Linda Pippen, on Behalf of Themselves and All Others Similarly Situated v. State

WATERMAN, Justice

(concurring specially).

I respectfully concur in the result only. I am unable to join the majority opinion’s affirmance of the district court’s thorough, well-reasoned decision on the merits in favor of defendants. See Pippen v. State, No. 05771 LACL107038, 2012 WL 1388902 (Iowa Dist. Ct. April 17, 2012), available at http://www.iowaappeals.com/wp-content/ uploads/2012/05/Pippen-Ruling.pdf. The majority’s discussion of the law includes dicta unnecessary to its holding. For example, the majority gratuitously undermines our court’s long-standing practice of relying on federal decisions to interpret equivalent provisions of the Iowa Civil Rights Act. Yet, the majority’s analysis of the dispositive issue is cursory. Given the importance of this case, I offer a more complete analysis of the grounds to affirm the judgment and respond to the dicta to help guide cases to follow.

Plaintiffs are a certified class of over 5000 members defined as “[a]ll African American applicants and employees who sought appointment to or held a merit-system position with an Executive Branch agency (not including Board of Regents) at any point from July 1, 2003, through April 17, 2012.” They brought disparate impact racial discrimination claims under Title VII of the Civil Rights Act of 1964 and the Iowa Civil Rights Act of 1965 (ICRA). These civil rights laws were enacted to eliminate discriminatory practices, provide remedies for discrimination, and allow equal opportunities in employment. Defendants are the State of Iowa and its thirty-seven executive branch departments. There are over 700 types of employment positions within the executive branch. During the relevant period, nearly 500,000 applications were submitted by 100,000 applicants for 20,000 openings. Approximately 2000 supervisors within the executive branch have authority in the hiring process. Plaintiffs sought to prove at trial that the State’s overall hiring system during the relevant time period discriminated on the basis of race.

Disparate impact law generally requires plaintiffs to prove a particular employment practice caused a disparate impact. The majority fails to discuss the purpose of this proof requirement, which is to enable the district court to fashion a remedy correct *33 ing the discriminatory practice without affecting other practices that are not discriminatory. Congress enacted a narrow exception to this general requirement. Specifically, plaintiffs do not have to prove a particular employment practice had a discriminatory impact if they prove the State’s hiring process was incapable of “separation for analysis.” See 42 U.S.C. § 2000e-2(k)(l)(B)(i) (2006). In that event, the law relieves them of the burden of attributing the discrimination to a particular employment practice. Here, the plaintiffs sought to proceed under that exception.

Following a month-long bench trial, the district court found the State’s employment decision-making process was capable of separation for analysis. For the reasons explained below, I conclude the “capable of separation” finding is supported by substantial evidence and is dispositive. I therefore agree the district court judgment must be affirmed.

I. Additional Background Facts and Proceedings.

As required under our standard of review in our appellate role, we review the trial evidence in the light most favorable to the judgment. Falczynski v. Amoco Oil Co., 533 N.W.2d 226, 230 (Iowa 1995). I will elaborate on the expert testimony discussed by the majority opinion.

As the majority notes, plaintiffs presented testimony from three expert witnesses. Anthony Greenwald and Cheryl Kaiser testified regarding the social science of implicit-bias stereotyping and how implicit biases affect decision-making. Kaiser explained the difference between explicit and implicit bias: explicit bias is “conscious, deliberate, controlled animosity,” whereas implicit bias is due to unconscious negative associations — stereotypes—that people have developed over time. 9 Implicit-bias theory helps explain how statistical disparities can result without intentional discrimination: individuals act on implicit biases without recognizing they are doing so. Greenwald testified unconscious bias leads to discrimination particularly in subjective decision-making and that most test groups demonstrate a seventy percent unconscious preference for whites over blacks. 10 And, as is customary in a disparate impact *34 case, the plaintiffs presented testimony from a statistical expert, labor economist Mark Killingsworth. 11

Killingsworth chose to limit his analysis to applications referred by DAS to the departments (thus effectively excluding applicants who did not satisfy minimum qualifications from his review). 12 Killings-worth’s system-wide review for the years 2004 through 2008 showed that African-American applicants were statistically less likely than white applicants to be interviewed and hired in every year. For example, in 2005, nine percent of African-American applicants were interviewed, as compared to twenty percent of white applicants. That same year, 1.8% of African-American applicants were hired, as compared to 4.2% of white applicants. Overall, his analysis showed that a minimally qualified white person had a forty percent greater chance of being hired than a minimally qualified African American. Splitting his analysis by department, Killings-worth testified there was a statistically significant disparity between the percentage of African-American applicants hired and the percentage of white applicants hired in many of the departments, though not all. Killingsworth further testified that, once hired, white and African-American employees were treated differently in performance evaluations. Killingsworth did not connect his findings to any particular State practice. 13

Robert Miller, another labor economist and the State’s statistical expert, also performed regression analyses. With regard to wages and promotion, he found no statistically significant differences between races after taking into account experience and the pay grade of the job for which the individual applied. He criticized Killings-worth’s regressions as inadequate because they failed to adjust for these other factors.

With regard to hiring, Miller subdivided his work into the three steps that the State followed to get to its actual hire. Thus, unlike Killingsworth, who considered only steps two and three and combined them, Miller separately considered steps one, two, and three. At step three, i.e., what occurs after the applicant is granted an in-person interview, Miller found no statistically significant difference between whites and blacks in their success in being hired.

