcom, supra, 671 F.2d at 860 n.33. On the other hand, 
  

  

the Eleventh Circuit's per curiam opinion in Adams v. Wainwright, 

supra, 709 F.2d at 1449, contained di€ta that "[olnly if the evidence 

of disparate impact is so strong that the only permissible inference 

is one of intentional discrimination will it alone suffice." More 

recently, the Eleventh Circuit in Spencer v. Zant, 715 F.2d. at 1582 
  

  

n.1l5, drawing directly from Arlington Heights, supra, 429 U.S. at 

266, suggested that the proper standard may require evidence of 

3/ 
"'a clear pattern, unexplainable on grounds other than race." 

  

2/ Petitioner contends that determination of whether the proper 

standard should be drawn from Smith, from Adams, from Spencer or 

from some other case need not be Fesolved here, since petitioner's 

statistical evidence accounts for every plausible rival hypothesis, 

thereby meeting or exceeding even the most stringent possible 

standard. See discussion at pp. 16-23, infra. 

 



  

Once petitioner has shown a prima facie case, the 

burden then shifts to the State to rebut the case in one of 

three ways: (i) "by showing that plaintiff's statistics are 

misleading.:;;[ii] by presenting legitimate non-discriminatory 

reasons for the disparity," Eastland v. TVA, supra, 704 F.2d at 
  

618-19; or (iii) by proving that the discrimination is justified 

by a compelling state interest (see Pet. Mem. 77-78, 115-23). A 

rebuttal case challenging a party's dots bass as misleading or 

inaccurate cannot succeed without strong evidence that the data 

are seriously deficient and unreliable: 

"[A] heavy burden must be met before a party can 

justify the rejection in toto of any statistical 

analyses on grounds of errors or omissions in the 

data . . . the challenging party bears the burden 

of showing that errors or omissions bias the data 

fand]l. . . . that this bias alters the result of the 

statistical analyses in a systematic way." 

Vuyanich v. Republic Nat'l Bank of Dallas, 505 F. Supp. 224, 255-56 
  

  

(N.D. Texas 1980); accord, Trout v. Lehman, 702 F.2d 1094, 1101 

(D.C. Cir. 1983); Detroit Police Officer's Ass'n v. Young, 608 F.2d 
  

  

671, 687 (6th Cir. 1979), cert. denied, 452 U.S, 938 (1l98l)(gee 

generally, Pet. Mem., 115-18). 
  

A rebuttal case predicated upon "legitimate non-discrim- 

inatory reasons for the disparity" cannot succeed merely by challeng- 

ing petitioner's prima facie case "in general terms,” Wade v. 

Mississippi Cooperative Extension Service, 528 F.2d 3508, 517 
  

(5th Cir. 1976). "[Ulnquantified, speculative, and theoretical 

objections to the proffered statistics are properly given little 

weight by the trial court," Trout v. Lehman, supra, 702 F.2d at 
  

    

1102; see, e.g., Castaneda v. Partida, supra, 430 U.S. at 499 n.l1l9; 

 



  

Jean v. Nelson, supra, 711 F.2d at 721, 730. Addressing this theme, 
  

Chief Judge Godbold recently noted in Eastland v. TVA, supra, 704 
  

F.2d at 622-23 n.1l4, citing D. BALDUS & J. COLE, STATISTICAL PROOF 

OF DISCRIMINATION §8.23 at 74 (1980): 

"A defendant's claim that the plaintiff's 
model is inadequate because a variable has 
been omitted will ordinarily ride on evidence 
[from the defendant] showing that (a) the 
qualification represented by the variable was 
in fact considered [by the defendant], and (b) 
that the inclusion of the variable changes the 
results of the regression so that it no longer 
supports the plaintiff. Both of these facts are 
established most clearly and directly if the defend- 
ant offers an alternative regression model similar 
to the plaintiff's except for the addition of the 

variable in question." 

Finally, while a rebuttal case might theoretically 

be made in support of racially discriminatory treatment in some 

limited area of the law, the Supreme Court in Furman v. Georgia, 
  

408 U.S. 238 (1972) made it perfectly clear that no purported state 

interest could ever justify discriminatory imposition of the death 

penalty. (The State in this case has never suggested that any valid 

State policy could be furthered by such discrimination, and there- 

fore this possible line of rebuttal need not detain the Court. (See 

Pet. Mem., 77-81A)). 

Petitioner should prevail under the analysis outlined 

above if his prima facie case -- discounted by any valid criticisms 

adequately proven by the State's rebuttal case, augmented by any 

surrebuttal evidence petitioner can muster to counter the State's 

rebuttal case -- establishes discrimination by a preponderance of 

the evidence. Petitioner need not produce statistical evidence 

 



  

which would fully explain the workings of the system so long as 

he can demonstrate that racial discrimination is a real and per- 

sistent characteristic of that system. 

IL. 

The Methods Employed By Petitioner To 
— Meet His Burden Of Proof 
  

Petitioner McCleskey employed well-accepted and rigorously 

controlled statistical methods in support of his constitutional 

claims of discrimination in capital sentencing. He first established 

through the comparison of unadjusted racial comparisons that sig- 

nificant race-of-defendant and race-of-victim disparities are 

characteristic of Georgia's capital sentencing system. (DB 62; DB 69; 

DB 70). Although such "unadjusted" racial disparities have been 

held legally insufficient to establish a constitutional violation in 

the context of capital sentencing systems, see, e.g., Spinkellink 
    

v. Wainwright, 578 F.2d 582 (5th Cir. 1978); Smith v. Balkcom, 680 
  

  

F.2d 573 (5th Cir. Unit B 1981), it is instructive to note that 

statistical evidence no more sophisticated than this has regularly 

been deemed sufficient to require reversal in other equal protection 

contexts such as jury cases, see, e.g., Castaneda v. Partida, supra, 
    

(statistically significant racial disparities, with no additional 

variables held constant); and employment discrimination cases, see 

e.g., Fisher v. Proctor & Gamble Mfg. Co., 613 F.2d 527, 544 (5th 
    

Cir. 1980), cert. denied, 449 U.S. 111% (1981)(a prima facie case 
  

established by "glaring" statistical disparities even without 

controlling for job qualifications). 

 



  

Petitioner obviously did not rest, however, with an 

identification of these unadjusted racial disparities. Instead, 

Professor David Baldus, petitioner's principal expert, testified 

that he drew on his own expert knowledge of the criminal justice 

. system, as well as the experience and knowledge of his professional 

colleagues, supplemented by extensive reading and review, to develop 

an extensive list of variables that might plausibly affect the 

sentencing outcome in a capital case. This list was incorporated 

into a questionnaire, completed for each case included in PRS and 

CSS studies, which contained over five hundred variables. For 

purposes of analysis, Professor Baldus employed more than 230 of 

these "recoded" variables that he judged to be plausible factors 

in conducting his sentencing analyses. (Professor Baldus specifically 

testified that he employed every relevant variable on which he could 

obtain information.) 

To proceed beyond unadjusted analysis, petitioner analyzed 

the effect on sentencing outcomes of the racial factors while "con- 

trolling" for, or holding constant, the effects of the other plausible 

explanatory variables. Professor Baldus and Professor George Woodworth 

both testified that, in conducting these analyses, they relied upon 

the two accepted statistical methods available to achieve such control: 

cross-tabulations, and multiple regression analysis. Cross-tabular 

analysis, Professor Baldus explained, proceeds by dividing cases 

into successively smaller subcategories, each distinguished by the 

presence or absence of a series of relevant variables. Cross-tabular 

analysis permits one to compare cases that are comparable or similar 

on all of the variables examined, observing changes in the variable of 

interest. (Professor Baldus reported upon the results of a number of 

10 

 



  

cross-tabular analyses he performed in which the racial effects 

remained influential (see DB 67; DB 76 ). The inherent limitation 

of cross-tabular analysis, Baldus and Woodworth explained, is that 

it cannot meaningfully account for a very large number of variables 

simultaneously, since at some point the number of cases possessing 

similar characteristics on each of the increasing number of variables 

becomes very small, and "cell sizes" decrease toward statistical 

and practical insignificance. 

Multiple regression analysis, Professors Baldus and 

Woodworth testified, avoids this inherent limitation of cross-tabular 

analysis by employing algebraic formulae to calculate the additional 

impact of the presence or absence of a variable of interest (e.g., 

the race of the victim) over and above the collective impacts of 

a host of other variables. Professor Woodworth explained that re- 

gression accomplishes this result, not by examining cases that are 

similar on all variables other than the variable of interest, but 

instead by assigning cases an index value along a scale determined 

by the presence or absence of other variables, and then calculating 

the comparative sentencing rates at each level. (See GW 9; GW 10). 

This use of regression analysis, Professor Woodworth testified 

(without contradiction from State's expert Dr. Joseph Katz), is 

mathematically sound and fully accepted as a valid means of statis- 

tical measurement. The algebraic formula for calculating a sample 

regression analysis with three variables was presented to the 

Court as GW 13 and GW 14. 

-—JO = 

 



  

The Fifth Circuit first adverted to the use of regression 

analysis in 1976, calling it "a sophisticated and difficult method 

of proof in an employment discrimination case," Wade v. Mississippi. 
  

Cooperative Extension Service, 528 F.2d 508, 517 (5th Cir. 1976). 
  

Five years later, the Court, having gained greater familiarity 

with the EY, observed that "[m]ultiple regression analysis is 

a relatively sophisticated means of determining the effects that 

any number of different factors have on a particular factor," 

Wilkins v. University of Houston, 654 F.2d 388, 402-03 (5th Cir. 
  

1981). The Court held in Wilkins that "if properly used, multiple 

regression analysis is a relatively reliable and accurate method of 

gauging classwide discrimination," id. at 402-03 n.18, indeed noting 

that "it may be the best, if not the only, means of proving classwide 

discrimination . . . in a case where a number of factors operate 

simultaneously to influence" the outcome of interest. Id. at 403. 

With proper attention to its possible misuse, the Eleventh Circuit 

has also embraced multiple regression analysis as an appropriate 

tool for the proof of discrimination claims. See, e.g., Eastland 
    

v. TVA, supra, 704 F.2d..at 621-22; Jean v. Nelson, supra; see also, 
  

  
    

Valentino v. United States Postal Service, 674 F.2d 56, 70 (D.C. 
  

Cir. 1982); see generally, Finkelstein, "The Judicial Reception of 
  

Multiple Regression Studies in Race and Sex Discrimination Cases," 

80 COLUM. L. REV. 737 (1980). 

Perhaps the most extensive judicial discussion of the 

nature and role of multiple regression analysis in the proof of 

discrimination claims is Judge Higgenbotham's influential and 

widely cited opinion in Vuyanich v. Republic Nat'l Bank of Dallas, 
  

i bf 

 



  

505 F. Supp. 224, 261-79 (N.D. Tex. 1980). Judge Higgenbotham 

observes that multiple regression techniques have been "long used 

by social scientists and more recently [have been] used in judicial 

resolution of antitrust, securities, and employment discrimination 

disputes," Vuyanich, supra, 505 F. Supp. at 261. He notes that these 
  

"mathematical models are designed to determine if there is any differ- 

ential treatment not entirely attributable to legitimate differences," 

id. at 265, calling them "an important addition to the judicial 

toolkit,” id. at 267, 

Drawing upon basic texts in econometrics nd regression 

analysis (including D. BALDUS & J. COLE, STATISTICAL PROOF OF 

DISCRIMINATION (1980)), Judge Higgenbotham then embarks upon an 

extensive mathematical and statistical discussion of regression 

methods, including the derivation of the basic regression formulae, 

id. 269-71, the calculation of the statistical significance of 

regression coefficients, id., 271-73, improper applications of 

regression methods, id., 273-75, and different methods of employing 

regression analysis to measure possible discriminatory behavior, id., 

275-79. 

The discussion in Vuyanich coincides with and confirms 
  

the teachings of Professor Franklin Fisher in his influential article 

"Multiple Regression in Legal Proceedings," 80 COLUM. L. REV. 702 

(1980). Both make clear that multiple regression analysis "is . . 

a substitute for controlled experimentation," Vuyanich, supra, 
  

505 F. Supp. at 269, and that "[t]he results of multiple regressions 

-- such as what we will call 'coefficients' in the ordinary least 

square methodology -- can be read as showing the effect of each 

independent variable on the dependent variable, holding the other 

“Yo 

 



  

independent variables constant. Moreover, relying on statistical 

inference, one can make statements about the probability that the 

effects described are due only to a chance fluctuation," id., at 

269; accord, Fisher, supra, 80 COLUM. L. REV. at 706. Chief Judge 
  

Godbold explicitly recognized the value of regression analysis in 

Eastland v. TVA, supra, 704 F.2d at 621, finding that "[m]ultiple 
  

regression analysis is a quantitative method of estimating the 

effects of different variables on some variable of interest.” 

