Amicus Curiae Brief — Bazemore v. Friday

Supreme Court, U.S.

FILED

eae { FEB 13 1986

So

IN THE rs

Supreme Court of the United'States

OCTOBER TERM, 1985

P. E. BAZEMORE, et al.,

rs Petitioners,

WILLIAM C. FRIDAY, et al.,

Respondents.

UNITED STATES OF AMERICA, et al.,

Petitioners,

Vv.

WILLIAM C. FRIDAY, et ai.,

Respondents.

On Writ of Certiorari to the United States Court of Appeals

for the Fourth Circuit

BRIEF AMICUS CURIAE OF THE

EQUAL EMPLOYMENT ADVISORY COUNCIL

ROBERT E. WILLIAMS

DouGLas S. MCDOWELL *

THOMAS R. BaaBy, P.C.

GAREN E. DODGE

McGUINEss & WILLIAMS

1015 Fifteenth Street, N.W.

Suite 1200

Washington, D.C. 20005

(202) 789-8600

Attorneys for Amicus Curiae

Equal Employment Advisory

Council

* Counsel of Record

LS OT ET SRNR SE TE SN SE AS RRNA KS ERR SEES

WILSON - Erxs PRINTING Co.. Inc. - 769-0096 - WASHINGTON, D.C. 20001

®

1"

TABLE OF CONTENTS

TABLE OF AUTHORITIES ........

INTEREST OF THE AMICUS CURIAE

STATEMENT OF THE CASE

SUMMARY OF ARGUMENT .........................

ARGUMENT

THE COURT OF APPEALS PROPERLY AL-

LOCATED THE BURDENS OF PROOF IN THIS

DISPARATE TREATMENT PATTERN OR

PRACTICE CASE INVOLVING THE USE OF

A MULTIPLE REGRESSION ANALYSIS

I. To Establish A Prima Facie Case Of An Em-

ployer’s Pattern Or Practice Of Disparate

Treatment, A Plaintiff’s Multiple Regression

Analysis Must Account For Variables Reason-

ably Viewed As Determinants Of Salary

A. A Plaintiff Has The Burden To Account

For Relevant Variables

1. The Plaintiff’s Regression Analysis Must

Include The Relevant Objective And

Quantifiable Factors Actually Used By

The Employer To Make Salary Deter-

minations

Where The Plaintiff’s Regression Does

Not Account For Subjective And Un-

quantifiable Variables, Evidence Revard-

ing These Relevant Variables Must Be

Introduced In Some Other Form To Es-

tablish A Prima Facie Case

B. The Requirement That A Plaintiff Using

A Regression Analysis Acccunt For Rele-

vant Variables As Part Of Its Prima

Facie Case Is Consistent With The Legal

Requirements And Logical Basis For Es-

tablishing A Pattern Or Practice Of Dis-

parate Treatment

ii

TABLE OF AUTHORITIES—Continued

Il. A Defendant Need Only Articulate A Non-

discriminatory Reason For Its Conduct By

Identifying A Relevant Variable Omitted From

Plaintiff’s Regression, And Need Not Disprove

Plaintiff’s Case At Rebuttal

CONCLUSION

Page

20

ili

TABLE OF AUTHORITIES

Cases: Page

Agarwal v. Arthur G. McKee & Co., 19 FE:/P Cases

503 (N.D. Cal. 1977), aff’d, 644 F.2d 803 (9th

GH: BD etltcecttincdeenninnentiitiincece ine, eae 14, 15

Bazemore v. Friday, C.A. No. 2879 (E.D.N.C.

Aug. 20, 1982), aff’d, 751 F.2d 662 (4th Cir.

TEED cccnicsnniiattibssticibtencicnssiibiaiene eee eee 5, 13, 20

Capaci v. Katz & Besthoff, 711 F.2d 647 (5th Cir.

1983), cert. denied, 466 U.S. 927 (1984) .......... 22

Coates v. Johnson & Johnson, 756 F.2d 524 (7th

REE EE pre enna aoe 17

Connecticut v. Teal, 457 U.S. 440 (1982) iinet 3

Cooper v. Federal ei Bank, —— U.S. —,

104 S.Ct. 2794 (1984)...... SS Oe eS 18

Coser v. Moore, 587 F. Supp. 572 (E.D.N.Y.

1983), aff’d, 739 F.2d 746 (2d Cir. 1984) 14, 15

Eastland v. Tennessee Valley Authority, 704 F.2d

613 (11th Cir. 1983), cert. denied sub nom..,

James v. Tennessee Valley Autherity, 465 U.S.

Be GED cetitebntitnitaaee: 13, 14

EEOC v. Federal Reserve Bank, 698 F.2d 633

(4th Cir. 1983), rev ’ and remanded on other

grounds sub nom., > oon v. Federal Reserve

Bank, U.S. , 104 S.Ct. 2794 (1984)......... 11-12

EEOC v. Hartford Fire Insurance Co., 31 FEP

Cases 531 (D.D.C. 19838) _................... 14

EEOC v. International Business Machines Corp..

583 F. Supp. 875 (D. Md. 1984) ” 14, 19

EEOC v. Sears, Roebuck & Co., C.A. No. 79C

1873 (E.D. Ill. Jan. 31, 1986), reprinted in

Daily Labor Report (BNA) D-1 (Feb. 6, 1986) .. 12, 13,

14, 15, 16, 17

Furnco Construction Corp. v. Waters, 438 U.S.

567 (1978) ......... 3,9

Grano v. Department of Dev elopment, 699 F. 2d 836

ee ee ee a

Greenspan v. Automobile Club of Michigan, 495

F. Supp. 1021 (E.D. Mich. 1980) _........... 17

Hazelwood School District v. United States, 433

U.S. 299 (1977) ...............-...

iv

TABLE OF AUTHORITIES—Continued

Page

International Brotherhood of Teamsters v. United

pe RE UL UU >», | ee passim

James v. Stockham Valves & Fittings Co., 559

F.2d 310 (5th Cir. 1977), cert. denied, 434 U.S.

I I is seca aniettadelenislanninsiamenennnndmenienneenetentn 16

McDonnell Douglas Corp. v. Green, 411 U.S. 792

a 9, 10, 20

Melani v. Board of Higher Education, 561 F.

