Professional Documents
Culture Documents
Discovery and The Evidentiary Foundations of Implicit Bias
Discovery and The Evidentiary Foundations of Implicit Bias
xperts who opine on implicit, or unconscious, bias as a source of discrimination pose unusual challenges for attorneys who must challenge
their testimony. 1 Implicit bias is dicult to measure using even sensitive
instruments in controlled testing environments, and to date courts have not ordered employees of defendant companies to submit to testing for implicit bias. 2
Thus, unlike other experts who commonly appear in employment cases, experts
who testify regarding implicit bias base their testimony on general social science
research that has no demonstrated connection to the present case but which the
experts contend is helpful in understanding the claims or defenses at issue in the
case. 3 For example, Dr. William Bielbys report in Dukes v. Wal-Mart Stores, Inc.
contains over 100 citations to the social-psychological literature, but fails to analyze
any decisions by Wal-Mart that are alleged to be discriminatory. In contrast, the
report in that same case by the plaintis statistician, Dr. Richard Drogin, contains
no citations to academic research but describes his analysis of Wal-Marts data on
pay and promotions. 4 The attack on Dr. Drogins testimony focused on his own
statistical analysis, which was rebutted with the statistical studies of Wal-Marts
own expert. On the other hand, the attack on Dr. Bielby focused primarily on
the attenuated connection between the research he cited, which was not specic
to Wal-Mart, and the employment decisions at issue in the case; the numerous
studies on which he based his testimony went largely unexamined.
Dr. Drogins testimony was based on Wal-Marts own data. During the course
of discovery, Wal-Mart was able to replicate his statistical work and expose what
it alleged to be its weaknesses and shortcomings. As a result, not only was Dr.
Drogins credibility as a witness in play, but so were the many judgments he made
as a statistician. His decision to include or exclude various segments of Wal-Marts
workforce in his data, which promotions he deemed worthy of study, how he
dened a promotion, any distinctions he drew between full and part time employment, the estimation techniques he used to obtain his statistical results, and his
2015 BY ALLAN G. KING, GREGORY MITCHELL, RICHARD W. BLACK,
CATHERINE A. CONWAY, AND JULIE A. TOTTEN
49
Impediments To Scrutinizing
Secondary Data
Two important points frame our discussion of discovery
of secondary data: (1) although often ignored in practice,
a strong norm exists within scientic communities that
the data behind research publications should be available
for review by others to the greatest extent possible without
jeopardizing the safety or privacy interests of those studied;
46
and (2) an expert who relies on secondary data places
SPRING 2015
Rule 45(c)(3)(B)(ii)
When a testifying expert bases his or her report on an
empirical study performed solely for a pending lawsuit, it
is customary to obtain the experts entire le that pertains
to the case, including the input le containing the data
on which the analysis was performed, the programs that
operated on that data, and the output le containing the
results of the statistical analysis performed by the expert. 49
Rule 703 authorizes the same searching inquiry regarding
secondary data relied on by testifying experts, including inspection of the data and code les underlying the reported
results. Unless a party is permitted access to secondary data,
crucial limitations and aws may never be discovered, and
the party may not be able to demonstrate why the foundation of the opposing experts opinion is unreliable.
The chief obstacle to discovery of secondary data is Rule
45(c)(3)(B)(ii) of the Federal Rules of Civil Procedure.
That Rule provides in pertinent part: To protect a person
subject to or aected by a subpoena, the issuing court may,
on motion, quash or modify the subpoena if it requires:
disclosing an unretained experts opinion or information
that does not describe specic occurrences in dispute and
results from the experts study that was not requested by a
party. Because the literature relied upon by implicit bias
experts may be authored by dozens of authors who have
not been retained in the particular case, and because the
data will rarely be drawn from the parties to the litigation, this discovery rule seemingly precludes the inquiry
mandated by Rule 703.
The Advisory Committee Notes to the 1991 amendments to Rule 45 indicate that this provision was intended
to provide appropriate protection for the intellectual
property of the non-party witness . A growing problem has been the use of subpoenas to compel the giving
of evidence and information by unretained experts. 50
According to the Advisory Committee, compulsion to
give evidence may threaten the intellectual property of
SPRING 2015
56
And in yet another case, the court extended the journalists privilege to research materials assembled by two
academics and protected the materials from disclosure.
