General - Baldus Study - Affidavit Drafts

    In the United States District Court 

Northern District of Georgia 

Atlanta Division 

  

WARREN McCLESKY, 

Petitioner 

- Against - CIVIL ACTION 

WALTER D. ZANT, Superintendent, NO. C81-2434A 

Georgia Diagnostic & 

Classification Center, 

N
r
?
 

N
t
?
 

a
?
 

N
t
?
 

N
e
’
 

wt
? 

N
a
?
 

N
a
?
 

S
a
?
 

a
d
 

S
d
 

Respondent 

Affidavit of David C. Baldus 

on behalf of Petitioner 

The undersigned, being duly sworn, deposes and says as 

follows. On or about August 18, 1983, the Court in this proceeding 

delivered to George Woodworth and me a specification of variables from 

the Charging and Sentencing Study styled the "Lawyers' Model," a copy of 

which is attached hereto as Appendix A. The Lawyers' Model includes 

three sets of variables -- aggravating factors (items 1-4), mitigating 

factors (item 5), and strength of the evidence measures (item 6). The 

Lawyers' Model specified that a case be included in the analysis if it 

(a) possessed a specified combination of the aggravating factors in 

items 1-4, and (b) showed the strength of the evidence as measured by 

the variables in item 6; the instructions further stated that cases with 

the mitigating factors in item 5 were to be excluded from the analysis. 

Pursuant to this request, we created a series of variables to 

identify the cases specified for inclusion in the Lawyers’ Model. A 

screen of the cases with these variables produced a sample of 29 cases, 

 



  

which, when weighted in the manner of our other analyses, constituted 44 

of the cases in the universe of the Charging and Sentencing Study. 

8/30/83 

overall death sentencing rate among these cases was .30 (13/44). 

Although the pool of selected cases was too small for regression 

analysis, we conducted crosstabulation analyses which showed the 

following association between selection rates at three stages in the 

Charging and Sentencing System and the defendant/victim racial 

combinations represented in the data: 

  

The 

Outcome All Cases Black Def. White Def. Black Def. 

Variable White Vic. White Vic. Black Vic. 

A. Death Sentence «30 «53 07 0 

Given A Murder (13/44) (12/23) (1/14) (0/7) 

Indictment 

B. Prosecutor Seeks . +68 1.0 «56 .0 

A Penalty Trial (19/28) (14/14) (5/9) (0/5) 

After Murder Guilt 

Trial Conviction 

a/ 
C. Jury Death +82 .80 +20 «0 

Sentencing (13/21) (12/15) (1/5) (0/1) 

Decision ;   
  

a/ This decision point includes 2d and 3d penalty trials for defendants 

whose original death sentences were reversed on appeal, as well as guilty 

plea cases that resulted in a penalty trial. 

An analysis of separate frequency distributions for the 

variables used to select the cases for the Lawyers' Model indicated that 

the small sample resulted from the exclusion of cases with the 

In a second mitigating circumstances specified in item 5 of Appendix A. 

analyses, the requirements of item 5 were partially relaxed by deleting 

item 5B and deleting the last item variables in item 5D (281-303). This 

 



“3m 8/30/83 

  

modification increased the sample of cases to 61 which when weighted 

represent 92 cases in the universe. The overall death sentencing rate 

for these cases was .28 (25/92) and they showed the following 

association between the three defendant/victim racial combinations 

represented in the data and selection rates at three steps in the 

  

process: 

Outcome All Cases Black Def. White Def. Black Def. 

Variable White Vic. White Vic. Black Vic. 

A. Death Sentence .28 «55 14 «0 

Given A Murder (25/92) (21/39) (4/30) (0/24) 

Indictment 

B. Prosecutor Seeks +57 «87 .14 +13 

A Penalty Trial (37/66) (26/30) (9/17) (2/19) 

After A Murder 

Guilt Trial 

Conviction 

C. Jury Death .64 «78 .46 .0 

Sentencing (25/39) (21/27) (4/9) (0/3) 
Decision   
  

To enlarge the pool of selected cases to a number which would 

allow us to conduct multiple regression analysis, we dropped all of the 

constraints imposed by the mitigating factors in item 5 of Appendix A. 

This resulted in a pool of 237 cases which when weighted represent 435 

of the cases in the universe of the Charging and Sentencing Study. The 

overall death sentencing rate among those cases was .15 and they showed 

the following association with the defendant victim racial combinations 

represented in the data. 

 



-4- 8/30/83 

  

  

Outcone All Cases ( Black Def. White Def. Black Def. White Def. 

Variable White Vic. White Vic. Black Vic. Black Vic. 

A. Death Sentence +15 .43 +15 wg OH «10 

Given A Murder (64/435) (27/63) (27/167) (8/195) (1/10) 

Indictment 

'B. Prosecutor Seeks .46 17 .46 .25 .40 
A Penalty Trial (102/219) (37/48) (45/99) (17/67) (2/5) 

After A Murder 

Guilt Trial 

Conviction 

C. Jury Death .60 «71 «57 .44 «50 

Sentencing (64/107) (27/38) (27/48) (8/18) (1/2)   
  

We next conducted weighted least squares multiple regression 

analyses which controlled for the 39 background variables in Schedule 4 

of the Technical Appendix (Petitioner's Exhibit DB96A). For each of the 

stages in the process analyzed, we controlled first for all 39 

background variables simultaneously and then controlled for those that 

showed a relationship with the outcome variable that was statistically 

significant at the .10 level. The following tabulation shows the 

results: 

 



od 8/30/83 

  

Weighted Least Squares Regression Coefficient for Race of Victim 

And Race of Defendant Controlling for 39 Non-Racial Background Variables 
  

A. B. : C. 

Outcome Variable & Race of Victim Regression Race of Victim Regression 

Background Variables Coefficient (with level of Coefficient (with level 

Included Statistical Significance) of Statistical 

Significance) 

  

I. Death Sentence 

Given a Murder 

Indictment 

(DPMURIDT) (n=237) 

A. 39 Non-Racial «13 .06 

Background Variables (.04) (.29) 

B. 14 statistically dD «10 

Significant {.01) (.07) 

Background Variables 

II. Prosecutor Seeks A 

Penalty Trial After A 

Guilt Trial Murder 

Conviction (PSEEKNGP) 

(n=157) 

A. 39 Non-Racial .18 .06 

Background Variables (.09) {.54) 

B. 14 statistically .20 .03 

Significant {.03) ; {.70) 

Background Variables 

III. Jury Death Sentencing 

Decision At Penalty 

Trial (DEATHSNT) (n=108) 

A. 39 Non-Racial «18 .08 

Background Variables (.23) {.58) 

B. 9 Statistically ‘13 .001 

Significant (.23) (.99) 

Background Variables 

 



  

tl 8/30/83 

We also repeated these regression analyses with a variable for the 

defendant's socio-economic status (LSTATDEF) included in each analysis. 

The inclusion of this variable in the analyses arhancad slightly the 

racial effects observed for the combined effects of the entire Charging 

and Sentencing process (Part I above) and the prosecutorial decision to 

advance a case to a penalty trial (Part II above). The analysis of the 

jury death sentencing decision, however, was not affected by the 

addition of the variable for the defendant's socio-economic status. 

The results of the analyses of all three samples selected with 

the Lawyers' Model, are consistent.with the analyses presented earlier 

i.e., when the analysis focuses on the more aggravated cases where there 

is a substantial risk of a death sentence the race of victim effects 

increase substantially. 

Finally, the Court suggested in the Lawyers' Model that we 

conduct separate analyses among cases with appointed private counsel on 

the one hand and either retained counsel or appointed counsel with an 

institutional affiliation such as a public defender, or Team Defense 

lawyer. BAmong the 237 cases selected for the overall analyses 43 had 

missing data on the status of defense counsel. Among the 194 cases 

where status of counsel was known 119 had private appointed counsel and 

75 had either retained counsel or appointed counsel with an 

institutional affiliation. We then conducted regression analyses on 

the two subsets of data defined by the status of defense counsel. The 

results presented below showed much stronger race of victim effects 

among the cases with private appointed defense counsel. 