Supp. 769 (B.D.N.Y. 19GB) ...........--....---.0-cecceeee---- 17

Penk v. Oregon State Board of Higher Educ.,

C.A. No. 80-436 (D. Or. Feb. 13, 1985) .........-..... passim

Pouncy v. Prudential Insurance Co., 499 F. Supp.

427 (S.D. Tex. 1980), aff'd, 668 F.2d 795 (5th

a a acllcciels 14

Segar v. Civiletti, 508 F. Supp. 690 (D.D.C.

1981), rev'd in part sub nom., Segar v. Smith,

738 F.2d 1249 (D.C. Cir.), cert. denied sub nom.,

Meese v. Segar, 105 S. Ct. 2357 (1985) ............. 22

Sobel v. Yeshiva University, 566 F. Supp. 1166

I a eelinennenens 14, i6

Spaulding v. University of Washington, 740 F.2d

686 (9th Cir. 1984), cert. denied, —— USS.

S% §: 8 Oe 18

Stastny v. Southern Bell Tel. & Tel. Co., 458 F.

Supp. 314 (W.D.N.C. 1978), aff'd in part, 628

F.2d 267 (4th Cir. 1980) -....... acta cattails 18, 19

Sweeney v. Research Foundation, 711 F.2d 1179

EEE aw perro ree 22

Texas Department of Community Affairs v. Bur-

Game, GO Us BE CIID cccecncccttcncnccssntesecnccsesccssses passim

Tolliver v. Yeargan, 567 F. Supp. 116 (S.D. Ark.

1983), aff'd, 728 F.2d 1076 (8th Cir. 1984) ........ 14

Trout v. Hidalgo, 517 F. Supp. 873 (D.D.C. 1981),

rev'd in part sub nom., Trout v. Lehman, 702

F.2d 1094 (D.C. Cir. 1983), vacated and re-

manded on other grounds, 465 U.S. 1056 (1984). 22

United Air Lines, Inc. v. Evans, 431 U.S. 553

ERE NT PR ETN a ee NOR I 18

V

TABLE OF AUTHORITIES—Continued

Page

Valentino v. United States Postal Service, 511 F.

Supp. 917 (D.D.C. 1981), aff’d, 674 F.2d 56

I 14, 18

White v. City of San Diego, 605 F.2d 455 (9th

i Panne =a 12

White v. Vathally, 732 F. 2d 103 37 (st Cir. ), cert.

denied, 469 U.S. , 105 S.Ct. 331 (1984) _....... 22

Wilkins v. University of Houston, 471 F. Supp.

1054 (S.D. Tex. 1979), vacated and remanded,

654 F.2d 388 (5th Cir.), vacated and remanded

on other grounds, 459 U.S. 822 (1982), aff’d on

remand, 695 F.2d 134 (5th Cir. 1983) _................ )S- 10-11

Federal Statute:

Title 42, United States Code (1982)

§ 2000e, et seq. (Title VII, Civil Rights Act of

a ae passim

Miscellaneous:

D. Baldus and J. Cole, Statistical Proof of Dis-

crimination (1980 & Supp. 1985)... are il

D. Bloom and M. Killingsworth, Pay Discrimina-

tion Research and Litigation: The Use of Re-

gression, 21 Indus. Rel. L. J. 318 (1983) ............ 11

T. Campbell, Regression Analysis in Title VII

Cases: Minimum Standards, Comparable Worth,

and Other Issues Where Law and Statistics Meet,

36 Stan. L. Rev. 1299 (1984) 00 . 12,18

M. Finkelstein, The Judicial Reception of Multiple

Regression Studies in Race and Sex Discrimina-

tion Cases, 80 Colum. L. Rev. 737 (1980)........ 15

F. Fisher, Multiple Regression in Legal Proceed-

ings, 80 Colum. L. Rev. 702 (1980) ..................... 11, 12

Note, Title VII, Multiple Linear Regression Mo-

dels, and the Courts: An Analysis, 46 Law &

Contemp. Probs. 283 (1983)... 15

B. Schlei and P. Grossman, Employment Discrim-

ination Law (BNA) (2d ed. 1983 & Supp. 1983-

IIIT satenessdeaddhsediniadimiiaiinnaibatiainipe denen ae nee ier 21, 22

IN THE

Supreme Court of the United States

OCTOBER TERM, 1985

Nos. 85-93 and 85-428

P. E. BAZEMORE, et al.,

Petitioners,

Vv.

WILLIAM C. FRIDAY, et al.,

Respondents.

UNITED STATES OF AMERICA, et al.,

Petitioners,

V.

WILLIAM C. FRIDAY, et al.,

Respondents.

On Writ of Certiorari to the United States Court of Appeals

for the Fourth Circuit

BRIEF AMICUS CURIAE OF THE

EQUAL EMPLOYMENT ADVISORY COUNCIL

The Equal Employment Advisory Council (EEAC)

respectfully submits this brief amicus curiae. The

2

parties’ written consents to file this brief have been

filed with the Clerk of this Court.

INTEREST OF THE AMICUS CURIAE

The Equal Employment Advisory Council (EEAC)

is a voluntary nonprofit associaticn organized to pro-

mote the common interest of employers and the gen-

eral public in sound government policies, procedures

and requirements pertaining to nondiscriminatory

employment practices. Its membership comprises a

broad segment of the employer community in the

United States, including both individual employers

and trade and industry associations. Its governing

body is a Board of Directors composed primarily of

experts and specialists in the field of equal employ-

ment opportunity, whose combined experience gives

EEAC a unique depth of understanding of the prac-

tical and legal considerations relevant to the proper

interpretation and application of equal employment

policies and requirements. EEAC’s members are

committed firmly to the principles of nondiscrimina-

tion and equal employment opportunity.

Substantially all of EEAC’s members, or their con-

stituents, are employers subject to various equal em-

ployment laws, including Title VII of the Civil Rights

Act of 1964, as amended, 42 U.S.C. § 2000e, et seg.