In particular, Microsoft sought, for use in its defense
against the civil antitrust case led against it by the U.S.
government, the notes, recordings, transcripts, and correspondence between two business school professors and
employees of Netscape in connection with a book on
Netscape and its competition with Microsoft. 58 After a
detailed analysis, the appellate court armed the district
courts ruling quashing the subpoena, on grounds that
ordering production would hamstring not only [these
researchers] future research eorts but also those of others
similarly situated scholars. 59 The court also stated that it
was noteworthy that the respondents [were] strangers to
the antitrust litigation; insofar as the record reects, they
have no dog in the ght. [C]oncern for the unwanted
burden thrust upon non-parties is a factor entitled to
special weight in evaluating the balance of competing
needs. 60 However, the district court did leave open the
possibility of revisiting the need for the requested materials if the testimony in the case conicts with quotations
contained in the book. 61
The following tables summarize additional pertinent
cases, with the rst table summarizing cases in which
the secondary expert was required to testify and/or
produce documents:
Case
Facts
Holding
Frazier v. Stryker Corp., No. MC-10-0059-PHXFJM, 2010 U.S. Dist. LEXIS 91550 (D. Ariz. Aug.
12, 2010)
SPRING 2015
In the following cases, the court protected secondary experts from discovery:
Case
Facts
Holding
The court denied the defendants motion to compel, nding defendant did not want to depose him
as a fact witness, but as an expert on the cold sore
medication industry.
The plaintiff alleged that the defendants vaccine infringed on one of its patents. The defendant subpoenaed an expert who consulted with
the plaintiff in another lawsuit. The defendant
argued that this consulting experts opinion in
a prior lawsuit differed from the position of the
plaintiffs current expert.
The court quashed the subpoena nding that beThe plaintiff alleged that challenged research
was false because of a deeply awed methodol- cause experts research was published, plaintiff did
ogy. The plaintiff sought to depose a secondary not have a substantial need for the testimony.
expert for purposes of establishing an appropriate benchmark of sound methodology to which
the challenged research could be compared. The
expert moved to quash the subpoena.
In re Bextra & Celebrex Mktg. Sales Pzer subpoenaed the New England Journal of
Practices & Prod. Liab. Litig., 249
Medicine to obtain its correspondence with the
F.R.D. 8 (D. Mass. 2008)
authors of articles published regarding two of
Pzers products.
Dow Chem. Co. v. Allen, 672 F.2d
1262 (7th Cir. 1982)
Plaintiff subpoenaed all notes, reports, and raw Court refuses to enforce subpoena because predata regarding research in progress by univermature disclosure of research in progress would
sity researchers.
undermine peer review process and jeopardize the
incomplete research.
Proprietary Issues
Research takes time and money. Scholars therefore may
object to producing the data on which they base their
research on the grounds that they have yet to exhaust the
fruits of the data they compiled (as Dr. Kaburakis did in
the case discussed above). If their data were disclosed,
they contend they could lose what essentially are their
monopoly rights in that data. For example, in Holloway
v. Best Buy, Inc., the University of California resisted a
subpoena served on one of its researchers by arguing that
disclosure of the data she compiled would appropriate
her intellectual property and preempt her opportunity to
publish based upon the data she collected. 62
However, courts routinely protect trade secrets and intellectual property by means of protective orders. Thus, a
protective order would ensure that the secondary experts
data is disseminated solely to the parties to the litigation,
and only scholars retained as experts by each party would
have access. The retained experts would be subject to the
courts contempt powers and would be precluded from
commenting or publishing articles that reference the data
in question. With such protection in place, there is little
likelihood that bootleg research based upon that data
would preempt the research of the scholar who initially
created that database.
The universitys objection in the Holloway case also raises
an important issue about the public nature of scholarship.
First, much of the research relied on by implicit bias experts
has been conducted with the aid of federal grant funds.
