Equal Employment Opportunity Commission v. Chicago Miniature Lamp Works

CUMMINGS, Circuit Judge. This case highlights the problems of implementing Title VII’s prohibitions against race-based employment discrimination in Chicago, which contains various segregated communities. 1 Defendant Chicago Miniature Lamp Works (“Miniature”) is a manufacturer of light bulbs located in a largely Hispanic and Asian neighborhood on the north side of Chicago. Plaintiff Equal Employment Opportunity Commission (“EEOC”) investigated Miniature, found reasonable cause to believe that Miniature had engaged in a pattern or practice of discrimination against blacks as a class in recruitment, hiring and promotion, and filed suit against Miniature on June 8, 1979. After a bench trial, the district court found that Miniature had violated Title VII by engaging in a pattern or practice of discrimination against blacks as a class. 2 Miniature appeals the trial court’s ruling that it was liable under both a disparate treatment and a disparate impact theory, arguing that these findings were based on a misapprehension of the law and were clearly erroneous. For the reasons stated below, we reverse. 1. FACTS Miniature manufactured small incandescent and neon lamps and related components at its plant located at 4433 North Ravenswood Avenue in Chicago. The EEOC alleged at trial that Miniature had discriminated against blacks in its recruitment and hiring of entry-level workers. Although the district court considered evidence concerning the representation of blacks in Miniature’s work force from 1970 to 1981, the bulk of the evidence presented at trial concerned its hiring practices from *295 1978 to 1981. Miniature last hired an entry-level factory worker on June 29, 1981. The entry-level jobs at Miniature involved light manufacturing work and did not require any prior experience, particular educational background, or special skills. Basic manual dexterity and the ability to speak some English were the minimal qualifications required by Miniature. The beginning pay was low and did not rise substantially over time. Because the jobs did not require English language fluency, they had some special attraction to those persons who did not speak English as a primary language. Miniature’s principal method of obtaining new entry-level workers merits particular attention. Miniature relied almost exclusively on “word-of-mouth” in order to fill its entry-level job openings. Employees would simply tell their relatives and friends about the nature of the job — if interested, these persons then would come to Miniature’s office and complete an application form. Miniature did not tell or encourage its employees to recruit in this manner. The evidence indicates that the only time Miniature initiated this word-of-mouth process was during 1977 when it adopted an Affirmative Action plan. At that time, Miniature asked one or two black employees to recruit black applicants from among their relatives and friends. Miniature received applications whether or not it currently had an entry-level opening to fill. Whenever Miniature needed to fill an opening, it would go through its applications on file in order, phoning applicants until one was reached at home. Until August 1980, Miniature would start with the applications that were four to five months old and process them forward in time. After that date, Miniature started with the most recent application and processed applications backward in time. Evidence indicated that only one person was hired for every fifteen that applied. Because of the success of this process, Miniature never advertised for these jobs, and only rarely used the State of Illinois unemployment referral service. The district court found that Miniature’s reliance on word-of-mouth for filling entry-level jobs “resulted in the exclusion of blacks from the network of information concerning jobs at Chicago Miniature, * * * and gross underrepresentation of blacks in, and their exclusion from, Chicago Miniature’s entry-level work force.” 622 F.Supp. 1281, 1288 (N.D.Ill.1985). In reaching this conclusion, the lower court relied almost exclusively on statistics presented by the EEOC’s expert, Dr. Hek-mat Elkhanialy. A central task for the district court was to define the relevant labor market — that area from which Miniature would be expected to draw workers. The court found that the factors that enter into this definition include the location and accessibility of the employer, commuting patterns, and the employer’s applicant flow. Based on its analysis of these factors, the trial court found that Chicago (the city only, not the metropolitan area) was the relevant labor market for Miniature’s entry-level workers. Although it concluded that Chicago was the relevant labor market, the district court examined a number of different “relevant labor markets” and concluded that illegal discrimination occurred in each of these labor markets. Specifically, the district court examined each of the following labor markets: the City of Chicago, a 12-zip code area from which Miniature drew more than 90% of its applicants, a 5-zip code area from which it drew more than 70% of its applicants, and Miniature’s own zip code area, from which it drew nearly 30% of its applicants. In 1970, the percentage of entry-level workers in Chicago who were black was 35%. In 1980, this percentage had increased to 36.4%. The judge used these percentages as a basis for comparison against Miniature’s work force in two ways. First, he compared them with the percentage of blacks in Miniature’s entry-level work force for the years 1970-1981. In addition, he compared them with the percentage of blacks hired by Miniature for entry-level positions for the years 1978-1981. *296 Between 1978 and 1981, Miniature hired 146 entry-level workers. Nine of these workers (6%) were black. The trial court concluded that “the statistical probability of Chicago Miniature’s hiring so few blacks in the 1978-81 period, in the absence of