 



  

= 8/30/83 

Race of Victim Regression Coefficients (with level of Statistical Significance) After 

Controlling for 39 Non-Racial Background Variables & the Race of Defendant 
  

  

  
  

Outcome All Cases Counsel Private Counsel Retained Or Appointed 

Variable Appointed (n = 194) With Institutional Affiliation 

(n = 119) 

A. Death Sentence «13 v2 .04 

: Given A Murder (.04) (.01) {.82) 

Indictment 

B. Penalty Trial .18 +33 «3D 

Held After (.09) {.05) {.78) 

Murder Guilt 

Conviction 

Ce Jury Death .18 +38 Eg 

Sentencing {(.23) (.45) 

Decision 

1l/ Too few cases for analysis. 

A similar analysis also show stronger race of defendant effects among 

cases with appointed private counsel. 

 



-8- 8/30/83 

  

Race of Defendant Regression Coefficient (with level of Statistical Significance) 

Controlling for 39 Non-Racial Background Variables and the Race of Victim 
  

  

A. B. : C. 

Outcome All Cases Counsel Private Counsel Retained or 

Variable : (n = 237) Appointed Appointed with 
(n = 194) Institutional Affiliation 

(n = 119) 

Death Sentence .06 .25 .03 
Given A Murder (.29) (.02) (.82) 

Indictment 

Penalty Trial Held .06 «11 07 

After Murder Guilt (.54) (.38) (.90) 

Conviction 

Jury Death » ..08 .07 -2/ 

Sentencing Decision (.54) {.72)   
  

a/ Too few cases for the analysis. 

These results show a strong interaction effect between the 

status of defense counsel and the degree of racial disparity observed in 

prosecutorial decisionmaking; with substantially stronger effects, 

particularly the race of victim effect, apparent in cases with private 

retained counsel. The interaction effect does not appear in the jury 

decisions however. This evidence suggests that private retained counsel 

are less likely to be independent in their representation and less 

likely to represent their client in a fashion that will incline a 

prosecutor to plead the case out or otherwise waive a penalty trial. 

 



  

LAWYERS' MODEL 
  

Include if: 

1. Indicted for murder, and if 

2. Defendant age greater than 18 and less than 65, and if 

3. One or more statutory aggravating circumstances present, and if 

4. Any of the following present: 

(A) Two or more statutory aggravating but not b(3) where b (2) 

present and not b(9) or b(10) where b(8) present. 

OR 

(B) One or more statutory aggravating circumstance and any one 

positive response in foils 

161, 163, 164, 168,:170,'176,.176A,:177, 1785, 180, 181, 134, 

OR 

{C). Foil 82, 86, 90, 94, 98, or 102 is.«<©oded 1, 5-or 6, 

OR 

(D) Responses in each foil numbered 82, 86, 90, 94, 98 if that 

response is 12 or less, 

OR 

(E) Any response in foils 172-175B if the response is numbered 1, 

3,: 4,9, 94, 13,14, 16, 

OR 

(F) Any response in foils 131-134A if the response is numbered 3, 

5, 6, 6A, and if 

5. None of the following are present: 

(A) A response in foil 121 carrying any of the following numbers: 

1,2, 4, 5,6, 7,8, 9, 310,.15,:16,317;, 18, 19, 20. 

OR 

APPENDIX A 

 



  

(B) A positive response in foils 135, 136, 137, 138, 139, 144A, 

145, 147, 148, 150, 151, 152,153, 154, 155, 156, 

OR 

(C) Foil 237, 242, or 247 coded two and that co-perpetrator is 

equally culpable (i.e., foil under 48C, code A or B is coded one or 

two) and his treatment is lenient (i.e., his sentence foil is not 

coded 99, 98, or greater than 19 years). 

OR 

(D) Positive Response in foils 261, 264, 265, 266, 268A, 269, 270, 

281, 282, 283, 288, 289,294, 298, 299,.302, 303, 

OR 

(E) There are a total of 4 or more positive responses in foils 

259-3210 excluding foils listed in 5.D., and if 

Any one of the following is present: 

(A) Foil 385 not coded 5 and 390 not coded 1, and positive 

response in foils 323, 324, 327, 328, and where any three of the 

following foils also have a positive response: 364, 365, 366, 367, 

368, 369, 385, (if coded 1-4), 389, 394-401, 446, 452, 453, 454, 

466, 467, 

OR 

(B) Positive Response in 394, 395, 395A, 396, 397 and if 430 not 

coded 2 and three or more positive responses in foils 322, 389, 

446, 452, 453, 454, 466, 467, 

OR 

(C) Foil 237, 242, or 247 coded 2 and positive response in foil 

405, 406, 406A, 407, 408, 409, or 410 and at least six positive 

responses in the following foils: 389, 394-401, 446, 451-457, 460, 

461, 462, 466-469, 471, 472, 

OR 

(D) At least six positive responses in the following foils: 322, 

389, 428, 446, 451, 466, 467. 

Run this Group of Cases once where foil ten is coded two or eight and 

once where foil ten is coded either 1, 3, 4, 5, 6, or 7. 

 



  

In the United Sta 

str 

Atlanta Divi 

Petitioner 

- Against - CIVIL ACTION 

WALTER D. ZANT, Superintendent, NO. C81-2434Aa 

Georgia Diagnostic & 

Classification Center, 

Respondent 

Affidavit of David C. Baldus 

on behalf of Petitioner 
  

The undersigned, being duly sworn, deposes and says as 

follows. On or about August 18, 1983, the Court in this proceeding 

delivered to George Woodworth and me a specification of variables from 

the Charging and Sentencing Study styled the "Lawyers' Model," a copy of 

which is attached hereto as Appendix A. The Lawyers' Model includes 

three sets of variables -- aggravating factors (items 1-4), mitigating 

factors (item 5), and strength of the evidence measures (item 6). The 

Lawyers' Model specified that a case be included in the analysis if it 

(a). possessed a specified combination of the aggravating factors in 

items 1-4, and (b) showed the strength of the evidence measured by the 

variables in item 6; the instructions further stated that cases with the 

mitigating factors in item 5 were to be excluded from the analysis. 

Part I of this affidavit presents the results of analyses 

conducted with four subsets of cases identified with variables in the 

Lawyers' Model. 

 



  

An alternative method for estimating racial effects while 

controlling for the non-racial background variables specified in the 

Lawyers' Model is to enter them into a multiple regression analysis as 

independent variables. Part II of this affidavit presents the results 

of four analyses using this approach. 

Finally, the Lawyers' Model requested that we account for the 

status of defense counsel in the case, specificzlly whether the 

defendant's counsel was (a) a private attorney appointed by the court, 

or (b) either a privately retained attorney or a court appointed 

attorney with an institutional affiliation (e.g., a public defender). 

Part III of this affidavit presents the results of the analyses 

addressing this issue. 

I. Analyses Using Cases Selected With Variables Specified in the 

Lawyers' Model. 

a. The Samples 

The first three variables in the Lawyers' Model (items 1-3) 

were already included in the file of the Charging and Sentencing 

1/ ; : : ia 
Study.— However, to identify the cases with the characteristics 

specified in items 4-6, it was necessary to create a series of new 

variables. The coding for these variables, (V4aA-V6D), whose names 

correspond to the paragraphs in the Lawyers' Model at Appendix A, is 

listed in Appendix B of this Affidavit. 

  

1l/ Item 1 (INDICT); Item 2 (DEFAGE); Item 3 (DEATHELG). 