These clear precedents establish that the multiple 

regression method has been judicially accepted as a principal 

analytic tool -- indeed, in cases involving a large number of 

simultaneously operative variables, perhaps "the only means of 

proving classwide discrimination,” Wilkins v. University of Houston, 
  

supra, 654 F.2d at 463. 

In evaluating regression analyses, the courts and commenta- 

tors have pointed to a number of problems that could arise if sufficient 

care is not taken in analysis. If data are totally inaccurate or 

are shown to be systematically biased for the variable of interest, 

the analysis may be flawed. Vuyanich v. Republic Nat'l Bank of 
  

Dallas, supra, 505 F. Supp. at 255-58, 273. Further, if a "model," 
  

or group of independent variables is employed that omits '"some 

relevant explantory variable . . . the regression coefficient . . . 

would be 'biased' . . . and the usual tests of significance 

concerning the included regression coefficient . . . will be invalid,” 

id. at 274. Fortunately, as Judge Higgenbotham notes, "[clertain 

statistical tests are available to suggestwhether this sin of 

onission has occurred.” 1d. 

- 1% = 

 



  

At the other extreme, "when one or more irrelevant 

variables are included in the model . . . [a] risk of 'multicolinearity'" 
  

arises. Id. Yet the effect of possible multicolinearity is not 

to increase but to deflate evidence of possible discriminatory 

impact, id. at 274-75: thus "if multicolinearity exists, the prob- 

ability will be increased that the net impact of [racial factors] 

wiv -Will De judged statistically nonsignificant, even in cases 

where there are actual differences in the treatment." Id. In short, 

multicolinear models may underestimate, but do not overestimate, 

the extent of possible discrimination. 

A third possible problem can arise "where ths analyst 

chooses to use a regression equation that is linear in the explanatory 

variables when the true. regression model is nonlinear,” id. at 275. 

Obviously, the means by which to avoid such a problem is to conduct 

Oh employing both linear and nonlinear (such as logistic) 

regressions. 

Finally, least squares. regression depends upon the 

assumption that the "error term," -- the "u" in a regression formula 

which stands for idiosyncratic or "random influences" that characterize 

virtually every social scientific model, id. at 269-70, 273 -- 

"follows the 'normal distribution, '" id. at 275, that is, displays. 

no systematic relation to other independent variables. However, 

Judge Higgenbotham observed that "[wlith respect to this assumption, 

basic least squares regression models are "quite" robust" in that 

they will tolerate substantial deviations without affecting the 

validity of the results.' D. Baldus & J. Cole, supra, n.55 §8A.41, 

at 284." Id. at 275. Moreover, he noted, "[n]lonnormality of errors 

can be detected through the use of [statistical] . . . techniques." 1d. 

EY 

 



  

Petitioner's experts testified without contradiction 

that they had carefully followed all of the requisite steps in 

conducting regression analysis, and that they had taken particular 

care to conduct statistical diagnostic tests to determine whether 

any of the assumptions of regression analyses had been violated 

in petitioner's analyses, and whether the results could possibly 

be biased thereby. Professor Woodworth offered his expert statis- 

tical opinion, without any contradiction by the State, that the methods 

employed by petitioner were appropriate, that models were not 

misspecified, and that no bias could be discerned in the reported 

results. Professor Berk, petitioner's reubttal expert, confirmed 

Professor Woodworth's expert opinion. He explicitly complimented 

petitioner's conduct of regression analysis as "state-of-the-art," 

and found both of petitioner's studies to be of "high credibility." 

In sum, the statistical methods employed by petitioner, 

including cross-tabular and regression analysis, have been expressly 

adopted by the Fifth and Eleventh Circuits as appropriate tools 

for the measurement of the possible effect of racial variables. 

The regression analyses relied upon by petitioner were properly 

conducted by leading experts in the field, were carefully monitored 

for possible statistical problems, and have been found to be both 

statistically appropriate and accurate in their assessment of the 

presence and magnitude of racial disparities in capital sentencing 

in Georgia. Methodological concerns, whether based in law or in 

statistics, thus pose no impediment to the Court's evaluation of 

petitioner's reported results. 

RL ah 

 



  

TIT. 

Petitioner's Proof of Discrimination 
  

To meet his prima facie burden of proof, petitioner has 

offered the Court a wide range of statistical analyses, virtually 

-all of which demonstrate or, at a minimum suggest, significant 

race-of-victim effects, as well as significant race-of-defendant 

effects within important subcategories. Petitioner reboried strong 

unadjusted racial disparities (see Pet. Mem., 24-25). He then con- 

structed a model which would take into account the statutory factors 

identified by the Georgia legislature as sufficiently important 

aggravating circumstances to permit the imposition of a death sentence, 

together with the "nonstatutory" aggravating circumstance of prior 

record (also expressly designated as relevant by Georgia statute). 

Professor Baldus reported the results of this analysis employing 

both a least squares®'analysis, which assumes a linear distribution 

of cases, and a logistic analysis, which depends upon no such 

assumptions. The results, as indicated below, demonstrate that the 

race-of-victim factor wields an independent effect on sentencing 

outcome at a highly significant level: 

  

  

  

  

w.L.S. Logistic Regression 

Regression Results Results 

Regression Coefficient & Regression Death Odds 

ae egal of Statistical Coefficient Multiplier 

Significance 

Race of Victim «07 iE 2.8 

{.0014) : 

Race of Defendant .04 .02 1.0 

(.09) {.93) 

(DB 78) 

 



  

Under this analysis, race of the victim is at least as 

important a determinant of sentence as such factors as that the 

defendant had a prior capital record, that the murder was vile, 

horrible or inhuman, that the victim was a policeman, or other 

serious aggravating factors. When Professor Baldus refined this 

model to incoprorate not only statutory aggravating factors, but 

75 mitigating factors as well, the relative impact of the race-of- 

victim variable actually increased: 

W.L.S. Logistic Regression 

  
  

  

  

Regression Results : Results 
Regression Coefficient & Regression Death Odds 
Level of Statistical Coefficient Multiplier 

Significance 

Race of Victim .10 : 2.1 8.2 
(.001) (.001) 

Race of Defendant | O7 v 26 1.4 

{.01) (ns) 

Professor Baldus thereafter employed a wide range of models 

(see,e.g., DB 80, DB 83, DB 96, DB 98) to see whether any constellation 

of variables would eliminate or substantially diminish the race-of- 

victim effect. None did. In effect, petitioner thereby "anticipated 

and adequately met the government's statistical challenge. Plaintiffs 

offered a variety of statistical and testimonial evidence to demonstrate 

that [other independent variables] . . . were irrelevant," Jean v. 

Nelson, supra, 711 F.2d at 1498, in explaining the persistence of 
  

the racial variables. 

-37 

 



  

Professor Baldus, as noted, conducted a number of 

analyses employing the 230+ variable model which included all 

known variables which plausibly might have affected sentencing 

outcome, and the racial factors remained significant (see, e.g., 
  

DB 80 "Race of Victim . . . After Simultaneous Control for 230+ 

Non-Racial Variables . . . .06(.01); Race of Defendant . . . .06 

(.01)). In one sense, this model operates most "realistically" 

since it includes and controls for the effects, however small, of 

any aggravating or mitigating factors that might affect a prosecutor's 

or jury's judgment. Yet, because of problems of multicolinearity, 

‘which as explained above can actually dampen or suppress the real 

impact of other independent variables, Professor Baldus, Professor 

Woodworth and Professor Burford, the State's expert, all indicated 

that the 39 variable or "mid-range" model probably provided the 

best statistical evaluation of the independent impact and significance 

of the racial variables. 

Using that mid-range model (as well as models with seven 

variables, eleven variables, all statutory aggravating circumstances, 

and all statutory plus 73 mitigating circumstances), Professor 

Woodworth conducted a comprehensive series of diagnostic tests to 

see whether problems in the weighting procedure employed, the selection 

of least squares or logistic regression, the existence of some 

"missing" data, the influence of the 48 most important cases, or the 

presence of possible "interaction" effects among the variables 

included might explain the racial disparities reported. Professor 

Woodworth's conclusion, amply supported by Table 1 from GW 4, is 

that the race-of-victim coefficient remains large (from .041 to 

Wty i 

 



  

.117) and statistically significant throughout the diagnostic 

analyses. The race of the defendant exhibited an unstable, 

although often important effect as well. In sum, the persistent 

racial effects reported in petitioner's regressions are not statis- 

tical artifacts, but reflect real-world disparities in capital 

sentencing treatment based upon racial factors. 

Professor Woodworth also explained that the npn 

calculations reported in his diagnostics did not mean that peti- 

tioner's models were inadequate or incapable of accurately measuring 

the racial effects. First, Professor Woodworth noted that, insofar 

as Georgia's capital sentencing system is in fact operating in an 

arbitrary and capricious pattern, no statistical model can explain 

all of the variance, since a part of it will necessarily be random 

and idiosyncratic. Secondly, Professor Woodworth stressed that 

large np or "u" terms do not affect the accuracy of the measurement 

of the effect of other variables, concurring with Professor Fisher's 

analogous observation that 

#iilt is very important . . . to realize that a 
large standard error of estimate does not tell 
one anything at all about the accuracy with 
which the effects of the independent variables 
are measured . . . The standard error of estimate 
is a way of assessing how important the random part 
of the model is; it does not tell one how large the 
affects of such randomness are on one's ability to 
measure the systematic part." 

Fisher, supra, 80 COLUM. L. REV. at 719. 
  

216. 

 



  

Both Professor Baldus and Professor Woodworth agreed 

that the figures which most accurately and completely summarized 

the racial effects they had observed were reflected in GW 5 and GW 6. 

Those figures, based upon the mid-range model with interactions 

and nonlinearities accounted for, show a disparity in the treatment 

of homicide cases by race-of-victim and race-of-defendant. which 

varies in magnitude depending upon the level of aggravation, or 

seriousness, of the homicide. Among the least aggravated cases, 

little racial disparity exists, because virtually 4 death sentences 

are imposed in any cases. Among the most aggravated cases, once 

again there exists little racial disparity, since nearly all of the 

cases receive a death sentence. Among the moderately aggravated 

‘cases, however, substantial and unchecked racial disparities exist. 

At petitioner McCleskey's level of aggravation, for example, the 

sentencing disparity Rowen white victim and black victim cases 

is .22 points. 

(See next page for GW 6,Table 2) 

Wt Ty Pr, 

 



GWo 

  

Figure 2: midrange?! Model With Interactions and Nonlinearities-- 

g Black Defendants 
  

  

    

  

  

100 T 

7% + 

£21 

25 + 

BO-~lsfonnig ; Mel eskey 2 . 

0 aa 8 1.0 1.2 

LEVEL OF AGGRAVATION 

2/ The curves reoresent 95% confidence bounds on the average death 

sentencing rate at increasing levels of aggravation (redrawn from 

computer output). 

 



  

Petitioner has set forth in his principal brief the parallel 

findings he obtained from both a statistical and a qualitative analysis 

of data from Fulton County, where petitioner was tried. Although 

the smaller sample size restricted the statistical significance 

of the results, the same pattern of influence of racial variables 

can clearly be ascertained. (See Pet. Mem., 36-40; DB 106-116). 

Beyond this statistical evidence in support of his prima 

facie case, petitioner introduced the deposition of District Attorney 

Lewis Slayton. That testimony, summarized in petitioner's principal 

brief at page 48, reveals a system for the processing of capital 

indictments in Fulton County EAR is decentralized among a dozen 

or more assistants, carried out with no written procedures or 

guidelines, and no central review of all decisions in homicide 

cases. Therefore, petitioner has shown a circuitwide system which 

affords an "opportunity for discrimination," since it leaves 

processing decisions up to a multitude of decisionmakers whose 

decisions are not routinely reviewed by a central authority for 

compliance with any objective criteria. 

In response to petitioner's prima facie case, the State 

offered nothing more than "unquantified, speculative, and theoretical 

objections to the proffered statistics," Trout v. Lehman, supra, 
  

702 F.2d at 1102, ignoring judicial warnings that "the most effective 

way to rebut a statistically based prima facie case is to present 

more accurate statistics.” Id. The State presented one untested 

hypothesis -- that the apparent racial disparities could be explained 

by the generally more aggravated nature of white victim cases -- but 

it offered not a single statistical analysis to confirm or deny the 

= 21 = 

 



  

hypothesis. (Petitioner's analyses reported at GW 5 and GW 6, by 

contrast, demolish the State's theory, proving by HT 

tion of cases at similar levels of aggravation that white victim cases 

are systematically more likely to receive capital sentences). 