racial bias against blacks in recruitment and hiring, is virtually zero.” Id. at 1295. The trial court also concluded that racial bias was the reason for the disparities between the percentage of blacks in Miniature’s entry-level work force for the years 1970-1981 and the percentage of black entry-level workers in Chicago. Although the differences between expected number of black hires and applicants and actual numbers were less when the judge restricted the relevant labor market to smaller geographical areas closer to Miniature’s plant, he concluded that racial discrimination must have been the cause of disparities even in geographically smaller relevant labor markets. The judge also commented that a number of black applicants (the record reveals that there were 11) had the letter “B” hand-written on their applications. The district court declined to attach significance to the fact that Miniature, on a percentage basis, hired more blacks than non-blacks who applied for the jobs in question. From 1978-1981, the percentage of black applicants who were hired was 16.4%; the overall percentage of applicants who were hired was 6.1%. 622 F.Supp. at 1319 (App. 3). The judge considered the higher percentage of black applicants hired an unreliable figure because of the small number of blacks involved (only 9 blacks were hired in the 1978-1981 period). II. ANALYSIS Miniature challenges the district court rulings as based both on a misapprehension of the law and as unsupported by the evidence. Although we revisit questions of law decided below and raised on appeal on the clean slate of de novo review, the Court sets aside only those factual determinations that were clearly erroneous. Mozee v. American Comm’l Marine Serv. Co., 940 F.2d 1036, 1044 (7th Cir.1991). The clearly erroneous standard for factual questions is especially appropriate in a discrimination case where the trial court weighed the competing statistical testimony of experts. Soria v. Ozinga Brothers, Inc., 704 F.2d 990, 995 n. 6 (7th Cir.1983). The Supreme Court has considerably altered the landscape of Title VII jurisprudence since the trial judge issued his opinion in October 1985. See Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 109 S.Ct. 2115, 104 L.Ed.2d 733 (1989); Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 108 S.Ct. 2777, 101 L.Ed.2d 827 (1988). Title VII itself remains unchanged — an employer may not discriminate against an employee or applicant on the basis of race: (a) It shall be an unlawful employment practice for an employer— (1) to fail or refuse to hire * * * any individual, or otherwise to discriminate against any individual with respect to his compensation, terms, conditions, or privileges of employment, because of such individual’s race, color, religion, sex, or national origin; or (2) to limit, segregate, or classify his employees or applicants for employment in any way which would deprive or tend to deprive any individual of employment opportunities or otherwise adversely affect his status as an employee, because of such individual’s race, color, religion, sex, or national origin. 42 U.S.C. § 2000e-2(a). Nevertheless, the Supreme Court has attempted in recent years to reconcile this general proscription with other considerations expressed elsewhere in Title VII — a caution to employers not to engage in hiring quotas, and a recognition that employers may rely on nondiscriminatory, objective standards in hiring their employees as shown in the following statutory provisions: (j) Nothing contained in this subchapter shall be interpreted to require any employer * * * subject to this subchapter to grant preferential treatment to any individual or to any group because of the race, color, religion, sex, or national origin of such individual or group on account of an imbalance which may *297 exist with respect to the total number or percentage of persons of any race, color, religion, sex, or national origin employed by any employer, * * * in comparison with the total number or percentage of persons of such race, color, religion, sex, or national origin in any community * * *. (h) Notwithstanding any other provision of this subehapter, it shall not be an unlawful employment practice for an employer to apply different standards of compensation, or different terms, conditions, or privileges of employment pursuant to a bona fide seniority or merit system * * *, provided that such differences are not the result of an intention to discriminate because of race, color, religion, sex, or national origin, nor shall it be an unlawful employment practice for an employer to give and to act upon the results of any professionally developed ability test, provided that such test, its administration or action upon the results is not designed, intended or used to discriminate because of race, color, religion, sex or national origin. 42 U.S.C. § 2000e-2. These three sections of Title VII show that Congress intended a delicate balance, strongly condemning discrimination on account of a protected characteristic, yet recognizing that racial imbalances in the work force may result from legitimate, nondiscriminatory factors. A recognition of this tension informs our analysis in this case. See Watson, 487 U.S. at 991-993, 108 S.Ct. at 2787-2788 (O’Connor, J.). The district court found Miniature liable based on both a disparate treatment and a disparate impact model. Although it is clear that the same set of facts can support both theories of liability, Teamsters v. United States, 431 U.S. 324, 336 n. 15, 97 S.Ct. 1843, 1854 n. 15, 52 L.Ed.2d 396 (1977), it is important to treat each model separately because each has its own theoretical underpinnings. The disparate treatment model is based most directly on Title VII’s statutory language, and requires an inquiry into the defendant’s state of mind. The defendant is liable under this model when the plaintiff can prove that the defendant subjectively intended to discriminate against the plaintiff on account of a protected