At step one, Miller found no statistically significant difference between white and black applicants in getting a referral. On a per application basis, though, African *35 Americans were less successful than whites in getting a referral. However, the data also showed that African-American applicants, on average, filed more applications than whites (approximately 5.9 versus 4.3 per applicant over a multiyear period — a thirty-eight percent difference).

At step two — referral to interview — Miller’s findings were more nuanced. On the whole, he found that African Americans were statistically less likely to receive a department job interview after their application was referred by DAS. Probing more deeply, he found this was only true for about one-third of State departments. It was not true for the remaining two-thirds of departments. As Miller explained,

[t]he overall conclusion with respect to step two is that a common factor could not or does not appear to be operating across all the departments in the same way, because our results show that there are clear departmental differences.

For the specific jobs that the thirteen named plaintiffs had applied for, Miller found that African Americans were not statistically less likely to get interviews than whites.

The plaintiffs reviewed paper hiring files produced by the State for 667 separate positions that were filled. Based on the examples in the record, these files contained items such as applicant résumés, score sheets for résumé reviews, score sheets for interviews, and letters asking for authority (and justifying the decision) to hire the successful applicant. For instance, for the position of workforce advis- or in the unemployment insurance service center, Iowa Workforce Development used a résumé review worksheet that awarded a maximum of fifty-nine points. There were a maximum of twenty points potentially available for education, ten points potentially available for unemployment insurance claims experience, fourteen points potentially available for possessing various computer skills, five points available for being a veteran (or ten for being a disabled veteran), and five points available for “ability to follow resume and cover letter submission directions.” These files were not provided to or reviewed by Killings-worth. As Killingsworth put it, “I don’t have any access or haven’t had any hiring files.”

Miller testified that the data were “absolutely” capable of separation for analysis, and in fact, he separated them to the extent noted.

II. Analysis.

On appeal, plaintiffs do not argue that the State’s failure to follow its own equal-employment-opportunity policies constituted a discrete employment practice. Instead, plaintiffs challenge the district court’s determination that the State’s overall hiring process was capable of separation for analysis. Plaintiffs argue the hiring process could not be analyzed in terms of separate practices. Plaintiffs also assert on appeal that the district court erred in determining they failed to prove causation. Because the district court correctly decided the dispositive separation issue, we need not reach the causation issue. I will provide an overview of the governing law to place the separation issue into context.

A. Disparate Impact Law. Title VII of the 1964 Civil Rights Act’s “central statutory purposes [are] eradicating discrimination throughout the economy and making persons whole for injuries suffered through past discrimination.” Albemarle Paper Co. v. Moody, 422 U.S. 405, 421 , 95 S.Ct. 2362, 2373 , 45 L.Ed.2d 280, 299 (1975). To that end, Title VII seeks “to eliminate those discriminatory practices and devices which have fostered racially stratified job environments to. the disad *36 vantage of minority citizens.” McDonnell Douglas Corp. v. Green, 411 U.S. 792, 800 , 93 S.Ct. 1817, 1823 , 36 L.Ed.2d 668, 676 (1973). “The ICRA was modeled after Title VII” and enacted to serve the same purposes: it “was passed in 1965 in an effort to establish parity in the workplace and market opportunity for all.” Vivian v. Madison, 601 N.W.2d 872, 873 (Iowa 1999).

Title VII and the ICRA each provide two principal ways to prove employment discrimination: disparate impact and disparate treatment. Hy-Vee Food Stores, Inc. v. Iowa Civil Rights Comm’n, 453 N.W.2d 512, 516 (Iowa 1990). The fundamental difference between the two theories is that “the disparate treatment theory focuses on the employer’s motivation; the disparate impact theory focuses on the consequences of the employer’s conduct.” Id. Disparate treatment requires a plaintiff to prove intentional discrimination. Id.

Disparate impact, the theory plaintiffs presented at trial, presents an avenue for addressing inequalities in the absence of intentional discrimination. In a disparate impact case, what matters is not the subjective motivation of the employer, but the effects of an employment practice. Griggs v. Duke Power Co., 401 U.S. 424, 432 , 91 S.Ct. 849, 854 , 28 L.Ed.2d 158, 165 (1971). Disparate impact prohibits employer practices “that are facially neutral in then-treatment of different groups but that in fact fall more harshly on one group than another.” Int’l Bhd. of Teamsters v. United States, 431 U.S. 324 , 335 n. 15, 97 S.Ct. 1843 , 1854 n. 15, 52 L.Ed.2d 396 , 415 n. 15 (1977). Meant to remove barriers to employment or promotion that are unrelated to job performance, the focus of disparate impact theory is on fairness in operation— not fairness in form. Griggs, 401 U.S. at 431 , 91 S.Ct. at 853 , 28 L.Ed.2d at 164 (commenting that “equality of opportunity merely in the sense of the fabled offer of milk to the stork and the fox” is insufficient under Title VII).