The State offered, in fact, not a single analysis in which it 

5 had controlled for any variable. It did not propose, much less 

test the effect of, any plausible explanatory variable that had 

not been included in petitioner's models. It did not propose 

any alternative model employing a different combination of 

petitioner's variables that might plausibly reduce the racial 

factors. It did not suggest any form of statistical analysis, apart 

from those employed by petitioner, that might yield a different 

result. It did not point to a single analysis conducted by peti- 

tioner in which the racial effects disappeared or ran counter wo 

petitioner's claims. 

The State, in short, presented no affirmative statistical 

case. on rebuttal at all. 

What the State attempted unsuccessfully to do was to 

attack the integrity of petitioner's data sources. On surrebuttal, 

however, petitioner presented strong evidence to defend those data 

(see Pet. Mem., 48-49, 54-58), and he showed that additional 

analyses conducted, on a worst case basis, to take full account 

of the State's criticisms, simply did not alter the racial effects 

consistently found by petitioner.(See Pet. Mem. 56-57; DB 120-DB 124). 

The uniqueness of petitioner's evidence, compared with 

that in most other constitutional SAE dependent upon analysis of 

statistical data, is the comprehensive and thoroughgoing presentation 

Lio9 Lo 

 



  

made by his experts, and the unanimity of results on the presence 

and persistence of the racial variables. Petitioner has uncovered 

no reported decision in which more methods of analysis, involving 

more alternative hypotheses, have been applied to the data. The 

problems that might confront a court in determining which of 

several statistical methods to credit -- if those methods yielded 

radically contrary results -- pose no problem here, where all of 

the methods agree, confirming the reality and persistence of the 

racial effect. Indeed, Professor Richard Berk, referring to these 

"triangulated" results, testified that they offered perhaps the 

strongest possible witness that racial factors play a real and 

genuine role in determining capital sentencing outcomes in Georgia. 

Faced with this overwhelmingly one-sided and unrebutted 

statistical case, which after accounting. for all plausible alternative 

variables nevertheless shows - the existence of strong racial factors 

that systematically influence the decision to impose sentences of life or 

death, this Court should apply the clear and controlling Fourteenth 

Amendment principles guaranteeing equal protection of the law to 

grant petitioner's requested relief and vacate his sentence of 

death. 

£31 

 



  

CONCLUSION 
  

The writ should therefore issue, ordering petitioner 

: to be released unless, within a reasonable time, he is resentenced 

to life imprisonment. 

Dated: November 1, 1983 

Respectfully submitted, 

ROBERT H. STROUP 
1515 Healey Building 
Atlanta, Georgia 30303 

JOHN CHARLES BOGER 

10 Columbus Circle 

New York, New York 10019 

TIMOTHY K. FORD 
600 Pioneer Building 
Seattle, Washington 94305 

ANTHONY G. AMSTERDAM 
New York University Law School 
40 Washington Square South 
New York, New York 10012 

ATTORNEYS,Z FOR PETITIONER 

BY: 
  

ich ge oF SAE 

 



  

CERTIFICATE OF SERVICE 
  

I hereby certify that I am one of the attorneys for 

petitioner and that I served the annexed Supplemental Memorandum 

of Law on respondent by placing a copy in the United States mail, 

first-class mail, postage prepaid, addressed as follows: 

Mary Beth Westmoreland, Esq. 
Assistant Attorney General 

132 State Judicial Building 

Atlanta, Georgia 30334 

Done this lst day of November, 1983. 

AUG IS 
  

\ JOHN CHARLES BOGER 

 



  

IN THE UNITED STATES DISTRICT COURT 

FOR THE NORTHERN DISTRICT OF GEORGIA 

ATLANTA DIVISION 

  

WARREN McCLESKEY, 

Petitioner, 

-against- - CIVIL ACTION 

NO. C81-2434A 

WALTER D. ZANT, Superintendent, 

Georgia Diagnostic & Classification 

Center, 

o
®
 

Respondent. 

  

LJ] 

PETITIONER'S SUPPLEMENTAL MEMORANDUM OF LAW 
  

ROBERT H. STROUP 
1515 Healey Building 
Atlanta, Georgia 30303 

JOHN CHARLES BOGER 

10. Columbus Circle 

New York, New York 10019 

TIMOTHY K. FORD 
600 Pioneer Building 
Seattle, Washington 94305 

ANTHONY G. AMSTERDAM 

New York University Law School 

40 Washington Square South 

New York, New York 10012 

ATTORNEYS FOR PETITIONER 

 



  

TABLE OF CONTENTS 
  

I. Petitioner's Burden of Proof On His Claim Of Racial 

DSO imitation os sie He es aa aie ee mie er Hie mee 2 

II. The Methods Employed By Petitioner To Meet His Burden 
OF Proof viv: uv os 6s veins wile Peli 3b vive wr eve steve 8 

ITI. Peritioner's Proof Of Discrimination. « civ ds oi oie nlB 

CONC ISI ON, yy tN ie Ne Ee rie ie iis Haale a nie Terie rel 

 



  

IN THE UNITED STATES DISTRICT COURT 
FOR THE NORTHERN DISTRICT OF GEORGIA 

ATLANTA DIVISION 

  

WARREN McCLESKEY, 

Petitioner, 

-against- CIVIL ACTION 

NO. C8l-2434A 

WALTER D. ZANT, Superintendent, 

Georgia Diagnostic & Classification 
Center, 

Respondent. 

  

LJ 

PETITIONER'S SUPPLEMENTAL MEMORANDUM OF LAW 
  

Petitioner Warren McCleskey ("petitioner") submits this 

supplemental memorandum of law at the invitation of the Court, 

following the hearing of October 17, 1983, to address certain 

questions arising from the evidence presented by the parties. 

This memorandum is designed, not as a comprehensive statement of 

petitioner's case, but rather to supplement petitioner's previous 

memorandum of September 26, 1983. That initial memorandum provided 

the Court with an overview of petitioner's case and addressed at 

length the constitutional foundations of petitioner's arbitrariness 

and racial discrimination claims. 

In this brief, petitioner will not retrace that ground; 

instead, having already demonstrated that proof of persistent and 

intentional disparities by race. in the treatment of capital cases 

in Georgia would suffice to make out a violation of the Equal 

 



  

Protection Clause of the Fourteenth Amendment, requiring petitioner's 

death sentence to be vacated, petitioner will now turn to the question 

of how such disparate racial treatment must be proven. Specifically, 

v. petitioner will address: (i) the burden of proof petitioner must 

shoulder to establish his evidentiary claims; (ii) the methods of 

proof petitioner has adopted to meet this burden; and (iii) the 

facts petitioner has established, measured by the prevailing legal 

standards. 

I. 

Petitioner's Burden of Proof 

On His Claim of Racial Discrimination 
  

Petitioner has shown in his initial brief (see Pet. Mem., 

86-92)1/ that intentional discrimination sufficient to establish an 

Equal Protection Clause violation under the Fourteenth Amendment 

can be proven by statistical evidence alone: "In some instances, 

circumstantial or statistical evidence of racially disproportionate 

impact may be so strong that the results permit no other inference 

but that they are the product of a racially discriminatory intent 

or purpose.” Smith v. Balkcom, 671 F.2d 858, 839 {5th Cir. Unit B 
  

1982) (on rehearing); accord, Spencer v. Zant, 715 F.2d 1562, 1581 
    

(11th Cir. 1983); cf. Adams v. Wainwright, 709 F.2d 1443, 1449 
  

(11th Cir. 1983).2/ 

  

/ Each reference to Petitioner's Post-Hearing Memorandum of Law, 

dated September 26, 1983, will be indicated by the abbreviation 

"Pet. Mem." 

2/ By denying as irrelevant petitioner's prehearing request for 
discovery on other actions that might demonstrate a pattern of 

cont'd. ] 

 



  

In Castaneda v. Partida, 430 U.S. 482 (1977), the Supreme 
  

Court has held that to make out a prima facie statistical case, at 

least "in the context of grand jury selection," requires a 

petitioner to establish that 

"the [racial] group is one that is a recognizable, 
distinct class . . . Next the degree of underrepre- 
sentation must be proved, by comparing the propor- 
tion of the group in the total population to the 

proportion called to serve as grand jurors . . . 

Finally, a selection procedure that is susceptible 

of abuse or is not racially neutral [must be shown 
which] supports the presumption of discrimination 
raised by the statistical showing." 

Castaneda v. Partida, supra, 430 U.S. at 494. At that point, the 
  

Court continued, petitioner "has made out a prima facie case of dis- 

criminatory purpose, and the burden then shifts to the State to 

rebut that case.” Id., at 495. The Eleventh Circuit recently 

adopted a virtually identical procedure in analyzing a Fourteenth 

Amendment equal protection claim stemming from the detainment of 

Haitian immigrants: 

Although the standard of proof in Title VII cases 
differs from that in constitutional equal protection 
cases, the framework for proving a case, i.e. prima 

facie case, rebuttal, ultimate proof, is the same. 

See, e.g., Castaneda v. Partida, 430 U.S. at 495-96, 

. . Because of the similar framework, and because 
there are few equal protection cases relying on 

statistics, when appropriate we draw upon Title VII 

cases." 

  

  

2/ cont'd. 

racial discrimination in the criminal justice system in Fulton 

County and the State of Georgia, the Court necessarily limited 

petitioner's proof to statistical evidence, supplemented by 

reported decisions evidencing racial discrimination of which the 

Court might take judicial notice. (Pet. Mem. 101-02; see also 
  

Petitioner's First Interrogatories to Respondent, dated April 

18, 1983, 9% 9-18; Order of June. 3, 1983, at 2.) 

 



  

Jean v. Nelson, 711 F.2d 1455, 1488 n.30 (11th Cir.), vacated and 
  

  

pending on reh'g en banc, 714 F.2d 96 (llth Cir. 1983); cf. 
  

Eastland v. Tennessee Valley Authority, 704 F.2d 613, 618 (llth 
  

Cir.. 1983). 

A proper analysis therefore requires, first, the 

determination of whether petitioner has established a prima facie 

case; second, the examination of respondent's rebuttal case, if 

any; and third, an assessment of whether, in light of petitioner's 

responsive evidence, he has ultimately met his burden of proof. To 

prevail, petitioner must demonstrate an Equal Protection violation 

"by a preponderance of the evidence." Jean v. Nelson, supra, 711 
  

F. 2d at 1487, citing Texas Dep't. of Community Affairs v. Burdine, 
  

450 U.8., 248, 252-53. (1981). 

Since petitioner asserts systemwide discrimination, the 
  

principal focus of analysis should be, not upon the evidence of 

discrimination in petitioner's individual case, as would be appro- 

priate in analogous individual Title VII cases, see McDonnell 
  

Douglas Corp. v. Green, 411 U.S. 792 (1973); Mt. Healthy Board of 
  

  

Education v. Doyle, 429 U.S. 274 (1977), but rather upon systemwide 
  

(or perhaps judicial circuitwide, see Pet. Mem. 104-09) statistical 

evidence of disparities, as in analogous jury cases and other cases 

alleging classwide or systemwide discrimination. See, e.g., 
  

Castaneda v. Partida, supra; see also Washington v. Davis, 426 
    

  

U.S. 229, 241-42 (1976); Arlington Heights v, Metropolitan Houglng 
  

Authority, 429 U.S. 252, 266 (1977); Hazelwood School District v. 
  

  

Uhited States, 433 U.S. 299, 307-08 (1977). 
  

The precise evidentiary burden necessary to establish a 

prima facie capital sentencing case has never been definitively 

wae 

 



  

established. In Smith v. Balkcom, on rehearing, the former Fifth 
  

Circuit strongly suggested by negative implication what might 

suffice to establish a prima facie case: 

3 "No data is offered as to whether or not charges 

or indictments grew out of reported incidents or 

as to whether charges were for murder under aggra- 

vating circumstances, murder in which no aggra- 

vating circumstances were alleged, voluntary man- 

slaughter, - involuntary manslaughter, or other 

offenses. The data are not refined to select 

incidents in which mitigating circumstances were 

advanced or found or those cases in which evidence 

of aggravating circumstances was sufficient to 

warrant submission of the death penalty vel non 
to a jury. No incidents resulting in not tT guilty 

verdicts were removed from the data. The unsupported 

assumption is that all such variables were equally 

distributed . . « 

Smith v. Balkcom, supra, 671 F.2d at 860 n.33. On the other hand, 
  

  

the Eleventh Circuit's per curiam opinion in Adams v. Wainwright, 

supra, 709 F.2d at 1449, contained dicta that "[o]nly if the evidence 

of disparate impact is so strong that the only permissible inference 

is one of intentional discrimination will it alone suffice." More 

recently, the Eleventh Circuit in Spencer v. Zant, 715 P.24 at 1582 
  

  

n.1l5, drawing directly from Arlington Heights, supra, 429 U.S. at 

266, suggested that the proper standard may require evidence of 

3/ 
"'a clear pattern, unexplainable on grounds other than race."-— 

  

3/ Petitioner contends that determination of whether the proper 

standard should be drawn from Smith, from Adams, from Spencer or 
  

from some other case need not be Fosolved here, since petitioner's 

statistical evidence accounts for every plausible rival hypothesis, 

thereby meeting or exceeding even the most stringent possible 

standard. See discussion at pp. 16-23, infra. 