EEAC’s members maintain wage and salary sched-

ules that comply with Title VII’s non-discrimination

requirements. Thus, EEAC has a direct interest in

one of the issues presented for the Court’s considera-

tion in the instant case—whether a district court

properly may dismiss allegations of a pattern or prac-

tice of disparate treatment which are based on multi-

ple regression analyses that fail to account for rele-

vant variables that are likely to affect salary deter-

3

minations. EEAC takes no position regarding

whether the plaintiffs in the instant case established

a timely claim under Title VII for current wage

disparities that result from pre-Act discrimination,

but rather limits its discussion herein to the issue of

the appropriate burdens of proof in cases involving

the use of multiple regression analyses.

Because of its interest in the issues of statistical

proof of discrimination under Title VII, and the rela-

tive burdens required of plaintiffs and defendants

thereunder, EEAC has filed amicus curiae briefs with

this Court in numerous cases involving the interpre-

tation of that Act, including /nternational Brother-

hood of Teamsters v. United States, 431 U.S. 324

(1977); Furnco Construction Co. v. Waters, 438 U.S.

567 (1978); Texas Department of Community Af-

fairs v. Burdine, 450 U.S. 248 (1981); and Connecti-

cut v. Teal, 457 U.S. 440 (1982).

STATEMENT OF THE CASE

Bazemore and more than fifty other employees

(“Bazemore” or “plaintiffs”, -<ued officials of the

North Carolina Agricultural Extension Service

(“NCAES” or “defendants’’), alleging a pattern or

practice of intentional race discrimination. Before

1965, NCAES agents, who provide agricultural and

other educational services to state residents, were

divided racially into separate branches. Prior to

1965, NCAES had paid black agents serving black

constituents a lower salary than white agents in a

similar capacity.

NCAES eliminated these separate branches in

1965, shortly after the effective date of Title VII, by

mcrging the two branches into a single organization.

4

NCAES attempted to eliminate pre-Act salary dis-

parities by, for exampie, setting uniform minimum

starting salaries for all agents. Wage disparities,

however, had not been entirely eliminated when this

suit was filed in 1971, or in 1972 when Title VII

was amended to require compliance by state agencies.

At trial, the Justice Department, as intervenor, in-

troduced multiple regression analyses comparing the

salaries of black and white agents for the years 1974,

1975 and 1981. These regression analyses, which con-

trolled for four variables allegedly determinative of

salary (race, education, tenure and job title), sug-

gested statistically significant racial effects in 1974

and 1975, but not 1981. These regressions did not

eontrol for across-the-board and percentage pay in-

creases. NCAES introduced regressions which also

added quartile rank, a measure of job performance,

yielding statistically significant disparities in 1975,

but not 1981.

The district court ruled in favor of NCAES on

all counts. First, the court ruled that a pre-Act dif-

ference in salaries need not be affirmatively elim-

inated. In addition, the district court rejected the

Justice Department’s regression analyses as flawed

because the regressions failed to account for a num-

ber of relevant variables, including across-the-board

and percentage pay increases. The court concluded

that while the plaintiffs had “probably made out a

prima facie case with respect to defendants’ promo-

tion and salary practices,” the defendant “articu-

lat{ed] plausible reasons for its actions . . . which

the court found convincing.” Pet. App. 190a.'

1 “Pet. App.” refers to the appendix filed with the petition

in hiv. 85-93. “Justice Br.” refers to the Brief for the Federal

5

The panel majority of the Fourth Circuit affirmed,

holding that while discrimination existed before the

enactment of Title VII, the law does not require that

a “pre-Act discriminatory difference in salaries

[. . . be] affirmatively eliminated.” Bazemore v.

Friday, 751 F.2d 662, 670 (4th Cir. 1984). Simi-

larly, the appeals court affirmed the district court’s

rejection of the plaintiffs’ regression analysis because

the figures analyzed “reflect the effect of pre-Act dis-

crimination” and fail to include “variables which

ought to be reasonably viewed as determinants of

salary.” Jd. at 672. Judge Phillips dissented in part,

reasoning that salary differentials continued “until

at least 1976,” and that the statistical analyses were

“wholly consistent” in showing a substantial race-

based disparity. Jd. at 693, 694.

SUMMARY OF ARGUMENT

The Fourth Circuit properly imposed upon the

plaintiffs the burden, in their prima facie case, to

account for those variables reasonably viewed as de-

terminants of salary. This Court should affirm that

standard, and also should require a plaintiff to prove

several specific elements, as part of its prima facie

case, when attempting to use a multiple regression

analysis to show a pattern or practice of disparate

treatment.

First, a plaintiff should be required to account for

the variables relevant to an employer’s salary deter-

Petitioners. “NAACP Br.” refers to the Brief for the Peti-

tioners, NAACP Legal Defense & Educational Fund, Inc.

“NCAES Br.” refers to the Brief for the Respondents, William

C. Friday, et al.

6

mination, a requirement met by initially ensuring

that the theoretical foundation of its regression

analysis “fits” the particular facts of the case. Thus,

a plaintiff’s regression analysis should include rele-

vant objective and quantifiable variables that are

likely to affect salary determinations. Second, if a

variable is truly unquantifiable, the plaintiff should

be required at a minimum to substitute a suitable

“proxy” for the relevant variable, or account for

the variable through other evidence. In addition, a

plaintiff should not be excused from the duty to

account for a variable merely because it is arguably

“tainted,” or allegedly masks an employer’s discrim-

ination, since the resolution of whether the variable

is in fact tainted is a question of fact to be decided

by the trial court. Also, a plaintiff’s regression nor-

mally should focus upon employment decisions made

during the relevant time period.

Legal and logical considerations favor adherence

to this standard. First, it is consistent with the

plaintiff’s customary burden in a disparate treatment

case to establish an employer’s discriminatory intent.

Imposition of this burden is particularly important

in a pattern or practice case, since a plaintiff under

that theory at all times maintains the responsibility

for proving by a preponderance of the evidence that

race discrimination “was the company’s standard op-

erating procedure—the regular rather than the un-

usual practice.” Teamsters, 431 U.S. at 336. Finally,

such burden is not onerous, but rather merely re-

quires the plaintiff to eliminate the most likely non-

discriminatory reasons for any observed disparity in

7

order to establish a prima facie case of discrimina-

tion.

