SCOTUS, No. 84-6811 - Social Science Amicus Brief Vol. 2 of 2 (Redacted)
Working File
July 1, 1986 - August 1, 1986
85 pages
Cite this item
-
Case Files, McCleskey Background Materials. SCOTUS, No. 84-6811 - Social Science Amicus Brief Vol. 2 of 2 (Redacted), 1986. f3830561-6acc-ef11-8ee9-7c1e5218011c. LDF Archives, Thurgood Marshall Institute. https://ldfrecollection.org/archives/archives-search/archives-item/23931eb3-af5b-499d-a4e6-08dfecd69381/scotus-no-84-6811-social-science-amicus-brief-vol-2-of-2-redacted. Accessed June 17, 2025.
Copied!
IN THE UNITED STATES COURT OF APPEALS
FOR THE ELEVENTH CIRCUIT
No. 84-8176
WARREN McCLESKEY,
Petitioner-Appellee, and
Cross-Appellant,
-against-
WALTER D. ZANT, Superintendent,
Georgia Diagnostic & Classification Center,
Respondent-Appellant, and
Cross—-Appellee.
ON APPEAL FROM THE UNITED STATES DISTRICT COURT
FOR THE NORTHERN DISTRICT OF GEORGIA
ATLANTA DIVISION
— ——— —— —— —— — — i WED WE WT WE — — —— — ——— CD — —— ——— —)—— — — w= —— wes wm we ——- ———
EE EE EE TE IE TE Tu Th Son wn nin — — — —— — — ——— ————— — — — — — — —— —— ——— — ——
EN BANC BRIEF FOR PETITIONER McCLESKEY
AS APPELLEE AND CROSS-APPELLANT
(HABEAS CORPUS)
—— —— — — —— — — — —— — —— — — — —— — — — = ——— 1 ——— —— = —— —— ln ——h = ww ctw w=. = =
EE EE EE EE TE EE EE ER EE NE EE ER EE EE EE Cm =m von =o 0 =n =m wos = — —. — — — —— — — — — — —— ———— — — —
ROBERT H. STROUP
L/ 1515 Healey Building
Aeoktery Atlanta, Georgia 30303
7) JACK GREENBERG
/ JAMES M. NABRIT, III
7 JOHN CHARLES BOGER
.- id 99 Hudson Street
New York, New York 10013
TIMOTHY K. FORD
600 Pioneer Square
Seattle, Washington 94305
ANTHONY G. AMSTERDAM
New York University Law School
40 Washington Square South
New York, New York 1012
ATTORNEYS FOR PETITIONER-APPELLEE
AND CROSS-APPELLANT
tions of the District Court's judgment denying relief on all
other claims should be reversed. Alternatively, the case should
be remanded to the District Court for a further evidentiary hear-
ing on petitioner's claims respecting the ineffective assistance
of counsel at trial, the arbitrary and racially discriminatory
»
imposition of his capital sentence, and the prosecution-proneness
4 of his trial jury.
: Dated: May 8, 1984 Respectfully submitted,
TIMOTHY K. FORD ROBERT H. STROUP
600 Pioneer Square 1515 Healey Building
Seattle, Washington 94305 Atlanta, Georgia 30303
ANTHONY G. AMSTERDAM JACK GREENBERG
New York University JAMES M. NABRIT, III
Law School JOHN CHARLES BOGER
40 Washington Square South 99 Hudson Street
New York, New York 10012 New York, New York 10013
ATTORNEYS FOR PETITIONER~APPELLEE
~ & CROSS-APPELLANT
ii a rt
By tee hiatal
te
V
a
n
a
¥
a 5
CERTIFICATE OF SERVICE
I hereby certify that I am one of the attorneys for
petitioner-appellee Warren McCleskey, and that I served the
annexed En Banc Brief for Petitioner McCleskey as Appellee and
Cross-Appellant on respondent by placing copies 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
All parties required to be served have been served. Done this
8th day of May, 1984.
” »
in C7 A
rd
go [UN en ate 2 Fi
\ 7 JOHN CHARLES BOGER
STATEMENT REGARDING PREFERENCE
These are cross-appeals by the parties from the judgment of
the United States District Court in a habeas corpus action,
entered February 2, 1984, seeking relief under 28 U.S.C. §§ 2241-
2254 from the judgment of a state court.
The cross-appeals should be given preference in processing
and disposition pursuant to Rule 12 and Appendix One (a)(3)
of the Rules of this Court.
STATEMENT REGARDING ORAL ARGUMENT
The Court directed oral argument en banc in its order of
March 28, 1984.
q %
Y
< re
-
»
x
)
Statemen
Statemen
Statemen
Statemen
(1)
(ii)
(iii)
Summary
Statemen
Argument
I.
Il.
111.
Iv.
TABLE OF CONTENTS
Regarding Preference ....:civrrvsvneevs
t Regarding Oral Argument ...icsesevese oy
t of the Issues Presented for Review ....
tof the Case ....ccceve Fe RAE a ie
Course of Proceedings in the State
Courts and in the Court Below ,....cesver
SEAL OMANT OF PaCS vt sc tv ts vc edinnissimcnns
Statement Of ReViGW ... ovis srsreniinsonse
of ‘Argument... ec... Gis wun ainiaieinis vane sy es
of of Jurisdiction 0. ® 9 ¢ O 0 0 0.0 0 0 0-0 O 0 0. 0 * oo ° 0
The District Court Correctly Concluded
That The State's Failure To Disclose To
Petitioner's Jury A Promise By Atlanta
Police Detective Dorsey To Witness Offie
Evans In Exchange For His Critical Test-
imony Against Petitioner Contravened
The Due Process Clause Of The Fourteenth
ANen@mant i... ees saiviires TI Pe EE Sr
A, Facts Supporting The Giglio Claim ...
B.. The Legal Standard ........ FER ING ELE
The district Court Erred When It Rejected
Petitioner's Claims That Trial Counsel
Failed To Render Effective Assistance Of
Counsel Jo. cv veo cee sa ieinnies iets sie eieis
The District Court Erred When It Rejected
Petitioner's Claim That His Jury Instruc-
tions Contravened The Due Process Clause.
In Rejecting Petitioner's Claim That
Substantial Racial Disparities Persist In
Georgia's Capital Sentencing System, The
District Court Misapplied Controlling
Legal Precedent, Misunderstood Basic
Statistical Principles, and Ignored Clear
Becord BV idence vans rr arse rinsss
A. Introduction: Petitioner's Constitu-
ional Cla imMS i i sear serena
10
12
24
25
25
Ba.
47
L
4
Y
D.
TABLE OF CONTENTS
Page
The Facts: Petitioner Has Made Out A
Compelling Prima Facie Case of Racial
Discrimination In Capital Sentencing 27
(i) Petitioner's Experts Were Well-
Qualified i. tics sce ra 27
(ii) Petitioner's Data-Gathering
Effort Was Carefully Conducted 31
(a) The integrity of the data
SOULCSS cts vivetesownosnsies 32
(b) The quality of the data-
gathering instrument ....... 36
(c) The care employed in
COBING ci ies vei chica 39
(d) The basic coding conven-
ol sb ERR aR I Cebus 41
(iii) The Statistical Methods Were
Valid and Appropriate .......... 44
(iv) The Results Make Out A Prima
Facie Case of Racial Discrimi-
RAEION iis esse tis ese 46
The Law: The District Court Misapplied
The Law In Rejecting Petitioner's Prima
PACLe CBSE i vis sins nsnivinnesvnnnos
(1)
(ii)
Petitioner's Data Clearly Exceed
Minimum Evidentiary Standards For
Use in Regression Analysis .....
Multiple Regression Analysis Is
An Appropriate Means of Proof ..
(iii) The Models Employed by Peti-
tioner Were Statistically
ADDO IBLE . ese savcies ancien
The State Presented No Significant
RebuttralrCase ive ieie ot onavssese
V. The District Court Improperly Rejected
Petitioner's Prosecution-Proneness Claim
Conclusion
LA I SE A A EE EE I JE REN BEY EN RE NEE EE REE BEE REN RE NEE UE NEY NEY REE TEE REY NEY NE NEE NEE NEY TY TEE NE TE NE
56
57
39
63
67
70
70
TABLE OF AUTHORITIES
Page
Cases:
Adams v. Wainwright, 709 F.2d 1443 (llth Cir. 1983) 27
Arlington Heights v. Metropolitan Housing Authority, 429
eS. 252: (1977) 27
Arnold v., State, 236 Ga. 534, 224 S.E.24 3856 (1776) 14
Bell v. Georgia, 554 P,24 1360 (5th Cir, 1977) 19
Castaneda v. Partida, 430 U.S, 482 (1977) 69
Columbus Board of Education v. Penick, 443 U.S. 449
{1979) 26
Curran v. Delaware, 259 F.2d 707 (1958) 12
Cuyler v, Sullivan, 446 U.S. 335 (1980) 6
Detroit Police Officers' Ass'n v. Young, 608 F.2d 671
(6th Cir. 1979), cert. denied, 452 U.S. 938 (1981) 58,68
Douglas v, Wainwright, 714 F.2d 1432 (llth Cir. 1983) 18
Eastland v. Tennessee Valley Authority, 704 F.2d 613
{llth Cir. 1983) 8,46
: 81,63,67
Enmund v, Florida, 458 U.S. 782 (1982) 34
Franklin v, Francis, 720 F.2d 1206 (llth Cir. 1983) 6
Preeman v. State of Georgia, 399% P.24 65 (Sth Cir.
1979) 12
A Purman v, Georgia, 408 U.S. 238 (1972) 25
3 Gaines v. Hopper, 375 P.2d 1147 (5th Cir, 1978) 19
Y Giglio v. United States, 405 U.S. 150 {1971) 1,5,6,6
: 32,20,70
Godfrey v. Georgia, 446 U.S. 420 (1980) 8,26
Goodwin v. Balkcom, 684 F.28 794 (llth Cir, 1932),
cert. denied, 0.8. 103 S.Ct. 1798 (1933) 17,18
Gregg v. Georgla, 428 U.S. 153 (1978) 25,29,3)
3 House v,. Balkcom, 725 P.2&8 608 (11th Cir. 1984) 18
International Brotherhood of Teamsters wv. United
States, 431 U.S, 324 (1977) 58
Jackson v, Virginia, 443 U.S. 307 (1979) 6
Johnson v. Uncle Ben's Inc., 628 F.2d 419 (5th Cir.
1980), cert. denied, B.S. 103 8.Ct. 293 (1982) 8,67
Jurek v. Estelle, 623 F.2d 929 (5th Cir. 1980) (en
banc) 6
Ring v. Strickland, 714 P.28 1481 (llth Cir, 1983) 18
Maxwell v. Bishop, 398 F.28 138 (8th Cir. 1968), rev'd
on other grounds, 398 U.S. 262 (1970) 29
Moore v. Zant, 722 P.24 640 (llth Cir. 1983) 12
Napue v. Illinois, 360 U.S. 264 (1959) 12
Nero v. Blackburn, 597 F.24:991 {sth Cir, .1979) 21
Palmes v. Wainwright, 725 F.2d 1511 (llth Cir. 1984) 6
Proffitt v. Wainwright, 683 F.2d 1227 (llth Cir, 1982) 26
Pullman-Standard v. Swint, 456 U.S. 273 (1982) 6
Pyle v. Kansas, 317 U.S. 213 (1942) 7.12
Rogers v. Lodge, 458 U.S. 613 (1982) ; 26
Rummel v. Estelle, 590 F.24 103 (3th Cir. 1979) 17
Schneider v. Estelle, 352 F.2d 393 (5th Cir. 1977) 12
Scott v. Wainwright, 698 F.2d 427 (llth Cir, 1983) 18
Smith v. Balkcom, 660 P.24 573 (52h Cir. Unit B (1981) 3
Smith v. Balkcom, 671 F.28 858 (5th Cir, Unit B 1982)
(on rehearing) 27
Smith v. Remp, 715 F.2d 1459 (llth Cir. 1983) 12
smith v. State of Plorida, 410 P.24 1349 (5th Cir.
1969) 12
Spencer v. Zant, No. 82-8408 (llth Cir.) 26
Spinkellink v, Wainwright, 578 P.2d4 ‘582 (5th Cir, 1978) 5,8,25
Trout vv. Lehman, 702 F.24 1094 (D.C, Cir. 1983) 58,68,69
Dnited States v. Antone, 603 P.24 566 (5th Cir. 1379) 14
-"vi =
United States v. Barham, 625 PF.24 1221 (5th Cir. 1980) 14
United States v. D'Antignac, 628 F.2d 428 (5th Cir. 1980) 14
United States v. Nixon, 634 F.2d 306 (5th Cir. 1981) 14
United States v. Poitier, 623 F.24 1017 (5th Cir. 1980) 14
United States v. Tucker, 404 U.S. 443 (1972) 21
Valentino v. United States Postal Service, 674 F.2d 56
(D.C. Cir. 1982) 63
Von Moltke v, Gillies, 332 U.S. 708 (1948) 17.
Vuyanich v. Republic Nat'l Bank of Dallas, 505 F. Supp.
224 (N.D. Tex. 1980), vacated, 723 P.24 1195 (Sth
Cir. 19384) 36,58
61,68
Wade v. Mississippi Cooperative Extension, 528 F.2d 508
{Sth Cir. 1976) 62
Washington v. Davis, 426 U.S. 229 (1976) 27
Washington v, Watkins, 655 F.28 1346 (5th Cir. Unit A 1981) 18
Weidner v. Wainwright, 708 P.24 614 (llth Cir. 1983). 18,19
Wilkins v, University.of Houston, 654 F.2d 383 (5th Cir.
1981), vacated & remanded, v.85. _ , 103 8.Ct.
34 (1932) 62
Williams v, Brown, 609 F.2d 216 (5th Cir. 1980) 12
Yick Wo v. Hopkins, 118 U.S. 356 (1886) 8,26
Young v., Zant, 677 F.2d 392 {llth Cir. 1982) 19
Zant v. Stephens, __u.s. Ces 103 8.00. 2733 (1983) 21
Zant v. Stephens, 456 U.S. 410 (1982) 8,26
Statutes:
PR. Civ. P. 52a) 8
Former GA. CODE ANN. § 77-512 33
0.C.G.A. § 15-1~-314) 20
o.C.G.A. § 17-10~2 20
0.C.C.A. § 42-55-50 20
- vii -
Other Authorities:
Baldus, "Welfare as a Loan: An Empirical Study of the
Recovery of Public Assistance Payments in the United
States,” 25 STAN. L., REV. 123 (1973) 28
Baldus & Cole, "A Comparison of the Work of Thorstein
Sellin and Isaac Ehrlich on the Deterent Effect of
Capital Punishment," 85 YALE L.J. 170 (1975) 29
D. BALDUS & J. COLE, STATISTICAL PROOF OF DISCRIMINATION,
(1980) 28
Baldus, Pulaski & Woodworth, "Proportionality Review of
Death Sentences: An Empirical Study of the Georgia
Experience," J. CRIM. L. & CRIMINOLOGY (forthcoming
1984) 29
Baldus, Pulaski, Woodworth & Kyle, "Identifying Compara-
tively Excess Sentences of Death," 33 STAN. L. REV.
801 (1977) 29
Bowers & Pierce, "Arbitrariness & Discrimination under
Post-Furman Capital Statutes," 26 CRIME & DELINQ. 563
(1980) 46
Finkelstein, "The Judicial Reception of Multiple Regression
Studies in Race and Sex Discrimination Cases," 80 COLUM.
Li. REV. 737 (1980) 63
Fisher, "Multiple Regression in Legal Proceedings," 80 COLUM.
L. REV, 702 (1980) 61
65,66
J. GUILFORD & B. FRUCHTER, FUNDAMENTAL STATISTICS IN
PSYCHOLOGY AND EDUCATION, 6th ed. (1969) 66
Halinski & Feldt, "The Selection of Variables in Multiple
Regression Analysis, 7 J. EDUC. MEASUREMENT 151 (1970) 53
H. RKALVEN & H. ZEISEL, THE AMERICAN JURY (1966) 51
J. NETER & W. WASSERMAN, APPLIED LINEAR STATISTICAL MCDELS
(1974) 61
Special Edition, "A Study of the California Penalty Jury in
Pirst Degree Murder Cases,” 21 STAN. L, REV, 1297 (1969) ag
Taylor, "Analyzing Qualitative Data," in P. ROSSI, J. WRIGHT
& A. ANDERSON, eds., HANDBOOK OF SURVEY RESEARCH (1979) 61
Wolfgang & Riedel, "Race, Judicial Discretion and the Death
Penalty," 407 ANNALS 119 (1973) 29
- viii -
IN THE UNITED STATES COURT OF APPEALS
FOR THE ELEVENTH CIRCUIT
No. 84-8176
WARREN McCLESKEY,
Petitioner-Appellee, and
Cross-Appellant,
-against-
WALTER D. ZANT, Superintendent,
Georgia Diagnostic & Classification
Center,
Respondent-Appellant, and
Cross-Appellee.