A three-step burden-shifting framework applies to disparate impact claims. In the first stage, to establish a prima facie case, a plaintiff must show that the employer “ ‘uses a particular employment practice that causes a disparate impact’ on one of the prohibited bases.” Lewis v. City of Chicago, 560 U.S. 205, 212 , 130 S.Ct. 2191, 2197 , 176 L.Ed.2d 967, 974 (2010) (emphasis omitted) (quoting 42 U.S.C. § 2000e-2(k)(1)(A)(i)). “Identifying a specific practice is not a trivial burden....” Meacham v. Knolls Atomic Power Lab., 554 U.S. 84, 101 , 128 S.Ct. 2395, 2406 , 171 L.Ed.2d 283, 296 (2008). Second, if the plaintiff makes this prima facie showing, the burden shifts to the employer to prove that the challenged employment practice reflects a business necessity. See 42 U.S.C. § 2000e-2(k)(1)(A)(i). Third, the plaintiff may rebut the employer’s business-necessity evidence by demonstrating there are “other reasonable alternatives that would have less adverse impact,” Hy-Vee, 453 N.W.2d at 518 , and the defendant “refuses to adopt such alternative employment practice,” 42 U.S.C. § 2000e-2(k)(1)(A)(ii). Only the first stage is at issue in this appeal because the district court held that plaintiffs failed to prove their prima facie case.

The identification of a particular employment practice in the first stage helps the court ascertain and remedy the cause of racial disparities. Proof focused on a particular employment practice enables the relevant comparison between “qualified persons in the labor market and the persons holding at-issue jobs.” Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 650 , 109 S.Ct. 2115, 2121 , 104 L.Ed.2d 733, 747 (1989), superseded by statute on other grounds, 42 U.S.C. § 2000e-2(k), as recog *37 nized in Wal-Mart Stores, Inc. v. Dukes, 564 U.S.-, 181 S.Ct. 2541 , 180 L.Ed.2d 874 (2011).

The United States Supreme Court cautioned that using overbroad statistics to prove a disparate impact claim “would result in employers being potentially liable for ‘the myriad of innocent causes that may lead to statistical imbalances in the composition of their work forces.’ ” Id. at 657, 109 S.Ct. at 2125 , 104 L.Ed.2d at 751 -52 (quoting Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 992 , 108 S.Ct. 2777, 2787 , 101 L.Ed.2d 827, 843 (1988)). “Title VII guarantees [individuals] the opportunity to compete equally with white workers on the basis of job-related criteria,” but does not guarantee equal outcomes. Connecticut v. Teal, 457 U.S. 440, 451 , 102 S.Ct. 2525, 2532-33 , 73 L.Ed.2d 130, 139 (1982). As the Wards Cove Court explained, if plaintiffs are allowed to use aggregated statistics alone to prove disparate impact, it is difficult for the Court to determine if the racial composition of hires is at odds with the relevant qualified labor market. 490 U.S. at 650-52 , 109 S.Ct. at 2121-22 , 104 L.Ed.2d at 747-48 .

Wards Cove highlighted that racial disparities revealed in aggregated statistics at times could be justified by a closer examination of the qualified labor pool. Similarly, disparities shown by statistics aggregated at the departmental level may be explained by the specific employment practices of a given department. See Wal-Mart, 564 U.S. at-, 131 S.Ct. at 2555 , 180 L.Ed.2d at 394 (“[I]nformation about disparities at the regional and national level does not establish the existence of disparities at individual stores, let alone raise the inference that a company-wide policy of discrimination is implemented by discretionary decisions at the store and district level.” (Internal quotation marks omitted.)).

The particularity requirement allows the court to fashion an appropriate remedy: if a particular employer practice is identified as causing discriminatory impact, the court can order the employer to correct it. Title VII “arm[s] the courts with full equitable powers” in order to address violations. Albemarle, 422 U.S. at 418 , 95 S.Ct. at 2372 , 45 L.Ed.2d at 297 . A court must be able to determine the cause of discrimination to effectively exercise these equitable powers. As the United States Supreme Court has long recognized, the usefulness of statistics “depends on all of the surrounding facts and circumstances.” Int’l Bhd. of Teamsters, 431 U.S. at 340 , 97 S.Ct. at 1856-57 , 52 L.Ed.2d at 418 . It is vital to determine the particular employment practice causing the disparate impact in order to fix the problem.

For example, the record shows that African Americans actually comprise a higher percentage of the State executive branch workforce than they do in the Iowa workforce as a whole. Yet, it would be wrong to conclude from that overall number that the State is not discriminating on the basis of race. One has to focus on actual employment practices.

While Congress generally required that a plaintiff identify particular employment practices that cause disparate impact, Congress also provided that the decision-making process could be challenged as a whole under certain circumstances. Specifically, Congress provided:

With respect to demonstrating that a particular employment practice causes a disparate impact ... the complaining party shall demonstrate that each particular challenged employment practice causes a disparate impact, except that if the complaining party can demonstrate to the court that the elements of a respondent’s decisionmaking process are not capable of separation for analysis, *38 the decisionmaking process may be analyzed as one employment practice.

42 U.S.C. § 2000e-2(k)(l)(B)(i). This law codifies an exception to the requirement that a plaintiff identify a particular employment practice, thereby ensuring that employers cannot avoid liability for disparate impact simply by making it difficult for a plaintiff to separately analyze the decision-making elements. See Griffin v. Carlin, 755 F.2d 1516, 1525 (11th Cir. 1985). This statutory exception balances the concern that employers could evade liability against the need for particularity by requiring the plaintiff to first demonstrate the employer’s process is incapable of separation. As the majority acknowledges, the burden was on plaintiffs to prove the State’s decision-making process is incapable of separation for analysis before proceeding to attack the process as a whole. See Grant v. Metro. Gov’t of Nashville, 446 Fed.Appx. 787, 740 (6th Cir. 2011).