 



  

The variables specified in the Lawyers' Model 

cases sharply limited the cases available for analysis, specifically, 

they identified only 31 of the 1066 cases in the sample and only 15 of 

the 128 death sentence cases. In order to obtain samples of sufficient 

size both to conduct multiple regression analyses and to obtain a 

substantial representation of death sentence cases, we relaxed the 

requirements of the Lawyers' Model in three successive stages with the 

results presented below. For Example, row 2 

  

Table 1 

2 B c D 
Sample Number apg Total Sample Cases in the Universe Proportion and 

Characteristics— of Cases Represented Number of Death 

by the Sample Sentence Cases 

1. Original Lawyers’ 31 46 «12 (15/128) 

iodel. 

2. Lawyers' Model with 66 104 «21 '{277/104) 

a Relaxation of the 

Exclusions Based on 

the Presense of 

Mitigating Factors. 

3. Lawyers' Model with 238 441 «51 (65/128) 

No Exclusion of 

Cases Because of 

Mitigating Factors. 

4. Lawyers' Model with 354 647 .76 (97/128) 

No Exclusion of 

Cases Because of 

Mitigating Factors 

and a Relaxation of 

the Required Level 

of Evidence Strength 

for Inclusion. 

  

1/ For Sample 2 the requirements of Item 5 were relaxed by deleting 

Item 5B and the last 10 variables in Item 5D. Sample 3 was produced by 

deleting entirely the requirement that a case be dropped 

analysis because of the presence of one of the variables 

S5A-5E. For Sample 4 we relaxed the strength of evidence 

changing Item 6D from "at least six" to "at least three" 

responses. 

from the 

listed in Items 

requirements by 

positive 

 



  

indicates that when the limitations of the original 

relaxed with respect to the mitigating 

from the analysis, the sample included 

in the universe and 21% of the death sentence cases. 

Lawyers 

factors used to exclude cases 

66 cases representing 104 cases 

The fourth and 

largest sample included 354 cases with 76% (27/128) of the death 

sentence cases. 

b. Crosstabulation Analysis 

For each sample, we calculated overall death sentencing rates 

and the rates among the four groups of cases produced by the 

"Defendant/victim racial combination" variable. 

follows: 

The results were as 

  

  

Table 2 

Death Sentencing Rates 

By By g Fol E r 
Sample Number Average— Black Def. White Def. Black Def. White Def. 

Rate White Vic. White Vic. Black Vic. Black Vic. 

1 .33 (15/46) | .55 (13/24) .13 (2/15) .0 (0/7) RE. 

2 .26 (27/104) | .54 (22/41) .17 (5/31) .0 (0/32) AL. 

3 .15 (65/441) .42 (27/65) ed’ 28/170) .04 (8/192) «07 (1/14) 

4 .15 (97/647) .35 (34/98) .18 (50/271) .05 (12/262) .06 (1/16)   
These analyses indicate that as the samples of cases are expanded beyond 

the original Lawyers' Model, the average death sentencing rate declines, 

but the race of victim and race of defendant effects persist in each 

analysis. 

  

1l/ The denominators are weighted figures. 

a/ No cases in this category. 

 



  

- 
sentence after a murder conviction was obtained at trial shows the same 

pattern. The results are as follows: 

Table 3 

Rates at Which Prosecutors Seek a Death 

Sentence After a Guilt Trial 
  

  

  

  

  

  

A By c ip E ir 
Sample Number Average— Black Def. White Def. Black Def. White Def. 

Rate White Vic. White Vic. Black Vic. Black Vic. 

1 «70 (21/30) 1.0°(15/15) .61 (6/10) «0 (0/5) on. 

2 .56 (38/68) «87427/31) .57 (10/18) 05 (1/19) ett 

3 .47 (102/214) .78 (38/49) .46 (46/101) .25 (15/60) .40 (2/5) 

4 .47 (161/345) «78 (58/75) .44 (79/177) .25 (22/89) .44 (2/5) 

1l/ The denominators are weighted figures. 

a/ No cases in this category. 

In contrast, an analysis of death sentencing rates at penalty trial 

shows substantially weaker race of victim and race of defendant effects. 

Those results were as follows: 

Table 4 

Death Sentencing Rates at a Penalty Trial 

a B c D E F 
Sample Number Average Black Def. White Def. Black Def. White Def. 

Rate White Vic. White Vic. Black Vic. Black Vic. 

1 .68 (15/22) 181: (13/16). 7.33 (2/6) BELA. hw. 

2 .68 (27/40) .79.422/28) +51 (5/10) +0 (0/2) tins wrth 

3 .61 (65/107) +69 (27/39) .57 (28/49) .50 (8/16) «50. (1/2) 

4 .57 (27/169) .58 (34/59) .59 (50/84) .50 (12/24) «50 (1/2)   
  

a/ No cases in this category. 

 



  

c. Multiple Regression Analysis 

We next conducted weighted least squares multiple regression 

analyses which controlled for the 39 background variables in Schedule 4 

of the Technical Appendix (Petitioner's Exhibit DB 96A). Analyses were 

conducted for Samples 3 and 4 with 238 and 354 cases respectively. For 

each analysis, we first controlled simlutaneously for the 39 background 

variables and then in a second analysis, for the background variables 

from the list of 39 that showed a statistically significant relationship 

with the outcome variable at the .10 level. The results are presented 

in Table 5 and Appendix C presents the complete regression results from 

which the Sample 4 race of victim and race of defendant coefficients 

reported in rows IB, IIB and IIIB were taken: 

 



  

- G/15 7/07 — - ¥ 3 / 0/12/83 

  

Table 5 

Weighted Least Squares Regression Coefficients for Race of Victim 

And Race of Defendant Controlling for 39 Non-Racial Background Variables 

A. B. Ce 

Outcome Variable & Race of Victim Regression Race of Defendant Regression 

Background Variables Coefficients (with level of Coefficients (with level 

Simultaneously Statistical Significance) of Statistical Significance) 

Controlled for in 

the Analysis 

  

I. Death Sentence 

    

Given a Murder Sample 3 Sample 4 Sample 3 Sample 4 

Indictment (n=238) (n=354) (n=238) (n=354) 

(DPMURIDT) 

2. 39 Non-Racial «13 «11 .04 «01 

Background Variables (.05) (.05) (.49) (.90) 

B. 13/16 statistically wf 15 .09 05 

Significant Backy (.01) (.01) (.10) (.29) 
ground Variables— 

II. Prosecutor Seeks A 

Penalty Trial After A 

Guilt Trial Murder 

Conviction (PSEEKNGP) 

A. 39 Non-Racial .24 vw23 +07 .06 

Background Variables (.04) (.02) (.50) (.49) 

B. 11/9 Statistically «23 .26 03 .08 

Significant Backs) (.02) (.01) {.75) {.27) 

ground Variables— 

III. Jury Death Sentencing 

Decision At Penalty 

Trial (DEATHSNT) 

A. 39 Non-Racial oi3 +11 .03 -.08 

Background Variables (.40) (.34) (.80) (.39) 

B. 10 Statistically .33 .07 .01 -.08 

Significant {.27) {.45) (.92) {.28) 

Background Variables 

  
1l/ The Sample 3 analysis included 13 background variables while the Sample 4 

analysis included 16 variables. 

2/ The Sample 3 analysis included 11 background variables while the Sample 4 

analysis included © variables. 