STATEMENT OF THE ISSUES PRESENTED FOR REVIEW
Whether the State's failure to correct false testimony
by a key prosecution witness, denying that he had been
promised assistance in exchange for his testimony,
deprived petitioner of due process under Giglio wv.
United States, 405.0.8. 150 (1972) at his gullt and
sentencing trials for murder and armed robbery?
Whether the failure of petitioner's trial counsel to
conduct any meaningful pretrial investigation, even
though he was aware that potentially damaging State's
evidence might be introduced and that useful defense
leads remained to be pursued on both guilt and sen-
tencing issues, violated petitioner's Sixth Amendment
right to effective assistance of counsel?
et
3. VWnether the trial court's instructions on the critical
element of malice could have been viewed by a reasonable
juror as shifting the burden of proof to petitioner in
violation of Sandstrom v. Montana, 442 U.S. 510 (1979),
and whether in a case where malice was seriously in
dispute, such a constitutional error could ever be
harmless beyond a reasonable doubt?
4. Whether the District Court erred in dismissing as a
matter of law petitioner's Eighth Amendment Claim that
the death penalty in Georgia is being applied in an
arbitrary and capricious manner? |
5S. Whether the District Court misapplied well-established
legal standards, charting an errant legal course in
rejecting petitioner's comprehensive statistical evidence of racial discrimination in Georgia capital sentencing
system?
6. Whether the District Court erred in dismissing petitioner's
challenge to the death-qualification of his trial jury
as a matter of law?
STATEMENT OF THE CASE
i (1) Course of Proceedings In The State Courts And In The
- Court Below
Petitioner Warren McCleskey, a young black man, was convicted
in the Superior Court of Fulton County, Georgia on October 12, 1978
-- by a jury comprised of eleven white jurors and one black juror
{Fed., Hab, 7T. 1316). -—- of the murder of Police Officer Frank
1/ Each reference to the transcript of the evidentiary hearing
(Continued)
Schlatt, and of two counts of armed robbery. (Petitioner McCleskey
alone, among seventeen defendants indicted in Fulton County between
1973 and 1980 for the murder of a police officer, received a death
sentence for his crime.) (Fed. Hab. Tr. 1050-52: DB 115, 116.)2/
He was sentenced to death for murder and to separate life sentences
for armed robbery.
After state habeas corpus proceedings had been completed, peti-
tioner filed a federal habeas corpus petition in the Northern
District of Georgia, Atlanta Division, on December 30, 1931. The
District Court entered an order on June 10, 1982 denying a motion
for an evidentiary hearing, and dismissing the petition without
prejudice, citing petitioner's ostensible failure to exhaust one
constitutional claim (R. 248). Petitioner filed a timely motion
to alter or amend the judgment (R. 261), and on October 8, 1982,
the District Court entered a further order, directing an evidentiary
hearing (R. 475).
Upon the State's motion (R. 498), the hearing was continued
while respondent Zant pursued discovery. On April 7, 1983, peti-
tioner also filed a motion for discovery, seeking to serve inter-
rogatories, to obtain production of documents, and to depose re-
in the United States District Court for the Northern District of
Georgia, Atlanta Division, on August 8-22, 1983, wil be indicated
by the abbreviation "Fed. Hab. T." Each reference to the transcript
of petitioner's state trial will be indicated by the abbreviation
Le Tr." References to the transcript of petitioner's state
habeas corpus hearing will be indicated by the abbreviation "St.
Hab, Tr."
2/ Each reference to petitioner's exhibits introduced during
the federal evidentiary hearing will be indicated by the initials
of the witness through whose testimony the exhibit was introduced
(e.g9., Professor David Baldus becomes "DB") followed by the
exhibit number.
°
spondent's Sinents. {(R. 541.) On June 3, 1983, the Digtrict
Court entered an order which: (i) granted the depositions peti-
tioner had sought; (ii) directed respondent to list all objections
or criticisms of petitioner's data base, and analyses; but (iii)
denied as "not relevant to any issue presented by the petitioner"
those interrogatories which sought information about discrimination
in the criminal justice system and about prosecutorial plea bargaining
: 4/
practices.”
Subsequently on July 21, 1983, petitioner was forced to file
a motion to compel respondent to respond in full to his request
for all objections to the Baldus studies. Thereafter, petitioner
filed a request for respondent to admit the accuracy of the data upon
which Professor Baldus conducted his analyses, except insofar as
respondent had noted objections (R. 593).
During a pretrial conference held on July 29, 1983, the
3/ continued
cases in which petitioner lacked this information; (ii) on the
race of the victim in designated cases; (iii) on whether a penalty
that had been held in designated cases; (iv) on all "private, judi-
cial, administrative or agency challenges made against [Fulton] County
grant or traverse juries" on grounds of racial discrimination, as well
as all actions challenging prosecutorial use of premptory challenges,
all challenges to police or prosecutorial actions in Fulton County
because of alleged racial discrimination in the administration of
criminal justice; and (v) on the percentage of blacks employed in a
range of city, county and state positions related to the criminal
justice system (R. 556-62). Petitioner also requested respondent to
"[l]ist all objections, criticisms or deficiencies of the data base,
date-gathering methods, analyses and conclusions" of petitioner's
expert Professor David Baldus (R. 566).
4/ The Court also initially denied as irrelevant petitioner's
request for access to documents revealing Fulton County prosecu-
torial practices and rationales in homicide case (R. 596).
Subsequently, in a pretrial conference, the Court reconsidered
this portion of its June 3rd order and allowed petitioner to
depose Lewis Slayton, the District Attorney for the Atlanta
Judicial Circuit.
“d=
District Court ordered respondent to file a complete list of its
objections by August 3, 1983. A modified order was entered on
August 5, 1983 (R. 716). Respondent did not fully meet this
schedule, as petitioner's counsel noted at the outset of the
August 8, 1983 hearing (Fed. Hab. Tr. 10-11) .2/
The evidentiary hearing itself comprised over two weeks of
testimony, from August 8-22, 1983. A further hearing on statisti-
cal issues was later held on October 17, 1983. The District Court
entered an order and judgment on February 2, 1984, granting habeas
relief on petitioner's due process claim under Giglio v. United
States, 405 U.S. 150 (1971), and otherwise dismissing petitioner's
Claims (R. 1129).
(ii) Statement of Facts
In view of the number and complexity of the issues presented
on these cross-appeals, the page constraints imposed, the need to
avoid repitition, and the statement of facts already presented by
respondent (see Resp. Br. 4-33 ,%/ petiticner will set forth facts
necessary for the consideration of his claims as part of the
argument on each of those claims.
(iii) Standard of Review
(a) As respondent concedes (Resp. Br. 8), petitioner's
St On several occasions during the hearing, respondent acknow-
Tedged that it had not identified alleged deficiencies in the
data base (which would have permitted petitioner's experts to
correct them and rerun its analyses) (Fed. Hab. Tr. 648-52;
gee algo id. 1385: 1417).
6/ Each reference to the Brief of Respondent-Appellant, dated
April 10, 1984, will be identified by the abbreviation "Resp.
Br." followed by the number of the page on which the reference
may be found.
due process claim under Giglio v. United States is a mixed ques-
tion of fact and law requiring independent review by this Court;
See e.g., Cuyler v., Sullivan, 446 US. 335, 341-42 (1980).
(b) Petitioner's claim that his counsel was ineffective
is also a mixed question of fact and law, requiring independent
review. See, e.g., Palmes v. Wainwright, 725 F.2d 1511, 1519
{11th Cir, 19384),
(c) Petitioner's due process challenge under Sandstrom
v. Montana, 442 U.S. 510 (1979), requires this Court's independent
application of legal principles to record facts. See, e.g.,
Franklin v. Prancis, 720 F.24 1206 (11th Cir. 1983).
(d) Petitioner's claim that the death penalty in the
State of Georgia is being imposed in arbitrary and discriminatory
patterns raised some factual issues, reviewable under F. R. Civ.
P. 52(a), see e.qg,, Pullman-3tandard v. Swint, 455 U.85. 273
(1982); many mixed questions of fact and law, requiring independent
review by this Court, see, e.g., Cuyler v. Sullivan, supra; and
several questions of federal constitutional law, requiring inde-
pendent appellate determination, see, e.q., Cuyler v. Sullivan,
supra.
(e) Two panels of this Court have previously held that
petitioner's challenge to the death-qualification of his jury
raised solely a legal issue. See Smith v. Balkcom, 660 F.2d 573,
578 & n.12 (5th Cir. Unit B 1981) Spinkellinlk v, Walnwright, 578
F.24 582, 593-96 (5th Cir, 1978). We respectfully disagree,
believing it raised both factual and legal questions requiring a
full evidentiary hearing and independent review by this Court.
SUMMARY OF ARGUMENT
The District Court correctly concluded that the State's fail-
ure to disclose a promise to one of the State's key witnesses
contravened the Due Process Clause. Giglio v. United States, 405
U.S. 150 (1972). The Supreme Court had held that the State has
an obligation to disclose such a promise, made as it was in this
case, by a police officer. Pyle v. Kansas, 317 U.S. 213 (1942).
Because of the critical nature of the witness' testimony, it
was likely that jury deliberations would have been affected if
the promise had been disclosed.
The District Court erred in rejecting petitioner's ineffec-
tive assistance of counsel claim. Trial counsel's failure to
interview State witnesses, to review the State's documentary
evidence, and to interview witnesses who supported a defense
theory presented at trial, all constituted ineffective assistance
of counsel which resulted in actual and substantial prejudice to
petitioner, Moreover, trial counsel's performance at the penalty
phase also fell below Sixth Amendment standards.
The District Court improperly rejected petitioner's Sandstrom
claim. The instruction given to his trial jury suggested that
petitioner had the burden of proof, and failed to specify what
burden lay on petitioner to rebut the presumption on intent and
malice which the trial court included in its charge. The en banc
court should hold that a Sandstrom violation is harmless only when
the instruction has no bearing on the offense for which the defend-
ant is convicted, or when the defendant has, by his own actions,
taken the issue of intent away from the jury. Alternatively, the
-
Court should hold that, on the facts presented herein, the
Sandstrom violation was not harmless beyond a reasonable doubt.
The District Court, relying in Spinkellink v. Wainwright, 578
PF.24 582 (5th Cir. 1978), rejected as a matter of law petitioner's
Eighth Amendment claim that Georgia's capital statutes are being
applied in an arbitrary and capricious manner. Yet the Supreme
Court has made clear the legislation valid on its fact can be
found discreiminatory in its application. Yick Wo v. Hopkins, 118
U.S. 356, 373-74 (1886). Capital statutes, even in the post-Furman
era, are no exception. See Godfrey v. Georgia, 446 U.S. 420, 428
(1980); Zant v. Stephens, 456 U.S. 410, 413 (1982). Therefore,
the refusal even to entertain petitioner's comprehensive statis-
tical evidence was clear constitutional error.
The District Court did permit petitioner to place his statis-
tical evidence before the Court in support of a Fourteenth Amend-
ment equal protection claim. However, the Court departed radically
from prior precedents in evaluating that evidence. Rejecting
multiple regression analysis as a valid mode of proof, refusing
to take seriously any but large-scale statistical models which
contained every possible variable, measuring petitioner's data
against unattainable standards of perfection, the District Court
adopted novel and unjustifiable standards totally at odds with
this Court's prior teachings on the evaluation of statistical
proof. See, e.g.; Eastland v. Tennessee Valley Authority, 704
F.24 613 (11th Cir. 1983): Jonnson-v. Uncle Ben's Inc., 628 F.24
419 (5th Cir. 1980), The Court's opinion on this claim is suf-
fused with both factual misunderstandings and legal misconceptions,
a
requiring this Court to vacate and remand for further proceedings
under appropriate legal standards.
The Court's rejection as a matter of law of petitioner's claim
that exclusion of death-scrupled jurors unconstitutionally dent ived
him of a fair and impartial jury, drawn from a representative cross-
section of his community, was error. See, e.g., Grigsby v. Mabry,
569 P. Supp. 1273 (E.D. Ark. 1983), appeal pending No. 83~2113-EA
{8th Cir.).
STATEMENT OF JURSIDICTION
The judgment of the District Court was entered on February 2,
1984. The District Court entered orders granting a certificate
of probable cause to appeal and leave to proceed in forma pauperis
on March 12, 1984. The Court has jurisdiction of these cross-
appeals pursuant to 28 U.S.C. § 2253.
ARGUMENT
1. THE DISTRICT COURT CORRECTLY CONCLUDED THAT THE
STATE'S FAILURE TO DISCLOSE TO PETITIONER'S JURY
A PROMISE BY ATLANTA POLICE DETECTIVE DORSEY TO
WITNESS OFFIE EVANS IN EXCHANGE FOR HIS CRITICAL
TESTIMONY AGAINST PETITIONER CONTRAVENED THE DUE
PROCESS CLAUSE OF THE FOURTEENTH AMENDMENT
The District Court concluded that the State's failure to dis-
close to the jury an understanding between Atlanta Police detective
Sidney Dorsey and trial witness Offie Gene Evans contravened the
Due Process Clause of the Fourteenth Amendment. (R. 1218-25.)
The District Court's conclusion is consistent with precedents
established by the United States Supreme Court and faithfully
followed by the panel decisions of this Court.
A. Facts Supporting the Giglio Claim
Offie Evans was a key State witness in Warren McCleskey's
trial. Evans' testimony explained certain inconsistencies in the
eyewitness identifications of McCleskey as one of the robbers of
the Dixie Furniture Store. More critically, Evans told the jury
that McCleskey had confessed to him that he had shot Officer
Schlatt and would have done the same thing if it had been twelve
police officers. Evans' testimony was the centerpiece of the
prosecutor's argument to the jury that McCleskey had committed
the shooting with malice. (R. 1222).
The District Court correctly found that Evans testimony
before the trial jury was false and evasive. Evan's federal
prisoner described his "escape" from the Federal Halfway House in
Atlanta as nothing more than a misunderstanding between himself
and the Halfway House administrators, and when he stated that no
promises had been made to him concerning his escape charges in
exchange for his cooperation with the McCleskey prosecution. (R.
1220.) Evans' misleading testimony to petitioner's jury was as
follows:
#1) You do have an escape charge still pending, is that
correct?
A: Yes, sir. I've got one, but really it ain't no escave,
what the peoples out there tell me, because something
went wrong out there so I just went home. I stayed at
home and when I called the man and told him that I
would be a little late coming in, he placed me on
escape charge and told me there wasn't no use of me
coming back, and I just stayed on at home and he come
and picked me up.
QO: Are you hoping that perhaps you won't be prosecuted for
that escape?
A: Yeah, I nope I don't but I don't -- what they tell me,
they ain't going to charge me with escape no way.
(TT. Tr. 3568),
30
This testimony is directly contradicted by federal records
7/
detailing the circumstances surrounding Evans' escape.”
Evans' trial testimony was also wholly misleading regarding
his expectations of help from State authorities in exchange for
. his cooperation in the McCleskey prosecution. As the District
Court correctly found, the jury was left with the impression by
Evans' testimony that no promises had been made concerning the
pending escape charges. (R. 1220). His actual testimony before
¢ the trial court was:
Q: [Assistant District Attorney] Have you asked me to try
to fix it so you wouldn't get charged with escape?
A: No, sir.
Qs: Have I told you I would try to fix it for you?
A: No, sir,
(T. Tr., 868-69). On cross-examination Evans expanded upon his
evasion regarding promises made by the State:
Ys Okay. Now, were you attempting to get your escape
charges altered or at least worked out, were you
expecting your testimony to be helpful in that?
A: I wasn't worrying about the escape charge. I wouldn't
have needed this for that charge, there wasn't no
escape charge.
{T. Tr. 882),
That testimony, as the District Court found, is directly
contradicted by Evans' testimony to the State habeas court
J that "the Detective told me that he would -- he said he was going
to do it himself, speak a word for me. That was what the Detective
told me." (St. Eab. Tr. at 122).