The ICRA does not have a counterpart to the 1991 congressional amendment. 14 Plaintiffs do not argue here — and did not argue below — that their burden to establish a prima facie case under the ICRA was lower than their burden under Title VII. I will therefore focus on federal law and on the question of whether the State’s employment practices were “not capable of separation for purposes of analysis.” But, first, I must respond to dicta in the majority opinion that misleads by omission and thereby unfairly disparages, sub silentio, our long-standing practice, followed in numerous decisions of this court, of relying on federal decisions to interpret equivalent provisions in the ICRA. The majority, in a discussion unrelated to its holding, distances itself from federal decisions.

The majority, relying on Hubbard v. State, 163 N.W.2d 904, 909 (Iowa 1969), suggests federal interpretations of Title VII are only useful if those interpretations preceded the enactment of the Iowa statute. I disagree. Hubbard was decided a half century ago and interpreted the Iowa Tort Claims Act (ITCA), not the ICRA. See id. at 905 . Since Hubbard , our court has repeatedly relied on subsequent federal interpretations of the Federal Tort Claims Act to construe the ITCA. See Walker v. State, 801 N.W.2d 548, 569 (Iowa 2011) (Mansfield, J., dissenting) (collecting Iowa opinions that rely on federal cases decided after the ITCA’s enactment). Indeed, our court has cited Hubbard as support for the proposition that “[interpretations of the federal act are instructive” — without limiting that observation to cases decided before the enactment of the ITCA. Annear v. State, 419 N.W.2d 377, 379 (Iowa 1988); see also Meier v. Sulhoff 360 N.W.2d 722, 728 (Iowa 1985) (McCormick, J., dissenting) (citing Hubbard and stating “[bjecause [the Iowa Occupational Safety and Health Act] is based on the federal model, the federal court interpretations constitute persuasive authority for giving a similar interpretation to our statute”); Adam v. Mount Pleasant Bank & Trust Co., 340 N.W.2d 251, 252 (Iowa 1983) (“Because our statute is based on the fed *39 eral Tort Claims Act, we assume our legislature intended it to have the same meaning as the federal statute. Hubbard, 163 N.W.2d at 911 . Federal decisions interpreting the federal act are therefore entitled to great weight. Id. at 909 .”). In any event, our court, before today, has never relied on Hubbard to interpret the ICRA.

The Iowa bench and bar has long understood federal authorities provide guidance to interpret the ICRA. This lends predictability to an important area of the law, particularly when the legislature has long acquiesced in our interpretations of the ICRA based on federal interpretations of the counterpart federal statutory language. See Ackelson v. Manley Toy Direct, L.L.C., 832 N.W.2d 678, 688-89 (Iowa 2013) (discussing legislative acquiescence). Remarkably, the majority fails to acknowledge that our court decided many cases by relying on federal interpretations of equivalent statutory language in the civil rights acts. See, e.g., Casey’s Gen. Stores, Inc. v. Blackford, 661 N.W.2d 515, 519 (Iowa 2003) (“[W]e have looked to the corresponding federal statutes to help establish the framework to analyze claims and otherwise apply [ICRA].”); Schlitzer v. Univ. of Iowa Hosps. & Clinics, 641 N.W.2d 525, 529 (Iowa 2002) (“The common goals of the Federal ADA and our civil rights act have encouraged us to look to the federal statutory and regulatory standards in applying our statute.”); Vivian, 601 N.W.2d at 873 (“The ICRA was modeled after Title VII of the United States Civil Rights Act. Iowa courts therefore traditionally turn to federal law for guidance in evaluating the ICRA.”); Vincent v. Four M Paper Corp., 589 N.W.2d 55, 59-60 (1999) (“[W]e have recognized the common purposes of the [F]ederal [ADA] ... and the ICRA as well as the similarity in terminology of the statutes. Moreover, we have looked to the ADA and federal regulations implementing that [A]ct in developing standards under the ICRA for disability discrimination claims.” (Citations omitted.)); Hulme v. Barrett, 449 N.W.2d 629, 631 (Iowa 1989) (“Our court has ruled that civil rights cases brought under chapter [216] will be guided by federal law and federal cases.” (Internal quotation marks omitted.)); Probasco v. Iowa Civil Rights Comm’n, 420 N.W.2d 432, 435 (Iowa 1988) (“On several occasions, our courts have looked to the federal system for guidance in construing our similar civil rights legislation. We employ this approach again today because, as demonstrated below, the civil rights legislation and implementing rules involved in this case mirror those adopted on the federal level.” (Citations omitted.)); Iowa State Fairgrounds Sec. v. Iowa Civil Rights Comm’n, 322 N.W.2d 293, 296 (Iowa 1982) (“The parties assume we will find federal cases persuasive in selecting the analytical framework for deciding discrimination cases under the Iowa civil rights statute. This assumption is warranted by our prior decisions.”); Wilson-Sinclair Co. v. Griggs, 211 N.W.2d 133, 139 (Iowa 1973) (noting “the similarity of language” of Title VII and the ICRA and relying on “numerous relevant federal decisions”); Iron Workers Local No. 67 v. Hart, 191 N.W.2d 758, 765 (Iowa 1971) (recognizing the ICRA contains “[a]nalo-gous language” to Title VII and “is another manifestation of a massive national drive to right wrongs prevailing in our social and economic structures for more than a century”). 15 None of these cases interpreting the ICRA limited consider *40 ation of federal authorities to those decided before enactment of the Iowa statute.

