1 By the term empirical data, I refer to quantitative information gathered systematically for the purpose of characterizing one or more populations or phenomena. My use of the term in this paper excludes descriptive and anecdotal studies, i.e., those that do not attempt to reduce their observations to quantitative form. It also excludes evidence that may be characterized as expert or scientific, but is not rooted directly in quantitative data. Empirical research may result from investigations in either the natural or social sciences.

2 For an overview of the range of one type of empirical data — social science research findings — utilized in recent litigation, see J. MONAHAN & L. WALKER, SOCIAL SCIENCE IN LAW: CASES AND MATERIALS passim (1985).

3 See, e.g.. Walker, & Monahan, , Social Frameworks: A New Use of Social Science in Law, 73 VA. L. REV. 559 (1987)CrossRefGoogle Scholar; Monahan, & Walker, , Social Authority: Obtaining, Evaluating, and Establishing Social Science in Law, 134 U. PA. L. REV. 477 (1986)CrossRefGoogle Scholar.

4 See, e.g., U.S. v. Leon, 468 U.S. 897 (1984); Mississippi Univ. for Women v. Hogan, 458 U.S. 718 (1982); Ballew v. Georgia, 435 U.S. 223 (1978); Castenada v. Partida, 430 U.S. 482 (1977); Virginia State Bd. of Pharmacy v. Virginia Citizens Consumer Council, 425 U.S. 748 (1976).

5 See, e.g., Bersoff, , Social Science Data and the Supreme Court: Lockhart as a Case in Point, 42 AM. PSYCHOLOGIST 52 (1987)CrossRefGoogle Scholar; Bersoff, , Psychologists and the Judicial Systems: Broader Perspectives, 10 L. & HUM. BEHAV. 151 (1986)CrossRefGoogle Scholar; Levine, , The Adversary Process and Social Science in the Courts: Barefoot v. Estelle, 12 J. PSYCHIATRY & L. 147 (1984)CrossRefGoogle Scholar.

6 See, e.g., McCleskey v. Kemp, 107 S. Ct. 1756 (1987); Barefoot v. Estelle, 463 U.S. 880 (1983). Both cases are discussed in detail below.

7 For example, biased sampling techniques or experimental methods that failed to reproduce the real-life situation ostensibly under study.

8 463 U.S. 880 (1983).

9 TEX. CODE CRIM. PROC. ANN. § 37.071(b)(2) (Vernon 1988).

10 Sociopathy is properly referred to as “anti-social personality disorder” according to the official psychiatric nomenclature. See AMERICAN PSYCHIATRIC ASS'N, DIAGNOSTIC AND STATISTICAL MANUAL OF MENTAL DISORDERS 342-46 (rev. 3d ed. 1987). For a critique of the diagnostic process used by the psychiatrists in this case, see Appelbaum, , Hypothetical!, Psychiatric Testimony, and the Death Sentence, 12 BULL. AM. ACAD. PSYCHIATRY & L. 169 (1984)Google Scholar.

11 Brief Amicus Curiae for the American Psychiatric Association, Barefoot v. Estelle, 463 U.S. 880 (1983).

12 These studies are best summarized in J. MONAHAN, THE CLINICAL PREDICTION OF VIOLENT BEHAVIOR (1981).

13 See id. (note especially the table demonstrating false positive rates at 48). The best results were obtained by Kozol, , Boucher, , & Garofalo, , The Diagnosis and Treatment of Dangerousness 18 CRIME & DELINQUENCY 371 (1972)CrossRefGoogle Scholar.

14 The brief cited Woodsen v. North Carolina, 428 U.S. 280, 305 (1976) (plurality opinion). See also Radin, , Cruel Punishment and Respect for Persons: Super Due Process for Death, 53 S. CAL. L. REV. 1143 (1980)Google Scholar.

15 Barefoot v. Estelle, 463 U.S. 880, 896 (1982).

16 Id. at 901.

17 I am indebted to Professor Monahan, whose discussions of the Barefoot opinion clearly have influenced my analysis of the case.