 



  

Once petitioner has shown a prima facie case, the 

burden then shifts to the State to rebut the case in one of 

three ways: (i) "by showing that plaintiff's statistics are 

misleading..;;[ii] by presenting legitimate non-discriminatory 

reasons for the disparity," Eastland v. TVA, supra, 704 F.2d at 
  

618-19; or (iii) by proving that the discrimination is justified 

by a compelling state interest (see Pet. Mem. 77-78, 115-23). A 

rebuttal case challenging a party's data base as misleading or 

inaccurate cannot succeed without strong evidence that the data 

are seriously deficient and unreliable: 

"[A] heavy burden must be met before a party can 
justify the rejection in toto of any statistical 

analyses on grounds of errors or omissions in the 

data . . . the challenging party bears the burden 
of showing that errors or omissions bias the data 
and} . . . that this bias alters the result of the 
statistical analyses in a systematic way." 

Vuyanich v. Republic Nat'l Bank of Dallas, 505 F. Supp. 224, 255-56 
  

(N.D. Texas 1980); accord, Trout v. Lehman, 702 F.2d 1094, 1101 
  

(D.C. Cir. 1983); Detroit Police Officer's Ass'n v. Young, 608 F.2d 
  

671, 687 (6th Cir. 1979), cert. denied, 4532 U.S. 938 (19381)(gec 
  

generally, Pet. Mem., 115-18). 
  

A rebuttal case predicated upon "legitimate non-discrim- 

inatory reasons for the disparity" cannot succeed merely by challeng- 

ing petitioner's prima facie case "in general terms," Wade v. 

Mississippi Cooperative Extension Service, 528 F.2d 508, 517 
  

(5th Cir. 1976). "[Ulnquantified, speculative, and theoretical 

objections to the proffered statistics are properly given little 

weight by the trial court," Trout v. Lehman, supra, 702 F.2d at 
  

1102; see, e.g., Castaneda v. Partida, supra, 430 U.S. at 499 n.19: 
    

 



  

Jean v. Nelson, supra, 711 F.2d at 721, 730. Addressing this theme, 
  

Chief Judge Godbold recently noted in Eastland v. TVA, supra, 704 
  

F.2d at 622-23 n.l4, citing D. BALDUS & J. COLE, STATISTICAL PROOF 

OF DISCRIMINATION §8.23 at 74 (1980): 

"A defendant's claim that the plaintiff's 
model is inadequate because a variable has 
been omitted will ordinarily ride on evidence 
[from the defendant] showing that (a) the 
qualification represented by the variable was 
in fact considered [by the defendant], and (b) 
that the inclusion of the variable changes the 
results of the regression so that it no longer 
supports the plaintiff. Both of these facts are 

established most clearly and directly if the defend- 

ant offers an alternative regression model similar 
to the plaintiff's except for the addition of the 
variable in question." 

Finally, while a rebuttal case might theoretically 

be made in support of racially discriminatory treatment in some 

limited area of the law, the Supreme Court in Furman v. Georgia, 
  

408 U.S. 238 (1972) made it perfectly clear that no purported state 

interest could ever justify discriminatory imposition of the death 

penalty. (The State in this case has never suggested that any valid 

State policy could be furthered by such discrimination, and there- 

fore this possible line of rebuttal need not detain the Court. (See 

Pet. Mem., 77-81A)). 

Petitioner should prevail under the analysis outlined 

above if his prima facie case -- discounted by any valid criticisms 

adequately proven by the State's rebuttal case, augmented by any 

surrebuttal evidence petitioner can muster to counter the State's 

rebuttal case -- establishes discrimination by a preponderance of 

the evidence. Petitioner need not produce statistical evidence 

 



  

which would fully explain the workings of the system so long as 

he can demonstrate that racial discrimination is a real and per- 

sistent characteristic of that system. 

11. 

The Methods Employed By Petitioner To 
Meet His Burden Of Proof 
  

Petitioner McCleskey employed well-accepted and rigorously 

controlled statistical methods in support of his constitutional 

claims of discrimination in capital sentencing. He first established 

through the comparison of unadjusted racial comparisons that sig- 

nificant race-of-defendant and race-of-victim disparities are 

characteristic of Georgia's capital sentencing system. (DB 62; DB 69; 

DB 70). Although such "unadjusted" racial disparities have been 

held legally insufficient to establish a constitutional violation in 

the context of capital sentencing systems, see, e.g., Spinkellink 
    

v. Wainwright, 578 F.2d 582 (5th Cir. 1978); Smith v. Balkcom, 860 
  

  

F.2d 573 (5th Cir. Unit B 1981), it is instructive to note that 

statistical evidence no more sophisticated than this has regularly 

been deemed sufficient to require reversal in other equal protection 

contexts such as jury cases, see, e.dg., Castaneda v. Partida, supra, 
  

(statistically significant racial disparities, with no additional 

variables held constant); and employment discrimination cases, see 

e.g., Fisher v. Proctor & Gamble Mfg. Co., 613 F.2d 527, 544 (5th 
    

Cir. 1980), cert. denied, 449 U.S. 1115 (1981)(a prima facie case 
  

established by "glaring" statistical disparities even without 

controlling for job qualifications). 

 



  

Petitioner obviously did not rest, however, with an 

identification of these unadjusted racial disparities. Instead, 

Professor David Baldus, petitioner's principal expert, testified 

that he drew on his own expert knowledge of the criminal justice 

system, as well as the guserience and knowledge of his professional 

colleagues, supplemented by extensive reading and review, to develop 

an extensive list of variables that might plausibly affect the 

sentencing outcome in a capital case. This list was incorporated 

into a questionnaire, completed for each case included in PRS and 

CSS studies, which contained over five hundred variables. For 

purposes of analysis, Professor Baldus employed more than 230 of 

these "recoded" variables that he judged to be plausible factors 

in conducting his sentencing analyses. (Professor Baldus specifically 

testified that he employed every relevant variable on which he could 

obtain information.) 

To proceed beyond unadjusted analysis, petitioner analyzed 

the effect on sentencing outcomes of the racial factors while "con- 

trolling" for, or holding constant, the effects of the other plausible 

explanatory variables. Professor Baldus and Professor George Woodworth 

both testified that, in conducting these analyses, they relied upon 

the two accepted statistical methods available to achieve such control: 

cross-tabulations, and multiple regression analysis. Cross-tabular 

analysis, Professor Baldus explained, proceeds by dividing cases 

into successively smaller subcategories, each distinguished by the 

presence or absence of a series of relevant variables. Cross-tabular 

analysis permits one to compare cases that are comparable or similar 

on all of the variables examined, observing changes in the variable of 

interest. (Professor Baldus reported upon the results of a number of 

nig NEO 

 



  

cross-tabular analyses he performed in which the racial effects 

remained influential (see DB 67 ; DB 76 ). The inherent limitation 

. of cross-tabular analysis, Baldus and Woodworth explained, is that 

it cannot meaningfully account for a very large number of variables 

simultaneously, since at some point the number of cases possessing 

similar characteristics on each of the increasing number of variables 

becomes very small, and "cell sizes" decrease toward statistical 

and practical insignificance. 

Multiple regression analysis, Professors Baldus and 

Woodworth testified, avoids this inherent limitation of cross-tabular 

analysis by employing algebraic formulae to calculate the additional 

impact of the presence or absence of a variable of interest (e.g., 

the race of the victim) over and above the collective impacts of 

a host of other variables. Professor Woodworth explained that re- 

gression accomplishes this result, not by examining cases that are 

similar on all variables other than the variable of interest, but 

instead by assigning cases an index value along a scale determined 

by the presence or absence of other variables, and then calculating 

the comparative sentencing rates at each level. (See GW 9; GW 10). 

This use of regression analysis, Professor Woodworth testified 

(without contradiction from State's expert Dr. Joseph Katz), is 

mahematically sound and fully accepted as a valid means of statis- 

tical measurement. The algebraic formula for calculating a sample 

regression analysis with three variables was presented to the 

Court as GW 13 and GW 14. 

 



  

The Fifth Circuit first adverted to the use of regression 

analysis in 1976, calling it "a sophisticated and difficult method 

of proof in an employment discrimination case," Wade v. Mississippi. 
  

Cooperative Extension Service, 528 F.2d 508, 517 (5th Cir. 1976). 
  

Five years later, the Court, having gained greater familiarity 

with the method, observed that "[m]ultiple regression analysis is 

a relatively sophisticated means of determining the effects that 

any number of different factors have on a particular factor," 

Wilkins v. University of Houston, 654 F.2d 388, 402-03 (5th Cir. 
  

1981). The Court held in Wilkins that "if properly used, multiple 

regression analysis is a relatively reliable and accurate method of 

gauging classwide discrimination," id. at 402-03 18 indeed noting 

that "it may be the best, if not the only, means of proving classwide 

discrimination . . . in a case where a number of factors operate 

simultaneously to influence" the outcome of interest. Id. at 403. 

With proper attention to its possible misuse, the Eleventh Circuit 

has also embraced multiple regression analysis as an appropriate 

tool for the proof of discrimination claims. See, e.g., Eastland 
    

v. TVA, supra, 704 F.2d at 621-22; Jean v. Nelson, supra; see also, 
  

  
  

Valentino v. United States Postal Service, 674 F.2d 56, 70 (D.C. 
  

Cir. 1982); see generally, Finkelstein, "The Judicial Reception of 
  

Multiple Regression Studies in Race and Sex Discrimination Cases," 

80 COLUM. L. REV. 737 (1980). 

Perhaps the most extensive judicial discussion of the 

nature and role of multiple regression analysis in the proof of 

discrimination claims is Judge Higgenbotham's influential and 

widely cited opinion in Vuyanich v. Republic Nat'l Bank of Dallas, 
  

- 1} 

 



  

505 F. Supp. 224, 261-79 (N.D. Tex. 1980). Judge Higgenbotham 

observes that multiple regression techniques have been "long used 

by social scientists and more recently [have been] used in judicial 

resolution of antitrust, securities, and employment discrimination . 

disputes," Vuyanich, supra, 505 F. Supp. at 261. He notes that these 
  

"mathematical models are designed to determine if there is any differ- 

ential treatment not entirely attributable to legitimate differences,” 

id. at 265, calling them "an important addition to the judicial 

toolkit,” id. at 267. 

Drawing upon basic texts in econometrics and regression 

analysis (including D. BALDUS & J. COLE, STATISTICAL PROOF OF 

DISCRIMINATION (1980)), Judge Higgenbotham then embarks upon an 

extensive mathematical and statistical discussion of regression 

methods, including the derivation of the basic regression formulae, 

id. 269-71, the calculation of the statistical significance of 

regression coefficients, 1id., 271-73, improper applications of 

regression methods, 1ld.,. 273-73, and different methods of employing 

regression analysis to measure possible discriminatory behavior, 1id., 

275-79. 

The discussion in Vuyanich coincides with and confirms 
  

the teachings of Professor Franklin Fisher in his influential article 

"Multiple Regression in Legal Proceedings," 80 COLUM. L. REV. 702 

(1980). Both make clear that multiple regression analysis "is . . . 

a substitute for controlled experimentation," Vuyanich, supra, 
  

505 F. Supp. at 269, and that "[t]he results of multiple regressions 

-- Such as what we will call ’'coefficients' in the ordinary least 

square methodology -- can be read as showing the effect of each 

independent variable on the dependent variable, holding the other 

ol 

 



  

independent variables constant. Moreover, relying on statistical 

inference, one can make statements about the probability that the 

effects described are due only to a chance fluctuation,” id., at 

269; accord, Fisher, supra, 80 COLUM. L. REV. at 706. Chief Judge 
  

Godbold explicitly recognized the value of regression analysis in 

Eastland v. TVA, supra, 704 F.2d at 621, finding that "[m]ultiple 
  

regression analysis is a quantitative method of estimating the 

effects of different variables on some variable of interest.” 

These clear precedents Ssuiblish that the multiple 

regression method has been judicially accepted as a principal 

analytic tool -- indeed, in cases involving a large number of 

simultaneously operative variables, perhaps '"the only means of 

proving classwide discrimination,” Wilkins v. University of Houston, 
  

supra, 654 F.240 at 463. 