7/ Those records show that Evans had been told by federal person-
nel that disciplinary measures would be taken against him because
of his use of drugs. In describing his activities during his
escape, Evans had told federal prison authorities that he had gone
to Florida as part of an investigation dealing with drugs, and
that he expected to be well paid for his part. (R. 333, R. 1206).
~11m
Bs The Legal Standard
Applicable Supreme Court authority supports the District
Court's conclusion that the State's failure to disclose its
promise to witness Offie Gene Evans, made by Atlanta Police
Detective Sidney Dorsey, contravened the Due Process Clause.
It is the obligation of the State =-- not simply the prosecuting
Te attorney -- to reveal all promises or understandings between a
5 witness and agents of the State which might affect the jury's judgment as to the witness' credibility. Napue v. Illinois, 360
U.S. 264, 269 (1959) expressly holds that
it is established that a conviction obtained through
use of false evidence, known to be such by representatives
of the State, must fall under the Fourteenth Amendment.
(Citations omitted). The same result obtains when the
State, although not soliciting false evidence, allows it to
be uncorrected when it appears.
(Emphasis added). Accord, Giglio v. United States, 405 U.S. 150
(1972); Pvle v. Ranszasg, 317 U.S. 213 (1942).
Panels of this Circuit have repeatedly held that an undis-
closed promise of favorable treatment, made by a pollce officer
to a government witness, is subject to the protections of the
Due Process Clause. Smith v. State of Florida, 410 F.2d 1349 (5th
Cir. 1969); Schneider v,. Estelle, 552 P.2d 893 (Sth Cir. 1977);
Freeman v, State of Georgia, 599 F.2d 65 (5th CIr. 1979); Williams
v, Brown, 609 F,.28 216 (5th Cir. 1980); Smith v. Remp, 713 F.24
1549 (11th Cir. 1983); Moore v. Zant, 722 F.24 B40 (11th Cir. 1983).
The District Court, furthermore, properly concluded that
there was a reasonable likelihood that disclosure of the promise
of favorable treatment to Evans would have affected the judgment
of the jury on the murder indictment (R. 1223-25), since Evans'
“fo
testimony was critical to the State theory at trial (R. 1222):
First, he [Evans] alone of all the witnesses for the prosecu-
tion testified that McCleskey had been wearing makeup on the
day of the robbery. Such testimony obviously helped the
jury resolve the contradictions between the descriptions
given by witnesses after the crime and their in-court
identifications of petitioner. Second, Evans was the
only witness, other than the co-defendant, Ben Wright, to
testify that McCleskey had admitted to shooting Officer
Schlatt. No murder weapon was ever recovered. No one
saw the shooting. Aside from the damaging testimony of
Wright and Evans that McCleskey had admitted the shooting,
the evidence that McCleskey was the triggerman was entirely
circumstantial.
* * *
The court's conclusion ... is bolstered by the fact that the
trial judge, in charging the jury as to murder, instructed
the jury that they could find the defendant guilty as to
either malice murder or felony murder. After approximately
two hours of deliberation, the jury asked the court for
further instructions on the definition of malice. Given the
highly damaging nature of Evans' testimony on the issue of
malice, there is a reasonable likelihood that disclosure
of the promise of favorable treatment to Evans would have
affected the judgment of the jurv on this issue.
{R. 1222-23)
The District Court also correctly found that Evans' testimony
resolved contradictions in the eyewitness' testimony that McCleskey
was one of the robbers. Evans testified that McCleskey told him
he had been wearing makeup at the time of the robbery, thereby
offering an explanation to the jury as to why the description
given to police officers by one of the primary eyewitnesses _
varied so radically from McCleskey's physical appearance at
trial.
Evans' testimony that McCleskey had admitted shooting
Officer Schlatt was obviously critical. No one saw the shooting,
and McCleskey's statement given to police indicated only that he
had participated in the robbery. While Ben Wright, one of the
-i3=
co-defendants, testified that iMcCleskey had admitted the shooting,
under Georgia law, the testimony of an accomplice must be corrobo-
rated. Arnold v, Stare, 236 Ga. 534, 224 s.EBE.2d 336- (1976),
and Wright nad a clear interest in casting blame on a co-defendant.
Because the State tried the case to the jury on the theory
that McCleskey was the triggerman and sought the death penalty on
2% that ground (T. Tr. 973), Evans' testimony was critical to the
Y State's case. The State's argument that Offie Evans was not a
key witness is contrary to the express findings of the District
8/ gourt. It is also contrary to the principal argument the
8/ This case is unlike those wherein this Court has found no
likelihood that the deliberations of the jury would have been
affected by revelation of the undisclosed material. In both United
States v. Poitier, 623 F.2d 1017 (5th Cir. 1980) and United States
V. Nixon, 632 7.24 306 (5th Cir. 1981), the State failed to
provide information regarding a promise, but the jury nonetheless
learned the details of the State's understandings with the witness
during the trial. In petitioner's case, the jury never learned
of the promise to Evans, nor of any of the evidence which the
District Court found to be highly impeaching in character. (See
R. 1255.) In United States v, D'Antignac; 628-P.24-428--(5th- Cir.
1930), this Court held that failure to disclose an understanding
would not have affected jury deliberations because a series of
other promises were revealed to the jury. In United States v.
Barham, 625 F.2d 1221 (5th Cir. 1980), the Court reached the same
result because most of the details of the agreement were revealed
to the jury, although certain inconsistencies in testimony
existed. The jury in petitioner McCleskey's trial heard no
evidence indicating any promises made to Evans.
Finally, in United States v. Antone, 603 F.2d S66 (5th Cir.
1979), a panel of this Court found no reasonable likelihood that
' jury deliberations would have been affected by the failure of the
State to correct false testimony by a witness regarding the State's
agreement to pay certain of his attorney's fees. The Court noted
that the witness "background was exposed to the jury in considerable
detail." Included in the evidence presented to the jury in Antone
was evidence of numerous felony convictions; the witness use of
addictive drugs; his previous treatment for mental illness, his
prior inconsistent statements about matters which were the
subject of his trial testimony; and the details of his plea
bargain with the Government. In this case, by contrast, Evans'
packground was not exposed to the jury. The jury did not learn
of the agreement with Dorsey, nor of Evans' history of drug
abuse, or the drug-related nature of his escape.
“ld
promise of favorable treatment would have affected the judgment
of the jury on the issue of petitioner's conviction for malice
murder. 22/
11. THE DISTRICT COURT ERRED WHEN IT REJECTED
PETITIONER'S CLAIMS THAT TRIAL COUNSEL FAILED
TO RENDER EFFECTIVE ASSISTANCE OF COUNSEL
In the District Court, petitioner contended that trial
counsel had failed to render effective assistance of counsel
before, during and after McCleskey's state trial. In rejecting
those claims, the District Court committed a number of legal and
factual errors. Prior to petitioner's trial, which began on October 9,
1978, petitioner's trial counsel had interviewed none of the 96
witnesses listed by the State as potential witnesses at McCleskey's
8A/ The District Court, however, concluded that there was no
reasonable likelihood that the jury would have reached a different
verdict on the charges of armed robbery. (R. 1222-23). The
Court reasoned that Evans' testimony regarding petitioner McCleskey's
participation in the armed robbery, as distinguished from Evans’
testimony regarding McCleskey's intent to commit malice murder,
was cumulative of other testimony. The Court refused to grant
petitioner a new trial or a new sentencing phase in light of the
jury's decision to impose consecutive life sentences at the
penalty phase. (R. 1266). Petitioner does not appeal from the
. Court's determination regarding the armed robbery convictions; he
: does, however, contend that the District Court erred when it
failed to set aside the consecutive life sentences for armed
robbery and order a new penalty phase on these convictions.
It is likely the jury deliberations on the sentencing for
armed robberies would have been different had the jury known of
the impeaching evidence regarding Evans. The jury's decision was
based on its judgments about petitioner, influenced as they were
by the testimony of Offie Gene Evans, that petitioner's role in the
armed robberies including the murder of Officer Sclhatt. If the
jury had disbelieved Evans on that point, it may well have imposed
a lesser penalty for the armed robbery convictions. The District
Court erred, therefore, in failing to grant a new penalty trial
on the armed robbery convictions.
-15-
or the co-defendant's crisis Among those whom trial counsel
failed to interview were: (i) the victims of the Dixie Furniture
Store robbery; (ii) a Fulton County Sheriff's Deputy and an
inmate at the Fulton County Jail, who testified regarding circum-
stances surrounding an alleged confession by McCleskey to the
inmate; and (iii) the State's Crime Lab expert, who gave his
opinion regarding the identity of the murder weapon.
Nor were potential defense witnesses interviewed prior to
trial. Among the victims of the robbery at the Dixie Furntiure
Store were four witnesses whose statements to police called into
question the State's circumstantial evidence that only one of the
robbers was in the front of the Store at the time of the shooting,
and therefore, he was the one who shot Officer Schlatt. None of
these witnesses, who could have supported petitioner's defense
was ever interviewed, much less presented at trial. Nor were
potential leads to character witnesses for the sentencing
phase pursued by defense counsel.
Defense counsel also failed to investigate the State's docu-
mentary evidence. Although the prosecution advised trial counsel
that it intended to introduce into evidence, as aggravating cir-
cumstances at the penalty phase, evidence of McCleskey's prior
convictions on armed robbery counts and sentences to three life
terms in Douglas County, Georgia in 1970 (State Court R. at 47),
trial counsel conducted on inquiry into the circumstances surround-
ing those sentences, thereby failing to discover that they had
9/ Of the 24 witnessed called by the State at McCleskey's trial,
only four had testified at the preliminary hearing. (St. Hab. Tr.
28.)
lf
been set aside by the prosecutor's agreement to a new trial in
1971, and failing to discover that the reason the prosecutor con-
sented to a new trial was that key evidence introduced at the
trial had been seized without a valid search warrant in contra- vention of the Fourteenth Amendment. (Fed. Hab. Tr 1817-18;
1821-24.)
The District Court held that, although McCleskey's trial
vy counsel had concluded, prior to trial, that the State's case
would likely include testimony of an alleged "jailhouse confession"
by McCleskey, trial counsel had no duty to investigate because
petitioner himself steadfastly denied making such a confession
{R. 1255) .18/ That conclusion was contrary to this Circuit's
principles regarding the obligations of trial counsel to conduct
an independent investigation into the facts of the case. Goodwin
v. Balkcom, 684 F.2d 794, at 805 (11th Cir. 1982); Rummel v.
Estelle, 590 F.24 103, 104 (5th Cir, 1979), quoting Von Moltke v,
10/ Trial counsel admitted that his "only conclusion" once he
Tearned that a Fulton Deputy had been listed by the State as a
witness was that someone was going to testify regarding a jailhouse
confession, (St. Hab. Tr. 76.) This conclusion by trial counsel
was bolstered by his knowledge that the prosecutor had a statement
from an undisclosed person which had not been given to defense
counsel. (St. Hab. Tr. 77.) Yet trial counsel never interviewed
the Fulton County Deputy nor took any other available steps to protect
his client against the risk of false testimony regarding a
subject as potentially damaging as a jailhouse confession.
The District Court's reasoning that it "would be anomalous
the, for this court to grant McCleskey habeas corpus relief on
the grounds that McCleskey's counsel was ineffective because he
did not disbelieve petitioner and undertake an independent
investigation" (see R. 1253) is faulty. Trial counsel's duty
to investigate the circumstances surrounding a confession does
not turn on whether his client admitted he gave the confession.
Rather, the duty to investigate arises when trial counsel has a
reasonable belief that a confession will be offered by the State,
which 1s what the District Court found to be the case here.
Ry
Gillies, 332 U.S. 708, 721 (1948).
Viewing the facts "from the perspective of counsel, taking
into account circumstances known to counsel at the time," Douglas
v. Wainwright, 714 P.24 1432, 1554 (11th Cir. 1983); Houss v,.
Balkcom, 725 P.24 808, 619 (11th Cir, 1984), trial counsel failed
to investigate when his own understanding about the facts called
for an investigation. This Court has repeatedly held that trial
counsel has an obligation to conduct an investigation into
possible defenses as well as evidence in support of the State's
case. Goodwin v. Balkcom, supra, at 810-12, 817; Scott v.
Walnwright, 698 P.2d4 427 (11th Cir. 1983); Weidner v. Wainwright,
708 P#.24 614 (11th Cir. 1983).
Furthermore, the District Court acknowledged that prejudice
resulted from trial counsel's failure to conduct an investigation
once he had concluded that testimony regarding a confession
would be part of the State's evidence:
Evans' testimony was certainly very damaging to petitioner
and a pretrial investigation as to what his testimony may
have uncovered the details of his escape from a halfway
house and the pending federal charges against him, his
"understanding" with an Atlanta police detective, his
history of drug abuse, and his imaginative story that he
had gone to Florida and participated in an undercover drug
investigation during his escape. Discovery of such
evidence would have had substantial impeachment value.
(R. 1255.) Presentation of this testimony to the jury would
likely have had a substantial impoact, both at the guilt and
sentencing phase.
The District Court also rejected petitioner's claim that
trial counsel was ineffective when he failed to interview and
subpoena for trial four crime scene witnesses whose testimony
“lB
"might have cumulatively created a reasonable doubt as to
whether petitioner was the triggerman." (R. 1254.) The
District Court incorrectly found that trial counsel, as a matter
of strategy, chose to offer only one defense at trial -- that of
alibi. (R. 1254.) Examination of trial counsel's testimony
before the State habeas court shows this finding is clearly
erroneous. Trial counsel testified that he chose to develop two
defenses at trial -- one of alibi, and another that more than one
of the co-defendants was in the front of the store at the time of
the shooting:
I would ... state that was part of the theory if the
defense in that information was developed during the
cross examination of several of the State's witnesses
and one of the theories that the defense put forth was
the fact that Ben Wright had come from the back into
the front and was in fact the person who shot Officer
Schlatt.
(St. Hab, Tr. 45-48).
The District Court's finding that trial counsel pursued
only one line of defense is thus erroneous. Rather than assess
trial counsel's failure to interview as though trial counsel had
rejected this line of defense, the appropriate analysis was one
based on trial counsel's decision to present the defense at
trial.
Judged in this light, it was ineffective to fail to inter-
view prior to trial, and subpoena for trial, available witnesses
who would have offered support for one of petitioner's
defenses. Davis v. Alabama, supra; Gaines v. Hopper, 575 F.2d
1147 (5th Cir. 1978); Weidner v. Wainwright, supra; cf, Young v.
Zant, 677 .F.28 392 {11th Cir. 1982) 1%
12/ As the District Court noted, the testimony of the four
witnesses who were not called by the defense could have created
(Continued)
a i Ta
The District Court agreed that trial counsel's failure to
object to introduction of evidence of three life sentences which
had been imposed upon McCleskey in September, 1970, constituted
ineffective assistance of counsel. (R. 1256-57). The Court
further concluded, however, that petitioner could not show
actual and substantial prejudice (apparently at the guilt phase)
and that, as to the penalty phase, while the petitioner was
prejudiced, the Court was unprepared to grant a new trial, at
least in part because the Court had done so on the Giglio claim.
(R. 1257). While the Court correctly concluded that counsel was
ineffective in failing to object to entry of the evidence of the
three life sentences, the District Court erred when it concluded
that no relief should be granted.
Evidence introduced during McCleskey's trial, and used at
the penalty phase, showed that a Douglas County jury had
imposed three life sentences upon petitioner in September, 1970.
(Exh. JT-2). Yet, before the federal court petitioner showed
that those three convictions and life sentences had been set
aside upon the consent of the District Attorney to a new trial
in December, 1971. (Exh. WM-2). Under both Georgia law, and
federal law, introduction of the evidence of the three life
sentences was error. Under Georgia law in effect at the time of
petitioner's trial, the sentences were not admissible because
they were not based upon final judgments. O0.C.G.A. § 16-1-3(4)
and 42-5-50 (R. 900-01).
(Continued)
a cumulative doubt as to whether McCleskey was the triggerman.
Statements by each of them contained in the prosecutor's file
indicated, contrary to the State's circumstantial evidence
presented at trial, that at least one of the co-defendants had
left the rear of the Store, and had gone to the front, prior to
the firing of the shots that killed Officer Schlatt. Relevant
excerpts are set out at R. 886-87.
