Statisticians

Currier v. United Techs. Corp., 393 F.3d 246 (1st Cir. 2004). After being discharged during reduction in force, worker brings age discrimination claim. District court admits testimony plaintiff's statistical expert, Dr. Sat Gupta. Jury finds for plaintiff. Admissibility affirmed. Defendant says statistician should have worked not from data on defendant's entire salaried workforce of 183 employees, but rather from sample comprising only six managers with whom plaintiff was similarly situated, or at most from sample of 44 employees actually susceptible to layoff. But despite having objected to statistician's testimony on numerous grounds before and during trial, defendant did not raise this particular argument until its post-trial motions, and jury could have found that age played role in defendant's selection of 44 layoff-eligible workers from its total payroll of 183 employees. Defendant raises other challenges to probative value of statistician's testimony, but defendant could and did bring those challenges to jury's attention. No abuse of discretion.

Citizens Fin. Group, Inc. v. Citizens Nat'l Bank, 383 F.3d 110 (3d Cir. 2004), cert. denied, 125 S. Ct. 1975 (2005). In trademark action, district court excludes survey evidence from trademark owner's expert Robert Reitter. Exclusion affirmed. In "reverse confusion" cases, where larger and more powerful company has allegedly infringed trademark of smaller and less powerful senior owner, the universe of survey respondents should comprise only customer base of senior owner. District court permissibly found that survey suffered from fatal methodological flaws.

Anderson v. Westinghouse Savannah River Co., 406 F.3d 248 (4th Cir. 2005), cert. denied, ___ S. Ct. ___ (2006). Plaintiff in Title VII suit alleging racial discrimination in salaries and promotions offers testimony from Dr. Edwin L. Bradley, who opines based on statistical analysis that minorities score consistently lower on instrument used by employer to evaluate employees for promotions. District court excludes testimony and awards summary judgment to employer. Exclusion affirmed. Expert's analysis did not compare similarly situated employees, because it relied on broad occupational categories embracing very different types of job.

Runnels v. Tex. Children's Hosp. Select Plan, No. 04-20825 (5th Cir. Jan. 26, 2005) (unpublished). Title VII defendant in race discrimination suit moves for summary judgment, relying in part on statistical report from psychologist Dr. P.R. Jeanneret and on report on defendant's organization table and job descriptions from biomedical engineer Dr. Joseph Dyro. District court accepts magistrate's recommendation and awards summary judgment to defendant. Admissibility affirmed. Plaintiffs complain that Jeanneret's report did not list all documents he reviewed, but data used in his analyses were set forth. District court did not abuse discretion in finding that he employed standard and reliable statistical methods. Defendants also complain that both experts opined on ultimate questions, but that is not prohibited under current evidentiary rules.

Tyler v. Union Oil Co. of Cal., 304 F.3d 379 (5th Cir. 2002). In age discrimination suit, employees offer statistical testimony from Dr. Blake Frank, industrial/organizational psychologist, who opines that Unocal employees over age fifty were more likely to be subject to adverse employment decisions. Jury returns verdict for plaintiffs. Admissibility affirmed. Defendants' five specific methodological quarrels go to weight, not admissibility. In particular, expert could permissibly create his own database from data supplied by Unocal, and did attempt to address factors other than age that might affect employment decisions.

Munoz v. Orr, 200 F.3d 291 (5th Cir.), cert. denied, 531 U.S. 812 (2000). Does civilian employee promotion program at Kelly Air Force Base operate to discriminate against Hispanic males? District court excludes testimony from plaintiffs' statistician who so opines in disparate impact suit under Title VII. Exclusion affirmed. Witness committed gross mathematical errors, displayed bias in commencing analysis with assumption that promotion program discriminated against Hispanics, did not take education and experience into account as variables affecting promotion, and performed no regression.