In evaluating regression analyses, the courts and commenta- 

tors have pointed to a number of problems that could arise if sufficient 

care is not taken in analysis. If data are totally inaccurate or 

are shown to be systematically biased for the variable of interest, 

the analysis may be flawed. Vuyanich v. Republic Nat'l Bank of 
  

Dallas, supra, 505 F. Supp. at 255-56, 275%. Purther, if a "model," 
  

or group of independent variables is employed that omits "some 

relevant explantory variable . . . the regression coefficient . . . 

would be 'biased' . . . and the usual tests of significance 

concerning the included regression coefficient . . . will be invalid,” 

id. at 274. Fortunately, as Judge Higgenbotham notes, "[clertain 

statistical tests are available to suggestwhether this sin of 

omission has occurred.” 1d. 

1% 

 



  

At the other extreme, "when one or more irrelevant 

variables are included in the model . . . [a] risk of 'multicolinearity'" 
  

arises. Id. Yet the effect of possible multicolinearity is not 

to increase but to deflate evidence of possible discriminatory 

impact, id. at 274-75: thus "if multicolinearity exists, the prob- 

ability will be increased that the net impact of [racial factors] 

« «..« Will De judged statistically nonsignificant, even in cases 

where there are actual differences in the treatment." Id. In short, 

multicolinear models may underestimate, but do not overestimate, 

the extent of possible discrimination. 

A third possible problem can arise "where the analyst 

chooses to use a regression equation that is linear in the explanatory 

variables when the true regression model is nonlinear," id. at 27S. 

Obviously, the means by which to avoid such a problem is to conduct 

Salis employing both linear and nonlinear (such as logistic) 

regressions. 

Finally, least squares. regression depends upon the 

assumption that the "error term,” -- the "u" in a regression formula 

which stands for idiosyncratic or "random influences" that characterize 

virtually every social scientific model, id. at 269-70, 273 -- 

"follows the 'normal distribution, '" id. at 275, that is, displays 

no systematic relation to other independent variables. However, 

Judge Higgenbotham observed that "[w]ith respect to this assumption, 

basic least squares reqression models are "quite" robust" in that 

they will tolerate substantial deviations without affecting the 

validity of the results.’ D. Baldus & J. Cole, supra, n.5S §8A.41, 

at 284." Id. at 275. Moreover, he noted, "[n]lonnormality of errors 

can be detected through the use of [statistical] . . . techniques." Id. 

da, 

 



  

Petitioner's experts testified without contradiction 

that they had carefully followed all of the requisite steps in 

conducting regression analysis, and that they had taken particular 

care to conduct statistical diagnostic tests to determine whether 

any of the assumptions of regression analyses had been violated 

in petitioner's analyses, and whether the results could possibly 

be biased thereby. Professor Woodworth offered his expert statis- 

tical opinion, without any contradiction by the State, that the methods 

employed by petitioner were appropriate, that models were not 

misspecified, and that no bias could be discerned in the reported 

results. Professor Berk, petitioner's reubttal expert, confirmed 

Professor Woodworth's expert opinion. He explicitly complimented 

petitioner's conduct of regression analysis as "state-of-the-art," 

and found both of petitioner's studies to be of "high credibility." 

In sum, the statistical methods employed by petitioner, 

including cross-tabular and regression analysis, have been expressly 

adopted by the Fifth and Eleventh Circuits as appropriate tools 

for the measurement of the possible effect of racial variables. 

The regression analyses relied upon by petitioner were properly 

conducted by leading experts in the field, were carefully monitored 

for possible statistical problems, and have been found to be both 

statistically appropriate and accurate in their assessment of the 

presence and magnitude of racial disparities in capital sentencing 

in Georgia. Methodological concerns, whether based in law or in 

statistics, thus pose no impediment to the Court's evaluation of 

petitioner's reported results. 

— 5 = 

 



  

III. 

Petitioner's Proof of Discrimination 
  

To meet his prima facie burden of proof, petitioner has 

of fered the Court a wide range of statistical analyses, virtually 

all of which demonstrate or, at a minimum suggest, significant 

race-of-victim effects, as well as significant race-of-defendant 

effects within important subcategories. Petitioner reported strong 

unadjusted racial disparities (see Pet. Mem., 24-25). He then con- 

structed a model which would take into account the statutory factors 

identified by the Georgia legislature as sufficiently important 

aggravating circumstances to permit the imposition of a death sentence, 

together with the "nonstatutory" aggravating circumstance of prior 

record (also expressly designated as relevant by Georgia statute). 

Professor Baldus reported the results of this analysis employing 

both a least squares analysis, which assumes a linear distribution 

of cases, and a logistic analysis, which depends upon no such 

assumptions. The results, as indicated below, demonstrate that the 

race-of-victim factor wields an independent effect on sentencing 

outcome at a highly significant level: 

  

  

  

  

W.L.S. Logistic Regression 

A Regression Results Results 

Regression Coefficient & Regression Death Odds 

Level of Statistical Coefficient Multiplier 

Significance 

Race of Victim «07 TH 2:8 

(.0014) : 

Race of Defendant .04 : «02 1:0 

(.09) (.93) 

(DB 78) 

 



  

Under this analysis, race of the victim is at least as 

important a determinant of sentence as such factors as that the 

defendant had a prior capital record, that the murder was vile, 

horrible or inhuman, that the victim was a policeman, or other 

serious aggravating factors. When Professor Baldus refined this 

model to incoprorate not only statutory aggravating factors, but 

75 mitigating factors as well, the relative impact of the race-of- 

victim variable actually increased: 

  
  

  

  

w.L.S. Logistic Regression 
Regression Results Results 
Regression Coefficient & Regression Death Odds 
Level of Statistical Coefficient Multiplier 

Significance 

Race of Victim .10 : 2.1 8.2 
{.001) : {.00}) 

Race of Defendant «07 +36 1.4 

(.01) (ns) 

; Professor Baldus thereafter employed a wide range of models 

{gee .e.g., DB 30, DB 83, DB 96, DB 98) to see whether any constellation 

of variables would eliminate or substantially diminish the race-of- 

victim effect. None did. In effect, petitioner thereby "anticipated 

and adequately met the government's statistical challenge. Plaintiffs 

offered a variety of statistical and testimonial evidence to demonstrate 

that [other independent variables] . . . were irrelevant," Jean v. 

Nelson, supra, 711 F.2d at 1498, in explaining the persistence of 
  

the racial variables. 

3 

 



  

Professor Baldus, as noted, conducted a number of 

analyses employing the 230+ variable model which included all 

known variables which plausibly might have affected sentencing 

outcome, and the racial factors remained significant (see, e.qg., 
  

DB 80 "Race of Victim . . . After Simultaneous Control for 230+ 

Non-Racial Variables . . . .06(.01); Race of Defendant . . . .06 

{.01)), In one sense, this model operates most "realistically" 

since it includes and controls for the effects, however small, of 

any aggravating or mitigating factors that might affect a prosecutor's 

or jury's judgment. Yet, because of problems of multicolinearity, 

which as explained above can actually dampen or suppress the real 

impact of other independent variables, Professor Baldus, Professor 

Woodworth and Professor Burford, the State's expert, all indicated 

that the 39 variable or "mid-range" model probably provided the 

best statistical evaluation of the independent impact and significance 

of the racial variables. 

Using that mid-range model (as well as models with seven 

variables, eleven variables, all statutory aggravating circumstances, 

and all statutory plus 73 mitigating circumstances), Professor 

Woodworth conducted a comprehensive series of diagnostic tests to 

see whether problems in the weighting procedure employed, the selection 

of least squares or logistic regression, the existence of some 

"missing" data, the influence of the 48 most important cases, or the 

presence of possible "interaction" effects among the variables 

included might explain the racial disparities reported. Professor 

Woodworth's conclusion, amply supported by Table 1 from GW 4, is 

that the race-of-victim coefficient remains large (from .041 to 

Iss 

 



  

.117) and statistically significant throughout the diagnostic 

analyses. The race of the defendant exhibited an unstable, 

although often important effect as well. In sum, the persistent 

racial effects reported in petitioner's regressions are not statis- 

tical artifacts, but reflect real-world disparities in capital 

sentencing treatment based upon racial factors. 

Professor Woodworth also explained that the np 

calculations reported in his diagnostics did not mean that peti- 

tioner's models were inadequate or incapable of accurately measuring 

the racial effects. First, Professor Woodworth noted that, insofar 

as Georgia's capital sentencing system is in fact operating in an 

arbitrary and capricious pattern, no statistical model can explain 

all of the variance, since a part of it will necessarily be random 

and idiosyncratic. Secondly, Professor Woodworth stressed that 

large npn or "u" terms do not affect the accuracy of the measurement 

of the effect of other variables, concurring with Professor Fisher's 

analogous observation that 

n{ilt is very important . . . to realize that a 
large standard error of estimate does not tell 
one anything at all about the accuracy with 
which the effects of the independent variables 
are measured . . . The standard error of estimate 
is a way of assessing how important the random part 
of the model is; it does not tell one how large the 

affects of such randomness are on one's ability to 
measure the systematic part." 

Fisher, supra, 80 COLUM. L. REV. at 719. 
  

id Te 

 



  

Both Professor Baldus and Professor Woodworth agreed 

that the figures which most accurately and completely summarized 

the racial effects they had observed were reflected in GW 5 and GW 6. 

Those figures, based upon the mid-range model with interactions 

and nonlinearities accounted for, show a disparity in the treatment 

of homicide cases by race-of-victim and race-of-defendant which 

varies in magnitude depending upon the level of aggravation, or 

seriousness, of the homicide. Among the least aggravated cases, 

little racial disparity exists, because virtually no death sentences 

are imposed in any cases. Among the most aggravated cases, once 

again there exists little racial disparity, since nearly all of the 

cases receive a death sentence. Among the moderately aggravated 

cases, however, substantial and unchecked racial disparities exist. 

At petitioner McCleskey's level of aggravation, for example, the 

sentencing disparity between white victim and black victim cases 

iz .22 points, 

(See next page for GW 6,Table 2) 

- 00 - 

 



Gwe 

  

A Figure 2: Midrange’ Model With Interactions and Nonlinearities-- 

f Black Defendants 
g # 

  

  

  
    

oo 4 

75 + 

£2 2 

25 4 

McCleskey 3 

00 +——— FR SE a EA Se 
6 1.5 8 1.0 1.2 

LEVEL OF AGGRAVATION 

2/ The curves reoresent 95% confidence bounds on the average death 

sentencing rate at increasing levels of aggravation (redrawn from 

computer output). 

 



  

Petitioner has set forth in his principal brief the parallel 

findings he obtained from both a statistical and a qualitative analysis 

of data from Fulton County, where petitioner was tried. Although 

the smaller sample size restricted the statistical significance 

of the results, the same pattern of influence of racial variables 

can clearly be ascertained. (See Pet. Mem., 36-40; DB 106-116). 

Beyond this statistical evidence in support of his prima 

facie case, petitioner introduced the deposition of District Attorney 

Lewis Slayton. That testimony, summarized in petitioner's principal 

brief at page 48, reveals a system for the processing of capital 

indictments in Fulton County shat is decentralized among a dozen 

or more assistants, carried out with no written procedures or 

guidelines, and no central review of all decisions in homicide 

cases. Therefore, petitioner has shown a circuitwide system which 

affords an "opportunity for discrimination," since it leaves 

processing decisions up to a multitude of decisionmakers whose 

decisions are not routinely reviewed by a central authority for 

compliance with any objective criteria. 

In response to petitioner's prima facie case, the State 

offered nothing more than "unquantified, speculative, and theoretical 

objections to the proffered statistics," Trout v. Lehman, supra, 
  

702° F.2d at 1102, ignoring judicial warnings that "the most effective 

way to rebut a statistically based prima facie case is to present 

more accurate statistics." Id. The State presented one untested 

hypothesis -- that the apparent racial disparities could be explained 

by the generally more aggravated nature of white victim cases -- but 

it offered not a single statistical analysis to confirm or deny the 

oe Cp 

 



  

hypothesis. (Petitioner's analyses reported at GW 5 and GW 6, by 

contrast, demolish the State's theory, proving by examina- 

tion of cases at similar levels of aggravation that white victim cases 

are systematically more likely to receive capital sentences). 

The State offered, in fact, not a single analysis in which it 

had controlled for any variable. It did not propose, much less 

test the effect of, any plausible explanatory variable that had 

not been included in petitioner's models. It did not propose 

any alternative model employing a different combination of 

petitioner's varishiss that might plausibly reduce the racial 

factors. It did not suggest any form of statistical analysis, apart 

from those employed by petitioner, that might yield a different 

result. If did not point to =a single analysis conducted by peti- 

tioner in which the racial effects disappeared or ran counter to 

petitioner's claims. 

The State, in short, presented no affirmative statistical 

case. on rebuttal at all. 