trait. Id. at 335 n. 15, 97 S.Ct. at 1854 n. 15. In a disparate impact case, however, motive is irrelevant. As the Supreme Court stated in the seminal case of Griggs v. Duke Power Company, 401 U.S. 424, 91 S.Ct. 849, 28 L.Ed.2d 158 (1971), “Under the Act, practices * * * neutral on their face, and even neutral in terms of intent, cannot be maintained if they operate to ‘freeze’ the status quo of prior discriminatory employment practices.” Id. at 430, 91 S.Ct. at 853. The line between disparate impact and disparate treatment cases is most blurred in “pattern and practice” cases such as this one, because statistics can be used to prove both disparate treatment and disparate impact. Teamsters, 431 U.S. at 339, 97 S.Ct. at 1856; Mister v. Illinois Cent. Gulf R.R. Co., 832 F.2d 1427 (7th Cir.1987), certiorari denied, 485 U.S. 1035, 108 S.Ct. 1597, 99 L.Ed.2d 911 (1988). A. Disparate Treatment A prima facie case for a pattern or practice of disparate treatment can be established by “statistical evidence demonstrating substantial disparities in the application of employment actions as to minorities * * *, buttressed by evidence of general policies or specific instances of discrimination.” Coates v. Johnson & Johnson, 756 F.2d 524, 532 (7th Cir.1985). The plaintiff must prove “more than the mere occurrence of isolated or ‘accidental’ or sporadic discriminatory acts.” Teamsters, 431 U.S. at 336, 97 S.Ct. at 1855. Instead, the plaintiff must show that racial discrimination was the “standard operating procedure— the regular rather than the unusual practice.” Id. The issue is straightforward: Did the district court clearly err in deciding that Miniature had intentionally discriminated against blacks in its recruiting and hiring practices? On appeal, Miniature attacks separately the determination that it engaged in discriminatory hiring from the *298 determination that it engaged in discriminatory recruiting. Although the district court did conclude that Miniature had discriminated against blacks “in both recruitment and hiring,” id. at 1313, it clearly based liability on a comparison between the number of blacks who were actually hired and the number of blacks in the relevant labor market. In this sense, the trial court considered recruiting and hiring as necessarily linked parts of Miniature’s procedure for obtaining new employees. Miniature would have us arbitrarily separate the district court’s finding of intentional discriminatory recruiting from intentional discriminatory hiring. According to Miniature, “hiring” is the process by which a company selects employees from an already identified applicant pool. Miniature focuses on the fact that under its limited conception of the term “hiring,” it actually hired a greater percentage of blacks than of non-blacks. This is largely a question of semantics that only obfuscates the issue before us. The conclusion below was that the percentage of blacks in Miniature’s entry-level applicant pool and work force was so low compared to the relevant labor market that intentional discrimination against blacks must have occurred. We must consider the propriety of this finding as a legal and factual matter based largely on the statistics introduced by the EEOC — it does not help a correct analysis to separate “hiring” from “recruiting” arbitrarily. As part of its attempt to separate its recruiting and hiring practices, Miniature argues that it did not recruit and therefore, as a matter of logic, it could not have recruited diseriminatorily. We reject this simplistic syllogism. First, as mentioned above, the lower court considered Miniature’s hiring activities as a whole, and based its finding of disparate treatment on disparities between the percentage of blacks in Miniature’s entry-level work force and the percentage of blacks in the relevant labor market. Thus even if we were to accept Miniature’s “lack of recruiting” argument, Miniature could still be liable for its hiring practices. Miniature certainly cannot claim it did not hire employees! In any event, Miniature has recruited: it made an intentional decision to rely on word-of-mouth to attract applicants for its entry-level openings. Miniature knew how workers were learning of its employment opportunities. When it adopted an Affirmative Action plan in 1977, it made an effort to tell its black employees to contact their friends and relatives in order to increase the effectiveness of its word-of-mouth network in the black community. Miniature also used other recruiting procedures at times. It used newspaper advertisements to attract clerical and secretarial applicants. On occasion it used a job referral service. But Miniature obviously intentionally chose not to use these two forms of recruiting for its entry-level workers. The central inquiry in disparate treatment cases is an evaluation of the intent of the employer. We cannot conclude that as a matter of law it is impossible for an employer to discriminate intentionally against blacks by relying on word-of-mouth to provide its applicants. Cf., American Nurses’ Ass’n v. Illinois, 783 F.2d 716 (7th Cir.1986) (reversing dismissal of sex-based discrimination complaint alleging intentional discrimination based on failure to equalize pay between traditionally male and female jobs). Of course, the fact that a plaintiff is unable to point to specific actions or practices by an employer that are arguably discriminatory is a probative factor that a fact-finder can and should rely upon in ruling on the ultimate issue of intentional discrimination. In other words, we reject defendant’s claim that “Miniature cannot be held liable because it did not commit any act” (Reply Br. at 8). It was the trial court’s finding that Miniature’s overall entry-level hiring decisions were made with racial animus. It is true that the trial court focused on Miniature’s reliance on word-of-mouth as evidence of its discriminatory intent. On review, we look at all the evidence on the record to determine if the trial court’s finding was