 



  

We also conducted weighted logistic regression analysis using 

sample 4 (n=354). The racial coefficients estimated in two analyses 

with "Death Sentence Given a Murder Indictment" as the dependent 

variable were as follows: 

    

  

  

Table 6 

Race of Victim Race of Defendant 

Death Odds Regression Death Odds Regression 

Multiplier Coefficient Multiplier Coefficient 

(with level (with level 

of Statistical of Statistical 

Significance) Significance) 

Background Variables 

Simultaneously 

Controlled For 

39 variables in 4.0 1.39 «57 -.56 

Schedule 4 of (.03) (+15) 

Petitioner's Exhibit 

DB S96A 

19 of the 39 variables 6.5 1.87 «71 -.39 

in a) above with a (.001) (.40) 

statistically signifi- 

cant relationship 

(.10 level) with the 

dependent variable 

  

Finally we used the 354 cases in Sample 4 to produce two 

figures which contrast the rise in death sentencing rates in white and 

or 
Coop 00 TA 

o I 

TTT ‘ : 
black victim cases, after—controiliing.for-the aggravation level ef—the 

  

¥ 

AANA SOA 
casesy For this purpose, the aggravation level of each case was 

estimated from the results of a weighted least square regression which 

controlled for the 39 non-racial background variables included in 

Schedule 4 of Petitioner's Exhibit DB 96A. The results are presented in 

figures 1 and 2. They provide additional support for the hypothesis 

(Figures 1 & 2 go here) 

that the State of Georgia operates a dual system for processing homicide 

cases in which white cases victim cases are in fact considered more 

 



1ibit igure 1 Midrange second order model for 354 death elibible cases. 
(Black Defendants) 

+r 

  

          

\ 

+ 
2 

St
d.

 
De

v.
 

N 

B
l
a
c
k
 

V
i
c
t
i
m
s
 

       

    

    
  

4 3 NI 

Sf E > 
— 0 

# 3 = 

> 
vo 100 T > 
U &J 

c the o - 

- McClesky's — oN 

3 level of = +] 
= 5 T Sgevgyation Sow 

f= 

X50 T 
dd ° . . 

o White victim cases a 

> at McClesky's level ’ 
Oo y —— 25 percentage point 

— 3 race of victim 
= 2573 disparity at McClesky's 
J a Black victim cases Tools 8 Tri 
3) at McClesky's level 88 
- 

£ { 
0 - Z , Ru EE 

  

or 0 40 60 80 100 
/ Level of Aggravation 

NOTES: Level of aggravation is the linear WLS Model leaving out the racial 

effects. The second order model includes race of victim, race of 

defendant, level of aggravation, all two-way interactions and the 
square of the level of aggravation. The model includes white 

defendants, but they are not graphed here. 

  

 



|) 
8) 
c 

¢ 
I= 
Qo 

wn 

Jr 
© 
oy 

an 

vo) 

wu 

Oo 

~ 
ad 
or 
i 
rd 

ao 
3°] 

vo 
o 
el 
a 

  

100 

No
 

wn
 

pon 

  

Midrange second order model for 
(White Defendants) 

  

   

Black Victims 

+ 2 Std. Dev. 

      
Level of Aggravation 

— tt ttt 
40 60 80 100 

W
h
i
t
e
 

V
i
c
t
i
m
s
 

+ 
2 

St
d.
 

D
e
v
,
 

IR
E 

Br
 

a
h
 
S
T
 

 



  

| 
(5
0)
 | 

N
N
.
 fe
t 

t ~ 

aggravated than similarly situated black victim cases. Figures 1 als 

icates that among cases with an aggravation level comparable to w x (o
f 

[9
 

Warren McClesky; there is a 25 percentage point race of victim 

disparity. 

The results of the analysis shown in the first section of this 

affidavit are consistent with the results presented in the hearing on 

this matter. There are persistent race of victim effects and when the 

analysis focuses on the more aggravated cases, where there is a 

substantial risk of a death sentence, those effects increase 

substantially. 

II. Racial Coefficients Estimated in Analyses Using Variables Created 

for the Lawyers' Model as Background Controls 

The 18 variables created to select cases for the analyses 

reported in Part I provide an alternative means of controlling for the 

non-racial background case characteristics identified as important in 

the Lawyers' Model. We estimated racial coefficients in regression 

analyses while controlling simultaneously for these 18 non-racial 

background factors. We also estimated racial effects in analyses using 

the variables from the Lawyers' Model and the 39 variables which we 

believe to be most important in the system. The results of weighted 

least squares analyses using the 354 cases in Sample 4 were as follows: 

 



  

-10- 

Table 7 

~~ Race of Victim 
  

Coefficient (with 

Race of Defendant 
  

Coefficient (with 

  

  

Adjusted level of Statistical level of Statistical 

R? Significance Significance 

Background Variables 

Simultaneously 

Controlled For 

b 
1. 18 variables in +31 .10%/ id 

the Lawyers’ (.04) {.27) 
Model 

2. 18 variables in .43 ew 05 

the Lawyers’ (.04) (.34) 

Model and the 39 

variables in 

Schedule 4 of 

Petitioner's 

Exhibit DB 96A 

3. 39 variables in .39 oll .01 

Exhibit DB 96A (.05) (.20) 

  

a/ The logistic regression coefficient and death odds multiplier for race of victim 

were 1.45 and 4.3 respectively (significant at the .003 level). 

b/ The logistic regression coefficient and death odds multiplier for race of 

defendant were .34 and 1.4 respectively (significant at the .42 level). 

These data show the same pattern of racial effects observed in our 

earlier analyses. 

III. Racial Effects Estimated After Adjustment For Status of Counsel 

and Defendant's Socio-Economic Status 

Finally, tihe—Geurt| suggested sn—=the Lawyers’ Model”iseparate 

\ 

analyses for defendants (a) with appointed private counsel, en-the-one 

hand, and (b) with retained counsel or appointed counsel with an 

institutional affiliation, such as a public defendery -on~the~other. 

 



  

-11- 

Among the cases in the universe, the 

is an estimated .10 (78/795) for defendants with appointed private 
A 

counsel, and .05 (48/1002) for defendants with retained counsel or 
/ 

appointed counsel with an institutional affiliation. 

olf 
Multiple 

regression analysis indicate, however, that the inclusion of a variable 

for the status of defense counsel does not explain the race of victim 

effects observed in the data. 

regression coefficients in weighted least squares analyses which 

The race of victim and race of defendant 

controlled for the 39 background variables in Schedule 4 of Petitioner's 

Exhibit DB 962A and the status of defense counsel— 

Dependent Variable 

and Unweighted 

Sample Size 

Table 8 

Race of Victim 

Coefficients (with 

level of Statistical 

were as follows: 

Race of Defendant 

Coefficients (with 

level of Statistical 

  

Significance) Significance) 

Death Sentence Given .0S 05 

a Murder Indictment. {.01) (.08) 

{n = 773) 

Prosecutor Seeks a +13 .04 

Penalty Trial after (.02) (.47) 

Murder Guilt Trial. 

(n = 366) 

Jury Death Sentencing +15 -.03 

Result. (n=232) {:11) (.67) 

  

These analyses are based on the entire sample and also included an 

interaction term between race of victim and status of defense counsel which 

is discussed below. 

££ a I . 4 Pi 
ee CL GAL Aa A Ap Co L, A A Ri 

—y 

aa ti J rt « pr ay, 7 A oY A oh AE op Spey CE Cg DO ty 
yt « Ne \ é 

pn & 7 ~ £ § 
~~ - Ny pr | £5 4 Ll » —- # 

< Ley ami AL ert ad a VE ¢ l] Letnl t ; 
«* § [ 4 

LJ 

J — 
Qa AAAS Sy 

pro— 

at 

A d Rh Ph a a 

N . 

A oi fF ES 
gE r— a # J i & 

 



  

-12- 9/12/83 

We also conducted a series of regression analyses in which racial 

effects were estimated while controlling for the 39 variables mentioned 

above and a variable for the defendant's socio-economic status 

(LSTATDEF) ;2/ the racial coefficients estimated in these analyses were 

virtually identical to those reported in Table 8, confirming that the 

race of victim and race of defendant effects observed in the data are 

not spuriously caused by the status of defense counsel or the 

defendants' socio-economic status. 