Accordingly, the notion that a private researcher has an exclusive property interest in data compiled with public funds
seems questionable. Apart from that is the larger question
of whether a researcher who, for the sake of argument,
nanced her research with private funds retains a proprietary interest once she publishes an article based upon that
data. Once the cat is out of the bag, the interest asserted
by this scholar is the right, in eect, to generate a series of
additional publications without sharing the data until their
value to her has been exhausted. At issue is whether there
is a right recognized in academia to monopolize data and
58
requirement includes adhering to ASAs stated policy on data-sharing: Sociologists make their data
available after completion of the project or its major
publications, except where proprietary agreements
with employers, contractors, or clients preclude
such accessibility or when it is impossible to share
data and protect the condentiality of the data or
the anonymity of research participants ( e.g., raw
eld notes or detailed information from ethnographic interviews) (ASA Code of Ethics, 1997).
The American Economic Review, the publication of the
American Economic Association, requires the following:
It is the policy of the American Economic Review
to publish papers only if the data used in the analysis are clearly and precisely documented and are
readily available to any researcher for purposes of
replication. Authors of accepted papers that contain empirical work, simulations, or experimental
work must provide to the Review, prior to publication, the data, programs, and other details of
the computations sucient to permit replication.
These will be posted on the AER Web site. The
Editor should be notied at the time of submission
if the data used in a paper are proprietary or if, for
some other reason, the requirements above cannot
be met.
Moreover, according to the American Economic Review,
the burden is on the submitting author to explain why
data are proprietary and how her results may be replicated:
If a request for an exemption based on proprietary
data is made, authors should inform the editors if the
data can be accessed or obtained in some other way by
independent researchers for purposes of replication.
As a general rule of scientic publishing, the right to
monopolize data is waived on publication of the rst
article based on that data. Therefore, the objection that
producing data in litigation would forfeit the researchers
valuable right has no merit, at least where the data has
served as the basis for a scientic publication. 63
Confidentiality Issues
Most data in the social sciences are collected from individual respondents or subjects, and often those participants
are anonymous or assured condentiality. For example,
the Census forms completed by households remain condential for a designated span of years. EEO-1 data that
SPRING 2015
employers provide to the EEOC is condential, as is information reported on tax returns led with the IRS. Yet,
research based on condential data is published routinely,
and the data underlying these studies can be made available without compromising their condential nature. 64
Most often condentiality is maintained by redacting
identifying information or else by aggregating data so
that the identity of any one respondent is merged with
all other members of the group. For example, a telephone
interview with survey participants can yield responses
that can be shared with third parties without revealing
the identity of the person who gave a particular response.
Under these circumstances, the responses of the person
interviewed may be recorded or noted by the researcher
who speaks to the subject, and these can be purged as well
of any identifying information.
Courts routinely inspect in camera documents containing condential information and assess whether the
identity of the survey respondents could not be redacted.
65
Occasionally, redaction may not suce, because a respondent may be so singularly identiable that just the
slightest information may reveal a respondents identify.
For example, an employer in Bentonville, Arkansas, with
hundreds of thousands of employees worldwide, is easily
recognized as Wal-Mart, no matter what else is redacted.
Therefore, courts may order aggregations of data to be produced, such as the average income of every 10 respondents,
to protect the privacy of any one respondent. However,
this determination is best made case by case, and blanket
representations that promises of condentiality preclude
production of data should be deemed insucient. Moreover, any argument that de-identication is not possible
should be viewed in light of the courts power to impose
a protective order and require that documents referencing
the data be led under seal.
Timing
Even when data are obtained without objection, reanalyzing published research is time-consuming. Whats
more, the relevant literature will often consist of more
than a single study, making it impossible both to replicate
these studies and incorporate any criticisms into a report
that meets a courts scheduling order. Yet, it may not be
until the opposing expert report is in hand that one knows
the research literature on which that expert is relying.
The question arises, therefore, how to obtain data from a
secondary expert in time to analyze it in a pending case.
One possibility is to anticipate the opposing experts testimony and incorporate a critique into ones own experts
report. For example, when testimony regarding implicit
bias is reasonably certain, the defense expert may oer
59
SPRING 2015
Conclusion
Discovery from secondary experts looms so large in challenging experts on implicit bias because the standards for
the admissibility of scientic evidence in federal court
are more stringent than the requirements for publication
in academic journals. A testifying expert will survive a
Daubert challenge under Rule 702 only if their data and
methods pass rigorous scrutiny of the courts, with the aid
of experts retained by the parties. Academic journals, in
contrast, may publish studies based upon data that face
no scrutiny whatsoever.