clearly erroneous. Nevertheless, Miniature’s passive reliance on word-of-mouth to generate applicants must be given minimal weight because it involved no *299 affirmative act by Miniature. Drawing the inference of intent from “non-action” is necessarily more difficult than drawing the inference of intent from particular actions. This is especially true since intent means more than knowledge that a certain action (or non-action) will cause certain discriminatory results. Intent means a subjective desire or wish for these discriminatory results to occur. American Nurses’ Ass’n, 783 F.2d at 722. Statistical evidence of disparities between minority representation in an employer’s work force and minority representation in the community from which employees are hired can prove disparate treatment in a pattern or practice case. Hazel-wood School Dist. v. United States, 433 U.S. 299, 307, 97 S.Ct. 2736, 2741, 53 L.Ed.2d 768 (1977). As indicated above, a pattern or practice of disparate treatment is shown through a combination of “statistical evidence demonstrating substantial disparities * * * buttressed by evidence of general policies or specific instances of discrimination.” Coates, 756 F.2d at 532. However, here the trial court clearly erred in determining that the EEOC had proved that Miniature had engaged in a pattern or practice of intentional discrimination. 1. Statistical Evidence The trial court made many factual findings regarding statistics in this case. The data underlying the statistical findings is found in Appendices 1, 3 and 4 to the trial judge’s opinion. Appendix 1 summarizes Miniature’s “static work force statistics” for the years 1970-1981. The static work force statistics set forth for each year an employment “balance sheet” that provides a snapshot of the employer’s work force for that year. For the twelve years reported, the percentage of blacks in entry-level jobs at Miniature was relatively constant, ranging from 4.5% to 6.5%. The total number of entry-level workers ranged from 202 to 396 over this period; the actual number of black entry-level workers ranged from 9 to 22. Unlike the percentage of black entry-level workers, the percentage of Asian and Hispanic entry-level workers during 1970-1981 was not static but rather increased. The representation of Hispanic workers ranged from 40.0% to 66.0%, while Asian workers ranged from 0.0% to 16.5% over the twelve years. Appendices 3 and 4 summarize Miniature’s hiring rates for the years 1978-1981. These figures present an employment “profit and loss” statement showing how many blacks applied and were hired over an entire year. The appendices include both the EEOC’s and Miniature’s figures. For convenience, the relevant portions are capsulized below: 3 Applicants (No.) Hires (No.) Hiring Rates Year Total Total Non-Blacks Blacks Non-Blacks Blacks PQ 1978 P,D 401 66 65 16.4% 20% CO to o lO 1979-P 227 18 17 7.6% 50% to to en (M -D — 18 17 . 1 I I 1980-P 879 38 34 3.9% 25% CO a w CD H -D 880 38 34 4.0% 12.1% CO rf*. CO CO 1981-P 807 24 21 2.7% 9.4% -a -a oi <M CO -D 820 24 21 2.7% 8.6% -3 co ai I» CO Totals (’78, ’80, ’81) 2034 -P 2087 128 120 5.9% 15.1% 2028 -D 2101 128 120 5.9% 11.0% The key task for the trial judge was to determine the relevant labor market so that he could have a basis for analyzing *300 Miniature’s employment data. The district court properly observed that: Statistical proof relevant in pattern-or-practice disparate-treatment suits involves a showing of ‘significant statistical disparity’ between the racial composition of the employer’s work force (or applicant flow or hiring pattern) and the racial composition of the community from which the workers are drawn. 622 F.Supp. at 1308. The applicant flow would normally figure into the determination of an employer’s relevant labor market (“the community from which the workers are drawn”) because it provides cogent evidence of what areas, in fact, contain persons who are attracted to a particular employer. The judge, however, found that Miniature’s applicant flow must be ignored because it “is not remotely representative of the racial and ethnic composition of the civilian work force in any area that might be deemed the relevant labor market,” since the recruiting process itself was discriminatory. Id. at 1289-1290. The district court instead found that Chicago was the relevant labor market for several reasons. First, the factory was conveniently located to both the elevated train and bus service, thereby connecting it to virtually every point in Chicago because of Chicago’s extensive public transportation system. In addition, Miniature itself had identified Chicago as its “basic recruiting area” in its Affirmative Action Plan in place from June 1977 to May 1978. Finally, specific testimony of applicants from the far South and West Sides of Chicago was presented at trial as evidence that Chicago was the proper relevant labor market. After concluding that Chicago was the relevant labor market, the trial judge compared both the static and dynamic figures discussed above with the expected figures based on the percentage of black entry-level workers in Chicago. For example, in 1980 36.4% of the entry-level workers in Chicago were black. However, only 6.0% of the factory workers (16 workers) at Miniature were black. Because the expected number of blacks would have been 98 (36.4% times the total number of factory workers — 268), the court concluded that the actual number of black entry-level workers was 10.4 standard deviation units (“SDUs”) below the expected number in 1980. Similar computations, presented by the EEOC’s expert and accepted by the judge, show that for all intervening years between 1970 and 1980, the actual number of Miniature’s black entry-level workers was always at least 9 SDUs below the expected number. 