=30~
Under federal law, evidence of convictions tainted with
unconstitutionality is inadmissible. Zant v. Stephens,
U.S. + 103. 8.Ct, 2733 at 2748, n. 23 (1983); United Staces
v. Tucker, 404 U.S. 443 (1972). Because trial counsel conducted
no investigation whatsoever into the State's evidence of the
three Douglas County life sentences (Fed. H. Tr., 1817-18,
1821), he failed to object to their admission. This constituted
ineffective assistance of counsel.
McCleskey suffered actual and substantial prejudice at
the sentencing phase from this failure. The District Court
recognized that knowledge of the invalid convictions and life
sentences would likely have affected jury deliberations. (R.
1257). The Court's determination in this regard is amply sup-
13
ported by the facts in the case
13/ The District Attorney's argument to the jury was grounded
on the three life sentences. He asked the jury to consider
particularly, in their deliberations, the three life sentences
which McCleskey had already received. (T. Tr., 1019-20). He
also advised the jury that if they wished the sentence on the
armed robbery counts to be served consecutively, they should
specify on their verdict. (7. Tr., 1018).
The jury imposed a sentence of death, rather than life,
on the murder count, and two consecutive life sentences on the
armed robbery counts (State Trial R., at 56). The sentences
imposed were the most severe options open to the jury.
Because the jury improperly considered in its deliberations
on penalty, for the robbery as well as the murder convictions,
the prior life sentences which had been set aside, it is likely
that the jury imposed a greater penalty than they would have
otherwise. A panel of this Court has previously held that per-
mitting the State to admit prior invalid convictions constituted
ineffective assistance of counsel and that
[wle can hardly imagine anything more prejudicial
to Nero [the petitioner] than allowing the jury in
his armed robbery case to hear the prosecutor's
comments that Nero had been convicted twice before
of burglary and once on drug charges.
Nsro v. Blackburn, 597 F.24 891 (5th Cir. 1979)
lf
The District Court also concluded that trial counsel had
not been ineffective at the sentencing phase, since he made
inquiries of McCleskey and McCleskey's sister about the avail-
ability of character witnesses who could have testified on
McCleskey's behalf (R. 1258). However, trial counsel failed to
pursue other avenues clearly known to him at the time, which
would have led to favorable character testimony from witnesses
willing to testify on McCleskey's behalf. The District Court's
conclusion ignoring that failure is error, and petitioner is
entitled to a new penalty phase hearing on the robbery and
murder convictions.
The State habeas record shows that trial counsel's only
efforts to contact character witnesses for the sentencing phase
amounted to conversations with petitioner while he was incar-
cerated, and a telephone conversation or meeting with one of
petitioner's family members. (St. Hab, 80). Although clearly
aware of other potential sources of information which he left
untouched--including the minister of the McCleskey family
church in the community wherein McCleskey grew up--trial
counsel apparently did not contact them and personnel at the
high school which McCleskey attended. (St. Hab. 90, 83).
Had he done so, he would have been immediately placed in
touch with church members who were neighbors of the petitioner
: 14
when he was growing up et They would have been able to
14/ Petitioner sought to expand the federal habeas record below
to include an affidavit from Reverend Johnson showing that he
had been in touch with trial counsel, but that trial counsel had
(Continued)
present a positive picture to the jury of McCleskey. Similar
testimony was available, had trial counsel sought it out, from
other church members as well as Lemon Street High School personnel.
Their affidavits are a part of the State habeas record. See,
2.3. (St. Hab. R. 225-28; 231-32; 227-30).
Trial counsel's failure to pursue these available avenues
to sentencing phase witnesses was not a matter of strategic
choice. He indicated his desire to have character witnesses for
the sentencing phase (St. Hab. 82). Rather, his failure was
simply a reflection of his abdication of his client's cause
at the pre-trial investigatory phase. +> Beyond conversations with
his client and one of his sisters trial counsel simply conducted
no pretrial investigation whatsoever.
For the foregoing reasons, the District Court erred when it
failed to grant petitioner a new trial on the ineffective assist-
ance of counsel claim.
14/ (continued)
never asked for names of persons who could have testified regard-
ing petitioner's background, which Reverend Johnson could have
supplied. (R. 264, 322-23), The Court denied that motion.
To the extent this Court finds the Reverend Johnson evidence
critical to disposition of this question, petitioner submits the
District Court's refusal to make it, and the accompanying affi-
davit of petitioner's former wife, a part of the record, was
erroneous.
15/ Had trial counsel conducted an independent investigation,
when he asked the jury on McCleskey's behalf to reflect on his
humanity (T. Tr. 1024), the jury would have been in a position
to consider McCleskey's devotion as a father; the close relation-
ship he had with his daughter; and the positive manner in which
acquaintances described him. Trial counsel's plea that the
jury consider his humanity would have had an evidentiary founda-
tion. In the absence of any pretrial investigation the plea
was a hollow one.
23
III. THE DISTRICT COURT ERRED WHEN IT REJECTED PETI-
TIONER'S CLAIM THAT HIS JURY INSTRUCTIONS CON-
TRAVENED THE DUE PROCESS CLAUSE
The District Court rejected petitioner's claim that jury
instructions contravened due process guarantees as defined in
Sandstrom v. Montana, 442 U.S. 510 (1979) and Mullaney v. Wilbur,
421 U.S. 684 (1975). (Relevant portions of the instructions
are set forth in the District Court's order, at R. 1266-68,
n.21.) Under Sandstrom v. Montana, 442 U.S. 510 (1979), a jury
instruction is unconstitutional if it ‘could be interpreted by
the jury as creating a mandatory, rather than permissive, presump-
tion on an element of the crime. Moreover, even if a trial court
suggests that the presumption may be rebutted, the instruction
ls unconstitutional if it fails to advise the jury that the
presumption may be rebutted by "the defendant's simple presenta-
tion of 'some' evidence," Id, ar 2455,
The jury instruction here created a mandatory presumption,
and thus the District Court erred when it concluded that no
Sandstrom violation was present.
The District Court concluded that, in the alternative if
there was Sandstrom error, it was harmless (R. 1230). That
holding was also error. Under the erroneous instruction, peti-
tioner's jury could well have concluded that the burden had
shifted to McCleskey to rebut the State's case once there was
circumstantial evidence, however weak, that the shooting occurred
with malice. Even if the jury had disbelieved McCleskey's
co-defendant and Evans, the presumption filled in the gap for the
State, shifting the burden to McCleskey.
-24-
It thus cannot be said, on the basis of the facts presented
to the jury, that the instruction was harmless beyond a reason-
able doubt. The District Court erred in failing to grant peti-
tioner a new trial on the murder count because of erroneous
instruction.
: IV. IN REJECTING PETITIONER'S CLAIM THAT SUBSTANTIAL
* RACIAL DISPARITIES PERSIST IN GEORGIA'S CAPITAL
SENTENCING SYSTEM, THE DISTRICT COURT MISAPPLIED
CONTROLLING LEGAL PRECEDENTS, MISUNDERSTOOD BASIC
STATISTICAL PRINCIPLES, AND IGNORED CLEAR RECORD
EVIDENCE
A. Introduction: Petitioner's Constitutional Claims
Petitioner alleged that Georgia's capital sentencing system,
under which his death was imposed, has been administered "arbi-
trarily, capriciously and whimsically," (R. 17), and "pursuant to
a pattern and practice ... to discriminate on the grounds of
race," in violation of the Eighth Amendment and the Equal Pro-
tection Clause of the Fourteenth Amendment (R. 18). The District
Court, relying on Spinkellink v. Wainwright, 578 PFP.2d4 582 (5th
Cir. 1978), held that petitioner's Eighth Amendment claim was
not cognizable, irrespective of his proof, since Georgia's
capital statutes had been found valid in Gregg v. Georgia, 428
U.S..153 (1976) (BR. 255, 1133).
In so holding, the District Court misread both Gregg v.
Georgia, supra, and Furman v. Georgia, 408 U.S. 238 (1972), upon
which Gregg is grounded. The Supreme Court did no more in Gregg
than to uphold Georgia's capital stautes on their face. It has
long been clear that a law "fair on its face and impartial in
appearance" may nevertheless violate the Constitution in its
-25=
application, Yick Wo v. Hopkins, 118 U.S. 356, 373-74 (188s),
and the Supreme Court has expressly held that the Eighth Amend-
ment imposes on a State the constitutional responsibility both
"to tailor and apply its [capital punishment] law in a manner
that avoids the arbitrary and capricious infliction of the death
penalty," Godfrey v. Georgia, 446 U.S. 420, 428 (1980) (emphasis
added); accord, Zant v. Stephens, 456 U.S. 410, 413 (1982); see
38180, Proffitt v. Wainwright, 685 P.24 1227, 1262 n.52 (11th
Cir. 1982). Since petitioner's challenge is to the application
of Georgia's capital statutes, the District Court erred in
rejecting his Eighth Amendment claim as a matter of 1a.
The District Court did permit petitioner to present evidence
in support of his Fourteenth Amendment claim. Yet, in denying
as "irrelevant" petitioner's discovery requests related to prior
discriminatory conduct in the criminal justice system in Fulton County and the State of Georgia, the District Court erred again,
for such anecdotal evidence is plainly relevant to an Equal
11/
Protection claim.—
16/ Page constraints prevent petitioner from setting forth here
a full constitutional argument in support of his Eighth Amendment
claim. He agrees with and adopts the arguments advanced to this
Court on behalf of James Lee Spencer. See Spencer v. Zant, No.
82-8408, First Supplemental Brief for Petitioner-Appellant on
Rehearing En Banc, dated December 28, 1983 (hereinafter "Spencer
ist Br.") 51-54 (11th Cir.), and Second Supplemental Brief for
Petitioner-Appellant on Rehearing En Banc, dated January 11,
1984 (hereinafter "Spencer 24 Br.") 21-23, 27-28 (11th Cir.).
17/ The District Court's insistance that prior discriminatory
conduct -- especially by actors integrally involved in the admin-
istration of the criminal justice system -- was "irrelevant" to
petitioner's Equal Protection claim constitutes clear legal error.
See, e.g., Rogers v. Lodge, 458 U.S. 613,624-27 (1983); Columbus
Board of Education v Penick, 443 U.S. 449, 464-65 (1979);
(Continued)
hw
Nevertheless, we submit that the statistical case alone is
sufficient to warrant relief. This Court has recognized that
"[i]ln 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, 859 (5th Cir. Unit B 1982)(on rehearing);
cf. Adams v. Wainwright, 709 F.2d 1443, 1449 (11th Cir. 1983).
Petitioner's comprehensive statistical evidence on the operation
of Georgia's capital statutes from their inception in 1973
through 1979, demonstrating substantial, pervasive disparities
based upon the race of the homicide victim and the race of the
defendant, constitutes just the sort of "clear pattern, unex-
plainable on grounds other than race," Arlington Heights v.
Metropolitan Housing Authority, 429 U.S. 252 266 (1977), that
the Supreme Court has held to establish an Equal Protection
violation. It is to petitioner's evidence that we now turn.
B. The Facts: Petitioner Has Made Out A Compelling
Prima Facie Case Of Racial Discrimination In Capital
Sentencing
(i) Petitioner's Experts Were Well Qualified
The statistical case-in-chief for petitioner was pre-
17/ (continued)
Washington v. Davis, 426 U.S. 229, 265-66 (1976). Having denied
petitioner access to the records from which such discriminatory
acts might have been proven, moreover, (R. 596; see Fed. Hab.
Tr. 1797-99), the District Court should not have faulted peti-
tioner for failure to introduce such non-statistical evidence as
part Of its case~in-chief. (See R. 1141). If this Court's
review of petitioner's substantial statistical evidence leaves
the Court with any doubts about petitioner's prima facie claim,
it should remand the case to the District Court for the receipt
of this significant nonstatistical evidence.
le
sented through the testimony of two experts eminently qualified
to investigate the very matters at issue. Professor David
Baldus, petitioner's chief researcher, testified concerning his
background and training in law as well as his extensive experi-
ence in the development and use of social science methods to
examine legal issues. Educated in political science at Pittsburgh
and in law at Columbia and Yale Law Schools (Fed. Hab. Tr.
39-42), Baldus has pursued a distinguished research and teaching
career, focused upon the applications of social science methods
to legal issues. His first major research effort, on the impact
of certain social welfare laws, has subsequently "been reprinted
in a number of books, and it's used in courses in sociology
departments and in law schools to illustrate [time series] ...
methodology as a way of trying to determine the impact the
enactment of laws ha(s]," (id. $3=53).18/
As a result of consultations on that first project with
Professor James Cole, a statistician, Baldus began an extended
research collaboration with Cole on how courts should employ
statistical evidence in evaluation of claims of discrimination
(id. 54-55). The ultimate fruit of that effort is an authorita-
tive tex: in the field, D., BALDUS & J. COLE, STATISTICAL PROOF
: OF DISCRIMINATION (1980) (id. 68), widely relied upon by the
federal courts in evaluating the quality of statistical evidence
(Fed, Hab, Tr. 74-75; See DB6).
As part of his research for that work, Baldus happened to
18/ Baldus, "Welfare as a Loan: An Empirical Study of the
Recovery of Public Assistance Payments in the United States," 25
STAN. L. REV. 123 (1973).
-28-
obtain and reanalyze an extensive data set on capital punishment
patterns collected in the mid-1960's by Professor Marvin Wolfgang
Subsequently, Baldus also obtained and reviewed a second major
data set on capital punishment patterns collected at Stanford
University during the late 1950's and early 1960's (id.). 20/
Baldus further pursued his interest in capital punishment in a
critical evaluation of the methodologies employed in two key
studies on the deterrent value of capital punishment, published in
a special 1975 symposium on the death penalty in the Yale Law
2Y/
Journal.—
Professor Baldus' research interest in capital punishment
intensified after Gregg v. Georgia in 1976 into a principal focus
of his work:
During the succeeding seven years, Baldus devoted a major
portion of his research (id. 84-100), writing (id. 85-90) 22/,
and teaching energies (id. 90) to the post-Gregg capital punish-
ment statutes and their administration, reviewing every Supreme
Court case on capital sentencing and studying the professional
19/ See Wolfgang & Riedel, "Race, Judicial Discretion and the
Death “Penalty, 407 ANNALS 119 (1973).
. 20/ See Special Edition, "A Study of the California Penalty
i Jury in First Degree Murder Cases," 21 STAN. L. REV. 1297
. (1969).
s 21/ Baldus & Cole, "A Comparison of the Work of Thorsten Sellin
and Isaac Ehrlich on the Deterrent Effect of Capital Punishment,"
85. YALE L. J. 170 (1975).
22/ See DB 1 at 2; e.g., Baldus, Pulaski, Woodworth & Ryle,
"Identifying Comparatively Excessive Sentences of Death," 33
STAN. L. REV. 601 (1977); Baldus, Pulaski & Woodworth, "propor-
tionality Review of Death Sentences: An Empirical Study of the
Georgia Experience," J. CRIM. LAW & CRIMINOLOGY (forthcoming
19833).
DG
literature on sentencing patterns in both capital and non-capital
cases (id. 130-31; see DB 13) as part of his preparation for the
two studies that formed the basis of petitioner's statistical
case Selo. oY
Petitioner's other expert on his case-in-chief was Dr.
George Woodworth, an Associate Professor of Statistics at the
Univeristy of Iowa. Dr. Woodworth testified that he had been
trained as a theoretical statistician (id. 1195), specializing
in nonparametric analysis of categorical data (the very sort of
data at issue in petitioner's two studies) (id. 1197, 1200).
While teaching at Stanford University, Dr. Woodworth developed
an interest in applied statistics (id. 1200), and was invited by
the National Research Council and its chief statistician,
Frederick Mosteller, to conduct a formal review of the statisti-
cal methodology used in a major national research project (id.
1200-01) (which employed many of the methods Baldus and Woodworth
ultimately incorporated into their own studies) (id. 154-58).
Dr. Woodworth also served as the Director of Iowa's Statistical
Consulting Center, advising researchers on appropriate statistical
techniques for over eighty empirical studies (id. 1203-04). He
has published widely in statistical journals (see GW 1, at 2-3),
and is a member of the Committee on Law and Justice Statistics of
the American Statistical Association (id. 1194) .2%/
23/ Baldus also served as a consultant on capital sentencing
review to two state supreme courts (id. 94-96) and was at the
time of the 1983 hearing a principal consultant to a Task Force
of the National Center for State Courts, charged with developing
appellate capital sentencing methods and standards (id. 97-100).
24/ The District Court qualified Professor Woodworth in the
"theory and application of statistics, and in the statistical
vse @8nalysis of discrete] outcome data," (id. 1206).