18 Could one utilize the known inaccuracy of predictions of dangerousness to construct a paradoxical system of prediction? For example, might there be a point at which psychiatric prediction was sufficiently inaccurate (for instance, 9 out of 10 predictions of future violence being incorrect) that it would make sense to predicate a predictive system on doing just the opposite of what psychiatrists — or any other predictors recommend? Before jumping to this delightfully perverse conclusion, it might be useful to remember that the failure of predictive accuracy being discussed relates to predictions that a person will be violent. Given that violence is a low frequency event in almost all populations, predictions that a person will not be violent are highly accurate — even in the absence of good measures of violence potential. Thus, although on a statistical basis it might make sense to treat all those predicted to be violent as non-dangerous, it makes no sense to treat those predicted not to be violent as dangerous. The dilemma exemplified by the majority opinion in Barefoot derives precisely from an unwillingness to follow the first of these courses of action, namely to discount specific predictions of violence, although that would minimize overall inaccuracy.

19 Several variables have been shown to correlate with future violence, including a history of past violence, age, sex, race, socioeconomic status, employment and living stability, and opiate and alcohol abuse. J. MONAHAN, supra note 12, at 71-76. Yet none account for a portion of the variance substantial enough to be of material assistance to decision-makers, as demonstrated in actuarial studies that have attempted to base predictions solely on these and other nonclinical factors. Id. at 67-71.

20 Barefoot v. Estelle, 463 U.S. 880, 921 (1982).

21 Id. at 928.

22 Id. at 929-30.

23 See Perlin, , The Supreme Court, the Mentally Disabled Criminal Defendant, and Symbolic Values: Random Decision, Hidden Rationales, or “Doctrinal Abyss”. 29 ARIZ. L. REV. 1, 71 (1987)Google Scholar.

24 107 S. Ct. 1756 (1987).

25 I am grateful to Professor Baldus for providing me with drafts of the report of his studies on discrimination and arbitrariness in Georgia's capital charging and sentencing system; BALDUS, WOODWORTH & PULASKI, DISCRIMINATION AND ARBITRARINESS IN GEORGIA's CAPITAL CHARGING AND SENTENCING SYSTEM: A PRELIMINARY REPORT (July 29, 1983) [hereinafter DISCRIMINATION] and for parts of his forthcoming book which analyzes the decision in McCleskey; D. BALDUS, G. WOODWORTH & G. PULASKI, EQUAL JUSTICE AND THE DEATH PENALTY (Ch. 11, July 23, 1987; chs. 12-13, Sept. 3, 1987) [hereinafter EQUAL JUSTICE].

26 McCleskey, 107 S. Ct. 1763.

27 Id. at 1763-64.

28 Id. at 1764. The summaries of the Baldus data in the majority and dissenting opinions only hint at the variety of analyses applied by the research group. The researchers explored the relationships of both race of victim and race of defendant with the imposition of the death penalty, finding both influences operative, but the former consistently present and the latter inconsistently so. Racial disparities were observed after controlling (in least squares multiple regression and logistic multiple regression procedures) for presumed aggravating and mitigating circumstances, statutorily defined aggravating factors, defendants’ prior record, and up to 230 background variables. In addition, sorting of the cases by degree of aggravation demonstrated that the effects of racial factors were greatest in the middle to upper range of cases, where the offense was neither so severe that the death penalty could be expected as a matter of course, nor so mitigated that its imposition was unlikely in any event. EQUAL JUSTICE, supra note 25, ch. 11 at 11-11 to 11-25.

29 McCleskey, 107 S. Ct. at 1766 n.7.

30 Id.

31 Id. at 1778.

32 The Court cited as examples Castenada v. Partida, 430 U.S. 482 (1977); Turner v. Fouche, 396 U.S. 346 (1970); and Whitus v. Georgia, 385 U.S. 545 (1967).

33 The Court cited Bazemore v. Friday, 106 S. Ct. 3000 (1986).

34 McCleskey, 107 S. Ct. at 1767-68. Justice Powell also offered two non-statistical arguments in an effort to distinguish the current case: first, that there is no common standard by which variations in the application of the death penalty can be compared; and second, that the state lacks the means to rebut statistical arguments when the decision-makers (i.e., jurors and district attorneys) either are unavailable to testify or would not be able to recall the circumstances of the case. Id. These arguments are susceptible to fairly persuasive rebuttal. See EQUAL JUSTICE, supra note 25, Ch. 13 at 13-4 to 13-7.