4 The trial court used a similar analysis for evaluating the racial composition of Miniature’s applicants, as summarized in the table above. For these calculations, data was available only for the years 1978 to 1981. The court multiplied the number of applicants for a given year by 36.4%, the percentage of black entry-level workers in Chicago in 1980. This number, according to the court, represented the “expected number” of black entry-level applicants. For example, using Miniature’s numbers, in 1980 there were 880 applicants. Therefore, the court concluded that the expected number of blacks in this applicant pool was 320 (36.4% times 880). The actual number of black applicants was 33-20.1 SDUs below the expected number. We begin our analysis of these statistics by noting that a statistical model need not be completely specified, and some arguably relevant variables can be omitted in certain cases. Bazemore v. Friday, 478 U.S. 385, 400, 106 S.Ct. 3000, 3008, 92 L.Ed.2d 315 (1986). This rule is largely a consequence of our clearly erroneous standard and our deference to the district court’s weighing of statistical evidence. Miniature argues that the EEOC’s statistics are clearly erroneous because they do not include the following considerations: relative commuting distance, shift prefer *301 ence, and lack of an English fluency requirement. 5 The EEOC responds that Bazemore precludes any attack on its statistical evidence, since Miniature did not include these variables in its own statistical analysis. A careful reading of Bazemore reveals that the EEOC’s argument that its statistics are unassailable is without merit. The Supreme Court in Bazemore rejected the proposition that “all measurable variables” must be included in the statistical analysis. Id. at 400, 106 S.Ct. at 3009. The ultimate question in Bazemore was whether the defendant had discriminated against blacks with regard to wages. The plaintiffs’ analysis included four independent variables — race, education, tenure, and job title. Id. at 398, 106 S.Ct. at 3007. The Court disagreed with the conclusion that an analysis “which accounts for the major factors” could be considered unacceptable simply because it did not include all factors. Id. at 400, 106 S.Ct. at 3009. As the Court noted, “There may, of course, be some regressions so incomplete as to be inadmissible as irrelevant; but such was clearly not the ease here.” Id. at 400 n. 10, 106 S.Ct. at 3009 n. 10. Indeed, this Circuit has already rejected an expansive reading of Bazemore that requires the defendant to include in its own statistical analysis variables that it asserts are missing from the plaintiff’s analysis. EEOC v. Sears, Roebuck & Co., 839 F.2d 302 (1988) (affirming the district court’s rejection of plaintiff’s statistical model because it failed to account for qualitative factors even though the employer did not present a competing statistical model); see also Mister, 832 F.2d at 1437 n. 3 (“it is necessary to review the inferential process case by ease; the biases of one study in one case may be avoided or reversed in the next.”). The missing neutral variables of relative commuting distance and language fluency requirements are too important to be ignored in this ease. Our analysis of the judge’s findings leaves us with the definite and firm conviction that a mistake has been committed. Anderson v. Bessemer City, 470 U.S. 564, 573, 105 S.Ct. 1504, 1511, 84 L.Ed.2d 518 (1985); Coates, 756 F.2d at 533. Thus we hold that the district court clearly erred in crediting the statistical evidence outlined above as proof of disparate treatment under Title VII. It is clear that Miniature was held liable because there were relatively few blacks in its applicant pool as compared to the number of entry-level black workers in Chicago. The EEOC’s statistical expert used a very simple demographic model. The applicant’s race was the only variable considered in the model, and Miniature’s racial composition was compared with the racial composition of Chicago as a whole. The trial court’s impressive litany of high standard deviations is only as impressive as the underlying hypothesis from which the deviations were generated. Mister, 832 F.2d at 1431. In this case, the underlying hypothesis is faulty because considerations of applicant preference were totally ignored. The decision in Wards Cove, 490 U.S. 642, 109 S.Ct. 2115, reveals the importance of applicant preference for the jobs at issue. 6 Wards Cove involved, in relevant part, a claim of racial discrimination in an Alaskan salmon cannery. The lower court in Wards Cove had held that the disparity between the number of minorities filling unskilled “cannery” positions and the number of minorities in the more attractive, but still unskilled, “noncannery” positions was a proper basis for Title VII liability. The Supreme Court reversed, stating that a direct comparison between these two labor pools was inappropriate “because the vast majority of [the unskilled] cannery workers did not seek jobs in unskilled non-cannery positions; there is no showing that many of them would have done so even if *302 none of the arguably ‘deterring’ practices existed.” Wards Cove, 490 U.S. at 653-654, 109 S.Ct. at 2122-2123. The Court noted that the noncannery positions were generally filled by a union that was largely nonwhite, and that this nondiscriminatory reason for the discrepancy between the cannery and noncannery workers could not be ignored. Id. at 654, 109 S.Ct. at 2123. Interest in the jobs at issue is therefore ordinarily an important consideration for proof of a Title VII pattern or practice case. See EEOC v. Sears, Roebuck & Co., 839 F.2d 302 (7th Cir.1988) (upholding trial court’s determination that probative value of plaintiff’s regression statistics was reduced because it failed to take into account relative lack of interest on the part of females in commission sales jobs); EEOC v. Consolidated Services Systems, No. 85-C8312, 1991 WL 181119, 1991 U.S. Dist. LEXIS 12481 (N.D.Ill. September 9, 1991) (holding that plaintiff’s statistical evidence was not sufficiently probative because it did not consider whether individuals would be interested in the part-time, low-paying