What the State attempted unsuccessfully to do was to 

attack the integrity of petitioner's data sources. On surrebuttal, 

however, petitioner presented strong evidence to defend those data 

(see Pet. Mem., 48-49, 54-58), and he showed that additional 

analyses conducted, on a worst case basis, to take full account 

of the State's criticisms, simply did not alter the racial effects 

consistently found by petitioner.(See Pet. Mem. 56-57; DB 120-DB 124), 

The uniqueness of petitioner's evidence, compared with 

that in most other constitutional ies dependent upon analysis of 

statistical data, is the comprehensive and thoroughgoing presentation 

i LR 

 



  

made by his experts, and the unanimity of results on the presence 

and persistence of the racial variables. Petitioner has uncovered 

. no reported decision in which more methods of analysis, involving 

more alternative hypotheses, have been applied to the data. The 

problems that might confront a court in determining which of 

several statistical methods to credit -- if those methods yielded 

radically contrary results -- pose no problem here, where all of 

the methods agree, confirming the reality and persistence of the 

racial effect. Indeed, Professor Richard Berk, referring to these 

"triangulated" results, testified that they offered perhaps the 

strongest possible witness that racial factors play a real and 

genuine role in determining capital sentencing outcomes in Georgia. 

Faced with this overwhelmingly one-sided anciunrotut ied 

statistical case, which after accounting. for all plausible alternative 

variables nevertheless shows - the existence of strong racial factors 

that systematically influence the decision to impose sentences of life or 

death, this Court should apply the clear and controlling Fourteenth 

Amendment principles guaranteeing equal protection of the law to 

grant petitioner's requested relief and vacate his sentence of 

death. 

HEEL an 

 



  

CONCLUSION 
  

The writ should therefore issue, ordering petitioner 

to be released unless, within a reasonable time, he is resentenced 

to life imprisonment. 

Dated: November 1, 1983 

Respectfully submitted, 

ROBERT H. STROUP 
1515 Healey Building 
Atlanta, Georgia 30303 

JOHN CHARLES BOGER 
10 Columbus Circle 
New York, New York 10019 

TIMOTHY K. FORD 
600 Pioneer Building 
Seattle, Washington 94305 

ANTHONY G. AMSTERDAM 
New York University Law School 

40 Washington Square South 
New York, New York 10012 

ATTORNEYS, FOR PETITIONER 

  

- DA 

 



  

: CERTIFICATE OF SERVICE 
  

I hereby certify that I am one of the attorneys for 

petitioner and that I served the annexed Supplemental Memorandum 

of Law on respondent by placing a copy in the United States mail, 

first-class mail, postage prepaid, addressed as follows: 

Mary Beth Westmoreland, Esq. 

Assistant Attorney General 

132 State Judicial Building 

Atlanta, Georgia 30334 

Done this 1st day of November, 1983. 

AUNT IN 
  

\JOHN CHARLES BOGER 

 



  

IN THE UNITED STATES DISTRICT COURT 

FOR THE NORTHERN DISTRICT OF GEORGIA 

ATLANTA DIVISION 

WARREN MCCLESKEY, 

Petitioner, CIVIL ACTION FILE 

VS. NO. C81-2434A 

WALTER ZANT, Warden 
Georgia Diagnostic and 
Classification Center, 

Respondent. 

P
R
 

X
K
 

X
X
 

X
X
X
 
X
X
X
 

  

PETITIONER'S PROPOSED FINDINGS OF FACT WITH RESPECT 
TO EVIDENCE THAT DEATH PENALTY IN GEORGIA 

IS IMPOSED IN RACIALLY DISCRIMINATORY MANNER 
  

Comes now the petitioner, WARREN MCCLESKEY, through his 

undersigned counsel, and files the following proposed findings of 

fact with respect to his claims that the death penalty in Georgia 

is imposed in a racially discriminatory fanner. These proposed 

findings of fact are submitted to assist the Court, in light of 

the recently-available transcript. Findings, including Yooota 

cites, for petitioner's other claims, are found within the text 

of the briefs previously submitted. 

I. The Two Studies Condticten By Professors Baldus, 
Woodworth and Pulaski 
  

1. A number of studies of death sentencing rates, including 

death sentencing rates in Georgia, had been conducted prior to 

those conducted by Dr. Baldus. (Tr. 134-144). 

 



  

2. Baldus relied upon these earlier studies to refine his 

work, and to improve upon methodological problems inherent in the 

earlier studies. (Tr. 145-46, 147-48). 

3. Prior to the start of his own studies, Dr. Baldus 

engaged 1n an extensive review of the legal and social science 

literature regarding the criminal justice system, and particular- 

ly, the sentencing process. (Tr. 130). 

4. Beginning in 1978 Dr. Baldus conducted two different 

studies related to the death penalty in Georgia. (Tr.. 121-22, 

127). 

A. The Procedural Reform Study, Generally. 
  

5. The first study, called the Procedural Reform Study, 

looked at offenders who had been convicted of murder at a guilt 

trial. (Tr. 122}. 

6. Baldus's Procedural Reform Study considered the dispari- 

ties in the rates at which prosecutors advanced murder cases to 

penalty trial so that a jury an given a choice as to whether or. 

not the death penalty would be imposed. (Tr. 122). The 

Procedural Reform Study also considered the disparities in the 

rates at which juries imposed death sentences in those cases 

which were advanced to penalty trials. (Tr. 122-23, 166, 167). 

7. The racial disparities considered in both studies were 

disparities based upon race of defendant and race of victim. 

(Tr. 123). 

 



  

8. The Procedural Reform Study focused upon the prosecu- 

torial decision to advance (or not to advance) a case to a 

penalty trial, and the jury decision to recommend a life sentence 

or to recommend death in those cases where the prosecutor had 

elected to seek the death penalty. (Tr. 122-23, 166-67). 

9. The Procedural Reform Study considered a universe of 594 

murder cases. (Tr. 169). 

10. The time period covered by the Procedural Reform Study 

was the effective date of the current Georgia Death Penalty 

statute, March 28, 1973 through June 30, 1978. (Tr. 170). Any 

persons arrested prior to June 30, 1978 were included within the 

study. (Tr. 170). 

11. The Procedural Reform Study relied, as principal 

sources of information on the crime and the defendant, the 

records of the Georgia Supreme Court, including briefs of the 

Attorney General and briefs of the defense counsel, transcripts 

of the trial, when available, and background information 

contained in the files of the Georgia Department of Offender 

Rehabilitation. (Tr. 175), 

12. Included in the records of the Department of Offender 

Rehabilitation was information on the age and race of the 

offender, and information on the prior record of the accused. 

{Tr. 176). 

13. Also available to Dr. BRaldus for both studies were 

records of the Bureau of Vital Statistics, which contained 

information regarding homicide victims, including their race, 

age, and occupation. {Ty. 1275, 591}. 

 



  

14. Included in the information gathered for the first 

study was information regarding the defendant, the victim, and 

circumstances surrounding the crime, such as contemporaneous 

offenses, method of killing, number of victims killed, role of 

co-perpetrators, and procedural history of the case. (Tr. 196- 

200, DB-27, DB-35). 

15. Included among the sources of information consulted by 

Br. Baldis were summaries of life sentence cases prepared by the 

Georgia Supreme Court and used in its proportionality reviews. 

(Tr. 203). a 

16. The Georgia Board of Pardons & Paroles maintains 

records which give detailed information on the characteristic of 

the offender, characteristics of the crime, and offender's prior 

record. (Tr. 177). 

17. This information from the Board of Pardons and Parole 

files was available for a portion of Baldus's first study, and 

for his entire second study. (Tr. 176, 293). 

18. In collecting the data for both of his studies, Dr. 

Baldus sought to create a data set that would allow him to 

control for legitimate characteristics of each case that one 

could reasonably expect to affect decision-making regarding 

imposition of the death sentence. (Tr. 195). 

19, Prior to development of the questionnaires used for 

gathering the date, extensive research was conducted in the 

criminal justice area in an effort to uncover all possible 

variables or case characteristics that might reasonably affect 

the decision-making process. (Tr. 195). 

 



  

20. Dr. Baldus, along with his co-researchers, sought the 

advise of a number of faculty members at their respective 

institutions who were involved in the criminal justice area. 

{Tr. > 195). 

21. For both studies, to obtain information regarding the 

status of defense counsel and to otherwise gather missing 

information, Baldus corresponded Aiveotly with the prosecution 

and defense counsel. (Tr. 206, 587-88). 

22. Efforts were also made to include as much information 

as possible redatding the circumstances of the crime,.-aggravating 

and mitigating circumstances, through the use of the question- 

naires developed for the two studies, and the use of a case 

narrative which described the maior features of the case and any 

factors of importance noted in the records and not otherwise 

reflected in the questionnaires. (Tr. 238-39). 

23. Dr. Baldus made extensive efforts to assure uniformity 

in entry on the questionnaires for both studies. (Tr. 221-301: 

DB-34, 240-41). 

24. After the information for the 594 cases in the 

Procedural Reform Study as gathered, a single file with all of 

the cases was created and used for Dr. Baldus's analysis. {Tr. 

245-47). 

B. Charging Sentencing Study. 
  

25. Planning for the second study, the Charging and 

Sentencing Study, was begun in late 1980. (Tr. 258-261). 

 



  

26. The Charging and Sentencing Study (hereinafter, C&S 

Study) differed from the Procedural Reform Study in that it 

considered people who were convicted of murder or voluntary 

manslaughter and sentenced to state prison. {Tr.« 123). 

27. Secondly, the C&S Study looked at additional points in 

the decision-making process leading up to imposition or non- 

imposition of the death penalty, considering any differentials 

that appeared along racial lines in (i) the indictment decision, 

(11) the decision to permit a defendant to plead out to a lesser 

included offense, that is voluntary manslaughter, or to plead to 

murder in exchange for a waiver or a death penalty trial, and 

(iii) the jury's decision to convict of murder or to a lesser 

offense, as well as the decision points covered by the Procedural 

Reform Study. 

28. The second study also differed from the first in that 

-Baldus developed measures of the strength of the evidence. (Tr. 

126). 

29. One of the primary goals of the Charging and Sentencing 

Study was to focus on each individual step in the criminal 

justice decision making process and be able to estimate racial 

effects at each stage of the process, as well as overall. (Tr. 

261-62). 

30. The universe of cases studies for the Charging and 

Sentencing study was a sample of offenders convicted of murder or 

voluntary manslaughter whose crime occurred after March 28, 1973, 

and who were arrested prior to December 31, 1978. (Tr. 263-65). 

 



  

31. To assure a representative sample, a sampling plan was 

developed which stratified the cases, to assure representative- 

ness in terms of outcome (i.e., whether murder-death: murder-1life 

or manslaughter), and by judicial circuit within the State of 

Georgia. (Tr. 267-69). 

32. The sampling plan used was consistent with generally 

accepted procedures for survey samples. (Tr. 269). 

33. The questionnaire developed for the second study 

included additional information over and above that gathered in 

the first. Including in the expanded information was=additional 

material on aa ravAL Ln and mitigating circumstances (Tr. 274), 

expanded information regarding prior criminal record, (Tr. 274), 

and additional information regarding the strength ot the 

evidence. (Tr. 275). 

34. Extensive efforts were made to gather all relevant 

information related to the death-worthiness of the particular 

case in the Charging and Sentencing Study, drawing upon the 

experience gained in the prior Procedural Reform Study. (Tr. 

274-77). 

35. Date gathering for the second study began in May 1981 

{Tr. 308). | 

36. Included in the information gathered for the Charging 

and Sentencing study were such matters as the procedural history 

of the case, the charaes brought against the defendant and their 

disposition, characteristics of the defendant, prior record of 

the defendant, contemporaneous offenses, characteristics regard- 

 



  

ing the victim(s), special aggravating and mitigating circum- 

stances of the crime, information on co-perpetrators, number of 

victims, and strength of the evidence. (Tr. 380-85, DB-38). 

37. A narrative summary of the highlights of each case was 

prepared in the Charging and Sentencing Study, as was done in the 

Procedural Reform Study. {Tr . 290-91). 

38. For the Charging and Sentencing Study, the principal 

source of information concerning the offender, the offense and 

the victim was the files of the Georgia Department of Pardons and 

Paroles. {Tr. 293). 

39. Extensive efforts were made with the Charging and 

Sentencing Study to assure uniformity in the information gather- 

ing process. (Tr. 308-313). 

40. Defense counsel and prosecutors were contacted after 

the close of the data-gathering which occurred in Georgia in the 

summer in 1981, for the purpose of providing information not 

available in the Roard of Pardons and Paroles files, (Tr. 587- 

88): and race of victim information was obtained from the Bureau 

of Vital Statistics. (Tr. H91}). 

41. Once the data was gathered in Georgia for the Charging 

and Sentencing Study, it was entered onto a computer tape by 

personnel with the Political Science Laboratory of the University 

of Iowa. Substantial precautions were taken to assure the 

accuracy of the entry process. (Tr. 595-609). 