3)
(ii) Petitioner's Data-Gathering Effort
Was Carefully Conducted
Petitioner's experts testified that they undertook
two overlapping studies of the administration of Georgia's
capital sentencing system in the post-Furman era. The first
of these, designated the Procedural Reform Study ("PRS"),
was designed to examine whether disparities in treatment,
based upon race, could be found at two key "decision points"
in the Georgia eysten: the prosecutor's decision, following
a murder conviction, on whether to proceed to a penalty |
trial, where a death sentence might be imposed, or to accept
the automatic life sentence that follows any murder conviction
under Georgia law; and the jury's decision, in those cases
advancing to a penalty trial, on life imprisonment or death. (id. 165-67)23/ The universe for the PRS was defined to
include all defendants arrested between the enactment of
Georgia's post-Furman capital statute on March 28, 1973 and
June 30, 1978, who were subsequently convicted of murder -
some 594 individuals (id. 170-71; 192).
The second study, designated the Charging and Sentencing
Study ("CSS"), was designed to examine possible racial discrimi-
nation at all decision points from indictment forward, including
prosecutorial plea bargaining decisions, jury decisions on
conviction or acquittal, and the sentencing decisions encompassed
in the PRS (id. 261). The CSS was framed to include a sample
of persons indicted for both murder and for voluntary manslaughter
25/ For a description of the statutory options available
under Georgia law upon conviction for murder, see Gregg v.
Georgia, supra, 428 U.S. at 162-66.
-3 fw
(id.) during the entire period from 1973 through 1978 (id.
263-84) .28/
The data-gathering procedures have been summarized
elsewhere (see Spencer lst Br., App. A 11-13, 17-23). We
will here confine our attention to four aspects of that
process: (a) the integrity of the data sources; (b) the
strengtns of the data-gathering instruments employed; (c)
the care and accuracy of the coding process; and the (4d)
coding conventions employed.
(a) The Integrity of the Data Sources
Professor Baldus testified that, in choosing a State for
study, he and his colleagues "were very much concerned about the
availability of data" (id. 160). Baldus dispatched a colleague
"to Georgia for a period of two weeks to find out what data were
here that we could get access to, and he returned to Iowa with a
glowing report about the many sources of data" (id. 174-75).
These included not only the records of the Supreme Court of
Georgia -- which typically contained trial transcripts, trial
judges' reports, appellate briefs, and a summary card on each
case (id. 175; 202-04; see, e.g., DB 29-33) -- but also back-
ground information on each defendant in the files of the De-
partment of Offender Rehabilitation (id. 175: 204-05) and victim
information from the Bureau of Vital Statistics (id. 176; 205-06;
see, e.g. DB 47).
26/ The PRS does not involve a sample; instead it includes
every individual within the universe. The CSS, by contrast,
embraces a universe of 2484 from which a weighted sample of
1066 cases was drawn by scientifically appropriate procedures
(id, 265-73).
“3D
Most importantly, Baldus and his colleagues eventually lo-
cated "an extensive file of information on all offenders" in the
Board of Pardons and Paroles (id. 176), which became the basic
source for the Charging and Sentencing Study.
The official Pardons and Parole files, petitioner demon-
strated to the District Court, are kept pursuant to a stringent
state statute that requires the Board "to obtain and place in its
permanent records as complete information as may be practically
available on every person who may become subject to any relief
which may be within the power of the Board to grant ... [inclu-
ding] A. A complete statement of the crime for which such person
is sentenced, [and] the circumstances of such crime ... E. Copy
of pre-sentence investigation and previous court record ... [and] H. Any social, physical, mental or criminal records of such
person." (Former GA. CODE. ANN. § 77-512). L.W. Warr, a former
field officer for the Board, now a field supervisor (Fed. Hab.
Tr. 1327), testified that field officers (all of whom are re-
quired to be college graduates) (id. 1329), are trained to "check
local criminal records ... go to the clerk of court, get sentence
information, indictments, jail time affidavits, we get police
reports from the agency that handled the case" (id. 1330-31) .2/
27/ The District Court noted that "the police reports were
missing in 75% of the cases [and] the coders treated the Parole
Board summary as a police report" (R. 11671; see 1157). Officer
Warr testified, however, that whenever the actual police reports
were not included in Parole Board files, they were always sum-
marized, and nothing "contained in the police reports ... would
[be] Toutinely omiti{ted]l" (Fed. Hab. Tr. 1332; accord, id. 1331).
Furthermore, Warr stated that, especially in homicide cases, field
officers often went beyond the report to "interview the [police]
officers that were involved in the case™ (id. 1332). Por this
reason, the Pardon Board summaries were typically superior sources
of information to the actual police reports themselves.
-33~
In homicide cases, moreover, Parole Board officers routinely
speak, not only with the investigating police officers (id. 1332),
but also with the District Attorney to obtain "his comments con-
cerning the case" and "his impression regarding what happened at
the, involving the particular crime" (id. 1333). The officers
were guided in their investigation by a Field Operations Manual
(LW 1), which contained the following instructions,
among others:
"3.02 ... The importance of this report cannot
be over-emphasized; and where the offender
has been convicted of crimes against the
person, it is imperative that the Officer
extract the exact circumstances surrounding
the offense. Any aggravating or mitigating
circumstances must be included in the report.
%* * *
"3.02 ... Circumstances of the offense -
This should be obtained in narrative form, it
should be taken from the indictment, the
District Attorney's Office, the arresting
officers, witnesses, and victim. A word
picture, telling what happened, when, where,
how and to whom should be prepared.”
* * *
The Parole Officer should be as thorough .as
possible when conducting post-sentences on
persons who have received ... "sentences in
excess of fifteen years. In cases where
arrest reports are incomplete the circum-
stances of the offenses should be obtained as
thoroughly as possible and the Parole Officer
should review the transcript of the trial if
available for detailed information. A per-
sonal interview with the arresting or investi-
gating officer is almost always a valuable
source of information as the officer may
recall important details and facts which were
not revealed in the arrest report."
(Id., at 2-4) The State offered no testimony to suggest that
these standards were not regularly followed, or that the official
«3d
Parole Board record contained any systematic errors or omissions
(id. 648: "we're not in a position at this point to challenge
the underlying data source ... from the Pardons and Paroles
Board") =-- much less any information that these files were sys-
tematically biased according to the race of the defendant or the
victin, 2
Baldus acknowledged that some data were occasionally missing
from the Pardons and Paroles files, as well as from the files of
other agencies =-- the Georgia Supreme Court, the Department of
Offender Rehabilitation, and the Bureau of Vital Statistics -- to
which he also turned (id. 205-06). The only important categories
of missing data, however, involved information on the race of the
victim, on whether a penalty trial had occurred, and on whether a
plea bargain had been offered (id. 586-88) .23/ Baldus took extra-
ordinary steps to obtain this information from official files,
even writing systematically to defense counsel and prosecutors to
secure it where official sources failed (id. 587-88; see DB
45, 46). Moreover, petitioner sought without success to secure
28/ In light of this uncontradicted testimony, the District
Court's findings that "[t]lhe information available to the coders
from the Parole Board Files was very summary," (R. 1160), and
that "[t]lhe Parole Board summaries themselves were brief" or
"incomplet(e] (id.), are at least misleading, if not clearly
erroneous.
29/ Despite extensive testimony explaining the rationale under
which the coders were instructed to code certain information as
"U" or "unknown" in Baldus' questionnaires (see id. 444-45, 524-
27, 1684-90), and further testimony on the scientific appropriate-
ness of Baldus' use of the "U" code (id. 1761-64), the District
Court suggests throughout its opinion that this accepted coding
convention represents "missing data" (R. 1163-67). We deal with
the "U" coding issue and its actual effort on Baldus' analyses at
pages .
-35-
these data from respondent during the discovery process (R.
556; 595-96; 399; 615).
In the end, the amount of missing data proved scientifically
insignificant. Only 5 of the 594 cases in the PRS lacked race-
of-victim information (id. 1096; 1705-06); for the CSS, the
‘number was 63 of 1066 (id.). Penalty trial information was missing
in only 23 of the 594 for the PRS (id. 1104), in an estimated 20
to 30 of 1066 cases in the CSS (id. 1119-21). Plea bargaining
information -- information not on record facts about whether
bargains were accepted and pleas entered, but rather more informal
information on whether pleas had been unsuccessfully sought or
offered (id. 1152-53) -- was obtained for sixty percent of the
cases (id. 1153). As petitioner's expert noted (id. 1765-66;
see Fed. Oct. Tr. 82) and as commentators have agreed, missing data at a rate of 10 to 12 percent normally does not produce any
systematic bias in ultimate outcomes, see, e.g. Vuyanich v.
Republic Nat'l Bank of Dallas, 505 F. Supp. 224, 257 (N.D. Tex.
1980), vacated on other grounds, 723 F.2d 1195 (5th cir. 1984).3%
(b) The Quality of the Data-Gathering Instrument
During the data collection effort for the PRS and the..CSS, -
Baldus and his colleagues developed and employed three separate
questionnairies -- two for the PRS, and a third, modified and
improved instrument for the CSS. The initial PRS "Supreme Court
>
30/ To confirm those theoretical judgment, Baldus testified
that he performed a wide range of alternative analyses, including
those specifically recommended as appropriate by respondent's
experts (id. 1501), precisely in order to see whether these
missing data might have affected the persistent racial disparities
that he found (id. 1101; 1694-1708). None did.
-36=-
Questionnaire" (see DB 27) was 120 pages in length , devised
through a lengthy drafting process. "We sought to identify,"
Baldus testified, "any variable that we believed would bear on
[the] matter of the death worthiness of an individual offender's
case ... relating to the nature of the crime, the personal charac-
teristics of offender, characteristics of the victim" (id. 194-95).
od The initial Supreme Court Questionnaire proved of unwieldy
length for use in the field (id. 208). Therefore, although 330
cases in the PRS study were eventually coding using this instru-
ment (id. 200; see DB 28, at 2), Baldus developed a revised version, designated the "Procedural Reform Questionnaire" (see
DB 35). The Supreme Court Questionnaire was actually coded in
Iowa, by coders who employed copies of original court documents
obtained from official Georgia files (see, e.g., DB 29-33), as
well from detailed abstracts of the files and a written case
summary provided on each case by Baldus' Georgia coders, (see DB
33) (Fed. Hab. Tr. 208-15). However, the 351 Procedural Reform
Questionnaires were all filled out in Georgia, in the offices of
the public agencies involved, with "the source document literally
at [their] fingertips when [they] did the coding" (id. 366).
One major feature of both PR3 questionnaires (as well as
the CSS questionnaire) was their inclusion of a "narrative
summary" section, in which the coders could register important
information that was not otherwise covered the questionnaire.
As Professor Baldus explained, "([w]e had no illusion that our
questionnaire could capture every Auancs of every case. But we
wanted to be able to record that somehow. So we entered that
“3 7-
information on these ... summaries" (id) 2A Baldus also
created an "other" category for certain questions to permit a
coder to include unforeseen but possibly relevant information.32/
Despite the comprehensiveness of the PRS instruments, the
CSS questionnaire (see DB 38) marked a substantial improvement
in several respects. First, Baldus included a number of vari-
ables to capture the strength of the evidence (Fed. Hab. Tr.
274-75). Second, he added additional variables on legitimate
aggravating and mitigating factors (id. 274). Third, Baldus
virtually abandoned the "foil entry" format employed in the PRS
questionnaires, under which a coder could occasionally find too
few foils on which to enter relevant data in response to partic-
ular questions id.) A
31/ The District Court apparently misconceived Baldus' testi-
mony concerning these summary documents, stating that "an
important limitation placed on the data base was the fact that
the questionnaire could not capture every nuance of every case.
R. 239" (R. 1159). In fact, the summaries were included pre-
cisely to permit Baldus to capture such nuances.
32/ The District Court also treated this "other" coding feature
as if it were a deficiency in the questionnaire design, not an
asset (R. 1168). In fact, it permitted Baldus to capture addi-
tional information and determine whether some unforeseen factor.
may have had a systematic impact on his analyses (id.- 1708-09).
Baldus re-analyzed the "other" response in some of—-his alterna- ~=—=- :-
tive statistical analyses, finding that their inclusion "had no
effect whatever. It in no way diminished the racial effects. In
fact, it intensified them slightly" (id. 1710).
33/ The District Court faulted the questionnaires for their use
of the foil method (R. 1159-60), without making clear that this
method was largely a feature of the PRS study =-- which played
only a minor role in Baldus' analyses. Almost all of the major
analyses were conducted on the CSS data (id. 1437). Even so, as
a check on the impact of the foils, Baldus identified some 50 PRS
cases in which there was "overflow information ... that wouldn't
fit into the original foils," recoded all of the important
variables from the PRS in which the foil method had been employed,
re-ran his analyses and "found that the results were identical,
and in fact, the race effects became somewhat intensified when
this additional information was included" (id. 1099-1100). A
recoding of the only two items on the CSS questionnaire that had
retained the foil method obtained identical results (id. 1101).
“38
The State's principal expert conceded that the CSS instru-
ment was "an improved questionnaire" (id. 1392); indeed, respon-
dent never proposed or identified any variables or set of vari-
ables, not included in the analyses, that might have eliminated
the racial disparities reported by Baldus (id. 1609).
(¢c) The Care Employed in Coding
The coding process for both studies employed "state-of-the-
art" procedures designed to ensure uniform, accurate collection
of data. Initial coding for the PRS study was overseen by a law
graduate (id. 207-05) who developed with Baldus a written
"protocol," a series of careful instructions to coders meant to
achieve consistent treatment of issues by regularizing coding
practices (id. 227-23; see DB 34).
To complete the questionnaire for the CSS study, Baldus
employed as his supervisor Edward Gates, one of the two coders
who had earlier worked on the PRS study (id.). He recruited
five coders in a nationwide law school search (id. 301); Baldus
flew to Georgia for a week in June of 1981 to train the students,
/ explain the extensive written protocol (id. 310-1133 ; See DB
43) and code practice questionnaires with them (id. 309).
Throughout the summer, Baldus maintained daily telephone contacts
with Gates and the coders to resolve any issues presented by the
coding (id. 400).
The State's expert purported to test the coders' accuracy,
not by checking questionnaires obtained through discovery
34/ The written protocol, as this Court can observe from even a
quick review (see DB 43), involved hundreds of instructions on
both general coding issues and specific issues for particular
questions. The District Court's statement that "the coders were
given two general rules to resolve ambiguities of fact" (R. 1157)
does hardly justice to the care taken in providing guidance to
the coders.
-30~
against files in the State's possession, but by running computer
comparisons on those cases included in both the PRS and CSS
studies. This computer check generated a list of ostensible
"mismatches" which the State implied were indicative of multiple
coding errors. The District Court apparently credited this
argument (R. 1162).
The State's expert admitted, however, that in compiling
"mismatches" he had made no attempt to compare the coding
instructions from the PRS and CSS protocols, to see whether in
fact coders had been following identical rules (id. 1447). In
fact, as Baldus and Gates both testified, instructions for cod-
ing items in the two studies were often quite different. As a
general example, in the PRS, coders were required to draw reason-
able inferences from the file (id. 367); in the CSS, they were
not (id.). By way of further example, protocols for the coding
of the (b)(3), (b)(7) and (b) (10) aggravating circumstances were
very different in the PRS and CSS studies. In short, as the
State was forced to concede, "I don't believe Dr. Ratz is indi-
cating either one is necessarily right or wrong in ‘his judgment.
He's just indicating he's done a computer count and found these
inconsistencies" (id. 1444).
Professor Baldus testified on rebuttal that he had performed
an extensive analysis of the State's alleged mismatches, employ-
ing the official file materials and the narrative summaries, to
determine whether the inconsistencies represented coding errors,
rather than differences in PRS and CSS coding instructions or
differences due to data sources relied upon (id. 1718-19) (many
of the PRS cases were coded from Georgia Supreme Court materials,
“ld One
whereas all of the CSS cases were coded from the Pardons and
Paroles Board files). Baldus reported that "the
average mismatch rate was 6 percent, of which one percent ...
were attributable to either a coding error or a keypunching
error Or data entry error of one sort or another" (id., 1719).
Baldus added
"that translates into an error rate of approximately
one-half of one percent in each of the two studies.
However, we found on further examination that ...
the error rate in the Procedural Reform Study was
higher than it was in the Charging and Sentencing Study.