The Fourth Circuit also applied the proper bur-

den of proof standards in finding that, even if the

plaintiffs established their prima facie case, the de-

fendants rebutted it. The appropriate rebuttal stand-

ard, as conceded by both plaintiffs herein, requires

a defendant only to articulate or “raise[] a genuine

issue of fact as to whether it discriminated against

the plaintiff,” not to “persuade the court that it was

actually motivated by the proffered reasons.” Texas

Department of Community Affairs v. Burdine, 450

U.S. at 254-55. This rebuttal standard, requiring a

defendant only to show that a plaintiff’s regression

is flawed in its assumptions, data or analysis, is

especially important in a case involving a pattern or

practice of disparate treatment. Of course, this artic-

ulation must be “clear and reasonably specific,” and

must have some basis in the facts as developed in

discovery or otherwise, but a defendant clearly is

not required to disprove a plaintiff’s case. Finally,

as this and other courts have noted, this standard is

appropriate since the defendant usually will attempt

to show at rebuttal that its practices were lawful, and

since liberal discovery rules permit a plaintiff access

to the relevant information needed to establish its

case.

8

ARGUMENT

THE COURT OF APPEALS PROPERLY ALLOCATED

THE BURDENS OF PROOF IN THIS DISPARATE

TREATMENT PATTERN OR PRACTICE CASE IN-

VOLVING THE USE OF A MULTIPLE REGRESSION

ANALYSIS

The Fourth Circuit below rejected the plaintiffs’

statistical proof because it believed that the plaintiffs

failed to establish a prima facie case of intentional

discrimination, and found that the defendants would

have rebutted any inference of discrimination. In

doing so, the court established appropriate burdens of

proof for both plaintiffs and defendants in a pattern

or practice disparate treatment case where multiple

regression analyses are used.’

While this case involves the use of a number of

complicated statistical analyses, including multiple re-

gression analyses, it is at bottom a relatively simple

burden of proof case, in which the parties’ respective

burdens are defined clearly under previous decisions

of this Court. In fact, all parties herein agree that

the defendant’s burden in such cases is that laid out

by this Court in Texas Department of Community

Affairs v. Burdine, 450 U.S. 248 (1981), and Inter-

national Brotherhood of Teamsters v. United States,

431 U.S. 324 (1977). See NAACP Br. at 53; Jus-

tice Br. at 25 n.23.

All parties concede, and the courts below properly

viewed this case to be a disparate treatment case. Dis-

2 EEAC takes no position on whether the plaintiffs in this

case established a timely claim under Title VII for current

wage disparities that result from pre-Act discrimination.

Rather, EEAC’s brief is limited to a discussion of the proper

burdens of proof in a class action or pattern or practice case

involving the use of multiple regression analyses.

parate treatment is concerned with those situations

where an employer “simply treats some people less

favorably than others because of their race, color,

religion, sex, or national origin.” Teamsters, 431

U.S. at 335 n.15 (emphasis added). Proof of dis-

criminatory motive is the central inquiry. /d. To

establish discriminatory motive in a disparate treat-

ment case, this Court has required a three-part order

and allocation of proof. McDonnell Douglas Corp. v.

Green, 411 U.S. 792 (1973). Briefly, a plaintiff may

establish a prima facie case by producing evidence

“adequate to create an inference that an employment

decision was based on a discriminatory criterion ille-

gal under the Act,” Teamsters, 431 U.S. at 358, and

by “eliminating the most common nondiscriminatory

reasons for the employer’s actions.” Burdine, 450

U.S. at 255.* A plaintiff carries the burden of show-

ing “actions taken by the employer from which one

can infer, if such actions remain unexplained, that

it is more likely than not that such actions were

‘based on a discriminatory criterion illegal under the

Act.’” Furneo, 438 U.S. at 576, citing McDonnell

Douglas (emphasis added).

* McDonnell Douglas permits the plaintiff to establish a

prima facie case by showing “(i) that he belongs to a [pro-

tected class]; (ii) that he applied and was qualified for a job

for which the employer was seeking applicants; (iii) that,

despite his qualifications, he was rejected; and (iv) that, after

his rejection, the position remained open and the employer

continued to seek applicants from persons of complainant’s

qualifications.” 411 U.S. at 802. The four McDonnell Douglas

requirements, however, are not an inflexible formula, but “a

sensible, orderly way to eveluate the evidence in light of com-

mon experience as it bears on the critical question of discrimi-

nation.” Furnco Construction Corp. v. Waters, 438 U.S. 567,

577 (1978).

10

The defendant then must “articulate,” or come

forward with, evidence of “some legitimate, nondis-

criminatory reason” for its actions, McDonnell Doug-

las, 411 U.S. at 802, or “provide a nondiscriminatory

explanation for the apparently discriminatory result.”

Teamsters, 431 U.S. at 360 n.46. If the defendant

carries this burden, the plaintiff has the opportunity

to show that the proffered reasons are “pretextual,”

that they were not the true reasons for the employ-

er’s action, and that the true reasons were discrimi-

natory. McDonnell Douglas, 411 U.S. at 804-05. The

burden of proof remains at all times with the plain-

tiff. Burdine, 450 U.S. 248. As detailed below, these

basic principles are fully and readily applicable to

pattern or practice cases involving the use of multi-

ple regression analyses.

I. To Establish A Prima Facie Case Of An Employer’s

Pattern Or Practice Of Disparate Treatment, A Plain-

tiff’s Multiple Regression Analysis Must Account For

Variables Reasonably Viewed As Determinants Of

Salary

This Court has recognized that “statistics can be

an important source of proof in employment discrimi-

nation cases.” Hazelwood School District v. United

States, 433 U.S. 299, 307 (1977). In fact, statistics

generally, and multiple regressions specifically, often

play an important role in proving a pattern or prac-

tice of discrimination. This Court has recognized,

however, that statistical proof’s “usefulness depends

on all of the surrounding facts and circumstances.”