In the majority’s view, if it does not like how federal decisions were decided, it can freely disregard them. The cost of this new approach is the stability and predictability of our law. See State v. Short, 851 N.W.2d 474, 515 (Iowa 2014) (Waterman, J., dissenting). After today, it is at best unclear what weight litigants and district court judges or the court of appeals should give federal cases when divining how our court will construe equivalent provisions in the ICRA. This is unfortunate. A more restrained majority would have deferred its pronouncements until a case in which they made a difference to the outcome.

B. Does Substantial Evidence Support the District Court’s Finding that the State’s Decision-Making Process Was Capable of Separation for Analysis? The district court specifically found the State’s decision-making process was capable of separation for analysis. I agree with the majority that this is a question of fact. See, e.g., McClain v. Lufkin Indus., Inc., 519 F.3d 264, 278 (5th Cir.2008). We are bound by the district court’s factual finding if it is supported by substantial evidence. Schlitzer, 641 N.W.2d at 529 . Plaintiffs argue on appeal there was no substantial evidence supporting this finding and that their evidence proved as a matter of law that the State’s decision-making process was incapable of separation for analysis. I disagree.

Plaintiffs assert that the State’s record-keeping practices precluded separate analysis of the decision-making process because (1) subjective assessments pervaded the decision-making process; (2) subjective assessments have a “ripple effect,” whereby the discriminatory impact accumulates as applicants move through the hiring process; and (3) the State did not retain some records regarding applicant evaluations.

The district court rejected all three theories based on the factual record developed at trial. To put the district court’s dispositive factual finding into its legal context, I construe the operative statutory language. The phrase “each particular challenged employment practice” in 42 U.S.C. § 2000e-2 (k)(l)(B)(i) indicates that a particular employment practice is distinct from a decision-making process. A particular employment practice is considered an element within the larger decision-making process. Congress’s choice to use the singular form, combined with the words “particular” and “each” demonstrates that the challenged practice must be individually identified. “This syntax would be strange if a plaintiff could bundle a number of discrete steps of a multi-phase hiring process together, based on a common characteristic.” Davis v. Cintas Corp., 717 F.3d 476, 496, 497 (6th Cir.2013) (holding plaintiff “did not identify a ‘particular employment practice’ within the meaning of Title VII by pointing to all of the subjective elements in the [employer’s hiring system]”).

I first consider plaintiffs’ contentions regarding subjective decision-making. How subjective decision-making plays into hiring and promotion depends on the type of job and the process used to fill it. Thus, subjective conduct may serve as the “particular employment practice” underlying a disparate impact claim if plaintiffs can prove that the conduct operates uniformly throughout an employer’s decision-making process to cause a disparate impact. For example, in Davis , the Sixth Circuit rejected plaintiffs argument that the defendant’s decision-making process was incapable of separation when “not all of the system’s subjective elements are the same.” Id. at 497 . The court noted “[e]ach different interview ... has a specific interview guide, and different manag *41 ers conduct interviews at different stages of the process.” Id. Likewise, in Bennett v. Nucor Corp., the Eighth Circuit concluded “this is not a case where the components of the employer’s selection process were incapable of separation.” 656 F.Bd 802, 817 (8th Cir.2011). The court pointed to the fact that the employer’s “five departments used a variety of measures to evaluate candidates for promotion, including objective criteria like experience, training, disciplinary history, and test scores, and subjective criteria such as interview performance and the opinion of the candidate’s current supervisor.” Id. at 817-18; see also Grant, 446 Fed.Appx. at 740 (“The problem, however, is that Plaintiffs make no effort to isolate any of these [decision-making] practices or to examine their individual effects on the promotions process.”).

Chin v. Port Authority is a good example of a subjective process that was incapable of separation for purposes of analysis. 685 F.3d 135, 154-55 (2d Cir.2012). In that case, the plaintiffs had passed exams and had been placed on eligibility lists for promotions to sergeant but had not received promotions. Id. at 142-43 . The plaintiffs proved that the decisions as to whom would be promoted from the eligible list were based on essentially subjective recommendations by commanding officers and subjective final decisions by the superintendent. Id. at 154-55 . The process was entirely discretionary, and the final decision rested with one person — the superintendent. See id.