Hopson v. DaimlerChrysler Corp., No. 04-2152 (6th Cir. Nov. 15, 2005) (unpublished). In support of his Title VII claim alleging racial discrimination, employee offers statistical testimony from John Sase, Ph.D. District court excludes testimony and awards summary judgment to employer. Exclusion affirmed. District court permissibly found that Sase knew nothing about employer's filling of jobs involved and that his statistical analysis did not take nondiscriminatory variables into account.

Cullen v. Ind. Univ. Bd. of Trs., 338 F.3d 693 (7th Cir. 2003) (see the briefs). To support her claim of salary discrimination, faculty member offers regression analysis performed by economics professor Paul Carlin, in which Carlin finds statistically significant gender differences, and also finds that faculty member's salary is more than one standard deviation below its predicted value. Carlin testifies he cannot rule out discrimination as the reason. University contends that faculty member's lower salary is explained by nondiscriminatory factors. District court admits regression study but awards summary judgment to university. Admissibility affirmed. University says that Carlin's regression study is inadmissible under Daubert, because factors like productivity are difficult to quantify. But under Bazemore v. Friday, 478 U.S. 385 (1986), omission of particular variables in regression analysis ordinarily goes to weight, not admissibility. However, regression study was insufficient to establish prima facie case.

Estate of Boncher v. Brown County, 272 F.3d 484 (7th Cir. 2001) (see the briefs). Man commits suicide in county jail. In resulting § 1983 action against county, suicide's estate offers testimony from reputable criminologist that jail's five suicides within five years is unusually high number. Trial court awards summary judgment for defendant. Admissibility reversed. Criminologist's testimony "was useless and should have been excluded under the Daubert standard." Relevant question would not be number of suicides, but suicide rate, as compared with background rate among persons residing in area but not incarcerated. Also, expert should have accounted for normal variance. It would not be sound to condemn jail administrators for suicide rates within one or two standard deviations from suicide rates at other jails. Every statistical distribution has upper tail, but jails unlucky enough to be in upper tail for inmate suicides should not automatically be subject to liability.

Kadas v. MCI Systemhouse Corp., 255 F.3d 359 (7th Cir. 2001) (see the briefs). Defendant obtains summary judgment in age discrimination case brought by one of three terminated employees. Affirmed. Plaintiff had no evidence to show discrimination. Witness who had himself sued for age discrimination and testified that company had age-discriminatory "culture" had obvious axe to grind, and his testimony was too vague and speculative to establish prima facie case. Also, company had independent noninvidious reason to terminate plaintiff, and plaintiff's statistical evidence actually favored company, because it tended to show likelihood that three terminated employees would be over forty even absent discrimination (since 27 out of company's 32 employees were over 40). This plaintiff, therefore, had no case, but court (by Posner, J.) wishes to seize opportunity to opine on issues of potential relevance in future discrimination cases: Some decisions elsewhere hold that statistical evidence is inadmissible to show discrimination unless it attains 95% confidence level, but no such rule applies in Seventh Circuit, because 95% confidence threshold is arbitrary. It is applied by scholarly journals to ensure that results are very probably not influenced by chance, but litigation is not fussy about evidence. Other forms of evidence (e.g., eyewitness testimony) are accepted without requiring showings of error rates under 5%. Low statistical significance may be attributable to noise in data, and high significance may be artifact of study design. Under Daubert, it is for judge to say, based on evidence from trained statisticians, whether particular significance level, in context of particular study in particular case, is too low for admissibility.

Chavez v. Illinois State Police, 251 F.3d 612 (7th Cir. 2001). In civil rights action alleging ethnic profiling by Illinois State Police, plaintiffs offer statistical analyses of traffic stops based on law enforcement databases, census data, and state records of licensed drivers. In recommending that summary judgment for defendants be granted on plaintiffs' equal protection claim, magistrate judge finds statistical evidence deficient because data on ethnicity has not been collected for every motorist stopped, and because such data as do reflect ethnicity do not constitute true random sample. Trial judge grants summary judgment without evaluating statistical evidence because statistical evidence would not sustain plaintiffs' claims even if valid. Two years later (other claims still having been pending), defendants move to strike plaintiffs' statistical experts under Daubert. Trial judge says it is unclear why defendants are raising Daubert only now and denies motion without prejudice. Plaintiffs appeal after remaining claims are dismissed. Summary judgment affirmed. Unnecessary to remand for consideration of plaintiffs' statistical evidence, because this statistical evidence cannot establish prima facie case, and because plaintiffs' statistical proof is (for reasons explained at length) neither relevant nor reliable. "[W]ithout reliable data on whom [defendant's] officers stop, detain, and search, and without reliable data indicating the population on the highways where motorists are stopped, detained, and searched, we can not find that the statistics prove that the [defendant's] officers' actions had a discriminatory effect on the plaintiffs."