Teamsters, 431 U.S. at 340. It is precisely “[b]e-

cause of the significant role that statistics can play

in discrimination cases” and because of their “inher-

ently slippery nature” that “it is imperative that

they be used properly.” Wilkins v. University of

11

Houston, 654 F.2d 388, 395 (5th Cir.), vacated and

remanded on other grounds, 459 U.S. 822, aff'd on

remand, 695 F.2d 184 (5th Cir. 1983); see also

EEOC v. Federal Reserve Bank, 698 F.2d 633, 645

(4th Cir. 1983), rev’d and remanded on other grounds

sub nom., Cooper v. Federal Reserve Bank, U.S.

, 104 S.Ct. 2794 (1984).

Multiple regression is a “relatively sophisticated

means of determining effects that any number of dif-

ferent factors have upon a particular variable.”

Wilkins, 654 F.2d at 402-03.* Thus, the “validity of

the [regression] and the significance of its results”

must be examined, id. at 403, to determine the “ ‘ex-

* A multiple regression analysis is concerned with the effect

that various independent variables (such as race, sex, seniority

or education) might have on a dependent variable (such as

salary). In other words, a regression analysis, in appropriate

cases, can help determine to what extent race may have been

a factor in the salary received by certain employees. Thus,

the purpose of a regression analysis regarding alleged pay

discrimination on the basis of race “is to determine whether a

[racial] difference in pay persists when one factors out the

possible effects of other variables.” D. Bloom and M. Killings-

worth, Pay Discrimination Research and Litigation: The Use

of Regression, 21 Indus. Rel. L.J. 318, 321 (1983). With a

multiple regression, “one first specifies the major variables

that are believed to influence the dependent [salary] variable.”

F. Fisher, Multiple Regression in Legal Proceedings, 80 Colum.

L. Rev. 702, 705 (1980). “The relationship between the de-

pendent variable and the independent variable of interest is

then estimated by extracting the effects of the other major

variables. .. .” Jd. at 706. Those results “can be read as show-

ing the effects of each variable on the dependent variable, hold-

ing the others constant. Moreover, those results allow one to

make statements about the probability that the effect described

has merely been observed as a result of chance fluctuation.”

Id.; see D. Baldus and J. Cole, Statistical Proof of Discrimina-

tion § 8.022[1] (1980 & Supp. 1985).

12

istence of proper supportive facts and the absence of

variables which would undermine the reasonableness

of the inference of discrimination which is drawn,’ ”’

Federal Reserve Bank, 698 F.2d at 645, quoting

White v. City of San Diego, 605 F.2d 455, 460 (9th

Cir. 1979). The Fourth Circuit properly stated gen-

erally that a plaintiff’s regression must account for

variables reasonably viewed as determinants of sal-

ary, but it is imperative to delineate several specific

elements of that standard which this Court should re-

quire for a plaintiff’s regression analysis to establish

a prima facie case of a pattern or practice of dispar-

ate treatment.

A. A Plaintiff Has The Burden To Account For Rele-

vant Variables

1. The Plaintiff’s Regression Must Include The

Relevant Objective And Quantifiable Factors Ac-

tually Used By The Employer To Make Salary

Determinations

The specific elements of the general standard out-

lined by the Fourth Circuit flow logically from that

standard. First, a plaintiff must relate the theoreti-

cal foundation of its regression analysis (in this case,

the variables believed to influence salary) to the

method actually used by the employer to make the

decisions under scrutiny. This requirement is often

referred to by statisticians as “goodness of fit.” See,

e.g., EEOC v. Sears, Roebuck & Co., C.A. No. 79C

4373 (E.D. Ill. Jan. 31, 1986), reprinted in Daily

Labor Report (BNA) D-1, D-6 (Feb. 6, 1986); Bal-

dus and Cole § 8.2; T. Campbell, Regression Analysis

In Title VII Cases: Minimum Standards, Comparable

Worth and Other Issues Where Law and Statistics

Meet, 36 Stan. L. Rev. 1299, 1803 (1984); Fisher at

716.

13

The statistical “model must be hypothesized to link

the influences with the outcome,” Campbell at 1302,

a hypothesis that usually is formed during discovery,

and which yields potential explanatory variables that

are believed to affect salary. See Eastland v. Tennes-

see Valley Authority, 704 F.2d 613 (11th Cir. 1983),

cert. denied sub nom., James v. Tennessee Valley Au-

thority, 465 U.S..1066 (1984); Stastny v. Southern

Bell Tel. & Tel. Co., 628 F.2d 267, 274 (4th Cir.

1980) ; Penk v. Oregon State Board of Higher Educ.,

C.A. No. 80-436, slip. op at 379-380 (D. Or. Feb. 13,

1985).° These variables in the instant case, as rec-

ognized by the Fourth Circuit, include education, ten-

ure, job title, work performance and county-by-county

disparities. Bazemore v. Friday, 751 F.2d 662, 672

(4th Cir. 1984).

The burden to construct a theoretical model that

“fits” places the concomitant burden upon a plaintiff

to include in its regression those objective and quan-

tifiable variables relevant in actually determining

wages. EEOC v. Sears at D-6 & n.8. Many courts

have imposed this relevance requirement, and have

examined closely those factors included or omitted

from a plaintiff’s regression. For example, the court

° The failure to establish a theoretical foundation repre-

sents a significant problem for plaintiffs attempting to use a

regression analysis to establish a prima facie case. In fact,

one commentator in 1984 noted that “in only eight reported

cases has the opinion reflected any attempt to measure

whether the model was appropriate for the data.” Campbell

at 1306-07. The commentator also noted that “evidence of

the model’s appropriateness and the closeness of its fit may

have been introduced at trial [in other cases] but not included

in the opinion. If so, judges are not exhibiting adequate con-

cern for a fundamental aspect of statistical proof, important

upon appellate review.” Campbell at 1307 n.22.

14

in EEOC v. International Business Machines Corp.,

583 F. Supp. 875, 906 (D. Md. 1984), found that

the plaintiff’s omission of relevant factors such as

seniority, job level and class groupings rendered the

results of its multiple regression analysis “useless.”’

The court in Tolliver v. Yeargan, 567 F. Supp. 116

(S.D. Ark. 1983); aff'd, 728 F.2d 1076 (8th Cir.