 



  

42. Extensive efforts were made to check the accuracy of 

the data relied upon by cross-checking information which was 

available for cases which appeared in both studies, as well as 

measuring for internal inconsistencies through frequency distri- 

bution analyses. (Tr. 615-16). 

Ce Data Gathering - Pardons and Parole Reports. 
  

43. The Board of Pardons and Paroles reports, which were 

relied upon by Dr. Baldus for development of factual information 

regarding the nature of the crime in his second study, including 

evidence of aggravating and mitigating circumstances, _.are 

developed for use by the Board of Pardons and Paroles in making 

decisions regarding parole, and the purpose of the reports is to 

provide as much relevant information as possible regarding the 

circumstances of the crime. (Tr. 1330, LwWw-1). 

44. Parole Board Guidelines in effect called for investi- 

gating officers to "extract the exact circumstances surrounding 

the offense. Any aggravating or mitigating circumstances must be 

included in the report." (LW-1, at $3.02). 

45. Among the public records routinely consulted by Pardons 

and Parole personnel in developing the reports are criminal 

records, clerk of court records, and police reports. (Tr. 1330~- 

31). 

46. The investigations by the Board of Pardons and Parole 

are routinely conducted shortly after a conviction. (Tr. 1331). 

 



  

47. In homicide cases, police officers who handled the 

case, as well as the district attorney, would routinely be 

consulted by Pardons and Parole investigators prior to writing 

their report. (Tr. 1332). 

48. As to circumstances of the offense, Pardons and Parole 

guidelines call for the information to be obtained from the 

indictment, the District Attorney's office, the arresting 

officers, witnesses, and victim. "A word picture, telling what 

happened, when, where, why, how and to whom, should be prepared." 

{Tr. 1336, L¥W~-1, 43.02,:%9). : - 

49. The Pardons and Parole guidelines also require that 

investigating officers be as thorough as possible with cases of 

persons convicted of more serious crimes, including the cate~ 

gories of convictions which were the subject of Dr. Baldus's 

study. {Tr. 1337). 

D. General Terms - Multivariate Study. 
  

50. The type of study used by Dr. Baldus and Dr. Woodworth 

in their studies, a retrospective examination of results, is a 

research design regularly used in scientific studies. (Tr. 

153-54.) 

51. The principal methods used by Drs. Baldus and Woodworth 

to control for backaround factors were cross-tabulation methods 

and regression methods. (Tr. 671). 

52. A background factor, as used in Dr. Baldus's analysis, 

is a factor that influences the outcome of interest. {152}. 

“310 

 



  

53. When one controls a background factor in multivariate 

analysis, one adopts a procedure which allows that factor (or 

factors) to be held constant with respect to the factor whose 

impact is being assessed. (Tr. 152). 

54. Multivariate procedures such as those used by Dr. 

Baldus and Dr. Woodworth in their studies will hold constant 

background factors and produce a situation where one can view all 

of the people considered in the analysis as being comparable in 

terms of specific factors having an obvious effect on the death 

sentencing results. (Tr. 143). a 

55. A cross tabulation method takes all the cases, and 

looks at two specific variables. The universe of cases is broken 

into four subgroups -- those being cases where both variables are 

present, cases where neither variable is present, and two other 

subgroups, each with one variable present and the other variable 

absent. {Tr. 683, DB=-64). 

56. Dr. Baldus and Dr. Woodworth used the last squares 

regression analysis and the logistic regression analysis through- 

out the Charging and Sentencing Study. (Tr. 671). 

57. The least squares regression analysis used by Dr. 

Baldus and Dr. Woodworth results in a regression co-efficient 

which is analogous to the arithmetic difference in death sentenc- 

ing rates for cases with and without the particular variable(s) 

being considered. (Tr. 670). 

-11- 

 



  

58. In contrast, the logistic regression procedure used by 

Drs. Baldus and Woodworth produces a regression coefficient 

‘which, upon final analysis, represents a measure of the number of 

times one's chances of receiving a death sentence are increased 

by the particular variable being considered. (Tr. 670). 

59. The value in a regression method is that it produces a 

single number which reflects the impact of the presence or 

absence of a variable in cases, while holding constant other 

background factors that need to be controlled for. (Tr. 692). 

60. The least squares regression coefficient represents a 

measure of the average increasing death sentencing rate across 

the cases with a variable present when compared with all cases 

lacking the particular variable. (Tr. 693-94). 

61 In the kind of least squares regression used by Baldus 

and Woodworth, an increase in the regression coefficient repre- 

sents a percentage increase in the probability that the death 

Penalty will be imposed. (Tr. 765). - 

62. A multiple regression analysis, through a method of 

algebraic adjustment, identifies subgroups of cases which are 

similar, and within those subgroups of cases, the racial effects 

are calculated. Conceptually, it is a process similar to trying 

to replicate, retrospectively, what would be done in a controlled 

experiment, so as to get cases where everyone is the same with 

respect to all factors, except the factor whose impact is being 

estimated. (Tr. 792-94). 

 



  

63. One of the purposes of multivariate analysis is to pull 

apart the causal effect of variables which are correlated. { TC. 

1780). 

E. Ceneral Statistics Regarding Homicides in Georgia. 
  

64. Petitioner's Exhibit DB-21 correctly traces the steps 

in the decision-making process with respect to processing of 

criminal charges arising out of homicides in the State of 

Georgia. (DB-21; Tr. 165-66). 

65. Petitioner's Exhibit DB-57 reflects the number of 

murders and non-negligent homicides in Georgia during _the period 

1974 through 1979, and the disposition of those cases. (Tr. 

629-633, DB-57). 

66. Petitioner's Exhibit DB-57 shows relative stability 

over this time period with respect to the observed outcomes. 

(Tr. 630-31). 

67. Petitioner's Exhibit DB-58 accurately depicts at each 

step of the criminal justice decision-making process the total 

number of cases remaining in the system for disposition. (Tr. 

634-638, DB-58). 

68. The figures in DB-58 suggest that a large proportion of 

the homicide cases in Georgia are disposed of at relatively early 

stages in the criminal justice system, and further, the most 

critical decision-making, in terms of risk of death sentence, is 

at the last two steps in the process--the decisions of prose- 

cutors to advance cases to penalty trial following a murder 

conviction and the decisions by juries at a penalty phase as to 

impose the death penalty or not. (Tr. 636-637). 

-13- 

 



  

69. Exhibit DB-58 accurately depicts the risk of receiving 

the death penalty at each of the stages of the criminal justice 

decision-making process. (Tr. 655-657). 

7. Race of Victim/Race of Defendant Effects - Statewide. 
  

70. The expert studies presented by petitioner show a 

pattern which runs throughout--that is, as the cases become more 

aggravated, because of the presence of aggravating circumstances, 

the death sentencing rate goes up. As the death sentencing rate 

rises, so does the disparity in sentencing based upon race of the 

victim. (Tr. 748). er 

71. The sentencing system in Georgia does not show uniform 

discrimination on the bases of race of defendant or race of 

victim; rather, in areas where most discretion is exercised by 

decision makers, that is the cases where aggravation levels are 

mid-range, the racial discrimination is greatest. (Tr. 841-42). 

72. Evidence of discriminatory application of the death 

penalty appeared, generally, in the middle range of cases with 

aggravated circumstances. In cases of very low aggravation, the 

death penalty was rarely imposed. And in cases with the very 

highest levels of aggravation, racial disparities in the 

frequency of imposition disappeared. (Tr. 777-78). 

73. Dr. Baldus's analysis shows that the race of victim is 

a factor which contributes to death sentencing outcomes in 

Georgia, even when the effect of the legitimate factors repre- 

sented by the statutory aggravating circumstances are considered. 

(Tr. 782-84, DB-78). 

lh 

 



  

74. Dr. Baldus's analysis shows that the race of victim has 

more impact on the sentencing outcome observed in the Georgia 

system than some of the statutory aggravating circumstances. 

(Tr. 786-87). 

75. In viewing overall death sentencing outcomes for 

defendants indicted for murder, Petitioner's Exhibit DB-78 

reflects the coefficients found for race of victim, race of 

defendant and the nine statutory aggravating factors, along with 

a variable for prior record, controlled for simultaneously. 

(DB-78}. 

76. When the nine statutorily-aggravating circumstances, 

and 75 mitigating circumstances are controlled for, race of 

defendant and race of victim disparities in sentencing outcomes 

remain evident. In fact, race of victim and race of defendant 

disparities appear stronger than when evidence of mitigating 

circumstances is not considered. (Tr. 797-99, DB-79). 

77. Petitioner's Exhibit DB-79 accurately sets forth the 

race of victim and race of defendant disparities, while control- 

ling for the 10 statutory aggravating factors and 75 mitigating 

circumstances. (DB-79). 

78. When sentencing outcomes across the universe are 

considered while controlling for nine important background 

factors, race of victim remains a strong influence on the system. 

The average likelihood that death would be imposed increased 7% 

with white victim cases when those nine background factors were 

held constant. The nine background factors considered in this 

phase of Dr. Baldus's study were (1) felony circumstances: (2) 

a 

 



  

serious prior record; (3) presence of a family-lover-liquor-bar 

room quarrel: (4) multiple victims; (5) whether the victim was a 

stranger; (6) whether the defendant was the triggerman; (7) 

whether there was physical torture; (8) whether there was mental 

torture, and (9) whether there was a serious aggravating circum- 

stance accompanying a contemporaneous offense. Those observed 

numbers are also statistically significant. (Tr. 801-802, DB- 

80). 

79. Similarly, when sentencing outcomes across the universe 

are considered while controlling for more than 230 nopn-racial 

background factors, race of victims remains a strong influence on 

the system. The average likelihood that death would be imposed 

increases 6% with white victim cases when those nine background 

factors are held constant. That observed outcome is statistic- 

ally significant. (Tr. 802-804, DB-80). 

80. The actual race of victim effects in the cases in the 

middle-range of aggravation (where the racial effect manifests 

itself) are substantially higher than the average 6-7% observed 

across the universe. This is because the less aggravated cases 

suppress the average, as no racial effect is shown in those 

cases. {Tr.. 503). 

81. Dr. Baldus ran a number of alternative studies of the 

data, using a number of different background variables which he 

believed, given his understanding of the criminal justice system, 

might offer an alternative explanation for sentencing outcomes, 

without race of victim continuing as an operative force in the 

i Gow 

 



  

system. In each of the alternative studies by Dr. Baldus, race 

of victim remains a strong influence on the sentencing outcome. 

(Tr. 808-809, R!20-31, DB-30, DB-83.) 

B2, ‘During the course of the trial, the District Court 

itself developed a "Lawyer's Model" of variables which, in the 

Court's view might explain in legitimate, non-discriminatory 

terms the functioning of the sentencing system. (Tr. 1475-76). 

However, when that model was used, race of victim and race of 

defendant were strong influences on sentencing outcome. (Baldus 

affidavit, filed 9/16/83.) = 

83. When Dr. Baldus selected 39 background factors which 

other studies suggested were the most plausible non-racial 

explanations for death sentencing results, and held those factors 

constant, race of victims effect remained strong and statistic- 

ally significant. (Tr. 815-819, DB-81, DB-82). 

84. Dr. Baldus found that the statutory aggravating 

circumstances contributing the largest number of death sentencing 

in Georgia were the B-2 and B-7 factors. When those cases are 

analyzed separately, strong race of victim effects appear in 

statistically significant numbers (increasing the probabilities 

of a death sentence, across the universe, 6%$-10% if race of 

victim was white). Significantly, in B-7 cases, where decision- 

maker discretion is relatively high, given the subjective nature 

of the B-7 circumstance, race of defendant effects appear high 

and in statistically significant numbers. In B-7 cases, race of 

a Bi 

 



  

defendant discrimination increases the probabilities of the death 

sentence, across the universe of B-7 cases, 13%. {DBR-85: Tr. 

839-42, B55). 

85. When B-2 cases are categorized into different groups, 

based upon level of aggravation, the greatest race of victim 

effect exists with black defendants. In those situations, the 

likelihood of death being imposed increases 24 to 36 percentage 

points, when a black defendant is charged with killing a white, 

as opposed to hlack victim. 

86. Similarity, when B-2 cases are categorized by level of 

aggravation, race of defendant disparities exist generally in 

white victim cases, with black defendants receiving the death 

sentences at a rate generally ranging from 5 to 32 percentage 

points greater than white defendants. {DR-87, Tr. 872-75): 

87. A study of cases at the highest levels of aggravation, 

controlling for relevant background factors, show statistically 

significant race of victim disparities. (DB-89, Tr. 876-80). 

88. When the 472 most aggravated cases are themselves 

grouped according the likelihood of receiving the death penalty, 

aiven the background factors present, race of victim disparities 

in the range of 16% to 33% exist. ‘It is in this category of 

cases that the evidence showed decision-maker discretion was 

being exercised, and being exercised in a racially discriminatory 

manner. (Tr. 8381-884, DBR-90). 