(Id. 1719-20). Since the CSS study was the basis for most of
Baldus' analyses (id. 1 ), it appears that the actual error 437
35/
rate was extremely low.
(d) The Basic Coding Conventions
The State vigorously attacked one coding convention relied on
by Baldus and his colleagues throughout the PRS and CSS studies:
the use of a "U" or "unknown" code. Edward Gates explained that
coders were instructed to enter a "1" if a fact were "expressly
stated in the file" (id. 444), a "2" if the fact were "suggested
by the file but not specifically indicated" (id. 444-45), a
blank if the fact were inconsistent with the file, and a "U" if
35/ The District Court noted that there were inconsistencies
between the coding of "several variables" for petitioner McCleskey
and his co-defendants (R. 1161). The Court's only reference is to
testimony indicating that in the PRS study, petitioner McCleskey
was coded as having three special aggravating factors while
co-defendant Burney is coded as having only two. Gates testified
that coding provisions for co-perpetrators in the CSS study were
"far superior ... in terms of precisely defining the differences
between the roles that the different actors in the crime played"
(id. 471). Once again the discrepencies appear to pose no threat
to Baldus' analyses, which were largely based on CSS data.
Indeed, although different coders were allowed to code the cases
Of co-perpetrators in the PRS (id. 1110-13), for the (CSS, Baldus
developed the practice of having a single coder complete ques-
tionnaires on all co-perpetrators (id. 1124-25).
ll Joe
the coder could not classify the item based on the file (id.).
As Professor Baldus explained:
"What an unknown means basically as it's coded in
the Charging and Sentencing Study is that the ...
information in the file, was insufficient to
support an inference as to the occurrences or the
non-occurrence of the event.... The idea was that if
the file would not support an inference of an occur-
rence Or non-occurrence, then we would further presume
that the person who created that file or who had the
information that was available in that file would be
in a state of ignorance with respect to that fact.
Furthermore, upon the basis of my knowledge of
decision making and also on the basis. of my prac-
tical experience, when people are ignorant about
a fact, that fact does not become a determinant
in the decision making.
(Id. 1684-85).
In sum, while the CSS instrument permitted the coders to
reflect the distinction between the affirmative non-existence of
a fact in the file (coded blank), and uncertainty about its
possible non-existence (coded "U"), once statistical analysis
began, the "U" was properly recoded as not present.
Baldus offered as an example of this logic the aggravating
variable that the "victim pled for his life." If there had been
witnesses present during the crime, a coder would code that .
variable either present or absent, depending on the witnesses’
accounts. But if there were no witnesses or other evidence,
Baldus reasoned there was no way to make an inference either
way, and the item would be coded "U" (id. 1685-86; see also id.
36
1155-58) . =
36/ The District Court's counter-example completely missed the
point. Twice the Court adverted to a case in which the defendant
told four other people about the murder, but in which the coder
was unable to determine from the file whether the defendant had
(Continued)
ag
This explanation casts in a radically different light the
District Court's ominous-looking list of variables coded "U" in
more than ten percent of the data (R. 1163-65). Many involve
either state-of-mind or relational variables that are often
unknown to any outside investigator. For example, while
"Defendant's Motive was Sex" may be important if known to a
prosecutor or jury, if the fact can be neither eliminated nor
confirmed from the evidence, Baldus' rule would be to code it
"unknown," and ultimately discount its impact either way by
treating it as non-existent. |
The District Court appeared to challenge the basic logic of
this coding treatment: "the decision to treat the "U" factors
as not being present in a given case seems highly questionable
«eo. it would seem that the more rational decision would be to
treat the "U" factors as being present" -(R. 1163). Yet no
expert in the case -- neither petitioner's (id. 1184-90 (Baldus);
36/ (continued)
been bragging or expressing remorse. (R. 1160, 1161-62). The
Court reasoned that "[als the witnesses to his statement were
available to the prosecution and, presumably, to the Jury, that
information was knowable and probably known. It was not,
however, captured in the study" (R. 1160).
The Court's reasoning assumes that the defendant must have
either been bragging Or expressing remorse, and that the prose-
cutor, by interviewing the four witnesses, must have ascertained
which. It is equally likely, however, that the defendant told
others about the murder without either bragging or expressing
remorse. In that case, the file would properly reflect the
contact with the witnesses, but would not reflect bragging or
remorse. Under Baldus' rules the coder would code "unknown" and
the bragging and remorse would ultimately be treated as not
having occurred. Only if the prosecutor and jury had known of
bragging or remorse, but the parole officer had somehow failed
to learn of it in his review of the transcript, in his talks
with the police and the District Attorney, or in his review of
police files, would "U" be a misleading code.
-i3~
1761-63 (Berk)) nor respondent's (id. 1503; (Katz); 1656-58
(Burford) ) suggested that a "U" should be coded as "1" or "present"
for purposes of analysis. Indeed, Dr. Berk, petitioner's
rebuttal expert, testified that the National Academy of Science
panel on sentencing had expressly considered this issue during
its two-year study of sentencing research and had endorsed the
very approach Baldus adopted (id. 1761-63). The District
Court's conclusion that a contrary code should have been used
is without foundation in the record. AY
(iii) The Statistical Methods Were Valid and Appropriate
Having gathered and compiled their data, Baldus
and his colleagues employed a wide variety of statistical
procedures to analyze it, including cross-tabular comparisons
(id. 683, 701-05), unweighted least squares regressions (id.
689-700) , weighted least squares regressions (id. 1222-25),
logistic regressions (id. 917-18), index methods (id. 1234-36),
and qualitative case comparisons, or so-called "cohort" studies
(id. 1043-59).
Baldus employed these methods on progressively more elaborate
"models," or groups of variables chosen to determine whether the
race-of-victim and race-of-defendant disparities could be reduced
37/ Moreover, Baldus testified that, among a series of alterna-
tives analyses he conducted to test the effects of his "U" coding
rules (see generally, Fed. Hab. Tr. 1194-1704 and DB 120-123),
he recoded unknowns as "1" or "present" just as the Court had
recommended. The effects on racial disparities "were within a
percentage point of one another and all the co-efficients that
were statistically significant in one analysis were in the
other" (id. 1701). Another alternative analysis, employing
"list-wise deletion" of all cases with "U" codes, recommended by
the State's principal expert, (id. 1501-02), also had no adverse
effect (id. 1695-96); see DB 120); indeed it increased the
race-of-victim coefficient by two percentage points.
4d
or eliminated: Baldus explained that no single methods of
statistical analysis, and no single model, was invariably infall-
ible, but that if statistical results could persist, no matter
what methods were employed, a researcher could have great confidence
that the "triangulated" results reflected real differences:
It's this widespread consistency that we see in the
i results ... it's this triangulation approach, if you
C- will, that provides the principal basis for our
opinions that there are real racial effects operating
in the Charging and Sentencing System."
(Id. 1082-83).
The District Court failed throughout to appreciate the
logic of this approach. Instead it rigidly, and petitioner
submits erroneously, refused to admit "except as to show process”
a series of relevant models, solely because they did not include
variables the Court thought should be included (see id. 742-46;
755; 760; 768; 771-73; 779: 981-82; 984). Indeed, the Court's
approach throughout the hearing was to fault Baldus' models for
failure to account for unspecified "unique" factors (e.qg., id.
38/
925; Fed. Oct. Tr. 92). = The Court reasoned -- contrary to
the expert testimony of Baldus (Fed. Hab. Tr. 808-19); Woodworth
(Fed, Oct, Tr. 55); and the State's expert Dr. Burford (id. 1673)
38/ The Court also overlooked in its opinion that, at the
invitation of petitioner's experts, it was able to test its
own "Lawyer's Model," constructed by the District Court during
the August 1983 hearing to reflect those factors it believed to
be most likely to predict the sentencing outcome (id. 810; 1426;
1475-76; 1800-03; see C-1). Baldus' subsequent analyses employ-
ing the Court's own model showed sharp differences in sentencing
outcomes by racial category (R., 735, 738). Strong and statis-
tically significant race-of-victim effects were reflected upon
regression analysis, whether employing the least squares (R.
738) or the logistic approach (R. 739), and Baldus averred that
these analyses further reinforced his earlier testimony. (See
generally, R. 731-752).
ll Ge
-- that since Baldus testified that he had identified 230
variables that might be expected to predict who would receive
death sentences, "it follows that any model which does not
include the 230 variables may very possibly not present a whole
picture” (R. 1171). If respondent had demonstrated that peti-
tioner's racial disparities only appeared in smaller models, but
disappeared or were substantially reduced whenever 230-variable
analyses were conducted, the District Court's position would
rest on logic and precedent. Since, however, as we will demon-
strate below, the race-of-victim disparities continue to show
strong effects in large models as well as small, the District
Court's position is without support. As a matter of fact, it is
clearly erroneous; not even the State's expert advanced such a
contention. As a matter of law, it has no allies. No prior case
has ever intimated that only large-scale models can constitute
relevant evidence in a statistical case. See, e.g., Eastland v.
Tennessee Valley Authority, 704 F.2d 613, 622-23 n.14 (11th Cir.
1983).
(iv) The Results Make Out A Prima Facie Case
Of Racial Discrimination
To begin his analysis, Baldus first calculated sentenc-
ing outcomes by race, unadjusted for any additional variables or
39/
background factors.— The pattern he found (DB 63) revealed
40/
marked racial disparities:
39/ Each of these analyses was conducted on the CSS data, unless
otherwise noted.
40/ These results closely parallel earlier Georgia findings.
Bowers & Pierce, "Arbitrariness and Discrimination under
Post-Furman Capital Statutes," 26 CRIME & DELINQ. 563, 599
(1980).
lif
Black Defendant/ White Defendant/ Black Defendant/ White Defendant/
White Victim White Victim Black Victim Black Victim
«22 .08 +0] +03
(50/228) (58/745) (18/1438) (2/64)
(id. 730-31)
However, Baldus made it clear that "[t]his table merely generates
an hypothesis ... it has no controls. There are many rival
hypotheses that could explain these relationships" (id. 731).
Baldus thus began a series of analyses, steadily adding
background variables to his multiple regression analyses,
thereby "controlling for" or holding constant the effect of those
factors, to see if an independent racial effect would persist.
Baldus found strong racial effects when he controlled for all of
Georgia's statutory aggravating circumstances (DB 78) and in
addition, for 75 mitigating factors (DB 79). In DB 80, Baldus
presented an important table which compared the racial effects
in several, increasingly complex models. Excerpts from that
table reveal the following:
After further After
Before After Adjust- Simultaneous Simultaneous
Adjustment ment for the Controls for Control for
for any Back- Other Vari- Nine Background 230 + Non Racial
ground Factors able Racial Variables Factors
Race of
Victim +10 «17 «07 .06
{.0001) (.0001) {.001) {.01)
Race of
Defendant -.03 + 10 .04 .06
{.03) (.001) {.10) {.01)
: il 41 ty
Baldus noted that while the coefficients™ for race-of-victim
declined somewhat as additional background variables were added
41/ Professor Baldus testified that a regression coefficient
is a summary figure that provides the average disparity, with
(Continued)
5 he
to the analysis, and that while the measures of statistical
significance also declined, 2% both figures remained signifi-
cant. Baldus explained that it is "quite unusual to see an
event like that," since so many of the 230 variables were
themselves correlated with both the race of the victim and the
sentencing outcome, a fact that could be statistically expected
to suppress the magnitude of the racial variable (id. 804).
To examine the relative power of the race-of-victim
+ and race-of-defendant variables in sentencing decisions, Baldus
compared them with other important sentencing variables, rank-
ordered by their coefficients (DB 81, 82). The impact of the
race-of-victim variable proved of the same order of magnitude
as major aggravating factors such as whether the defendant had
a prior record of murder, or whether the defendant was the
prime mover in the crime (id. 812-15).
Baldus then continued his analyses, looking at other
models that might eliminate the racial effects. Petitioner's
Exhibit DB 83 includes a variety of such models, some employing
all 230 of Baldus' recoded variables. All of these models show
41/ continued
and without the presence of a variable, across all the cases
(id. 690-94). A coefficient of .06 for a variable means that
the presence of that variable, after controlling for all other
factors in the model, would increase the outcome of interest
(here, a death sentence) by an average of six percentage points
(id. 692-93},
42/ Statistical significance, Baldus explained, is a measure
of the likelihood that, if in the universe of cases as a whole
there are in fact no disparities, one could have obtained
disparate results merely by chance (id. 712-15). Normally
expressed in "p" values, a figure of .01 means the likelihood
that the coefficient is merely a chance finding is 1-in-100; a
figure of .0001 would mean 1-in-10,000.
-lG-
43/
strong race-of-victim and race-of-defendant effects.
I. W.L.S. REGRESSION RESULTS
A B c
Coefficients and Level
Non-Racial of Statistical Significance
Variables in
The Analysis Race of Victim Race of Defendant
a) 230 + aggravating, .06 .06
mitigating, evidenti- {.02) (+02)
ary and suspect
factors
b) Statutory aggravating «07 .06
circumstances and {.01) (.01)
126 factors derived
from the entire file
by a factor analysis
¢) 44 non-racial vari- +07 .06
ables with a statisti- (.0002) (.0004)
cally significant
relationship (P<.10)
to death sentencing
d) 14 legitimate, non- .06 .06
arbitrary and statis- (.001) {.001)
tically (P<.10) sig-
nificant factors
screened with W.L.S.
regression procedures
e) 13 legitimate, non- .06 .05
arbitrary and statis- {.001) {.02)
tically significant :
(P<.10) factors
screened with logistic
regression procedures
Baldus adopted yet a different approach to analyze
precisely where in the system the racial effects were having
their impact. Employing a recognized social science technique,
43/ In light of DB 81 and DB 83, as well as DB 102 and DB 105,
the District Court was clearly erroneous in asserting that
"[tlhe best models which Baldus was able to devise which
account to any significant degree for the major non-racial
variables, including strength of the evidence, produce no
statistically significant evidence that race plays a part in
either of those decisions in the State of Georgia" (R. 1187)
-td G
the "index method," (see id. 877, 1234-36) he sorted the cases
into roughly equal groups based upon their predicted likelihood
of receiving a death sentence (id. 877-79); he then analyzed
racial disparities within those groups, which included increas-
ingly more aggravated cases (see DB 89). Noting that the likeli-
hood of a death sentence rises dramatically in the most aggravated
groups, Baldus further divided the top groups into eight subgroups
for analysis. As the excerpted portion of that table (DB 90)
reveals, there are clear race-of-victim differences -- especially
in the middle range of cases -- which are statistically significant
overall at a .01 (1=-in-100) level.
A | B c 2 E
Predicted Average
Chance of Actual Death Sentencing Arithmetic
a Death Sentencing Rates for Black Difference
Sentence Rate for Defendant Involving in Rate of
1 (least) the Cases White Black the Victim
to 8 at Each Victim Victim Rates
(highest) Level Cases Cases {Col. C~ Col. DD)
1 .0 «0 0
(0/33) (0/9) (0/19) 0
2 «0 0 .0 0
(0/55) (0/8) (0/27)
3 .08 30 11 «19
(6/76) {3/10) (2/18)
4 «07 a3 «0 vos
(4/57) (3/13) (0/15)
5 $27 «33 vi? +18
{15/58) (9/26) (2/12)
6 +17 «35 05 «33
(11/64) (3/8) (1/20)
7 .88 «91 75 +16
(51/58) 120/22) (6/8)
Baldus observed that there was little disparity in the less
aggravated cases, "[b]lut once the death sentencing rate
begins to rise, you'll note that it rises first in the white
5 J
victim cases. It rises there more sharply than it does in the
44
black victim cases" (id. ss2-83) . 2 Baldus testified that,
in his opinion, these data supported an hypothesis first advanced
by Harry Kalven and Hans Zeisel in their work, THE AMERICAN JURY
164-67 (1966),
"what they call the liberation hypothesis and in short
what it was, that the exercise of discretion is concen-
trated in the area where there's real room for choice.
[Wlhen you look at the cases in ... the midrange,
where the facts do not call clearly for one choice or
another, that's where you see there's room for exercise
of discretion ... the facts liberate the decision
maker to have a broader freedom for the exercise
of discretion, and it is in the context of arbitrary
decisions that you see the effects of arbitrary or
possibly impermissible factors working.