We next conducted separate regression analyses for the cases 

in Sample 4, first for defendants with private appointed counsel and 

then for defendants whose counsel was privately retained or appointed 

with an institutional affiliation. The status of defense counsel was 

known in 84% of the cases in Sample 4 and the results were as follows: 

  

2/ These analyses were also based on the entire universe included an 

interaction term between the defendant's socio-economic status and 

the race of victim. An analysis limited to the cases in Sample 3 

without an interaction showed slightly enhanced race of victim and 

race of defendant effects when the status of defense counsel is 

included in the analysis. 

 



  

Table 9 - 

RACE OF VICTIM EFFECTS 

Race of Victim Regression Coefficients (with level of statistical significance) After 

Controlling for 329 Non-Racial Background Variables & the Race of Defendant 
  

  

  
  

Outcome All Cases Private Appointed Counsel Retained Or Appointed 

Variable (n=354) Counsel With an Institutional Affiliation 

(n=169) (n = 127) 

A. Death Sentence “11 «23 +05 

Given A Murder {. 05) (.02) (.68) 

Indictment 

B. Penalty Trial 23 «31 .04 

Held After {.02) {.01) (.85) 

Murder Guilt 

Conviction 

C. Jury Death vl} «07 +35 

Sentencing (.34) {.72) (.43) 

Decision 

JI. RACE OF DEFENDANT EFFECTS 

Race of Defendant Regression Coefficient (with level of statistical significance) 

Controlling for 39 Non-Racial Background Variables and the Race of Victim 
  

    

A. B. C. 

Outcome All Cases Private Appointed Counsel Retained or 

Variable (n = 354) Counsel Appointed with an 

(n = 169) Institutional Affiliation 

(n = 127) 

Death Sentence .01 +10 -.07 

Given A Murder (.20) {.25) {.51) 

Indictment 

Penalty Trial Held .06 .03 «02 

After Murder Guilt (.49) {.73) (.89) 

Conviction 

Jury Death -.08 -.13 -.10 

Sentencing Decision {.39) {.27) {.72)   
  

 



  

-14~ 9/12; 

These data show a strong interaction between the status of 

defense counsel and the race of victim, i.e. the race of victim effect 

is much stronger in cases with private appointed counsel than in cases 

where defendant's counsel is retained or appointed with an institutional 

affiliation. These results tell us that prosecutors are more inclined 

to be punitive in cases involving white victims (and to a lesser degree 

black defendants) if the defense attorney is in private practice and 

court appointed. 2A possible explanation for this pattern is that 

private appointed attorneys put up less of a fight and otherwise develop 

less pressure on the prosecutor to accept a plea or unilaterally waive 

the death penalty. Under such circumstances, the system is more likely 

to respond to the pressures for a death sentence that are generated when 

the victim is white. These data have particular significance since over 

75% (98/128) of death sentences are imposed in cases in which defendant 

was represented by private appointed counsel. 

The status of defense counsel is also a proxy for the 

defendant's socio-economic status since counsel are appointed only for 

indigent defendants; the strong race of victim effect in appointed 

counsel cases therefore may also reflect a greater perception of 

culpability when the defendant is poor and a greater willingness of 

prosecutors and juries to respond to the pressures for a death sentence 

that arise when the victim is white. To test the extent to which the 

latter hypothesis is supported by the data, we conducted separate 

multiple regression analyses which included variables ("interaction 

terms") reflecting (a) the interaction between the race of victim and 

the status of counsel, and (b) the race of victim and the defendant's 

 



-]5m 9/12/83 

  

socio-economic status. 

    

  

follows: 

Table 10 

A. B. Ce. D. E. 

Race of Victim Race of Victim 

Interaction Main Effect 

Term With 

Status of Defendant's Status of Defendant's 

Defense Socio-Economic Counsel Socio-Economic 

Outcome Measure Counsel Status Analysis Status Analysis 

1. Death Sentence .04 .04 .09 07 

Given a Murder {.26) {.15) {.01) (.001) 

Indictment 

2. Prosecutor Seeks +13 «14 +13 217 

a Death Sentence (.10) (.09) (.03) {.01) 

After a Murder 

Trial Conviction 

3. Penalty Trial .04 .05 «15 «13 

Death Sentencing {.81) (.78) {-11) (.19) 

Decision 

  

The coefficients in columns B and C are for the interaction terms which 

reflect how much larger on average the race of victim effect is when 

counsel is appointed (Col. B), or when the defendant's socio-economic 

status is low (Col. C). Columns D and E show the magnitude of the race 

of victim "main effect," which indicates the average race of victim 

  

l/ The first analysis included variables for status of defense 

counsel, the race of victim/status of defense counsel interaction 

effect and the 39 non-racial background factors in Petitioner's 

Exhibit DB 96A. The racial coefficients from these analyses are 

reported in columns B & D. The second analysis included variables 

for defendant's socio-economic status, the race of 

victim/defendant's socio-economic status interaction effect and the 

39 non-racial background factors. The racial coefficients from 

this analysis are reported in columns C & E. 

 



  

-16~ 9/12/83 

effect across all cases. Thus for the outcome measure "Death Sentence 

Given a Murder Indictment" (row 1), column D indicates that in the 

analysis including the status of defense counsel variable an average 

race of victim effect of 9 percentage points and column B indicates a 4 

point interaction effect with the status of counsel. These results 

indicate that the overall race of victim effect is 13 points in the 

cases with appointed counsel and 5 points overall in the cases in which 

counsel is retained or appointed with an institutional affiliation. 

The results of the analysis in Table 10 indicate that the race 

of victim interaction effects are comparable with both the "status of 

defense counsel" and the "low socio-economic status" variables. One is 

left therefore with the impression that both the competence, 

independence and energy of defense counsel and the defendant's status as 

an indigent affect the extent to which the system will respond to the 

pressures for a punitive response that arise when the victim is white. 

IV. Summary 

wa 
osed 2'1- 

Our analysis of the questions baby the Lawyers' Model 

indicate that race of victim and race of defendant disparities in death 

sentencing rates persist among the most death worthy cases, both before 

and after adjustment for the leading non-racial background factors 

operating in the system (pp. 4-9). Similar disparities were also 

observed when the mitigating, aggravating and strength of evidence 

variables suggested by Lawyers' Model were introduced as background 

controls (pp. 9-10). The third phase of our analysis focused on the 

 



-17- 9/12/83 

  

status of defense counsel and the defendant's socio-economic status. 

Our analysis clearly demonstrated that the racial disparities in death 

sentencing rates observed in the system are explained by neither the 

status (and competence) of defense counsel nor the defendant's 

: : ‘ ‘ a Ss 
socio-economic status. Our analysis of these two variables indicate, 

however, that these racial disparities, particularly the race of victim 

effect, are most pronounced among cases in which the defendant is 

indigent or has a private appointed counsel, a category of cases 

accounting for over 75% of the death sentences imposed. The system 

wo = - \Jotiellsy - 
therefore fot only \appears’/ recialiy discriminatdzy /in operation but also 

to eperate-in-a-manner—thatl allocated the prindipal burden of that 

discrimination on the poor and underprivileged. 

The results reported in this affidavit reinforce the opinions 

I expressed at the August 8-19, 1983 hearing about the role of racial 

factors in Georgia's capital Charging and Sentencing System. our AL 

testimony in that hearing indicates that as the cases became more death 

worthy, according to both statutory and non-statutory criteria, the 

observed racial disparities in death sentencing rates would also 

increase. For example, Petitioner's Exhibit DB 83 showed an average 

race of victim disparity (as measured in least squares analysis) of 

Oerrad 
approximately 6 points, vinisd Ehat disparity increased to 10 points 

(Petitioner's Exhibit DB 85) when the analysis was limited to death 

eligible cases under the two leading statutory aggravating factors (B-2 

& B-7). The Lawyers' Model considered in this Affidavit limited the 

analysis to an even a worthy set of cases, and the 

IC eB 8 BE ot SO 

race of victim disparities were aimost- identical, ~whethermeasured- with 
—— 

- # PY ~ goon ( Lam YE, os 75 0 8 4 ijfecces - 

Py ate BE Ve? A &LAAL Ok (tn = aia die > IP 

ne ~~ . ’ ” : / 
| FA f a . ALY Fo oy 4 / 

Pr, = | atl i i 7 AL! ait Mp dl A A ae 

 



~18- 9/12/83 

  

PR 
4 / 

least-sguares-or logistic regression coefficients (compare Table 5 Part | C J Pp 

I and Table 6 of this Affidavit with Petitioner's Exhibit 85). 

