dissenting:
I respectfully dissent, because I believe that a remand for a clarification of the district court’s findings, as opposed to a reversal, is appropriate in this case. As the majority notes, the district court based its finding of discrimination in the issuance of derelictions on three factors: (1) statistical evidence; (2) a 1977 disciplinary matter involving a white officer; and (3) the “sincere impressions” of appellees’ witnesses that minority patrolmen were treated unequally. While I find fault with Judge Sifton’s analysis, and therefore would require a clarification of his findings, I am not prepared to conclude that any finding of discrimination in the issuance of derelictions on the record in this case would be clearly erroneous.
I do not believe, as the majority has concluded, that there is a “complete absence of creditable statistical evidence” in the record to support a finding of discrimination, although I agree that the court’s analysis of the statistical evidence raises problems. My main concern is with Judge Sifton’s conclusion that the difference between the expected and actual proportion of minority patrolmen receiving derelictions in 1979 was statistically significant. Judge Sifton reached this conclusion by looking to appellants’ expert’s analysis of aggregate work force data for 1979. Looking at this generalized date, Judge Sifton found that the difference between the expected and actual proportion of minority patrolmen receiving derelictions “in terms of standard deviations is 4.02,” a statistically significant result for the purposes of Title VII, Hazelwood School Hist. v. United States, 433 U.S. 299, 308 n. 14, 97 S.Ct. 2736, 2742 n. 14, 53 L.Ed.2d 768 (1977). Judge Sifton apparently rejected appellants’ contention that a statistical analysis of individual patrol districts (aggregated across patrol districts through the application of an aggregation technique known as Fisher’s Omnibus test), as opposed to a generalized analysis of the aggregate work force data, was appropriate. Appellants argue that an analysis of the decentralized data is warranted in light of the fact that derelictions are issued on a decentralized basis. Such an analysis results in a standard deviation of 1.29, a deviation which is plausibly explained by chance. Judge Sifton failed to adequately explain his reliance on the generalized data, and thus his choice of the 4.02 figure, and therefore I believe that a remand for clarification is appropriate.
I disagree, however, with the majority’s apparent conclusion that the choice of the appropriate standard deviation is solely a matter of interpretation over which appellants’ expert’s testimony is binding. The relevance of generalized data, as opposed to decentralized data, raises questions of law and fact, which must be decided, at least initially, by the district court. Furthermore, while the decentralized nature of the disciplinary process at issue may warrant attention to decentralized data, see Coser v. Moore, 739 F.2d 746, 750 (2d Cir.1984) (Title VII case involving university tenure decisions), Judge Sifton did not ignore such data. He examined 1977 and 1979 data on a patrol district basis, concluding that the generalized statistical discrepancy in 1979 could not be explained as the *773result of isolated instances of misconduct in the individual patrol districts. Coser does not suggest that generalized data must be ignored in evaluating statistical proof concerning independent, uncoordinated employment decisions; it merely requires that where generalized data is presented in a case involving decentralized decisions, such data must be analyzed “in light of the [fact] that independent, uncoordinated decisions are involved....” Id.
It should be noted, moreover, in evaluating the sufficiency of the statistical evidence in the record, that appellees introduced evidence showing that 40.39% of the derelictions issued in 1978 through 1980 were issued to minority patrolmen when such patrolmen represented only 28% of the work force. Appellants’ expert declined to examine the data from 1978 and 1980, claiming an absence of available data as to the racial composition of the department in those years. Nor did appellants attempt to prove that the discrepancy shown over those three consecutive years could plausibly have occurred by chance.
The majority points to other flaws in Judge Sifton’s discussion of the statistical evidence. While it may be that the judge misconstrued the 1979 data as setting forth the number of derelictions issued to minorities, when the data actually sets forth the number of minority officers receiving derelictions, this alleged mistake, if it is one, does not warrant a reversal. One of appellants’ own exhibits indicates that the discrepancies are even greater when one looks at the number of derelictions issued.
The majority also argues that the testimony credited by the district court “virtually precludes a finding of excessive lenience by white superior officers toward other white officers.” I agree that Judge Sifton failed to appreciate the logical implication of his statement that “It thus appears that white officers are not issuing derelictions to white officers with the same regularity that both white and black officers are issuing them to black officers,” namely, that black, as well as white, supervisors are issuing a disproportionate number of derelictions to minority patrolmen. Several reasons, however, may exist for the disproportionate issuance of derelictions by black supervisors, including: (1) acclimation of black supervisors to a system of racial bias; or (2) apprehension by black supervisors over issuing a greater percentage of derelictions to white patrolmen than that issued by white supervisors. Thus, while a remand is appropriate for the district court to clarify its findings, I do not believe that we should rely on the statistics to reverse.
In addition to statistical evidence, the court cited anecdotal evidence “suggesting tolerance of misconduct by white officers.” The most forceful of such anecdotal evidence, according to the court, was a 1977 incident involving a racial slur by a white lieutenant. This incident does not suffice as a specific instance of discrimination since it occured outside the applicable time period. Nonetheless, it would not be error for the court to take this incident into account, provided it was considered only as a background factor. See United Air Lines, Inc. v. Evans, 431 U.S. 553, 558, 97 S.Ct. 1885, 1889, 52 L.Ed.2d 571 (1977). Judge Sifton also credited witnesses who testified that appellants dealt unequally with the races, but he did not identify who these witnesses were. Appellees have been unable to point to an abundance of evidence in the record of specific instances of unequal treatment. Therefore, I would also remand for Judge Sifton to state specifically what he relied upon for his findings of anecdotal evidence of discrimination and the degree to which such anecdotal evidence formed the basis of his finding of discrimination. While in a proper case discriminatory intent may be established by statistical proof alone, anecdotal evidence is important in bringing “the cold numbers convincingly to life.” Rossini v. Ogilvy & Mather, Inc., 798 F.2d 590, 604 (2d Cir.1986), quoting International Brotherhood of Teamsters v. United States, 431 U.S. 324, 339, 97 S.Ct. 1843, 1856, 52 L.Ed.2d 396 (1977).
While it is not clear to me whether a disparate impact analysis is appropriate to the facts of this case, see Rossini, 798 F.2d at 605, I would also ask Judge Sifton on *774remand to clarify the legal standard or standards upon which any relief is granted.