Consequently, courts must be cautious of permitting
testifying experts to launder the research of others; that
is, allowing experts to base their testimony on research
so lacking in transparency that it would not be admitted if it had been collected for the case at hand. This
article has emphasized the importance of subjecting this
secondary data to the same scrutiny that is usually given
to case-specic data presented by testifying experts, challenging the foundational studies on which social science
testimony is based, and excluding primary evidence
that relies upon inaccessible secondary research. A rule
permitting testifying experts to rely only on data that
can be shared with the opposing party would ensure
these data are made available to maximum extent, and
subject the foundation of an experts opinion to the
appropriate scrutiny.
61
ENDNOTES
1
62
10
11
12
13
14
15
of Gender Discrimination: The Ascendance of Social Frameworks, 94 VA. L. REV. 1705 (2008).
To distinguish the social scientists who conducted
the research from retained consulting experts,
who are subject to the protections of Fed. R.
Civ. P. 26(4)(D), we may refer to the former as
secondary experts.
A separate objection may arise under Federal
Rule of Evidence 703 if an expert unreasonably
relies on social science data for her opinions. For
instance, if an expert makes a case-specic claim
about causation based on general social science
research that has no connection to the case at
hand, that would be an unreasonable use of the
research given that general social science research
cannot support specic causation claims in particular cases. See David L. Faigman, John Monahan
& Christopher Slobogin, Group to Individual (G2i)
Inference in Scientific Expert Testimony, U. CHI. L.
REV. (forthcoming).
External validity refers to whether results from
one setting with one sample will generalize to
another setting or sample. For a discussion of the
limited external validity of much psychological
research, see Gregory Mitchell, Revisiting Truth
or Triviality: The External Validity of Research in
the Psychological Laboratory, 7 PERSPECTIVES
ON PSYCHOL. SCI. 109 (2012).
See Hart Blanton, James Jaccard, Jonathan Klick,
Barbara A. Mellers, Gregory Mitchell & Philip E.
Tetlock, Strong Claims & Weak Evidence: Reassessing the Predictive Validity of the IAT, 94 J. APPLIED
PSYCHOL. 567 (2009); Gregory Mitchell, What Is
Wrong With Social Psychology?, 26 DIALOGUE 12
(Fall 2012).
Puffer v. Allstate Ins. Co., 255 F.R.D. 450, 468
(N.D. Ill. 2009), affd, 2012 U.S. App. LEXIS 6213
(7th Cir. Ill., Mar. 27, 2012).
Dr. R. F. Martell, Expert Rebuttal Report Responsive to Report of Wal-Mart Expert Nancy Combs.
EEOC v. Wal-Mart Stores, Inc., 6:01-cv-00339KKC, (E.D. Ky. Aug. 15, 2008) (citations omitted).
Id. at 26 n.30.
Id. at 26 & n.30.
The summaries of social science research offered by Dr. Martell and other experts often do
not accurately portray the body of research. For
instance, meta-analytic studies show that subjective personnel criteria are not a good predictor of
bias. We do not agree with or approve of any of
the opinions we offer as examples.
Expert Report of William T. Bielby, Ph.D., Betty
Dukes et al. v. Wal-Mart Stores, Inc., February 3,
2003, available at http://www2.law.columbia.
edu/fagan/courses/law_socialscience/documents/Spring_2006/Class%2017-Gender%20
Discrimination/BD_v_Walmart.pdf .
The Daubert factors are (1) whether the theory or
technique has been tested, (2) whether the theory
or technique has been subjected to peer review
and publication, (3) whether the technique has
a high known or potential rate of error, and (4)
whether the theory or technique is generally accepted within a relevant scientic community.