1984), also held that the plaintiff had not established

a prima facie case in part because its regression

failed to account for school-by-school differences in

salary. Another court found that the plaintiff’s re-

gression was “rife with methodological and analytical

flaws, and therefore is not credible,” partly because

the analysis failed to account for the defendant’s

across-the-board percentage increases, as well as dis-

cretionary or merit increases. Coser v. Moore, 587

F.Supp. 572, 599 (E.D.N.Y. 1983), aff'd, 739 F.2d

746 (2d Cir. 1984).°

It is clear, therefore, that a plaintiff does not

establish its prima facie case when it relies solely

upon a regression analysis that does not account for

® See also EEOC v. Sears at D-37 (failure to account for,

inter alia, job performance and the employer’s decentralized

decision-making made plaintiff’s statistics “‘so flawed that they

lack any persuasive value”); EEOC v. Hartford Fire Insur-

ance Co., 31 FEP Cases 531, 545 (D.D.C. 1983) (failure to

account for several relevant variables which might have influ-

enced an employee’s salary, including age, employment experi-

ence, work performance and tenure); Pouncy v. Prudential

Insurance Co., 668 F.2d 795, 803 (5th Cir. 1982); Sobel v.

Yeshiva University, 566 F. Supp. 1166 (S.D.N.Y. 1983) (fail-

ure to control for rank) ; Eastland, 704 F.2d at 623 (failure to

control for job category casts doubt on regression) ; Valentino

v. United States, 511 F. Supp. 917, 945 (D.D.C. 1981), aff’d

674 F.2d 56 (D.C. Cir. 1982); Agarwal v. Arthur G. McKee

& Co., 19 FEP Cases 503 (N.D. Cal. 1977), aff’d, 644 F.2d 803

(9th Cir. 1981).

15

relevant objective and quantifiable variables. More-

over, to shift the burden of persuasion to the defend-

ant based upon a poorly fitted or incomplete regres-

sion would violate the standards of proof articulated

by this Court in Teamsters and Burdine, as discussed

infra. :

2. Where The Plaintiff’s Regression Does Not Ac-

count For Subjective And Unquantifiable Vari-

ables, Evidence Regarding These Relevant V ari-

ables Must Be Introduced In Some Other Form

To Establish A Prima Facie Case

In some instances, a regression analysis may be

unable to account for truly unquantifiable variables,

such as job prefereces or motivation. Note, Title

VII, Multiple Linear Regression Models, and the

Courts: An Analysis, 46 Law & Contemp. Probs. 282,

286-7 (1983); See M. Finkelstein, The Judicial Re-

ception of Multiple Regression Studies in Race and

Sex Discrimination Cases, 80 Colum. L. Rev. 737, 742

(1980). It is obvious, however, that employers do,

in fact, make employment decisions based upon un-

quantifiable differences between empioyees (job per-

formance, in many instances), a practice that is “‘per-

missible and necessary under Title VII.” Penk, slip

op. at 387; see also EEOC v. Sears at D-38. These

unquantifiable variables, in fact, may significantly

affect the salaries paid to particular employees. It

follows, therefore, that a court cannot correctly hold

that plaintiffs have established a prima facie case

where their regression fails to account for the actual

“processes used . . . to make employment decisions.”’

Penk, slip op. at 388; see EEOC v. Sears at D-38;

Agarwal, 19 FEP Cases at 506; Coser, 587 F. Supp.

at 599.

Plaintiffs, certainly, can use non-statistical evi-

dence to account for the actual processes used by an

16

employer to make the decisions under scrutiny. In

fact, a regression may not be an appropriate means

by which to infer discrimination from an employer’s

compensation scheme, if the factors affecting compen-

sation cannot be quantified. EEOC v. Sears at D-39.

in the situation where a plaintiff fails to account for

the factors affecting compensation, either with a re-

gression or with other evidence, the district judge

should dismiss the claim for failure to establish a

prima facie case.

Plaintiffs will not be unduly hindered by the im-

position of this requirement to account for relevant

unquantifiable variables. If a plaintiff wishes to use

a regression analysis, instead of other relevant evi-

dence, suitable “proxies” sometimes can be used to

estimate variables such as “productivity.” ’ The court

in Sobel v. Yeshiva University, 566 F. Supp. 1166

(S.D.N.Y. 1983), for example, substituted rank as a

proxy for productivity. In fact, NCAES in the in-

stant case used the quartile ranking system for rro-

ductivity, a factor which plaintiffs chose not to in-

corporate into their regression. Of course, a proxy

may not represent a relevant variable exactly. Where

an appropriate proxy is not available, it is imperative

that a plaintiff prove its case with other evidence.

Plaintiffs often object to any requirement that

they account for subjective factors which are alleged

to be “tainted,” or which allegedly mask an employ-

‘A proxy attempts to measure a variable which seems un-

quantifiable in its pure form. For example, while a regres-

sion analysis can account for productivity in a job where the

measurement is widgets produced per hour, it may not be

able to measure research quality or student teaching in a

university setting. See, e.g., Penk, slip op. at 385. The proxy

purports to measure the apparently unquantifiable factor by

substituting another measurement.

17

er’s discrimination. See James v. Stockham Valves

¢ Fittings Co., 559 F.2d 310, 332 (5th Cir. 1977),

cert. denied, 434 U.S. 1034 (1978); Greenspan v.

Automobile Club of Michigan, 495 F. Supp. 1021, 1064

(E.D. Mich. 1980) ; Melani v. Board of Higher Educ.,

561 F. Supp. 769, 778 (S.D. N.Y. 1983). This objec-

tion lacks persuasive appeal, however, since whether

a variable is tainted depends upon the district court’s

ultimate ruling on whether there has been discrim-

ination in the employer’s application of that factor.

Absent persuasive extrinsic evidence that the variable

was in fact “tainted” by unlawful discrimination,

the plaintiffs should be required to include such a

variable in their regression analysis. See EEOC v.

Sears at D-41. Allowing the plaintiffs to exclude the

variable from their regression merely because they

have alleged that the factor was applied discrimina-

torily would presuppose that discrimination occurred.

See Coates v. Johnson & Johnson, 756 F.2d 524, 544

(7th Cir. 1985).