“Yo 

 



  

89. Similarly, when the 472 most aggravated cases are 

considered, race of defendant disparities exist in statistically 

significant numbers, enhancing the probability the death sentence 

will be imposed in white victim cases 15% to 27%. (DB-91, Tr. 

885-86). 

90. Although white victim cases are themselves generally 

more aggravated, that does not explain the racial effects shown 

in the system. When black and white victim cases with similar 

legitimate aggravating circumstances are considered, the results 

show that the System. teenonds more severely to white victim cases 

then to comparable black victim cases with the same legitimate 

aggravating circumstances. (DB-92, Tr. 887-893). 

91. The evidence shows that substantial race of victim 

effects are caused by the prosecutor's decision whether or not to 

seek a death penalty after a quilt verdict is returned on a 

murder indictment. Race of defendant disparities also exist at 

that decision-making point in the process. (DB-95, Tr. 906-14). 

92. Additional evidence exists showing that race of victim 

discrimination is contributed by the jury decision on whether to 

impose the death penalty or not. (DB-95, Tr. 906-14, DB-97, Tr. 

934). 

93. Dr. Baldus's analysis shows that the race of victim and 

race of defendant effects remain even after the appellate 

function of the Georgia Supreme Court is considered. {Tr. 953 

56). 

-19- 

 



  

94. The evidence shows that the pattern of race of victim 

and race of defendant discrimination is not isolated to any 

particular time period within the 1973-79 period studies, but 

rather, remains present throughout that period. {DR-103, Tr. 

961-64). 

95. The evidence shows that black and white victim cases at 

the same level of aagravation exhibit different death sentencing 

rates. {Tr. 3207, GW~-5). 

96. The "midrange" model, in which petitioner's experts 

have the greatest ehnridencs. shows that, on average,.-the white 

victim-black defendant case is exposed to an additional 7 to 15% 

higher death sentencing rate than if the victim were black. (Tr. 

1294, 1300-1302, GW-5). 

07 Existing technical literature in the area of criminal 

sentencing suggests that the retrospective study conducted by 

Baldus would likely understate racial effects. (Tr. 17683). 

G. Race of Victim/Race of Defendant Effect in Fulton 

County. 
  

98. As for Fulton County, the jurisdiction in which the 

petitioner was sentenced, the evidence shows that the likelihood 

of a death sentence is higher in white viotin cases than in black 

victim cases. (Tr. 978-993, DB-1046, DB-109). 

99. When thirty-two of the most aggravated Fulton County 

cases are studied, the evidence shows disparities based upon race 

of the victim within groups of cases with similar levels of 

aggravation. (DR-109, Tr. 992-914) 

-D 

 



  

100. The Fulton County disparities reflect patterns similar 

to that found statewide, including evidence that disparities are, 

in part, the result of prosecutorial discretion to advance the 

case to a penalty trial after the return of a guilty verdict at a 

murder trial. (Tr. 1002-1005, DB-111)., 

101. The evidence showing racial effect in Fulton County 

sentencing is supported by a regression analysis taking into 

account the non-racial factors with a statistically significant 

relationship to outcome at each particular stage in the process. 

(DB~-114, Tr. 1037-43). an 

102. In Fulton County since 1973, there have been 10 police 

homicides, with ChArges in those homicides brought against 17 

defendants. Only one person, the petitioner, has been sentenced 

to death. (Tr. 1051-1052, DR-115). 

103. Of seven persons, including petitioner, who were 

accused of being involved in a serious contemporaneous offense, 

as well as bheing the triggerman resulting in the homicide of a 

police officer, three of those charged pled guilty to murder, and 

no penalty trial followed the guilty plea; two went to trial on 

murder charges and were convicted, but no penalty trial was held 

thereafter, and two were convicted and had a penalty trial. The 

one case which went to a penalty trial and life sentence was 

imposed involved a black victim; the one case which went to a 

penalty trial and a death sentence was imposed [petitioner's] 

involved a white victim. (Tr. 1050-57: DB-115). 

-d 

 



  

104. No written guidelines exist which set forth standards 

for decision-making within the Fulton County District Attorney's 

office regarding the disposition of death-eliaqible cases from 

indictment through the decision to seek a death penalty after a 

jury returns a verdict of quilty in a murder trial. (Lewis 

Slaton Deposition, 10-12, 26, 31, 41, 58-59). 

105. Murder cases in the Fulton County District Attorney's 

office are assigned at different stages to one of a dozen or more 

assistant district attorneys, and there is no one person who 

invariably reviews all decisions on homicide disposition. 

(Slaton Deposition, 15, 45-48, 12-14, 20-22, 28, 34-38). 

106. The decision-making process in the Fulton County 

District Attorney's office, with respect to seeking a death 

penalty, has changed little or none from the pre-Furman time 

period. (Slaton Deposition, 59-61). 

107. The District Attorney who prosecuted Warren 

McCleskey's case was white (Deposition 43, Exhibit P-1), and 

generally, all but one or two of the trial district attorneys 

responsible for handling death cases in Fulton County since 1973 

have been white. (Slaton Dep., 43-50). 

108. All but three of the judges in Fulton Superior Court 

since 1973 have been black, and during the 1973-79 time period, 

not more than one of the eleven judges on the Fulton Superior 

Court bench at one time was black (Slaton Deposition, 52-58, 

Exhibit P-1). The judge who presided at Warren McCleskey's trial 

was white. (Deposition Tr. 52-58, Exhibit P-1). 

-22- 

 



  

H. Warren McCleskey's Case And Racial Effects Observed In 

Cases Of Comparable Aagravation. 
  

109. Petitioner Warren McCleskev, who is black, was tried 

by a jury consisting of eleven whites and one black. (Tr. 1318). 

The persons seated to hear petitioner's case, (including one 

white juror who was excused) were twelve whites and one black. 

(Tr. 136). 

110. The evidence showed that at the time of Warren 

McCleskey's trial, the population of Fulton County was 47-50% 

black (Tr. 1776-78). The probability that a panel of_.jurors 

which was eleven white and one black would be drawn from such a 

population, by chance, is approximately .005. The probability 

that a panel of jurors which was twelve white and one black would 

be drawn from such a population, by chance, is approximately 

003, Tr. 1777). 

111. Petitioner Warren McCleskey's case fell within a class 

of cases where there is roughly a twenty percentage point 

disparity between black victim cases and white victim cases in 

sentencing outcomes. {T™r. 1735, GW=83). 

112. Warren McCleskey experienced two increments in the 

probabilities that a death sentence would be imposed in his case, 

over and above the average homicide case. One increment occurred 

because of the aggravated nature of the crime; the other 

increment occurred because the victim was white. The increment 

incurred because the victim was white was comparable in maanitude 

to the increment observed because of the aggravating circum- 

stances. (Tr. 1744-45, GW-8). 

Fo 

 



  

I. Confidence In The Observed Results. 
  

113. Drs. Baldus and Woodworth used an approach to their 

data termed "triangulation." That approach suggests that if a 

researcher employs a variety of different methods to analyze the 

same question, and each of the different methods reaches a 

similar conclusion, one has greater confidence in the conclusion 

reached. In this case, two different data sets were used, though 

similar results were generally obtained. Further, different 

rearession analyses were used, focusing upon varying subsets of 

cases at varying stages in the decision-making process, with 

generally a similar conclusion reached, which is, that in the 

mid-range of the highly-aggravated cases, where the greatest 

amount of discretion is being exercised, racial effects are 

evident. (Tr. 1081-1083, 1736-40). 

J. Diagnostic Tests. 
  

114. Appropriate statistical tests were conducted upon 

regression coefficients found in Dr. Baldus's and Dr. Woodworth's 

study to assure that the observed disparities did not occur by 

chance. These statistical tests were a method of testing the 

rival hypothesis that the observed results occurred by chance. 

(Tr. 1244-46). 

115. Statistical tests, or diagnostic measures, were also 

run to assure that the observed results were not due to errors in 

statistical technique or to bias built into the regression 

analyses used. {GW-40, Tr. 1248-52, 1265, 1300). 

wie 

 



  

116. Diagnostic tests run assured that the results obtained 

showing race of victim discrimination were not a result of the 

weighting system used in the analysis. (Tr. 1253-54, 1711-1715, 

1727). 

117. A number of different analyses were conducted which 

assured that unknown information did not alter the racial effects 

observed. In those analyses, race of victim effects remained 

strong and statistically significant. (GW-4, Tr. 1255-1256, DB- 

120-124, Tr. 1694-1708). 

718. Diagnostic tests also showed that the race of victim 

effects were not caused by a few anamalous cases, but rather, 

were systemic. (GW-4, Tr. 1256-60). 

K. State's Objections. 
  

119. The State did not demonstrate or seek to demonstrate, 

that the data on Dr. Baldus's tapes used for his analyses did not 

substantially reflect reality. (Tr. 652). 

120. Tests run by petitioner's experts indicated that the 

respondent's criticisms of Dr. Baldus's coding practices did not 

affect the race of victim coefficients. (Tr. 1677-1678). 

121. The evidence showed that, to the extent there were 

errors in the data base, they were likely to have occurred on a 

random basis, and therefore, could not have created the race of 

victim or race of defendant effects observed. {Tr. 1727-28). 

122. Dr. Richard Berk, one of petitioner's expert called in 

rebuttal, indicated that, in his experience, if missing data were 

a problem, two things would have happened that were not observed 

in Baldus's and Woodworth's work: (1) the aggravating and 

-25- 

 



  

mitigating circumstances wouldn't have shown effects on outcome; 

and (2) very minor changes in which variables were included would 

have resulted in flipping the important effects from positive to 

negative and positive back again. (Tr. 1764-65). 

123. Dr. Berk further testified that in other research 

studies he has observed in the criminal justice area, missing 

data of the magnitude of 10 to 15 percent almost never makes a 

difference in the analysis. (Tr. 1766). The missing data in 

Baldus's study were well below this range. (Tr. 1766). 

124. In comparison to the hundreds of studies on sentencing 

reviewed by Dr. Berk, he was of the opinion that Dr. Baldus's 

study was far and away the most complete and thorough. (Tr. 

1766). 

L. Multicolinearity. 
  

125. To the extent that race of victim and level of 

aggravation are related to each other (i.e., colinear), the 

effect of such on the coefficients would be to produce a lower, 

rather than higher, race of victim effect. The existence of a 

statistically significant race of victim coefficient, even though 

there is some relationship between race of victim and level of 

aggravation, indicates that the race of victim effect cannot be 

explained by the fact that white victim cases tend to be more 

aggravated. {Tr. 1281-1283, 1659), 

126. The major risk in a study with many variables which 

are correlated with each other (i.e., there is much "multi- 

colinearity") is that it will be difficult to distinguish the 

“dw 

 



  

regression co-efficients from chance; in other words, one loses 

statistical power. But, the regression LR SRL 

will be unbiased. (Tr. 1782). 

M. Summary. 

127. On the basis of Dr. Baldus's study of sentencing 

patterns in the State of Georgia, he was of the opinion that: 

(1) systematic and substantial disparities exist in the 

penalties imposed upon homicide defendants in the State of 

Georgia, based upon the race of the homicide victims: 

(2) disparities in death sentencing rates do exist based 

upon the race of the defendant, but they are not as substantial 

and not as systematic as the race of victim disparities: 

(3) disparities in both race of victim and race of: 

defendant persist, even when the aggravating circumstances 

defined by Georgia's capital punishment statute are taken into 

account; 

(4) these disparities exist even when one controls 

simultaneously for statutory and non-statutory aggravating and 

mitigating circumstances and measures of the strength of the 

evidence: 

(5) that race of victim disparities persist within the 

jurisdiction of Fulton County, Georaia, wherein petitioner Warren 

McCleskey was tried; 

(6) that other leaitimate factors not controlled for in 

Dr. Baldus's analysis could not plausibly explain the persistence 

of these racial disparities in the State of Georgia and in Fulton 

County: and 

“37 

 



  

(7) that racial factors have a real effect in the capital 

charaing and sentencing system of the State of Georgia and in 

Fulton County. {Tr. 725-728, DBE-12). 

128. The Court finds the opinions of Dr. Baldus to be 

supported by the evidence in this case. 

Respectfully submitted, 

ROBERT H. STROUP 

1515 Healey Building 
Atlanta, Georaia 30303 

JACK GREENBURG 

JOHN CHARLES BOGER 

10 Columbus Circle 
New York, New York 10019 

TIMOTHY XX. PORD 

600 Pioneer Building 
Seattle, Washington 98136 

ANTHONY G. AMSTERDAM 

New York University Law School 
40 Washington Square South 
New York, New York 10012 

Baer 70 cep. 
ROBERT H. STROUP 
  

ATTORNEYS FOR PETITIONER 

«38