16
17
18
19
20
21
22
23
SPRING 2015
24
25
26
27
SPRING 2015
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
See, e.g., Joshua Carp, The Secret Lives of Experiments: Methods Reporting in the fMRI Literature,
63 NEUROIMAGE 289 (2012); Deborah Kashy, M.
Brent Donnellan, Robert A. Ackerman & Daniel
W. Russell, Reporting and Interpreting Research
in PSPB: Practices, Principles, and Pragmatics, 35
PERSONALITY SOCIAL PSYCHOL. BULL. 1131
(2009); Jelte M. Wicherts, Denny Borsboom, Judith Kats & Dylan Molenaar, The Poor Availability
of Psychological Research Data for Reanalysis, 61
AM. PSYCHOL. 726 (2006).
See Marjan Bakker, Annette van Dijk & Jelte M.
Wicherts, The Rules of the Game Called Psychological Science, 7 PERSPECTIVES PSYCHOL. SCI.
543 (2012); Daniele Fanelli, Negative Results Are
Disappearing From Most Disciplines and Countries, 90 SCIENTOMETRICS 891 (2012); Daniele
Fannelli, Positive Results Increase Down the
Hierarchy of the Sciences, 5 PLOS ONE e10068
(2010); John P. Ioannidis, Why Most Published
Research Findings Are False, 2 PLOS MED. E124
(2005); Jeffrey D. Scargle, Publication Bias: The
File-Drawer Problem in Scientific Inference, 14
J. SCI. EXPLORATION 91 (2000).
h t t p : / / r e t r a c t i o n w a t c h .w o r d p r e s s .
com/2013/07/08/time-for-a-scientific-journalreproducibility-index/.
http://www.nature.com/news/replicationstudies-bad-copy-1.10634#/b2 .
http://www.nature.com/nature/journal/v483/
n7391/full/483531a.html .
https://www.scienceexchange.com/reproducibility ; see also Carl Zimmer, A Sharp Rise In
Retractions Prompts Calls For Reform, The
New York Times, April 16, 2012, http://www.
nytimes.com/2012/04/17/science/rise-in-scientific-journal-retractions-prompts-calls-for-reform.
html?_r=0 .
See, e.g., Jens B. Asendorpf et al., Recommendations for Increasing Replicability in Psychology, 27
EUROPEAN J. PERSONALITY 108 (2013); Etienne
P. LeBel et al., PsychDisclosure.org : Grassroots
Support for Reforming Reporting Standards in
Psychology, 8 PERSPECTIVES PSYCHOL. SCI. 424
(2013); Open Science Collaboration, An Open,
Large-Scale, Collaborative Effort to Estimate the
Reproducibility of Psychological Science, 7 PERSPECTIVES PSYCHOL. SCI. 657 (2012).
See Hart Blanton et al., Strong Claims and Weak
Evidence: Reassessing the Predictive Validity of the
IAT, 94 J. APP. PSYCHOL. 567 (2009).
RAND Retracts Report about Medical Marijuana
Dispensaries and Crime, available at http://www.
rand.org/news/press/2011/10/24.html (last visited November 5, 2103).
See, e.g., Hart Blanton & Gregory Mitchell,
Reassessing the Predictive Validity of the IAT:
II. Reanalysis of Heider & Skowronski (2007), 13
N. AM. J. PSYCHOL. 99 (2011); Uri Simonsohn,
Just Post It: The Lesson From Two Cases of
Fabricated Data Detected by Statistics Alone,
24 Psychol. Sci. 1875 (2013); Uri Simonsohn,
Just Posting It works, leads to new retraction
in Psychology, available at http://datacolada.
org/2013/09/17/just_posting_it_works/ (last
visited November 5, 2013).
63
45
46
47
48
49
50
51
52
53
54
55
56
57
58
64
59
Id. at 717.
Id.
61
Id.
62
See Letter Brief to the Honorable Mari-Elena
James, July 24, 2007, Holloway v. Best Buy, Co.,
Inc., No. 3:05-cv-05056-PJH (N.D. Cal. 2005).
63
If journals and their editors enforced requirements
that data be publicly archived or made available
to other researchers, then the need for data
subpoenas would be lessened. Unfortunately,
many social science journals do not require public
archiving as a condition of publication or police
data-sharing requirements, and based on the
experiences of the second author and others who
have made data requests, psychologists share
their data at disappointingly low rates. See Hart
Blanton, James Jaccard, Jonathan Klick, Barbara
A. Mellers, Gregory Mitchell & Philip E. Tetlock,
Strong Claims & Weak Evidence: Reassessing the
Predictive Validity of the IAT, 94 J. APP. PSYCHOL.