janitorial jobs at issue). The identification of a relevant labor market— the key issue in a class-based Title VII case — means not only identifying qualified potential applicants for the job at issue but also identifying interested potential applicants. Wards Cove makes clear that properly identifying the relevant labor market is the key ingredient in proving Title VII discrimination through the use of statistics. “The proper comparison is between the racial composition of the at-issue jobs and the racial composition of the qualified population in the relevant labor market.” Wards Cove, 490 U.S. at 650, 109 S.Ct. at 2121, citing Hazelwood, 433 U.S. at 308, 97 S.Ct. at 2741. Thus the trial court’s initial task is to compare the racial composition of the qualified persons in the labor market with the persons holding at-issue jobs. Id. at 650-651, 109 S.Ct. at 2121-2122. In this case, the use of Chicago as a relevant labor market, with no weighting of distance, meant that no consideration was given to the fact that blacks living on the South and West Sides would be less willing to apply at Miniature because of the commuting time involved. In addition, no consideration was given to the fact that non-English speaking persons would be more willing to apply because of their inability to obtain jobs with English fluency requirements. The statistical evidence accepted by the lower court treated each entry-level worker in Chicago the same, no matter how far they lived from Miniature’s plant and no matter what their primary language. Yet Miniature’s entry-level jobs were low-paying and required no special skills. Low-paying, unskilled jobs are more likely to be filled by those living closer to the site of the job, simply because the cost (including the opportunity cost of time lost) of commuting cannot be justified. Mister, 832 F.2d at 1432. Commuting time is thus especially important in this case. Unlike Mister, 832 F.2d 1427, where this Court rejected the defendant’s claim that an applicant’s distance from a job site was erroneously ignored by the plaintiff’s expert, this case involves one factory on the north side of Chicago, not a cluster of work-sites in a 200-mile geographical area. 7 The district court uncritically accepted the assertion of the EEOC’s expert that one and one-half hours was a reasonable commuting time for a low-paying entry-level job. 8 The trial court ignored Miniature's lack of a fluency requirement when considering the relative attractiveness of its jobs to *303 different ethnic groups. Several Miniature employees testified at trial that Miniature’s position was one of only a few jobs open to them because of their limited English fluency (these witnesses testified through an interpreter). Common sense dictates that a nondiscriminatory employer with no English fluency requirement will receive a disproportionate amount of applications from non-English speaking persons. 9 If the EEOC’s model is accepted, it would show not only intentional discrimination against blacks but also against whites and males, since the majority of Miniature’s entry-level work force were Hispanics and Asians, and the vast majority were women (over 80%). Without more specific proof, it would be mere speculation to infer that discrimination, and not special interest on the part of women, Hispanics and Asians, caused the lack of representation for whites and males. Interest for the jobs at issue is a crucial element in the definition of the relevant labor market. The trial court erred by ignoring the two important factors of commuting time and language preference that significantly affected interest in the jobs at issue. In an attempt to counter criticism that Chicago was too expansive a relevant labor market, the court also conducted its “applicant flow analysis” for smaller geographic units — a 12-zip code area, a 5-zip code area, and a “home” zip code area surrounding Miniature’s plant. However, the court computed its statistics for these smaller areas somewhat differently than those it computed when it considered Chicago the relevant labor market. First, the court used as the basis for its comparison the percentage of black civilian workers, not black entry-level workers as it did when Chicago was the relevant labor market. The court did not discuss the appropriateness of using this different statistic. In addition, when making its comparisons based on smaller areas, the court counted as the actual number of black applicants only that portion coming from the area being considered as the “relevant labor market.” However, when considering Chicago as the relevant labor market, the court did not restrict the number of black applicants to only those living in Chicago. The effect of this restriction was to make these smaller areas more like “sub-markets” rather than a focal relevant labor market. Miniature received no credit in these “sub-analyses” for any applicants that came from outside the sub-markets, and was in effect penalized because a large percentage of its black applicants came from outside these closer areas. Because of this flaw, and because the comparison was not restricted to black entry-level workers, we believe that the court erred in accepting these comparisons based upon “relevant labor markets” smaller than Chicago. 2. Anecdotal Evidence The EEOC might have overcome deficiencies in its statistics if it had presented more compelling anecdotal evidence. The anecdotal evidence presented in this case, unfortunately for the EEOC, was not persuasive. In his opinion, the trial judge focused almost exclusively on statistical disparities in concluding that Miniature had intentionally discriminated *304 against entry-level blacks. In his findings of fact, the judge did refer to several practices that might be considered “specific instances of discrimination.” Although this anecdotal evidence was referred to in the findings of fact section, not in the conclusions of law section, we treat them as support for a legal conclusion of disparate treatment. 