(Id. 844)
Baldus and Woodworth marshalled a substantial body of
evidence in support of this liberation hypothesis during the
evidentiary hearing. The most striking illustration was the
figure constructed by Woodworth to illustrate the differential rates at which the likelihood of receiving a death sentence rises
in Georgia for black victim and white victim cases, given similar
levels of aggravation. Woodworth noted that, according to this
graph, petitioner Warren McCleskey's level of aggravation
"place([s] him in a class of defendants where there is roughly a
44/ The District Court apparently misunderstood those tables.
It noted, as if the fact were contrary to Baldus' testimony,
that DB 89 reports "higher racial disparities in the most aggrav-
ated level of cases," (emphasis added). It also discounted the
results in DB 90 because, unlike DB 89, it was purportedly not
"predicated on a multiple regression analysis" (R. 1205). In
fact, the liberation hypothesis predicts that disparities would
exist only at the higher levels of DB 89, a table that includes
all cases -- most of them very unaggravated. It is only in DB 90,
which comprises the subset of cases in which the risk of a death
sentence becomes significant, that the disparities in the middle
range appear. (Fed. Hab. Tr. 882-83) Like DB 89, moreover, DB 90
was built by employing regression analysis; the Court's surmise to
the contrary is clearly erroneous.
-5 lw
Gwe
Figure 2: Midrange’ Model With Interactions and Nonlinearities--
Black Defendants
25. 4
pod ; McCleskey
» T » be ie be 5
: 0 .6 8 Lo og
™ LEVEL OF AGGRAVATION
a/ re ;
=’ The curves represent 95% confidence bounds on the average death
sentencing rate at increasing levels of aggravation (redrawn {rom
computer output).
twenty percentage point of greater disparity between black victim
cases [and] ... white victim cases," (id. 1734-35).
[See GW 8]
Baldus performed a wide variety of further analyses
which we cannot fully review within the confines of this brief.
« A few, however, require additional attention. The District Court,
unguided by experts for either petitioner or respondent, suggested
that DB 95 was "perhaps the most significant table in the Baldus
study," since it "measures the race of the victim and the race of
the defendant effect in the prosecutorial decision to seek the
death sentence and the jury decision to impose the death sentence"
- (R. 1185). The Court noted that "[t]lhe coefficients produced by
the 230-variable model on the Charging and Sentencing Study data
base [in DB 95] produce no statistically significant race of the
victim effect either in the prosecutor's decision ... or in the
jury sentencing decision" (R. 1186).
The Court's statement in a literal sense is accurate.
It disregards, however, that the CSS figure, P=.06, is in fact
marginally significant; that the equivalent PRS model does pro-
duce a statistically significant resulta’ that the smaller
46
model results were highly significant; 1¢/ and that an analysis
45/ The Court discounted this figure as "totally invalid
for [the PRS Model] contains no variable for strength
of the evidence" (R. 1185). In so doing, it ignored Baldus'
obvious point that strength of the evidence was substantially
controlled for in the PRS, since the universe was limited by
definition to cases in which a conviction -- presumably based
on evidence sufficient beyond a reasonable doubt -- had
been obtained (Fed. Hab. Tr. 124-25).
46/ The Court stated that it "knows of no statistical convention
which would permit a researcher arbitrarily to exclude factors on
she basis of artificial criteria” (R., 1186). 'Baldus in fact
testified without contradiction that such a procedure is commonly
used in statistical analyses. (The State's principal expert
employed a variant of it throughout his testimony.) (See, e.g9.,
Resp, Ex. 26, 43, 45, 50).
ly
of the combined effect of the prosecutorial and jury decision
(see DB 98) showed a series of highly statistically significant
race-of-victim effects. In truth, what the Court has done is to
identify one of the very few large model coefficients for the
race-of-victim variable in either study that is not statistically
significant, brand it as a key figure, and then disparage all
collateral evidence that places it in context. Such an approach
to petitioner's comprehensive statistical evidence constitutes a
legally insufficient basis to reject petitioner's persistent
racial tindings 2
The second series of analyses that require comment
are those directed toward Fulton County (where petitioner was
tried) and toward petitioner's own case. Baldus conducted both
quantitative and qualitative studies of death sentencing rates
in Fulton County which were reflected in DB 106 through DB 116.32
Baldus testified that a repetition in Fulton County of the pro-
gressively more elaborate analyses he had conducted statewide
"showed a clear pattern of race-of-victim disparties in death
sentencing rates among the cases which our analyses suggested .
were death eligible" (id. 983). Regression analyses~at ‘succes= =
47/ The District Court also chose to impugn the integrity
of petitioner or his experts in discussing this exhibit,
noting that "we are given no outcomes based on _the larger . ..
scaled regression," although the Court "does not understand.
that the analysis was impossible, but instead ... that
because of the small numbers the result produced may not have
been statistically significant" (R. 1187). The Court is
wrong; such analyses employing these small numbers are
statistically inappropriate. See e.g., Halinski and Feldt,
"The Selection of Variables in Multiple Regression Analysis,"
7 J. EDUC. MEASUREMENT, 151 (1970). We note, morever, that
both in this table and elsewhere, petitioner and his experts
regularly reported non-significant findings even when statisti-
cal procedures could be appropriately conducted upon them.
48/ The District Court refused to admit DB 106 (id. 979),
DB 107 (id. 981-92), and DB 108 (id. 984), holding that because
they did not sufficiently control for background variables they
were irrelevant. This holding is legally erroneous.
-53
sive stages in the charging and sentencing process revealed highly
significant racial disparities at two points: the prosecutor's
plea bargaining decision and the prosecutor's decision to advance
a case to the penalty phase (id. 1038-39). While Baldus necessar-
ily tempered his evaluation of these results because of the small
size of the universe (id. 1040-43), he noted that "these coeffi-
cients are very large, it's not as if we're dealing with small
coefficients, these are substantial. So that leads me to believe
that what you're seeing is evidence of a real efficient" (id. 1044).
To supplement this statistical picture, Baldus con-
ducted two eohort studies; one of the "near neighbors" cases,
those which scored most like petitioner McCloskey in an overvall
"aggravation index" (id. 986-91). Having identified 32 near
neighbors, Baldus sorted them into typical, more aggravated, and
less aggravated groups (id. 991). Computing death sentencing
rates by race of victim and race of defendant, Baldus found sig-
nificant disparities; in McCleskey's group, the disparity was
«40. (14. .993).
In a second cohort study Baldus examined 17 defendants
involved in the homicides of police officers. Two among the
seventeen, including petitioner McCleskey, went to a penalty
trial. The other defendant, whose police victim was black,
received a life sentence (id. 1050-62; DB 116); petitioner's
sentence was, of course, death. "[T]he principal conclusion
that one is left with," Baldus testified, "is that ... this
death sentence that was imposed in McCleskey's case is not
consistent with the disposition of cases involving police
officer victims in this county" (see also 1085-86).
Finally, Dr. George Woodworth, petitioner's expert
statistician, testified concerning the likely impact of the
Bl
racial variables on a case at petitioner McCleskey's level of
aggravation. Woodworth noted that, using his exhibit GW-8,
he had computed the race-of-victim disparity at petitioner's
level of aggravation to be 22 percentage points (id. 1738).
He then turned to DB 90 and observed an 18 percentage point dis-
parity by race at petitioner's level (id. 1739). Calculated by
use of an unweighted logistic regression, the racial disparity
was 23 percent (id. 1740). Woodworth concluded:
So it would seem that at Mr. McCleskey's level of
aggravation the average white victim case has approxi-
mately a twenty percentage point higher risk of receiv-
ing the death sentence than a similarly situated black
victim case.
(1d. 1740). 2%
Petitioner's final expert was Dr. Richard Berk,
a highly qualified social scientist (see RB 1) and a frequent
consultant on criminal justice matters to the United States
Department of Justice (id. 1753). Berk in fact had served on
a distinguished National Academy of Sciences panel charged
with reviewing all previous research on criminal sentencing
issues in order to set standards for the conduct of such
research (id. 1761-62). After receiving Baldus' studies, = ==.
49/ Beyond this statistical and qualitative evidence on cases
like petitioner's, petitioner introduced the deposition of Dis-
trict Attorney Lewis Slayton (id. 1319). In that deposition,
Slayton acknowledged that his office has no express written =
or unwritten policies or guidelines to govern the disposition of
homicide cases at the indictment stage (Dep., 10-12), the plea
stage, (Dep. at 26) or the penalty stage (Dep., 31, 41, 58-59).
Moreover, murder cases in his office are assigned at different
stages to one of a dozen or more assistant district attorneys
(Dep., 15, 45-48), and there is no one person who invariably
reviews all decisions on homicide dispositions (Dep., 12-14,
20-22, 28, 34-38). Slayton confessed that his office does not
always seek a sentencing trial in a capital case, even when
statutory aggravating circumstances are present (Dep., 38-39).
Slayton testified further that the decisionmaking process in his
office for seeking a death sentence is "probably ... the same" as
it was in the pre-Furman period (Dep., 59-61).
55m
analyzing the data, and reviewing Baldus' preliminary report,
Berk's opinion on Baldus' study, especially its findings on
race, was virtually unqualified:
This has very high credibility, especially compared
to the studies that [The National Academy of Science
panel] ... reviewed. We reviewed hundreds of studies
on sentencing over this two-year period, and there's
no doubt that at this moment, this is far and away
the most complete and thorough analysis of sentencing
that's been done. I mean there's nothing even close.
(Id. 1766.)
Berk's conclusion is fully warTAnted. The data was
reliable and carefully compiled. The regression analyses
relied upon by petitioner were properly conducted by leading
experts in the field. These analyses were carefully monitored
for possible statistical problems, and they 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. These disparities are real
and persistent; they establish petitioner's prima facie case.
C. The Law: The District Court Misapplied the Law
In Rejecting Petitioner's Prima Facie Case
We have already pointed out many instances in- which
the District Court misread the record, overlooked testimony,
or made findings contrary to the evidence presented by
both parties -- petitioner and respondent alike. Yet the
: principal errors committed by the District Court on this
record stem from its apparent misunderstanding of statistical
proof, and its misapplication of controlling legal authority.
In effect, the District Court created for itself a roster of
new legal standards and principles to judge the quality of
petitioner's data, the admissibility of his exhibits, the
appropriateness of his models, and even the usefulness of
-56~-
multiple regression as a mode of proof. In fashioning this new
jurisprudence, the District Court departed from well-established
Supreme Court and Circuit precedent, requiring this Court to
vacate and remand for further proceedings in compliance with
appropriate legal standards.
(i) Petitioner's Data Clearly Exceed Minimum
Evidentiary Standards For Use In Regres-
sion Analysis
In assessing petitioner's statistical case, the
District Court announced that "[m]Jultiple regression requires
complete correct data to be utilized" (R. 1169); it clearly pro-
ceeded to hold petitioner to such a standard. Although acknowl-
edging that "the researchers attempted to be careful in [their]
data-gathering," the Court complained that "the final data base
was far from perfect," (R. 1159), noting that "[a]n important
limitation" of the data was "that the questionnaire could not
capture every nuance of every issue" (id.). The Court faulted
the data sources (erroneously, see id. 1117) because they "con-
talin] no information about what a prosecutor felt about the
credibility of any witness" (R. 1160). Indeed, the Court appeared
to hold that virtually any retrospective study would necessarily
be insufficient:
: To the extent that the records of the Parole Board
- accurately reflect the circumstances of each case,
they present a retrospective view of the facts and
circumstances. That is to say, they reflect a view
of the case after all investigation is completed,
after all pretrial preparation is made, after all
evidentiary rulings have been handed down, after
each witness has testified, and after the defendant's
defense or mitigation is aired.... Further, the file
does not reflect what was known to the jury but only
what was known to the police.... Consequently, the
court feels that any model produced from the data base
available is substantially flawed because it does not
measure decisions based on the knowledge available to
the decision-maker.
(Rs 1172.)
BE
The Court's insistence on such a standard of proof
was plain error. Statistical evidence of systematic racial
discrimination in violation of the Fourteenth Amendment simply
is not held to this pristine standard. To the contrary, it is
the respondent who must shoulder a "heavy burden" to come forward
with affirmative evidence that any "errors or omissions bias the
data [and] ... that this bias alters the result of the statistical
analysis in a systematic way," Vuyanich v. Republic National Bank,
supra, 505 F. Supp. at 255-56, vacated on other grounds, 732 F.2d
1195 (3th Cir. 1984); accord, Trout v. Lehman, 702 F.2d 1094,
1101-02 (D.C. Cir. 1983); Detroit Police Officers Ass'n v. Young,
608 P.24 671, 687 (6th Cir. 1979), cert, denied, 452 U.S. 938
(1981); cf. International Brotherhood of Teamsters v. United
States, 431 U.S. 324, 340 .n.20 (1977).
This legal standard is well justified as a matter
of scientific principle. Dr. Richard Berk indicated that missing
data or errors "of the order of 10, 15 percent, almost never make
a difference" in the outcome (Fed. Hab. Tr. 1766) .23/ Both
Baldus and Woodworth stressed that "errors which are randomly =
distributed with respect to the race of the defendant, or-the race
of the victim, are not going to create a race of the victim dis-
parity. In other words, they're not going to bias the effects"
(id. 1727-28; gee id. 1720; Fed. Oct. Tr. 82).
Here respondent offered no plausible explanation of
how the use of its own official State files might have resulted
in racial bias. Nor was there evidence that the data base was
grossly flawed. The Court's chief criticism of the data stemmed
from its misunderstanding of the logic and appropriateness of the
"U" or "unknown" coding convention, which we have earlier shown to
-58-
be scientifically appropriate. The alleged "mismatches" have been
largely accounted for as well. As for the small percentage of
missing data with respect to the race of the victim, and on whe-
ther plea bargaining had been sought and penalty trials held,
Baldus himself called these to the Court's attention, and pro-
ceeded to conduct a series of analyses =-- including every test
suggested by the Court or the State, and others besides -- to
discern whether "missing data" might affect his principal con-
clusions. He found the changes in outcome to be uniformly
insignificant.
In short, since the District Court applied an incorrect
legal standard in assessing petitioner's data base, its conclusion
that the data "is essentially untrustworthy" must be rejected by
this Court.
(ii) Multiple Regression Analysis Is an
Appropriate Means Of Proof
The single gravest error into which the District
Court fell was its conclusion -- only partially acknowledged
during many points in its opinion, although plainly stated at
others -- that multiple regression analysis is an inappropri-
ate and unacceptable method of proof. At the outset of its __
opinion, the Court frankly expressed its preference for a
simpler form of data analysis, the cross-tabular method:
To determine whether or not race [is] .... being con-
sidered [in the Georgia capital sentencing system], it
is necessary to compare very similar cases. This
suggests the use of a statistical technique known as
cross tabulation. Because of the data available, it was
impossible to get any statistically significant results
method. R. 705. Accordingly, the study principally
relies on multivariate analysis.
(R. 1153.) Petitioner indeed presented the District Court with
several cross-tabular analyses examining the impact of the racial
50
variables (See DB 66, 68, 71, 72, 76) (nearly all of which
were statistically significant =o
Yet petitioner's claim required more sophisticated
proof, which petitioner also placed before the Court. Despite
a careful attempt to illustrate the principles of regression
analysis (see Fed. Hab. Tr. 639-700, 709-18, 317-18), the Court
expressed skepticism throughout the hearing about whether regres-
sions could actually work. (E.g. "I have: a threshhold concern
about whether regression analysis is applicable to this aspiry
[sic]") (Fed. Oct. Tr. at 61); "[Flundamentally, what I am trying
to say, I don't understand regression analysis well enough to be
convinced that it demonstrates that equals are treated unequally
as a reference to any factor" 1d. at 63; see also id. 67, 101).
That basic skepticism, never dispelled, explicitly
colored the District Court's entire evaluation of petitioner's
proof. In a subsection entitled, "What a Multivariate Regression
Can Prove," (R. 1190), the Court purports to contrast what:
"[Baldus] means when he says that he has 'controlled' for other
independent variables" with "[w]lhat these terms usually mean"
(R. 1196), proceeding to describe the "usual" meaning of controls~ -
solely in terms of the cross tabular method -- "that a researcher
has compared cases where the controlled-for variables are present
in each case and where the cases are divided into groups where the
variable of interest is present where the variable of interest is
21/
not present" (R. 1196-97). The Court continues:
50/ The principal limitation of the cross-tabular approach, as
Baldus explained, is that it must subdivide its sample as each new
variable is added to the analysis, quickly depleting all but
extraordinarily large samples. (Fed. Hab. Tr. 705).