Adams v. Ameritech Servs., Inc., 231 F.3d 414 (7th Cir. 2000). In "disparate treatment" age discrimination suit, district court excludes plaintiffs' statistical evidence and awards summary judgment to defendant employers. Exclusion reversed. Plaintiffs' statistical evidence may not have been sufficient by itself to show disparate treatment, but that does not render their statistical proof irrelevant or inadmissible -- "a brick is not a wall." Likewise, it may be "odd" that plaintiffs' statistician did not perform regression to isolate age as factor in termination decisions, but multiple regression is just one reliable statistical method, with problems of its own, and other reliable statistical methods exist. District court's exclusion of plaintiffs' statistical evidence must be reversed notwithstanding district court's "herculean" efforts to wade through voluminous record. Lengthy discussion of statistical proof in age discrimination cases.

Bone Shirt v. Hazeltine, No. 05-4010 (8th Cir. Aug. 22, 2006). Native American Indians bring voting rights claim, relying in part on expert regression evidence to show minority political cohesion and majority bloc voting. District court finds in their favor. Admissibility affirmed. Regression evidence is reliable and widely accepted in this context.

Marvin Lumber & Cedar Co. v. PPG Indus., Inc., 401 F.3d 901 (8th Cir. 2005). Marvin Lumber buys wood preservative known as PILT (preservative in-line treatment) from PPG, for use in Marvin Lumber's millwork products (doors and windows), on strength of PPG's representations that PILT will last longer than alternative preservatives containing pentachlorophenol, a.k.a. "Penta" (which poses environmental concerns). When PILT fails to prevent premature rot and decay in Marvin Lumber's products, it sues on claims including breach of warranty of future performance. To prove breach and causation, it offers testimony from statistician Dr. Frank Martin, who opines based on Marvin Lumber's wood-rot data that PILT-treated products do not outlast Penta-treated products, and also that company's wood-rot problems are attributable to PILT's deficiencies as preservative. Jury awards $156 million to Marvin Lumber. PPG appeals. Admissibility affirmed. PPG argues that Dr. Martin deviated from methods sanctioned in Reference Manual on Scientific Evidence (2d). But judges are not required to follow Reference Manual, which does not have force of law and is not intended to instruct judges on standards for admissibility. PPG also faults Dr. Martin because study was prepared for litigation, relied on data collected by Martin Lumber's legal department, relied on small sample not drawn from representative geographical cross-section, and failed to account for other potential causes of wood deterioration. But those points to go factual basis of Dr. Martin's opinions, and thus involve questions of weight, not admissibility. No abuse of discretion.

Morgan v. United Parcel Serv. of Am., Inc., 380 F.3d 459 (8th Cir. 2004), cert. denied,125 S. Ct. 1933 (2005). Plaintiffs in Title VII action allege pay discrimination, offering regressions performed by Dr. Hilary Weiner and Dr. David C. Stapleton. Defendants move to exclude plaintiffs' expert testimony and for summary judgment. Trial court finds that plaintiffs' regressions omitted material variables and awards summary judgment, denying Daubert motions as moot. Affirmed. Choice of variables in regression ordinarily goes to weight, not admissibility. But even assuming admissibility, plaintiffs' regressions were insufficient to support reasonable inference of discrimination.