10 The problem with the anecdotal evidence touched upon by the court, and perhaps the reason that it was not specifically relied upon, is that it relates to instances where specific black applicants were not hired. However, the statistical evidence as well as the trial court’s legal conclusions all highlight deficiencies in the applicant pool as the linchpin of Miniature’s Title VII violation. Indeed, for applicants, the hiring rate for blacks was higher than the hiring rate for non-blacks. For example, the trial court stated that “[njumerous entry level job applications have the letter “B” written by hand on them.” 622 F.Supp. at 1293. The court did not specifically find that this was invidious “race-coding,” but instead obliquely cited another case in which that conclusion had been made. The EEOC concedes in its brief that the labeling of the applications “could have had some legitimate purpose.” (EEOC Br. 48). In addition, ordinarily the probative value of any race-coding would relate to its effect on Miniature’s hiring rate vis-a-vis its applicant pool. 11 The record also reveals that only 11 applications of blacks (out of either 55 or 75 total applications between 1978 and 1981, depending on whether the EEOC’s or Miniature’s data is used) had the letter “B” written on them. The trial court also noted a “computer analysis” that matched black entry-level applicants who were not hired with non-blacks who applied at approximately the same time and were hired. This evidence is unpersuasive as well because it does not relate to the ultimate reason relied upon by the court for Miniature’s liability — a lack of black applicants. The computer analysis only proves that certain black applicants did not get hired; it is uncontested that only 1 out of 15 applicants was hired. The central prerequisite for being hired was being home and near a phone when a position became open. That some black applicants were not hired is simply unpersuasive in the circumstances of this case without further evidence of discriminatory intent. In sum, the trial court erred in concluding that Miniature was liable under Title VII for disparate treatment of blacks. This is one of those rare cases where the statistical evidence credited by the lower court does not support liability. Anecdotal evidence of intentional discrimination, only tangentially relied upon by the lower court, also fails to carry the day for the EEOC. B. Disparate Impact When conducting its disparate impact analysis, the trial court again focused on the statistics put forward by the EEOC’s expert and on Miniature’s reliance on word-of-mouth recruiting. Underscoring the basic similarity in the two approaches, the court began by stating that the “EEOC’s statistics are not open to rebuttal under the disparate-impact model any more than under the disparate-treatment model.” 622 F.Supp. at 1310 n. 10. The district court recognized that under a disparate-impact approach a plaintiff must identify a particular practice that caused the disparate impact. In this case, the district court considered the “particular practice” to be reliance on word-of-mouth recruiting. Since “Chicago Miniature has not even sought to offer any evidence tending to show its reliance on word-of-mouth recruiting is business-related,” the court concluded that the EEOC’s disparate impact case was established and unrebutted. Id. at 1311. *305 We concluded above that it was clearly erroneous for the trial court to rely on the statistical evidence put forward by the EEOC. Since the EEOC relied on the same evidence to uphold liability under its disparate impact claim that it relied upon for its disparate treatment claim, it was also clearly erroneous for the trial court to support its findings of disparate impact with this statistical evidence. There is another reason that the holding of disparate impact liability against Miniature cannot stand. In a disparate impact case, “plaintiff is * * * responsible for isolating and identifying the specific employment practices that are allegedly responsible for any observed statistical disparities.” Wards Cove, 490 U.S. at 656, 109 S.Ct. at 2124, citing Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 994, 108 S.Ct. 2777, 2788, 101 L.Ed.2d 827 (1988) (O’Connor, J., concurring). The EEOC does not allege that Miniature affirmatively engaged in word-of-mouth recruitment of the kind where it told or encouraged its employees to refer applicants for entry-level jobs. Instead, it is uncontested that Miniature passively waited for applicants who typically learned of opportunities from current Miniature employees. The court erred in considering passive reliance on employee word-of-mouth recruiting as a particular employment practice for the purposes of disparate impact. The practices here are undertaken solely by employees. Therefore, disparate impact liability against Miniature must be reversed. As stated above, the reliance of word-of-mouth to obtain applicants for jobs does not insulate an employer from a finding of disparate treatment of minorities. However, for the purposes of disparate impact, a more affirmative act by the employer must be shown in order to establish causation. “[A] Title VII plaintiff does not make out a case of disparate impact simply by showing that, ‘at the bottom line’, there is a racial imbalance in the work force.” Wards Cove, 490 U.S. at 657, 109 S.Ct. at 2124. The EEOC here, in essence, is attacking Miniature’s overall hiring procedure by pointing to the “bottom line” results; it has not made the more focused allegation required by Wards Cove that a specific, affirmative employment practice caused the disparity between entry-level workers at Miniature and entry-level workers throughout Chicago. Cf. Bartman v. Allis-Chalmers Corp., 799 F.2d 311 (7th Cir.1986) (holding that failure of defendant to declare its intentions regarding the expiration of a pension is not an employment practice for the purposes of disparate impact liability under the Age Discrimination in Employment Act), certiorari denied, 479 U.S. 1092, 107 S.Ct. 1304, 94 L.Ed.2d 160 (1987). III. CONCLUSION There is no doubt that racial discrimination in employment remains widespread in Chicago. 