Wal-Mart, filed shortly before this case went to trial, is instructive. There, the Supreme Court decertified a nationwide class of 1.5 million current and former employees of Wal-Mart alleging gender discrimination. Wal-Mart, 564 U.S. at -, -, 131 S.Ct. at 2547, 2561 , 180 L.Ed.2d at 385, 400 . Pay and promotion decisions at Wal-Mart were generally committed to local managers’ broad, subjective discretion. Id. at-, 131 S.Ct. at 2547 , 180 L.Ed.2d at 385 . However, plaintiffs alleged that a strong and uniform corporate culture led that discretion to be exercised, even subconsciously, against the hiring and advancement of women. Id. at -, 131 S.Ct. at 2548 , 180 L.Ed.2d at 386 . Yet, the Court held “[respondents [did] not identif[y] a common mode of exercising discretion that pervades the entire company.” Id. at-, 131 S.Ct. at 2554-55 , 180 L.Ed.2d at 393 . The district court observed Wal-Mart “highlights both the need to identify a particular employment practice, the pertinence of discretionary decisionmaking in the employment process, and the interconnection with statistical proof.”

The district court here found that “[t]he State’s system has both objective and subjective components” which are “not so confused ... as to prevent Plaintiffs from honing in on one particular employment practice.” This finding is supported by the record. For example, the DAS screen that occurred at step one, the résumé score sheets that were part of step two in some departments, the second résumé screens and spelling and grammar screens that were part of step two in some departments, and the interview score sheets that were part of step three in some departments were objective components that could have been separately analyzed. This was not a purely subjective process.

Furthermore, while there undoubtedly was subjectivity and — as the plaintiffs credibly demonstrated — implicit bias in multiple State hiring decisions during the relevant time period, this case is a far cry from Chin, in which the decision-making process ultimately came down to a single individual’s discretion. By contrast here, the State’s hiring decisions were in the hands of numerous department managers. *42 The State of Iowa Executive Branch employs a far more diverse range of job categories than any of the defendants in the cases in which subjective decision-making has served as a basis for attacking the decision-making process as a whole. Those working for the State include corrections officers, parole officers, registered nurses, food service workers, power plant engineers, state troopers, and DOT road-maintenance crew members, to name a few. As noted by the district court, “The State system can be dissected into numerous decision-making stages among numerous independent agencies of the executive branch,” and there is “inconsistency in results among the numerous agencies.” Miller showed that, at step two, African Americans had a lower statistical likelihood of getting an interview in some departments but not in others. These outcomes could be due to either objective or subjective considerations, but either way they do not suggest the existence of a common practice (even a subjective one) that would justify class-wide relief. See Wal-Mart, 564 U.S. at-, 181 S.Ct. at 2555-56, 180 L.Ed.2d at 393 (noting that discretion when exercised by different managers in different ways is not in itself an employment practice).

No witness affirmatively testified the process was incapable of separation for analysis. To the contrary, the State’s expert, Miller, “emphatically” testified that the State’s hiring system was capable of separation for analysis. Plaintiffs on appeal do not challenge the admissibility of Miller’s testimony. His testimony alone constitutes substantial evidence supporting the district court’s finding. Furthermore, Miller actually did separate the process into the three steps for purposes of his analysis.

Additionally, plaintiffs’ own experts testified that the State’s decision-making process could be separated for analysis. The district court noted that Greenwald conceded the State’s employment process could be separated to individually analyze each step of the process and commented “that one ‘could determine whether or not there was bias at each one of the independent stages of the hiring process.’ ” Kaiser discussed how written résumés and in-person interviews can trigger implicit racial biases differently. The State’s statistical expert, Killingsworth, utilized a regression analysis to evaluate the State’s hiring process using data from different stages of the process and different departments. The district court summarized:

Dr. Killingsworth was capable of separating data for the referral stage, the interview stage, and the hiring stage for African Americans as compared to whites over a period of years. His work permits a fact finder to analyze the departments of the executive branch in each of those years at each of those stages. This charting of State data allows a fact finder to compare the various departments and draw important conclusions as to how the individual departments compare to each other at the various stages. While he elected not to begin his analysis at the application stage, the data available would permit this. And it could be used to track applicant flow from that first stage to the hiring of one applicant for the specific job opening in any given department — including the progress of each applicant through the various stages and examining the particular screening-devices used.

Further, Killingsworth “could limit his models by new applicants or incumbent State employees, initial pay, and performance evaluations.” However, he did not correlate any of his findings to a particular screening device.

*43 Significantly, Miller performed, separate statistical analyses by department on the three steps in the hiring process. In fact, as I discuss below, the NAACP relies on that analysis in pursuing an alternative argument for reversal. Furthermore, Kill-ingsworth, as quoted above, agreed with Miller that the three steps could be separated for purposes of analysis. Additionally, in four pages of findings on separability, the district court repeatedly referred to expert testimony that indicated the stages of the process could be separated.

This does not foreclose the possibility of further separation, as the district court found. For example, the record revealed other employment practices with potentially discriminatory effects such as résumé screens could have been isolated and separately analyzed. But, in' any event, the record supports' the district court’s finding that the plaintiffs did not meet their burden. Plaintiffs did not even attempt to challenge an actual employment practice and simply argued that the “total result” (to use Killingsworth’s phrase) was discriminatory.

In response to plaintiffs’ “ripple effect” argument, the district court found “the fact that one errant practice compounds a problem at a later stage of the process does not prevent investigation of either the earlier or later separate stage or practice.” I agree. Though the use of a discriminatory criterion at one stage may impact applicants throughout a decision-making process, such a “ripple effect” does not preclude separate analysis. In some cases, subjective and objective criteria may be so intertwined as to prevent separation. See McClain v. Lufkin Indus., Inc., 187 F.R.D. 267, 275 (E.D.Tex.1999) (finding an employer’s actions inseparable for analysis when “[t]he disparate impacts begin on the day one is hired and are potentially magnified each time one’s career ... intersects a subjective decision-making process”). But, plaintiffs here have not proven this is such a case. The very point of regression analysis is to allow isolation of particular elements and determine whether there is a “ripple effect.”