 



  

In the United States District Court 

Northern District of Georgia 

Atlanta Division 

WARREN McCLESKY, 

Petitioner 

- Against - CIVIL ACTION 

WALTER D. ZANT, Superintendent, NO. C81-2434A 

Georgia Diagnostic & 

Classification Center, 

N
t
 

N
t
 

at
l 

S
t
 

a
t
 

N
e
l
 

S
u
 

N
l
 

N
o
t
 

N
t
”
 

N
u
?
 

Respondent 

Affidavit of David C. Baldus 

on behalf of Petitioner 
  

The undersigned, being duly sworn, deposes and says as 

follows. On or about August 18, 1983, the Court in this proceeding 

delivered to George Woodworth and me a specification of variables from 

the Charging and Sentencing Study styled the "Lawyers' Model," a copy of 

which is attached hereto as Appendix A. The Lawyers' Model includes 

three sets of variables -- aggravating factors (items 1-4), mitigating 

factors (item 5), and strength of the evidence measures (item 6). The 

Lawyers' Model specified that a case be included in the analysis if it 

(a) possessed a specified combination of the aggravating factors in 

items 1-4, and (b) showed the strength of the evidence measured by the 

variables in item 6; the instructions further stated that cases with the 

mitigating factors in item 5 were to be excluded from the analysis. 

Part I of this affidavit presents the results of analyses 

conducted with four subsets of cases identified with variables in the 

Lawyers' Model. 

 



-2- 9/13/83 

  

An alternative method for estimating racial effects while 

controlling for the non-racial background variables specified in the 

Lawyers' Model is to enter them into a multiple regression analysis as 

independent variables. Part II of this affidavit presents the results 

of four analyses using this approach. 

Finally, the Lawyers' Model requested that we account in an 

analysis for the status of defense counsel in the case, specifically 

whether defendant's counsel was (a) a private attorney who had been 

appointed by the court, or (b) either a privately retained attorney or a 

court appointed attorney with an institutional affiliation (e.g., a 

public defender). Part III of this affidavit presents the results of 

analyses addressing this issue. 

I. Analyses Using Cases Selected With Variables Specified in the 

Lawyers' Model. 

a. The Samples 

The first three variables in the Lawyers' Model (items 1-3) 

were already included in the file of the Charging and Sentencing 

Study 2 However, to identify the cases which had the characteristics 

specified in items 4-6, it was necessary to create a series of new 

variables. The coding for these variables, (V4A-V6D), or whose names 

correspond to the paragraphs in the Lawyers' Model at Appendix A, is 

listed in Appendix B of this Affidavit. 

  

i/ Item] (INDICT); Item 2 (DEFAGE); Item 3 (DEATHELG). 

 



35 9/13/83 

  

The variables specified in the Lawyers' Model for selecting 

cases sharply limited the cases available for analysis, i.e., they 

identified only 31 of the 1066 cases in the sample for inclusion in the 

analysis and only 15 of the 128 death sentence cases. In order to 

obtain samples of sufficient size both to conduct multiple regression 

analyses and to obtain a sufficient representation of death sentence 

cases, we relaxed the requirements of the Lawyers' Model in three 

successive stages with the results presented below. For Example, row 2 

  

Table 1 

A B c D 
Sample Number apg Total Sample Cases in the Universe Proportion and 

Characteristics— of Cases Represented Number of Death 

by the Sample Sentence Cases 

1. Original Lawyers’ 31 46 +312 415/128) 

Model. 

2. Lawyers' Model with 66 104 .21 (27/104) 

a Relaxation of the 

Exclusion Based on 

the Presense of 

Mitigating Factors. 

3. Lawyers' Model with 238 441 «51 (65/128) 

No Exclusion of 

Cases Because of 

Mitigating Factors. 

4. Lawyers' Model with 354 647 .76 (97/128) 

No Exclusion of 

Cases Because of 

Mitigating Factors 

and a Relaxation of 

the Required Level 

of Evidence Strength 

for Inclusion. 

  

1l/ For Sample 2 the requirements of Item 5 were relaxed by deleting 

Item 5B and the last 10 variables in Item 5D. Sample 3 was produced by 

deleting entirely the requirement that a case be dropped from the 

analysis because of the presence of one of the variables listed in Items 

5A-5E. For Sample 4 we relaxed the strength of evidence requirements by 

changing Item 6D from "at least six" to "at least three" positive 

responses. 

 



le 9/13/83 

  

of the tabulation on page 3, indicates that when the limitation of the 

original Lawyers' Model were relaxed with respect to the mitigating 

factors used to exclude cases from the analysis, the sample included 66 

cases which represented 104 cases in the universe and 21% of the death 

sentence cases. The fourth and largest sample included 354 cases with 

76% (97/128) of the death sentence cases. 

b. Crosstabulation Analysis 

For each sample, we calculated overall death sentencing rates 

and the rates among the four groups of cases produced by the 

"Defendant/victim racial combination" variable. The results were as 

  

  

follows: 

Table 2 

Death Sentencing Rates 

2 Bay c Lin E of 
Sample Number Average— Black Def. White Def. Black Def. White Def. 

Rate White Vic. White Vic. Black Vic. Black Vic. 

1 .33 (15/46) .55 (13/24) .13 (2/15) .0 (0/7) hut. 2/ 

2 .26 (27/104) .54 (22/41) «d17: (5/31) .0 (0/32) avn 

3 .15 (65/441) .42 (27/65) .17 (28/170) .04 (8/192) .07 (1/14) 

4 .15 (97/647) .35 (34/98) «18 (50/271) 05 (12/262) .06 (1/16) 

These analyses indicate that as the samples of cases are expanded beyond 

the original Lawyers' Model, the average death sentencing rate declines, 

but the race of victim and race of defendant effects persist in each 

analysis. 

  

1l/ The denominators are weighted figures. 

a/ No cases in this category. 

 



=5- 9/13/83 

  

An analysis of the prosecutorial decision to seek a death 

sentence after a murder conviction was obtained at trial shows the same 

  

  

  

  

  

pattern. The results are as follows: 

Table 3 

Rates at Which Prosecutor Seeks a Death 

Sentence After a Guilt Trial 

A RLY c D E F 
Sample Number Average— Black Def. White Def. Black Def. White Def. 

Rate White Vic. White Vic. Black Vic. Black Vic. 

1 .70 (21/30) 1.0 (15/15) .61 (6/10) .0 (0/5) wit 

2 .56 (38/68) +87.:1217/31) .57 (10/18) 05 (1/19) suchas! 

3 .47 (102/214) .78 (38/49) .46 (46/101) .25 (15/60) +40 (2/5) 

4 .47 (161/345) «78.{58/75) 44 (79/177) .25 (22/89) .44 (2/5) 

1l/ The denominators are weighted figures. 

a/ =---- means no cases. 

In contrast, an analysis of death sentencing rates at penalty trial 

shows substantially weaker race of victim and race of defendant effects. 

Those results were as follows: 

Table 4 

Death Sentencing Rates at a Penalty Trial 

A BE c D x r 
Sample Number Average Black Def. White Def. Black Def. White Def. 