12 Without probative evidence of discriminatory intent, however, Miniature is not liable when it passively relies on the natural flow of applicants for its entry-level positions. Miniature’s entry-level hiring practices were straightforward, simple, and effective. The EEOC’s mis-specified statistical model that ignored commuting distance and language fluency requirements, when unaccompanied by more probative anecdotal testimony, cannot support a ruling that Miniature violated Title VII by discriminating against blacks. Therefore, the trial court erred in finding Miniature liable under Title VII. Its disparate treatment and disparate impact findings were clearly erroneous because they credited statistics that did not take into account applicant preference and because the anecdotal evidence presented at trial was not sufficiently probative. In addition, the EEOC’s disparate impact theory fails because the EEOC did not specifically identify a particular practice by the employer that caused any disparity. Because of our *306 holding, we do not reach the various issues raised on appeal regarding the propriety of the remedies awarded by the trial court. The judgment below is reversed. Footnotes [1] Chicago has been labeled one of ten "hyper-segregated” metropolitan areas in the United States. Turner, Fix & Struyk, Opportunities Denied, Opportunities Diminished: Discrimination in Hiring 7 (1991) (Urban Institute Project Report, preliminary version) [hereafter Opportunities Denied J, citing Massey & Denton, Hyper-segregation in U.S. Metropolitan Areas: Black and Hispanic Segregation Along Five Dimensions, 26 Demography 373 (1989). [2] The lower court also considered the individual complaint of Ed Randolph, who alleged that Miniature discriminatorily denied him a promotion on account of his race. The court found Randolph’s claim wanting, and neither he nor the EEOC appeals that determination. [3] The EEOC’s data is labeled “P” and Miniature’s data is labeled "D”. In 1978 the EEOC and Miniature agreed on the underlying data. In 1979 Miniature did not present any applicant data because it considered the data unreliable. [4] The Supreme Court has indicated that the hypothesis that decisions were made without regard to a protected characteristic is suspect if there are more than two standard deviations between the actual value and the expected value. Hazelwood, 433 U.S. at 308 n. 14, 97 S.Ct. at 2742 n. 14. [5] Shift preference is not discussed below because it is closely related to relative commuting distance (an applicant who lives relatively far from a job site is less likely to want to work on the night shift). [6] Although Wards Cove dealt with a disparate impact claim, its discussion of statistics is appropriate in evaluating a “pattern or practice" case where disparate treatment is proven primarily by the use of statistics. [7] In addition, unlike this case, the plaintiff in Mister introduced evidence showing gross disparities between the number of black applicants hired and the number of white applicants hired. Where interest on the part of black applicants is clearly shown, the relevance of their distance from the job site is considerably diminished. [8] The plaintiff’s statistics are also skewed because the boundaries of the relevant labor market were arbitrarily limited to the City of Chicago. Defendant’s plant is located 5 miles from the northern border of the city and 20 miles from the southern border. It is thus reasonable to conclude that at least some areas of the surrounding suburbs would be as likely to attract applicants as some far-away, southern sections of Chicago. This arbitrary limiting of the relevant labor market is suspect. [9] The following hypothetical, presented by defendant at trial, is compelling. Assume that five employers are equidistant to 300 job-seekers: 100 blacks, 100 whites, and 100 Hispanics. Assuming that all employers are equally attractive and have identical requirements and conditions, then one would expect each employer to receive approximately 20 applications from each ethnic group. Each employer would have a racially balanced applicant flow. However, if employer # 1 had no English fluency requirement and the other four employers did have an English fluency requirement, a different pattern would emerge. Assuming that all of the blacks and whites are fluent, as well as 50 of the Hispanics, and that the other 50 Hispanics are not fluent, then employer # 1 would receive 60% of its applications from Hispanics (all of the non-fluent Hispanic applications and its share of the fluent Hispanics), 20% from blacks, and 20% from whites. Thus, if the non-fluency characteristic of employer # 1 was ignored, it would appear to be discriminating against blacks and whites. Conversely, the other four employers would appear to be discriminating against Hispanics, because the majority of Hispanics, for nondiscriminatory reasons, would apply to employer # 1. [10] Indeed, the court below specifically stated that any findings of fact that are actually conclusions of law should be considered as such. [11] Of course, if an employer gets a reputation that it will not hire black workers, because of actions such as race-coding, then black applicants will be deterred from applying. Mister, 832 F.2d at 1436. The EEOC does not point to any evidence indicating such a reputational effect in this case. [12] The Urban Institute’s Opportunities Denied report describes a "hiring audit” conducted in 1990 in Washington, D.C. and Chicago. Two young men, one black and one white, were carefully matched on hiring characteristics and applied for entry-level jobs that were advertised in the newspaper. The report concluded that unfavorable differential treatment occurred twenty percent of the time. See Opportunities Denied at 31-33.