Finally, substantial evidence supports the district court’s finding that the plaintiffs failed to show the condition of the paper files precluded separate analysis of specific employment practices within the State’s hiring process. Killingsworth never looked at the hiring files. Significantly, as the majority notes, the district court found “the hiring files themselves permit a focused view of the different screening-devices and practices in referral, interview or hiring of applicants for any given job between the departments.” Plaintiffs make much of the missing documents from the files, but eighty-six percent of the files included interview questions, eighty-one percent had interview notes, and seventy-three percent provided an interview scoring matrix. Half of the files had reference checks. Eighty-four percent of the files also contained an individual’s application, résumé, and cover letter. As the majority recognizes, Greenwald commented, “The hiring files of the State are a gold mine that hasn’t been analyzed.” There is no evidence that the plaintiffs took even one of the objective standards the State employed and tried to determine whether it had a disparate impact using the available records.

Plaintiffs argue that the sample sizes would get smaller and statistical reliability would decline as one tries to analyze the effect of a practice that was only employed in certain areas at certain times. This may be correct, but does not excuse the failure to try. The statute does not permit courts to aggregate a collection of different hiring practices across different times and *44 departments just to increase the size of the sample.

The only case plaintiffs cite in support of their argument that a lack of records can prevent separate analysis is the district court opinion in Chin. Notably, the Second Circuit on appeal relied on a different rationale from the district court — i.e., that the process was basically entirely subjective and the final decision rested with one person. Chin, 685 F.3d at 154-55 . Stepping back and looking at Chin with the benefit of both opinions, the lack of records and the subjectivity of the process appear to be two sides of the same coin: No one documented why someone received a promotion because there was nothing to document. Port Auth. Police Asian Jade Soc. of N.Y. & N.J. Inc. v. Port Auth., 681 F.Supp.2d 456, 460-61, 464-65 (S.D.N.Y.2010) (finding the decision-making process could not be separated “both because records do not exist for every step and because the causal role of each step is called into doubt by the records that do exist”). Here, by contrast, it is undeniable that the records were incomplete, but equally irrefutable that no one who tried to analyze the records was unable to do so. As the district court found,

[t]he State’s data — its recordkeeping— while not perfect, was sufficient for both Dr. Killingsworth and Dr. Miller to conduct their analyses. The presence in the record of their models and opinions dispels the argument that the State’s recordkeeping is such that it precludes anything but a “systemic employment practice.”

I conclude substantial evidence supports the district court’s findings regarding the State’s record keeping.

I reiterate the importance of separately analyzing the different processes used by the various departments. In some departments, African-American applicants fared better than white applicants at certain stages; in others they fared worse, even much worse. This suggests that the different screening processes used by the departments may have had different impacts on applicant success. As the district court noted, these “[vjarying outcomes between the departments and stages of the process invite[ ] localized scrutiny.”

For example, based on the record in this case, I have concerns about the various résumé screening devices used by State departments at the step-two level. It is certainly possible that inappropriate screening devices may have been used in some of the departments in which Miller found a statistically significant disparity between blacks and whites at step two. But, it is just a possibility and not an aspect of the case that the plaintiffs chose to pursue.

Here, the district court observed that “the hiring files themselves permit a focused view of the different screening-devices and practices in referral, interview or hiring of applicants for any given job between the departments.” For example, as the court pointed out, DAS has a system-wide applicant screening manual, and an analysis could have been performed based on the manner in which DAS instructs managers on the use of the manual.

The district court went on to comment that “one can focus on any number of discrete employment decisions made as individual, separable, identifiable particular employment practices” and then gave two more examples:

One example of the separability of the process is the “second résumé screen” that had been utilized by some departments. It was a particular employment practice that was evaluated, determined to be inappropriate, and curtailed at the suggestion of DAS. Similar refinement of the hiring process by focusing on the *45 inappropriate use of “spelling and grammar screening” is another example of DAS having addressed a particular employment practice. The record reflects not only the ability to focus on these particular employment practices but when and which separate agencies responded to the suggested changes by DAS.

The foregoing has shown why I am confident the court reached the right conclusion. The district court methodically went through the record, focused appropriately on the testimony of statistical experts for each side, and identified various employment practices that could have been separately analyzed, including the three steps in the employment practice (separately analyzed by Miller) and more specific practices within those steps.

For all these reasons, I conclude substantial evidence supports the district court’s finding that the plaintiffs failed to show the State’s employment practices are not capable of separation for purposes of analysis. I close with three observations.

First, it is significant that the NAACP, in a well-argued amicus brief, relies on some of the same data that plaintiffs dismiss as inadequate. Thus, the NAACP asserts that Miller’s findings show there was an adverse impact at step two in eight departments that employed approximately fifty-eight percent of the State workforce. On that basis, it asks us to reverse the district court.