567 (2009); Gregory Mitchell, What Is Wrong With
Social Psychology?, 26 DIALOGUE 12 (Fall 2012).
The National Science Foundation has recently
strengthened its data sharing requirement as a
condition of the receipt of grant funds, a move
that may improve the data sharing practices of
social scientists.
Recently, The Economist observed:
A study published last month in PeerJ by Melissa Haendel, of the Oregon Health and Science
University, and colleagues found that more than
half of 238 biomedical papers published in 84
journals failed to identify all the resources (such
as chemical reagents) necessary to reproduce the
results. On data, Christine Laine, the editor of the
Annals of Internal Medicine, told the peer-review
congress in Chicago that ve years ago about
60% of researchers said they would share their
raw data if asked; now just 45% do. Journals
growing insistence that at least some raw data be
made available seems to count for little: a recent
review by Dr. Ioannidis which showed that only
143 of 351 randomly selected papers published
in the worlds 50 leading journals and covered by
some data-sharing policy actually complied.
Trouble at the lab: Scientists like to think of science
as self-correcting. To an alarming degree, it is not,
The Economist (October 13, 2013), supra.
64
Indeed, on occasion researchers have taken it
upon themselves to testify based upon condential data without obtaining permission from
the government agency providing that research.
For example, in the high-prole case of Dukes v.
Wal-Mart, Dr. Marc Bendick based his testimony
60
65
66
67
68
69
70
71
72
73
on condential EEO-1 reports which he was provided by the EEOC with the understanding that it
would be used for research and not for litigation.
Dukes v. Wal-Mart Stores, Inc., 222 F.R.D. 189, 193
(N.D. Cal. 2004 ), affd, 2010 U.S. App. LEXIS 8576
(9th Cir. Apr. 26, 2010) (en banc), revd, 131 S.Ct.
2554 (2011). The district court rejected a motion
to strike Dr. Bendicks declaration based on this
data and did not require production of the full
data set on which Dr. Bendick relied, in part on the
grounds that the full data set was not requested
in a timely manner. Dukes, 222 F.R.D. at 194.
Craig v. Rite Aid Corp., 2012 U.S. Dist. LEXIS
46274, *36 (M.D. Pa. March 30, 2012) (redacting
portion of document after in camera inspection).
For instance, However, biased treatment is not
always conscious. The EEO laws prohibit not
only decisions driven by animosity, but also
decisions infected by stereotyped thinking. Because viewing a video may trigger unconscious
bias, especially if opportunities for face-to
face conversation are absent, covered entities
should implement proactive measures, or best
practices, to minimize this risk. Letter from Assistant General Counsel of EEOC to member of
the public, dated September 21, 2010, available
at http://www.eeoc.gov/eeoc/foia/letters/2010/
ada_gina_titlevii_video_resumes.html .
957 F. Supp. 873, 877-880 (S.D. Tex. 1997).
Id.
611 F. Supp. 1223, 1245 (E.D.N.Y. 1985).
Id. at 1238.
Heckman v. Ryder Truck Rental, Inc., 2013 U.S.
Dist. LEXIS 21035, *2 (D. Md. Feb. 11, 2013) citing Mirchandani v. Home Depot, U.S.A., Inc., 235
F.R.D. 611, 614 (2006).
Mirchandani v. Home Depot, U.S.A., Inc., 235
F.R.D. at 615.
The destructive testing cases provide a helpful
analogy when data, like the allegedly defective
product, potentially are available. However,
there are commonly-cited studies in every eld
for which data may no longer be available simply
due to the age of the studies. Oftentimes these
studies will have spawned research that builds on
their initial insights or that conceptually replicates
the prior results. Consequently, an expert whose
reliance on past studies for which data are not
reasonably available should be able to point to
other studies that conrm or elaborate on the
earlier ndings, and for which data are available.
If the earlier studies have not been replicated
directly or indirectly, then parties and courts
should inquire into why that is the case.
SPRING 2015
Copyright of Labor Law Journal is the property of CCH Incorporated and its content may not
be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for
individual use.