51/ All of the expert testimony at the hearings confirmed,
of course, that multiple regressions do control for other
(Continued)
50
With these difficulties, it would appear that multi-
variate analysis is ill suited to provide the court
with circumstantial evidence of the presence of dis-
crimination, and it is incapable of providing the
court with measures of qualitative difference in
treatment which are necessary to a finding that a prima
facie case has been established with statistical evi-
dence. Finally, the method is incapable of producing
evidence on whether or not racial factors played a
part in the imposition of the death penalty in any
particular case. To the extent that McCleskey contends
that he was denied either due process or equal pro-
tection of the law, his methods fail to contribute
anything of value to his cause.
(Continued)
background variables (see e.g., Fed. Hab. Tr. 691-96; Fed.
Oct. Tr. 6-60; 77-79). Multiple regression analysis "is ...
a substitute for controlled experimentation," Vuyanich v.
Republic National Bank, supra, 505 F. Supp. at 269; accord,
Eastland v. TVA, supra, 7/04 F.2d at 621; Fisher, supra, 80
COLUM. L. REV. at 706.
The District Court raised two other concerns about
regression analysis: first, that it "requires continuous
dependent and independent variables" (R. 1194); and
secondly, that "the size of a regression coefficient ...
says nothing about the specific degree of disparity" (R.
1197), and thus "one cannot use the size of the regres-
sion coefficient as an indication of the relative
strength of one variable to another" (R. 1197-98). Both
concerns are misplaced. Dr. Woodworth, an expert in
dichotomous outcome analysis, testified unequivocally
that "there is no problem in controlling for a dichotomous
independent variable" in regression analysis (Fed. Oct.
Tr. 30). See, e.g., Fisher, "Multiple Regression in Legal
Proceedings," 80 COLUM. L. REV. 702, 722 (1980) (The
State's experts did not dispute this point). As for
the second issue, the experts cited by the Court on the
difficulties of interpreting regression coefficients say
only that one must be careful in comparing coefficients
for continuous variables (e.g. age, number of convictions),
. with coefficients for dichotomous variables (presence or
absence of a factor). No such problems attend compar-
isons solely among dichotomous variables, and Baldus
carefully observed the distinction in his analyses
(id. 1782-84) Berk fully endorsed the principle that
Baldus' coefficients were meaningful. This testimony is
consistent with the scientific literature. See, e.g., J.
NETER & W. WASSERMAN, APPLIED LINEAR STATISTICAL MODELS,
at 229 (1974); Taylor, "Analyzing Qualitative Data," in
P. ROSSI, J. WRIGHT & A. ANDERSON, eds., HANDBOOK OF
SURVEY RESEARCH, at 576.
-)
2
(R. 1190-1200) (underscoring opliindy So
In staking out this extraordinary legal position,
the District Court has set itself squarely in opposition to
established legal authority in this Circuit. The former
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, having gained greater familiarity
with the method, the Court observed that "[m]Jultiple regres-
sion analysis is a relatively sophisticated means of deter-
mining 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), vacated and remanded on
other grounds U.S. , 103 S.Ct. 34 (1982). 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 Aiiorintnation ... in a case where a number
of factors operated simultaneously to influence" the outcome
of interest. Id. at 403.
With proper attention to its possible misuse,
this Court has thus also embraced multiple regression
52/ One of the major reasons why the Court misunderstood
some of these basic statistical principles may have been
its refusal to admit Baldus' extensive report into
evidence (see Fed, Hab. Tr. 1008-31; DB 113), even
though a copy had been provided to the State's experts
during discovery, and though Baldus and Woodworth both
stood ready to submit full cross-examination on its
contents. Petitioner contends that this exclusion
was clear error, on each of the grounds he presented to
the District Court (id. 1008-31).
fs J
analysis as an appropriate tool for the proof of discrimination
claims. See, e.g., Eastland v. TVA, supra, 704 F.2d at 621-22; see
also, Valentino v. United States Postal Service, 674 F.2d 56, 70
(D.C. Cir. 1982); see generally, Finklestein, "The Judicial
Reception of Mutiple Regression Studies in Race and Sex Discrim-
ination Cases," 80 COLUM. L. REV. 737 (1980).
Whether multiple regression is legally appropriate as
a means of proof does not, in sum, depend on whether a district
court is able to understand of the underlying regressions. When
real, persistent racial differences are presented in a carefully
conducted statistical analysis, a federal court must take them
seriously =-- not dismiss them summarily with the wish that a less
mathematically sophisticated method of proof were available.
(iii) The Models Employed By Petitioner Were
Statistically Appropriate
The District Court's third major error in its eva-
luation of petitioner's case was its set of unfounded assumptions
about the "models," or sets of variables, that Baldus should have
used to conduct his analyses.
On the one hand, the Court apparently believed that
no model is sufficient unless it can control for all possible
racially neutral variables, since "any model which does not
. include the 230 variables may very well possibly not present
. a whole picture" (R. 1171). The Court did not merely treat
models with fewer variables as less reliable -- it discounted
them altogether. Thus Baldus' "worst case" reanalyses,
conducted to check the effect of "missing data", were dismissed,
since Baldus "used a 39-variable model [and] ... did not test
—63=
any of his larger regressions" (R. 1167). Similarly, Woodworth's
extensive series of diagnostic tests to see if possible statistical
biases had crept in was Pranded virtually worthless, because
"[n]Jone of Woodworth's models in which he performed his diagnostics
are large order regression analyses" (R. 1179). The "triangula-
tion" approach is unsound, the Court stated, because "Baldus is
taking his bearings off of many models, none of which are adequately
inclusive" (R. 1182). Baldus' 39-variable model "is by no means
acceptable" (R. 1205) 2
What evidence or legal principle warrants the District
Court's position? None we can locate. Baldus expressed the
expert opinion that a 39-variable model best "captured the
essence of this system in terms of the main determinants ...
those [39] variables constituted the most plausible rival hypo-
thesis ... in the sense that they were statistically important
and also they were important in terms of our general understanding
of how systems of this type operate" (Fed. Hab. Tr. 808, 815).
Expanding on this point, Baldus explained that no model is per-
53/ while insisting that only petitioner's large-scale models
were appropriate, the District Court nevertheless stressed the
- ostensibly severe problems of multicollinearity associated with
g large models (see R. 1176-80). The Court's treatment of this
issue is marked by confusion and error. The Court confuses multi-
collinearity =-- which "reduces the precision of the standard
. [error of] estimates with respect to the variables that are in-
terrelated to one another” (Fed. Hab. Tr. 1166-67) =-- with the
possibility that "racial variables are serving as proxies for
other permissible factors" (R. 1176). Multicollinearity affects
only the statistical significance of a result; it does not bias
the coefficients reported. See, e.g., Fisher, supra, 80 COLUM.
L. REV, at 713. (Fed, Hab. Tr. 1281-82; accord, 1782.)
Indeed, since the impact of multicollinearity is to reduce
the level of statistical significance, racial variables which
show a statistically significant result in large models are
2l] the more impressive (id. 1136).
The issue of whether race is serving as a proxy for other
variables is precisely what a regression analysis, by controlling
for such other variables, is designed to resolve. Petitioner did
control for 230 variables, and racial effects remained strong.
fd
LJ
AJ
fect. Very large models present certain "technical concerns" as
a statistical matter; smaller models raise the concern that an
important variable has been omitted. It was for that reason,
Baldus testified, that he employed a variety of models (id.
830-21).
Dr. Woodworth testified that the 39-variable or
"mid-range" model, which he used in all of his diagnostic
tests, "does an excellent job of predicting the [sentencing]
outcome ... predicting about 93 percent of the outcomes
correctly," (id. 1290). Even the State's expert admitted that
"[i]lt is theoretically possible that a twenty-variable model
with no interdependency could be more reliable than a 230-vari-
able model if a lot of the variables were interdependent, or
interrelated" (Feb. Hab. Tr. 1673). Franklin Fisher makes a
similar point in his influential article on the legal use of
regressions:
Without some theory about which variables are likely to
matter, throwing a great number of variables into the
hopper is likely to lead to spurious results ... [W]lhen
having a study done by an expert, one should not be too
insistent about covering every possibility at once.
Rather one should make sure that the expert proceeds
by estimating a reasonable model including the major
variables and then goes on to test other possibilities.
If one insists that all possible variables are likely to
be of equal importance, one is likely to end up with a
rather doubtful result."
Fisher, supra, 80 COLUM. L. REV. at 715.54/
The District Court's second erroneous assumption
about statistical models is, ironically, tied to the first.
54/ Petitioner of course stood ready throughout the August
1983 hearing, and again at the October hearing, to respond to
all challenges to its claims. If the Court had then indicated
any serious reservations about the reliability of petitioner's
diagnostic tests because of the 39-variable model, petitioner's
experts had the means available and would have been quite will-
ing to conduct important analyses with larger models.
-65=-
Even the 230-variable models, the Court held, were "insuffi-
ciently predictive to support an inference of discrimination"
(R. 1173; id. 1174; 1181), since they had a low ve? and did
not predict all of the outcomes. Indeed, the Court even faulted
its own "Lawyer Model" on this account: "I tried to suggest [what
factor are important] in the model I designed and obviously I
didn't do well because I didn't even predict half of the death
penalties" (Fed. Oct. Tr. 65).
The flaw in the Court's reasoning, as Woodworth
explained, is its assumption that there are no arbitrary or
random features to the system. If arbitrary or unique factors
play a part, no number of variables could account for them.
Woodworth testified that it was "possible for a set of variables
to have an r-square of .39 or .46 and yet still account for all
the significant variation apart from the randomness." Moreover,
Woodworth, an expert on dichotomous outcome analysis, added that
2 numbers dichotomous data typically produced artificially low r
2
-- that, in fairness, the appropriate "biserial r " for such
ie
models is roughly double the unadjusted r (id. 1289).33/
Employing such recalculations, most of petitioner's models seem
to have rg of between .70 and .90.
To summarize, the District Court's unwarranted assump-
tions about regression models led it into an erroneous legal and
factual judgment of petitioner's evidence. On this third major
ground, the judgment of the Court must be reversed.
55/ See, e.9., J. GUILFORD & B. FRAUCHTER, FUNDAMENTAL STATIS-
TICS IN PSYCHOLOGY AND EDUCATION, 6th ed., at 311 (1969); Fisher,
supra, 80 COLUM. L. REV. at 7319;
B86
D. The State Presented No Significant Rebuttal Case
Because petitioner made out a prima facie case, the legal
burden shifted to respondent to rebut that case in one of three
ways: (i) "by showing that plaintiff's statistics are mislead-
ing," Eastland v. TVA, supra, 704 F.2d at 613-19; (il) "by pre-
senting legitimate non-discriminatory reasons for the disparity,"
id.; or (iii) by proving that racial discrimination is justified
in this context by a compelling state interest. See generally,
Johnson v. Uncle Ben's, Inc., 628 F.24 419, 423-24 (5th Cir.
1980), cert. denied, u.S. 103 S.Ct. 233 (1982).
what did the State do to meet these burdens? It offered
no alternative model which might have reduced or eliminated the
racial disparities (Fed. Hab. Tr. 1609). It did not even propose
-- much less test the effect of -- any additional explanatory
variables (id.), admitting that it did not know whether the addi-
tion of other variables "would have any effect or not" (id. 1567).
The State declined petitioner's invitation to propose a recombina-
tion of Baldus' variables that might yield a different result
(id. 1475-76). It performed no multivariate analyses of its own
to test whether white and black victim cases are being treated
differently (id. 1615). Nor did it point to any analysis by
Baldus in which the racial effects disappeared or ran counter to
L
[]
petitioner's claims.
The State, in short, presented no affirmative statis-
tical case on rebuttal at aly, 25
56/ In one regard, at least, the failure is unsurprising.
Neither of the State's experts had background or training in
any area related to the criminal law (Fed. Hab. Tr. 1356-57;
Resp. Ex. 75). In its opinion, the District Court clearly
misread the record by reciting that Dr. Katz "was qualified
as an expert in analyzing data, in research design, in sta-
tistics, statistical analysis and quantitative methods.
(Continued)
-]
1
Instead, the State attempted unsuccessfully to attack
the integrity of petitioner's data sources. But a rebuttal case
challenging a party's data base as misleading or inaccurate can-
not 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 the ground
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 National Bank supra, 505 F. Supp. at, 255-56;
accord, Trout v. Lehman, 702 P.24 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 (1981), The State's
challenges were either trivial in their overall significance or
based on misunderstandings of petitioner's coding conventions,
No serious proof was offered by the State to "show|[] that errors
or omissions bias([ed] the data" or "that this bias alter([ed] the
result of the statistical analyses in a systematic way."
The State's expert did present one hypothesis =-- that
the apparent racial disparities could be explained by the gene-
rally more aggravated nature of white victim cases. However,
after assembling a wealth of material to prove that white victim
cases are more aggravated on the average, Dr. Katz never addressed
the critical question -- whether white and black victim cases
(Continued)
R. 1346" (R. 1149). While the State proposed Dr. Katz as
qualified in all of those areas (Fed. Hab. Tr. 1346), voir
dire exposed that he had never had training or experience
gathering empirical data on human behavior, had never taken
a course in multivariate analysis, and had virtually no
training in the design of empirical research (id. 1370-74).
The Court declined to qualify him in these areas (id.
1375-773 1383).
-68~
at the same level of aggravation are being treated similarly (id.
1664). Dr. Katz, in the other words, never tested his own hypo-
thesis (id. 1760-61), although on cross-examination he admitted
"that would have been desirable" (id. 1613).
This hypothesis is thus precisely the sort of "[u]lnqual-
ified, speculative, and theoretical objection[] to the proffered
statistics [that is] properly given little weight by the trial
court," Trout v. Lehman, supra, 702 F.2d at 1102; see e.g.,
Castaneda v. Partida, 430 US. 1182, 499 n.19 (i977). The Dis-
trict Court here declined fully to credit the State's theory
(see R. 1178-79); 1200-01), but it nevertheless called it "plau-
sible" (R. 1200), and held that it constituted "direct rebuttal
evidence [which] ... stands to contradict any prima facie case"
{R. 1201),
Even standing alone, that assessment would violate
Trout's cautionary injunction. However, petitioner's experts
directly addressed this hypothesis (id. 1297; 1729-32), tested ._ .
it themselves (id. 1291-96; see GW 5, 6, 7, 8; see also DB 92),
and conclusively proved that it does not explain the racial dis-
parities (id. 1732). Untested by its proponents, refuted by __
petitioner, the hypothesis is clearly entitled to no weight as
rebuttal.
The State's expert advanced one additional hypothesis,
that because the statistical significance of Baldus' racial coef-
ficients dropped when the first variables were added to his
model, the addition of still further, unspecified variables
beyond 230 identified by Baldus "might" eventually eliminate all
the statistical significance of the coefficients (id. 1564-69).
Dr. Katz once again acknowledged he had not performed any tests
to confirm this speculation (id. 1566).
69
Petitioner's statistical experts testified that Dr.
Katz's "extrapolation is not a reliable statistical tool" (id.
1729), and that the demonstration "in no way invalidates the
original results" (id. 1758). Moreover, Dr. Woodworth illus-
trated the point by presenting an equally plausible demonstra-
tion under which statistical significance was maintained (id.
1728-29; see GW 6).
In sum, the State's rebuttal arguments were themselves
nothing but uninformed speculations, which petitioner firmly
countered. Because petitioner's prima facie case was not seri-
ously challenged, much less refuted, he should have prevailed
on his constitutional claims. The District Court erred seriously
in denying relief.
V. THE DISTRICT COURT IMPROPERLY REJECTED PETITIONER'S
PROSECUTION-PRONENESS CLAIM
The District Court rejected as a matter of law peti-
tioner's claim that the exclusion of death-scrupled jurors vio-
lated his right to be tried by an impartial and unbiased jury
drawn from a representative cross-section of his community
(R. 1244). That holding, we respectfully submit, was error. This
Court sitting en banc should adopt the findings and conclusions
set forth in Grigsby v. Mabry, 569 P. Supp. 1273 (E.D. Ark. 1983),
hearing en banc ordered, No. 83-2113 EA (8th Cir., November 8,
1983) and Keeten v. Garrison, 578 F. Supp. 1164 (W.D.N.C. 1984),
appeal pending, No. 84-6139 L (4th Cir.).
CONCLUSION
That portion of the District Court's judgment granting
habeas corpus relief to petitioner on his due process claims
under Giglio v. United States should be affirmed. Those por-
=7 0
7
}
-
»
JERE
:
p
J
E
»
r
r
-
#
4
A
\
e