In my view, the NAACP’s brief raises serious questions as to whether the State committed unlawful discrimination. The problem with this argument is that it is not the case the plaintiffs elected to pursue. For instance, we do not know what practices were followed in those eight departments during step two. This seems like a relatively straightforward inquiry that could have been pursued in discovery. We also do not know which representative plaintiffs — if any — applied for jobs with those departments. And, the remedies sought by the plaintiffs would apply not only to those departments but to the State as a whole.

Instead of narrowing their focus, plaintiffs brought a class action alleging a common pattern of discrimination by the entire state executive branch of government. Having brought such a large case, it was then up to the plaintiffs to undertake the considerable work required to prove it. Under the prevailing law, this included analysis of specific hiring practices and their impact. Plaintiffs did not meet their burden.

Second, I do not downplay what this case has shown. Even according to Miller, it appears African Americans on the whole were disadvantaged in getting job interviews from some agencies, including some large departments like the department of human services and the department of transportation. This conclusion, from a defense expert, is disturbing although inconclusive. The district court, in my view correctly, questioned why “given all this data held by the State, it did not on a regular basis review it, as did these experts, with an eye toward measuring impact.”

Third, it bears emphasis that the defeat of this class action does not bar a person who believes he or she was a victim of discrimination from bringing an individual lawsuit on his or her own against the State for new acts of discrimination. What is clear here is that plaintiffs failed to prove, because they ultimately did not try to prove, that the State of Iowa engaged in specific employment practices that had discriminatory effects against African-American job applicants and that would allow for class-wide relief.

*46 For the foregoing reasons, I agree the district court’s judgment must be affirmed.

MANSFIELD and ZAGER, JJ., join - this special concurrence. ■

. Research into the process of socialization and development of social norms [has] led to an understanding that the development of stereotypes — and consequent biases and prejudices — is not a function of an aberrational mind, but instead an outcome of normal cognitive processes associated with simplifying and storing information of overwhelming quantity and complexity that people encounter daily.

Melissa Hart, Subjective Decisionmaking and Unconscious Discrimination, 56 Ala. L.Rev. 741, 746 (2005) (internal quotation marks omitted). Implicit-bias research and its application to legal theories has been thoroughly reviewed in legal scholarship. See Jerry Kang, Trojan Horses of Race, 118 Harv. L.Rev. 1489, 1515 & n. 122 (2005) (providing summary of employment-related implicit-bias studies, including experiment where résumés with "white names” received fifty percent more callbacks than résumés with "black names”); see also Judge. Mark W. Bennett, Unraveling the Gordian Knot of Implicit Bias in Jury Selection: The Problems of Judge-Dominated Voir Dire, the Failed Promise of Bat-son, and Proposed Solutions, 4 Harv. L. & Pol'y Rev. 149, 151-158 (2010) (discussing implicit-bias research as it relates to jury selection); L. Song Richardson, Arrest Efficiency and the Fourth Amendment, 95 Minn. L.Rev.2035 (2011) (discussing implications of implicit bias for police-citizen interactions and Fourth Amendment jurisprudence, and proposing "debiasing strategies" for police departments).

. Notably, Greenwald is an inventor of the Implicit Association Test, a widely used method of measuring implicit bias. See Project Implicit, About Us, https://www.projectimpli cit.net/about.html (last visited July 10, 2014).

. Notably, the CPS study mentioned by the majority did not perform regression analyses that excluded other possible variables that could account for the differences it reported. After pointing out the disparity between referrals and interviews, CPS acknowledged, “There could be ... very legitimate reasons why the percentage of African Americans is reduced so dramatically between referral and interview” and acknowledged that "the team was unable to determine a definitive reason(s) for these outcomes.”

. Both Killingsworth and the State’s expert, Robert Miller, used applications — as opposed to applicants — as their unit of analysis. As the CPS study noted, it is difficult to identify the exact makeup of the applicant pool or the actual number of applicants because the State’s tracking system did not track individual people, but rather applications. Both the State's and plaintiffs' experts acknowledged African-American applicants applied on average for 1.6 more jobs than white applicants. In short, both parties agreed that African-American applicants, on average, followed a strategy of casting a wider State job search net than white applicants.

.Killingsworth acknowledged that he looked only at ”[t]he total result,” not any particular employment practice other than "hiring” as a whole. He did not offer any opinions that the disparities he observed were the result of subjective as opposed to objective hiring practices.

. An act of Congress amending Title VII does not amend the ICRA. The ICRA requires the plaintiff to prove a specific employment practice caused the disparate impact. Hy-Vee, 453 N.W.2d at 518 . The Iowa legislature has not amended the ICRA to add a provision in response to our decision in Hy-Vee or the congressional amendment to Title VII over twenty-three years ago. I would not read such an exception into the ICRA in the guise of judicial interpretation. See Ackelson v. Manley Toy Direct, L.L.C., 832 N.W.2d 678, 689 (Iowa 2013) (declining to reinterpret the ICRA to allow punitive damages in light of precedent disallowing punitive damages and legislative acquiescence). Whether to create such an exception in the ICRA is a policy choice to be made by the legislature.

. The same majority has ignored an even longer line of cases adhering to our court’s long-standing practice of relying on federal decisions under the Fourth Amendment when interpreting the nearly identical search-and-seizure provision in the Iowa Constitution. See State v. Short, 851 N.W.2d 474, 511 (Iowa 2014) (Waterman, J., dissenting).