Lanphere Enters. v. Jiffy Lube Int'l, Inc., No. 03-35617 (9th Cir. June 1, 2005) (unpublished). Trial court excludes plaintiff's survey evidence in trade disparagement case and awards summary judgment to defendant. Exclusion affirmed. District court permissibly excluded evidence as irrelevant or only marginally relevant, unduly prejudicial, and unreliable by virtue of its analytical gaps. No abuse of discretion.

Obrey v. Johnson, 03-16849 (9th Cir. Mar. 4, 2005). Plaintiff in Title VII suit offers expert testimony from statistician James Dannemiller to show that shipyard discriminates against Pacific-Americans in promotions. District court excludes testimony and jury finds for defendants. Exclusion reversed. Plaintiff's statistical evidence may have failed to account for disparities in qualifications, but that did not render it irrelevant or inadmissible.

United States v. Blaine County, 363 F.3d 897 (9th Cir. 2004), cert. denied, 544 U.S. 992 (2005). United States sues county under section 2 of Voting Rights Act, claiming that county's at-large election system impermissibly dilutes vote of American Indians. To show racially cohesive voting by American Indians in county, United States offers Dr. Theodore Arrington, who testifies over county's objection that in all fourteen county-wide elections he examined, American Indian voters exceeded 67% cohesion. United States also calls two other experts, again over county's objection. District court rules against county and enjoins is at-large election system. Admissibility affirmed. County complains that district court did not rule on its objections to experts' testimony. But district court did evaluate reliability of Dr. Arrington's testimony, and court's ruling on the merits did not rely on testimony from United States' other experts in any essential way. County also contends that Dr. Arrington's testimony was unreliable, because he relied on race-identified registration lists. But county's own expert testified that such reliance is customary and appropriate, and Dr. Arrington's and county expert's bivariate ecological regression analysis and homogenous precinct analysis yielded similar results.

Hemmings v. Tidyman's, Inc., 285 F.3d 1174 (9th Cir. 2002), cert. denied, 537 U.S. 1110 (2003). Two women are denied promotion to corporate management and sue for sex discrimination. District court admits testimony from their statistician, Dr. Nyak Polissar, and jury returns verdict for plaintiffs. Admissibility affirmed. Defendant says statistician erred in using comparison pool comprising defendant's management employees, because data set (a) was not representative of entire available pool of qualified labor and (b) improperly included store management employees as well as corporate management employees. But (a) data set was representative, because defendant promotes from internal ranks, and (b) defendant does sometimes promote store management personnel to corporate management; moreover, inclusion of store management was necessary to make gender comparisons. Defendant also says statistician did not eliminate other causal variables (e.g., variations in education or qualifications), but regression analysis need not achieve near-impossible goal of ruling out all other causal factors. Ordinarily, failure to include relevant variables will affect regression's weight, not its admissibility.

Cooper v. Southern Co., 390 F.3d 695 (11th Cir. 2004), cert. denied, 126 S. Ct. 478 (2005). Plaintiffs in Title VII suit allege racial discrimination. In opposing summary judgment, they offer statistical evidence from expert John Del Roccili. District court strikes expert's affidavit as sanction for noncompliance with disclosure requirements of Fed. R. Civ. P. 26(a)(2)(B) and awards summary judgment to defendants. Exclusion affirmed. Plaintiffs initially designated Del Roccili as fact witness. They later identified him as expert witness in discovery responses, but they never provided his expert report as required under Rule 26. District court did not abuse discretion in excluding his testimony as sanction under Fed. R. Civ. P. 37(c)(1).

Johnson v. DeSoto Board of County Commissioners, 204 F.3d 1335 (11th Cir. 2000). Defendant in voting discrimination suit presents expert who relies on voter registration figures to opine on county population increase following 1990 census. Admissibility affirmed. Plaintiffs say voting registration data are inherently unreliable as measure of voting age population and cannot be used to contradict census figures. But there is no per se rule against use of voter registration data, and evidence derived through statistically valid sampling techniques has previously been held admissible.

Statisticians