Rate White Vic. White Vic. Black Vic. Black Vic. 

a/ a/ 
0) +68 (15/22) .81 (13/16) +33. (2/6) ———— —— 

2 .68 (27/40) .79 (22/28) .51 (5/10) .0 (0/2) Ao. 

3 .61 (65/107) .69 (27/39) .57 (28/49) .50 (8/16) «50 :{1/2) 

4 .57 (97/169) .58 (34/59) .59 (50/84) «50 (12/24) «50 11/2) 

  

a/ =---- means no cases. 

 



  

c. Multiple Regression Analysis 

We next conducted weighted least squares multiple regression 

analyses which controlled for the 39 background variables in Schedule 4 

of the Technical Appendix (Petitioner's Exhibit DB 96A). Analyses were 

conducted for samples 3 and 4 with 238 and 354 cases respectively. For 

each analysis, we controlled simlutaneously in one regression for the 39 

background variables and then in a second analysis for the background 

variables that showed a statistically significant relationship with the 

outcome variable that was at the .10 level. The results are presented 

in Table 5. Also Appendix C presents the actual regression results for 

the Sample 4 analysis reported in row IA: 

 



  

-"]- 9/12/83 

Weighted Least Squares Regression Coefficient for Race of Victim 

And Race of Defendant Controlling for 39 Non-Racial Background Variables 
  

A. B. Ca 

Outcome Variable & Race of Victim Regression Race of Defendant Regression 

Background Variables Coefficient (with level of Coefficients (with level 

Simultaneously Statistical Significance) of Statistical Significance) 

Controlled for in 

the Analysis 

  

I. Death Sentence 

    

Given a Murder Sample 3 Sample 4 Sample 3 Sample 4 

Indictment (n=238) (n=354) (n=238) (n=354) 

(DPMURIDT) 

A. 39 Non-Racial .14 «15 05 .05 

Background Variables (.03) 1.01) (.46) (.28) 

B. 13/16 Statistically 37 vi2 «09 +05 

Significant Back- {.01) (.01) (.10) {.31) 

ground Variables— 

II. Prosecutor Seeks A 

Penalty Trial After A 

Guilt Trial Murder 

Conviction (PSEEKNGP) 

A. 39 Non-Racial .26 $23 D7 .06 

Background Variables (.03) {.02) (.50) (.49) 

B. 11/9 statistically «28 «26 .07 .08 

Significant Backs, {.01) (.01) (.43) {.27) 

ground Variables— 

III. Jury Death Sentencing 

Decision At Penalty 

Trial (DEATHSNT) 

A. 39 Non-Racial «13 ell .04 -.08 

Background Variables (.40) (.34) (.79) {.39) 

B. 10 Statistically +13 07 01 -.08 

Significant te27) (.44) (.92) (.28) 
Background Variables 

  

1/ The Sample 3 analysis included 13 background variables while the Sample 4 

analysis included 16 variables. 

2/ The Sample 3_analysis included 11 background variables while the Sample 4 
- analysis included 9 variables. 

 



-= 9/12/83 

  

We also conducted weighted logistic regression analysis using 

Sample 4 (n=354). The racial coefficients estimated in an analysis with 

"Death Sentence Given a Murder Indictment" as the dependent variable 

were as follows: 

    

  

  

Table 6 

Race of Victim Race of Defendant 

Death Odds Regression Death Odds Regression 

Multiplier Coefficient Multiplier Coefficient 

(with level (with level 

of Statistical of Statistical 

Significance) Significance) 

Background Variables 

Simultaneously 

Controlled For 

a) 39 variables in 4.0 1.39 «57 -.56 

Schedule 4 of (.03) {.15) 

Petitioner's Exhibit 

DB 96A 

b) 19 of the 39 variables 6.5 1.87 «71 -.39 

in a) above with a (.001) (.40) 

statistically signifi- 

cant relationship 

(.10 level) with the 

dependent variable 

  

Finally we used the 354 cases in Sample 4 to produce two 

figures which contrast the rise in death sentencing in white and black 

victim cases after controlling for the aggravation level of the cases. 

For this purpose the aggravation level of each case was estimated from 

the results of a weighted least square regression which controlled for 

the 39 non-racial background variables included in Schedule 4 of 

Petitioner's Exhibit DB 96A. The results are presented in figures 1 and 

2. They provide additional support for the hypothesis that the State of 

Georgia operates a dual system for processing homicide cases in which 

white cases victim cases are in fact considered more aggravated than 

similarly situated black victim cases. Figures 1 also indicates that 

 



al got 9/12/83 

  

among cases with an aggravation level comparable to Warren McClesky; 

: (Figures 1 & 2 go here) 

there is a 25 percentage point race of victim disparity. 

The results of the analysis shown in the first section of this 

affidavit are consistent with the results presented in the hearing on 

this matter. There are persistent race of victim effects and when the 

analysis focuses on the more aggravated cases, where there is a 

substantial risk of a death sentence, those effects increase 

substantially. 

II. Racial Coefficients Estimated in Analyses Using Variables Created 

for the Lawyers' Model as Background Controls 

The 18 variables created to select cases for the analyses 

reported in Part I provide an alternative means of controlling for the 

non-racial background case characteristics identified as important in 

the Lawyers' Model. We estimated racial coefficients in regression 

analyses while controlling simultaneously for these 18 non-racial 

background factors. We also estimated racial effects in analyses using 

the variables from the Lawyers' Model and the 39 variables which we 

believe to be most important in the system. The results of weighted 

least squares analyses using the 354 cases in Sample 4 were as follows: 

 



Figure 1 Midrange second order model for 354 death elibible cases. 
(Black Defendants) 

  

     

     

  

    

   

  

  
  

  

[4p] 
BE iw 

=E 7 
= 
or 

227 
o 

~NoD 
£} LD 

© wn 
— + | E > 
/m - © 

Hi 
Ph 

o 100 FT So 
Q +J 

: 25 
o McClesky's — ON 

k: level of = +] 
LIST aggravation or 4 
= . 5 i 37 
3] 
a Wg 

7 

Wi 50 T 
oS € -/- - - —/-— White victim cases ‘ 

Ps / at McClesky's level : o —— 25 percentage point 
- 25 = race of victim 

0 } hi, disparity at McClesky's 
2 — = — === Black victim cases wn level of ag ravation 
° : at McClesky's level g 
9 ’ 

i al Y 4 4 { i 4 i <d 4 — 
Re PRR es, | 1 i J L 3 13 i   0 Aj 7 ¥ 1] 

es 20 40 60 80 100 
Level of Aggravation / 

/ 
NOTES: Level of aggravation is the linear WLS Model leaving out the racial 

effects. The second order model includes race of victim, race of 

defendant, level of aggravation, all two-way interactions and the 
square of the level of aggravation. The model includes white 

defendants, but they are not graphed here. 

  

 



  

Figure 2 Midrange second order model for 354 death elibible cases. 

P
r
o
b
a
b
i
l
i
t
y
 

of
 

a 
D
e
a
t
h
 

S
e
n
t
e
n
c
e
 

  

(White Defendants) 

  

   W
h
i
t
e
 

V
i
c
t
i
m
s
 

+ 
2 

S
t
d
.
 

   

  

100 4 

75 a 

50 
\ 

N 

25 4 >, 
~ Black Victims 

4+ 2'5td. Dev. 

SUN Srecivs Mts Gs SU Sn Be S SE 
40 60 80 100     

Level of Aggravation 

D
e
v
.
 