Thus, a plaintiff has the specific burden to include

in its regression analysis the relevant factors used

by the employer to determine salaries. Seemingly

subjective or unquantifiable factors may be repre-

sented in the analysis by suitable proxies, but where

suitable proxies are not available, unquantifiable fac-

tors affecting salary must be taken into account by

the production of other evidence of discriminatory

intent. Plaintiffs also must account for evidence that

is arguably tainted unless plaintiffs can prove, with

persuasive extrinsic evidence, that the variable actu-

ally masks the employer’s discrimination. Ultimately,

therefore, plaintiffs must account for the factors used

18

by the employer to make its decisions, or risk failing

to establish a prima facie case.*_

B. The Requirement That A Plaintiff Using A Regres-

sion Analysis Account For Relevant Variables As

Part Of Iis Prima Facie Case Is Consistent With

The Legal Requirements And Logical Basis For

Establishing A Pattern Or Practice Of Disparate

Treatment

Legal and logical reasons favor imposing upon a

plaintiff the burden to establish, at the prima facie

case stage, that its statistics account for relevant

variables that may have affected salary determina-

tions within the relevant tirae period. This burden is

not overly stringent; rather, it is simply a specific

application of the plaintiff's normal burden in a

disparate treatment case, which is to establish proof

of an employer’s “discriminatory motive” or intent.

Spaulding v. University of Washington, 740 F.2d 686

(9th Cir. 1984), cert. denied, U.S. —— (1984) ;

Penk, slip op. at 403.

Adherence to this standard is especially important

where a plaintiff alleges a pattern or practice of

disparate treatment, since under that theory a plain-

tiff must show by a preponderance of the evidence

that the alleged discriminatory practice is the “com-

pany’s standard operating prucedure—the regular

rather than the unusual practice.” Cooper, 104 S. Ct.

at 2800. citing Teamsters at 336. Such a wholesale

8 While EEAC takes no position regarding whether the

plaintiffs in the instant case established a timely claim under

Title V1I for current wage disparities resulting from pre-Act

discrimination, it is important to stress that regression analy-

ses normally should focus on employment decisions made

during the relevant time period. See Valentino, 511 F. Supp.

at 944, 955; see generally Hazelwood, 433 U.S. at 309; United

Air Lines, Inc. v. Evans, 431 U.S. 553 (1977).

19

attack on an employer’s standard operating proce-

dure “must be one based upon a specific intent to

discriminate against an entire group, to treat it as

a group less favorably because of [race].” HEHOC

v. International Business Machines Corp., 583 F.

Supp. at 903, citing Stastny, 628 F.2d at 274 n.10.

A plaintiff alleging disparate treatment, of course,

always bears the prima facie burden to “eliminate ]

the most common nondiscriminatory reasons for the

employer’s action.” Burdine, 450 U.S. at 254. But

where the attack is directed at an employer’s entire

operating procedure, it is particularly important that

the plaintiff’s evidence account for variables that

- might legitimately explain differences in measured

results.

This standard has intuitive appeal given the ulti-

mate factual issue before the district court in a case

of this type: whether the defendant both intention-

ally and as a standard operating procedure, discrim-

inated against the plaintiffs. Thus, it makes little

sense to permit a plaintiff to establish its initial bur-

den with statistics that fail to account for the pri-

mary factors that are likely to affect salary.

Requiring a plaintiff to account for such factors

is entirely reasonable. The Justice Department, a

plaintiff herein, has noted that the Fourth Circuit

“clearly would have erred if it had held, as private

petitioners assert in their Questions Presented, ‘that

statistics may not be treated as probative evidence of

discrimination unless the statistical analysis con-

siders every conceivable non-racial variable.’” Jus-

tice Br. at 21. Of course, the court below did not

impose this onerous “every conceivable variable” bur-

den on plaintiffs. Rather, as the Justice Department

itself acknowledges, “the appellate court articulated

— 20

the correct rule, that a plaintiff’s statistical analyses

in a disparate treatment case must include ‘variables

which ought to be reasonably viewed as determinants

of salary.’”’ Jd. This burden is one which properly

allocates the relative burdens between plaintiffs and

defendants and, with the further requirement that

a plaintiff account, either through its regression anal-

ysis or otherwise, for relevant variables, will pro-

vide adequate guidance to future litigants.

Il. A Defendant Need Only Articulate A Nondiscrimina-

tory Reason For Its Conduct By Identifying A Rele-

vant Variable Omitted From Plaintiff's Regression, And

Need Not Disprove Plaintiff’s Case At Rebuttal

The Fourth Circuit below rejected the plaintiffs’

regression analysis because the court determined that

plaintiffs failed to establish a prima facie case. Baze-

more, 751 F.2d at 672. The court went on to find that

even “assum[ing] that a prima facie case was made

. it has been completely refuted [by NCAES].”

Id. at 687. It is important for the present discussion

to examine what the law currently requires a defend-

ant to produce to rebut a plaintiff’s prima facie case,

and how that theoretical burden should be applied in

a regression case.

Once a plaintiff establishes an inference of inten-

tional discrimination by proving a prima facie case,

the burden shifts to the employer to “articulate some

legitimate, nondiscriminatory reason” for the chal-

lenged action, McDonnell Douglas, 411 U.S. at 802, or

to “provide a nondiscriminatory explanation for the

apparently discriminatory result.” Teamsters, 434

U.S. at 360 n.46; Burdine, 450 U.S. at 255. This

Burdine standard, conceded as applicable by NAACP

in its brief at 53 and by the Justice Department in its

brief at 25 n.23, requires that the defendant “clearly

21

set forth, through the introduction of admissible evi-

dence, the reasons for [its actions].” Burdine, 450

U.S. at 255. This concession by the petitioners is im-

portant, because defendant’s burden is then merely to

“raise[] a genuine issue of fact as to whether it dis-

criminated against the plaintiff,” not to “persuade

the court that it was actually motivated by the

proffered reasons.” Id. at 254-55.

Thus, the Burdine standard, as applied to a pat-

tern or practice case involving regression analyses,

permits a defendant to meet its burden either (1) by

showing that a plaintiff’s regression is flawed in its

assumptions, data or analysis; or (2) by demonstrat-

ing, through its own analysis, that there does not

exist a significant statistical disparity. Teamsters,

431 U.S. at ©“ 9-40; See B. Schlei and P. Grossman,

Employment Discrimination Law (BNA) 1323 (2d

ed. 1983).