 



“10- 9/12/83 

  

  
  

  

  

Table 7 

Race of Victim Race of Defendant 

Coefficient (with Coefficient (with 

Adjusted level of Statistical level of Statistical 

R2 Significance Significance 

Background Variables 

Simultaneously 

Controlled For 

b 1. 18 variables in 31 10% 06 
the Lawyers' (.04) {.27) 

Model 

2. 18 variables in .43 +11 05 

the Lawyers' (.04) (.34) 

Model and the 39 

variables in 

Schedule 4 of 

Petitioner's 

Exhibit DB 96A 

3. 39 variables in «39 +11 01 

Exhibit DB 96A {.05) (.90) 

  

a/ The logistic regression coefficient and death odds multiplier for race of victim 

were 1.45 and 4.3 respectively (significant at the .003 level). 

b/ The logistice regression coefficient and death odds multiplier for race of 

defendant were .34 and 1.4 respectively (significant at the .42 level). 

These data show the same pattern of racial effects observed in our 

earlier analyses. 

ITI. Racial Effects Estimated After Adjustment For Status of Counsel 

and Defendant's Socio-Economic Status 

Finally, the Court suggested in the Lawyers' Model that we 

conduct separate analyses both for defendants with appointed private 

counsel, on the one hand, and for defendants with retained counsel or 

appointed counsel with an institutional affiliation, such as a public 

defender on the other. 

 



-11- 9/12/83 

  

The status of defense counsel was known in an estimated 72% of 

the cases in the entire universe. B2Among the cases with appointed 

private counsel, the death sentencing rate is an estimated .10 (78/795) 

and .05 (48/1002) for defendants with retained counsel or appointed 

counsel with an institutional affiliation. Multiple regression analysis 

indicate, however, that the inclusion of a variable for the status of 

defense counsel does not explain the race of victim effects observed in 

the data. The race of victim and race of defendant regression 

coefficients in weighted least squares analyses which controlled for the 

39 background variables in Schedule 4 of Petitioner Exhibit DB 96A and 

the status of defense counsels! were as follows: 

  

Table 8 

Dependent Variable Race of Victim Race of Defendant 

and Unweighted Coefficient (with Coefficient (with 

Sample Size— level of Statistical level of Statistical 

Significance) Significance) 

Death Sentence Given .09 +03 

a Murder Indictment. {.01) (.08) 
(n = 773) 

Prosecutor Seeks a «13 .04 

Penalty Trial after (.02) (.47) 
Murder Guilt Trial. 

(n = 366) 

Jury Death Sentencing «15 -.03 

Result. (n=232) {.11) (.67) 

  

1l/ These analyses also included an interaction term between race of 

victim and status of defense counsel which is discussed below. 

 



=12- 9/12/83 

  

We also conducted a series of regression analyses in which racial 

effects were estimated while controlling for the 39 variables mentioned 

above and a variable for the defendant's socio-economic status 

2/ : Fag : : 
(LSTATDEF) ;— the racial coefficients estimated in these analyses were 

virtually identical with those reported in Table 8. 

We next conducted separate regression analyses for the cases 

in Sample 4, first for defendants with private appointed counsel and 

then for defendants whose counsel was private appointed or appointed 

counsel with an institutional affiliation. The status of defense 

counsel was known in 84% of the cases in Sample 4 and the results were 

as follows: 

  

2/ This analysis also included an interaction term between the 

defendant's socio-economic status and the race of victim. 

 



=13~ 9/12/83 

  

Table © 

I. RACE OF VICTIM EFFECTS 

Race of Victim Regression Coefficients (with level of statistical significance) After 

Controlling for 39 Non-Racial Background Variables & the Race of Defendant 
  

  

Outcome All Cases Private Appointed Counsel Retained Or Appointed 

Variable (n=354) Counsel With Institutional Affiliation 

(n=169) (nh = 127) 

A. Death Sentence +311 +23 +05 

Given A Murder (.05) {.02) (.68) 

Indictment 

B. Penalty Trial 23 «31 .04 

Held After (.02) (.01) {.85) 

Murder Guilt 

Conviction 

C. Jury Death +11 .07 «35 

Sentencing {.34) (72) (.43) 

Decision 

  

II. RACE OF DEFENDANT EFFECTS 

Race of Defendant Regression Coefficient (with level of statistical significance) 

Controlling for 39 Non-Racial Background Variables and the Race of Victim 
  

  

A. B. Ce 

Outcome All Cases Private Appointed Counsel Retained or 

Variable (n = 354) Counsel Appointed with 

(n = 169) Institutional Affiliation 

(n = 127) 

Death Sentence +01 .10 -.07 

Given A Murder (.90) {.25) {+.51) 

Indictment 

Penalty Trial Held .06 .03 «02 

After Murder Guilt (.49) {e73) {.89) 

Conviction 

Jury Death -.08 -.13 -.10 

Sentencing Decision (.39) £.27) {.72) 

  

 



  

~14- 9/12/83 

These data show a strong interaction effect between the status 

of defense counsel and the race of victim, die. the race of victim 

effects are much stronger in cases with private appointed counsel than 

they are in cases where defendant's counsel is retained or appointed 

with an institutional affiliation. These results tell us that 

prosecutors are more inclined to be punitive in cases involving white 

victims (and to a lesser degree black defendants) if the defense 

attorney is in private practice and court appointed. A possible 

explanation for this pattern is that private appointed attorneys put up 

less of a fight and otherwise develop less pressure on the prosecutor to 

accept a plea or unilaterally waive the death penalty. Under such 

circumstances the system is more likely to respond to the pressures for 

a death sentence that are generated when the victim is white. These 

data have particular significance since over 75% (98/128) of death 

sentences are imposed in cases in which defendant was represented by 

private appointed counsel. 

The status of defense counsel is also a proxy for the 

defendant's socio-economic status since counsel are appointed only for 

indigent defendants; the strong race of victim effect in appointed 

counsel cases therefore may also reflect a greater perception of 

culpability when the defendant is poorof prosecutors and juries in a 

greater willingness to respond to the pressures for a death sentence 

that arise when the victim is white. To test the extent to which the 

latter hypothesis is supported by the data, we conducted separate 

multiple regression analyses with variables ("interaction terms") which 

reflect the interaction between the race of victim and (a) the status of 

 



-15~- 9/12/83 

  

counsel, and (b) the defendant's socio-economic status. The results of 
I 

these separate analyses were as follows: 

    

  

Table 10 

A. B. C. D. FE. 

Race of Victim Race of Victim 

Interaction Interaction 

Term With lain Effect 

Status of Defendant's Status of Defendant's 

Defense Socio-Economic Counsel Socio-Economic 

Decision Point Counsel Status Analysis Status Analysis 

1. Death Sentence .04 «04 .09 «07 

Given a Murder (.26) {(.15) £.01) (.001) 

Indictment 

2. Prosecutor Seeks «13 «14 «13 wl? 

a Death Sentence {.10) (.09) {.03) (.01) 

After a Murder 

Trial Conviction 

3. Penalty Trial .04 «05 +15 13 

Death Sentencing {.81) {.78) {.11) {.19) 

Decision 

  

The coefficients in columns B and C are for the interaction terms which 

reflect how much larger on average the race of victim effect is when 

counsel is appointed (Col. B) or when the defendant's socio-economic 

is low (Col. C). Columns D and E show the magnitude of the race of 

victim "main effect" which indicates the average with the race of victim 

effect across all cases. Thus for the outcome measure "Death Sentence 

Given a Murder Indictment" (row 1), column D indicates in the analysis 

of the status of defense counsel variable an average race of victim 

effect of 9 percentage points and column B indicates a 4 point 

interaction effect with the status of counsel. These results indicate 

that the race of victim effect is 13 points in the cases with appointed 

 



  

-16- 9/12/83 

counsel and 5 points in the cases with counsel retained or appointed 

with an institutional affiliation. 

The results of the analysis in Table 10 show that the race of 

victim interaction effects are comparable for the "status of defense 

counsel" and the "low socio-economic status" variables. One is 

therefore left with the impression that both the competence, 

independence and energy of counsel and the indigence of the defendant 

affect the likelihood that the system will make the system more likely 

to respond to the pressures for a punitive response that arise when the 

victim is white.