The defendant’s articulation, of course, must be

“clear and reasonably specific.” Burdine, 450 U.S. at

258. Typically, this burden of specific articulation in

a regression case will be met when the defendant

identifies a relevant variable that is likely to have

affected salary, which is omitted from the plain-

tiff’s regression or otherwise is unaccounted for in

the plaintiff’s evidence. This identification of an omit-

ted variable will “raise[] a genuine issue of fact as

to whether [the defendant] discriminated against

the plaintiff,” id. at 254-55, and will also provide the

plaintiff with a full and fair opportunity to demon-

strate pretext at the next stage of proof. Id. at 258;

see Sweeney v. Research Foundation, 711 F.2d 1179

(2d Cir. 1983); Grano v. Department of Develop-

22

ment, 699 F.2d 836, 837 (6th Cir. 1983); White v.

Vathally, 732 F.2d 1037 (1st Cir.), 469 U.S. ;

cert. denied, 105 S.Ct. 331 (1984); See also Schlei

and Grossman (Supp. 1983-84) at 250.

The two alternative methods of rebutting a plain-

tiff’s case—identifying an omitted variable or pre-

senting a more appropriate regression—will not re-

sult in a “nonexistent” level of proof as plaintiffs

fear. Under the first method, a defendant’s articula-

tion of an omitted variable will raise more than theo-

retical or hypothetical objections to a plaintiff’s re-

gression. Trout v. Lehman, 702 F.2d 1094, 1102

(D.C. Cir. 1983), vacated and remanded on other

grounds, 465 U.S. 1056 (1984) ; Penk, slip op. at 404.

Normally, this articulation will be grounded in the

facts as determined during discovery. Alternatively,

if a defendant chooses to present a more appropriate

regression, its model normally will focus the issues

by demonstrating that the inclusion of different or

additional] variables may affect the outcome.

Under no circumstances, however, should the de-

fendant be required at rebuttal to “rework plaintiff’s

statistics” by “incorporating the omitted factor,” or

to demonstrate “how specific errors” actually have

affected the statistical results obtained from the plain-

tiff’s model, as some courts have required. See, e.g.,

Trout, 702 F.2d at 1102; Capaci v. Katz & Besthoff,

Inc., 711 F.2d 647, 654 (5th Cir. 1983), cert. denied,

466 U.S. 927 (1984) ; Segar v. Civiletti, 508 F. Supp.

690, 712 (D.D.C. 1981), rev’d in part sub nom.,

Segar v. Smith, 738 F.2d 1249 (D.C. Cir.), cert.

denied sub nom., Meese v. Segar, 105 S.Ct. 2357

(1985). As noted by the Justice Department, these

23

cases rely on Teamsters for the relevant burden

standards. Justice Br. at 25 n.23. “But the pattern

of proof in Teamsters does not differ significantly

from that stated in Burdine,”’ a standard which

should only require a showing of “inaccurate or insig-

nificant” proof or providing a “nondiscriminatory

explanation,” not the more burdensome standard of

persuasion. Jd. Thus, while the employer always has

the option of offering its own regression analysis to

offset the plaintiff’s prima facie showing, it is not

required to do so.

Other policy reasons stated by this Court favor

imposing the traditional Burdine rebuttal standard

upon a defendant in a pattern or practice disparate

treatment case. As this Court stated in Burdine, a

limited evidentiary obligation will not “unduly hinder

the plaintiff.” 450 U.S. at 258. At the “pretext”

stage, the plaintiff will have the opportunity to show

that the defendant’s stated reason has no basis in

fact, and that its real reason was a discriminatory

one, which, if proven, ends the inquiry.

This Court has noted a final reason why a plaintiff

will not be disadvantaged by a defendant’s limited

evidentiary obligation at the rebuttal stage. “([L]ib-

eral discovery rules applicable to any civil suit in

federal court” allow the plaintiff to become aware, at

an early stage, of the relevant facts making up the

employer’s decision. Jd; see also Penk, slip op. at

398. This rationale is especially appropriate in the

instant case, where the extensive discovery prior to

the plaintiff’s development of its statistical regres-

sion should have alerted it to the fact that NCAES

would claim that across-the-board increases, county-

24

by-county disparities, and merit differentials would

account for significant wage disparities.’

The Burdine and Teamsters rebuttal standard,

therefore, requires the defendant to identify clearly

and specifically a relevant variable omitted from a

plaintiff’s regression, or otherwise not accounted for

by the plaintiff’s evidence. This identification prop-

erly focuses the inquiry, yet reinforces plaintiff’s ulti-

mate burden of proving, either with a regression

analysis or with other evidence, that the employer

engaged in a pattern or practice of intentional dis-

crimination.

* The Justice Department argues that it “should be excused

from including” the county-by-county variable in its regres-

sion even though it is a “major salary-affecting variable,”

because this factor was not revealed “until the brink of trial.”

Justice Br. at 30 & n.30. NCAES responds that depositions

conducted before 1975, and again in July 1981 revealed this

factor as important. See NCAES Br. Regardless of whether

plaintiffs were actually notified, the argument of the Justice

Department may well be legally irrelevant, since lack of actual

notice does not obviate a plaintiff’s responsibility to account

for a relevant factor that it should have been aware of through

discovery. See discussion, supra, at 8-17.

25

CONCLUSION

For the foregoing reasons, EEAC respectfully

urges this Court to clarify the respective burdens of

proof that rest upon plaintiffs and defendants in cases

involving the use of multiple regression analyses, as

set forth herein.

Respectfully submitted,

ROBERT E. WILLIAMS

DOUGLAS S. MCDOWELL *

THOMAS R. Baapy, P.C.

GAREN E. DODGE

McGUINESS & WILLIAMS

1015 Fifteenth Street, N.W.

Suite 1200

Washington, D.C. 20005

(202) 789-8600

Attorneys for Amicus Curiae

Equal Employment Advisory

Council

* Counsel of Record

February 13, 1986