35 The key logistic regression analysis examining the effect of race while controlling for thirty-nine of the most relevant variables, for example, found defendants who had murdered white victims were 4.3 times more likely to be put to death than those who had murdered blacks. This finding was significant at the .005 level, indicating that the likelihood that this was a chance finding is 1 in 200. The generally accepted level of significance is .05; when an association is likely to occur by chance only 1 out of 20 times scientists will generally accept it as proven. An alternative analysis of this comparison utilized multiple linear regression and found an association between race of victim and the death sentence that was significant at the .001 level. There is only 1 chance in 1000 that this was a random finding, unrelated to a real association between race and outcome. EQUAL JUSTICE, supra note 25, ch. 11 at 11-13 to 11-14. Further, the regression analyses demonstrate that none of the other included variables account for the race-sentencing association.

36 This is certainly true for such purported safeguards as providing the jury with statutorily mandated mitigating and aggravating criteria. The situation is a bit more complex for the protections afforded by mandatory review by the state supreme court. Although the Georgia Supreme Court did not overturn any of the cases in the Baldus studies on the grounds of bias in sentencing, 27% of the cases were overturned on procedural grounds. Although exclusion of these cases from the analyses substantially reduces the racial disparities found in the larger study, the associations between race of victim and race of defendant and the death penalty remain statistically significant in all but one of the reanalyses performed. No decrement in effect was seen when similar analyses were conducted on the sample from the smaller study. DISCRIMINATION, supra note 25, at 98-99, Tables 61 and 62.

37 McCleskey, 107 S. Ct. at 1803 (quoting Arlington Heights v. Metropolitan Hous. Dev. Corp., 429 U.S. 252, 266 (1977)).

38 Fisher, , Multiple Regression in Legal Proceedings, 80 COLUM. L. REV. 701 (1980)CrossRefGoogle Scholar.

39 See, e.g., supra note 4.

40 Justice Blackmun's unique contributions to empirical jurisprudence during his early years on the Supreme Court bench are reviewed in Schlesinger, & Nesse, , Justice Harry Blackmun and Empirical Jurisprudence, 29 AM. U. L. REV. 405 (1980)Google Scholar. See also Note, The Changing Social Vision of justice Blackmun, 96 HARV. L. REV. 717 (1983). He authored the majority opinions in, among others, Ballew v. Georgia, 435 U.S. 223 (1978); Castenada v. Partida, 430 U.S. 482 (1977); and Virginia State Bd. of Pharmacy v. Virginia Citizens Consumer Council, 425 U.S. 748 (1976).

41 See, e.g., Lockhart v. McCree, 106 S. Ct. 1758 (1986). Lockhart is controversial because the decision goes against the thrust of the available empirical data. Bersoff (1987), supra note 5. Nonetheless, the Court's efforts to weigh an admittedly problematic body of data differ substantially from its approach in Barefoot and McCleskey.

42 Bersoff (1987), supra note 5.

43 Sperlich, , Social Science Evidence and the Courts: Reaching Beyond the Adversary Process, 63 JUDICATURE 280 (1980)Google Scholar.

44 Justice Blackmun has had the advantage of having been an undergraduate mathematics major at Harvard, where he graduated summa cum laude. Schlesinger & Nesse, supra note 40.

45 Ballew v. Georgia, 435 U.S. 223, 246 (1978) (Powell, J., concurring).

46 P, ROSEN, THE SUPREME COURT AND SOCIAL SCIENCE (1972), passim. Clark's study involved asking black children to choose between black and white dolls in response to instructions to select the doll “you like best,” “that is the nice doll,” “that is a nice color,” “that looks bad.” The children overwhelmingly selected white dolls for the first three conditions and black for the fourth. Clark explained the data as indicating a loss of self-esteem in black children. Both the adequacy of the methods and the bearing of the findings on the school segregation issue have been challenged. See also J. MONAHAN & L. WALKER, supra note 2, at 83-93.

47 O'Brien, , The Seduction of the Judiciary: Social Science and the Courts, 64 JUDICATURE 8 (1980)Google Scholar.

48 Youngberg v. Romeo, 457 U.S. 307 (1982).

49 Similar suggestions have been made in the past, see e.g., Sperlich, supra note 43; Leventhal, , Environmental Decision-making and the Role of the Courts, 122 U. PA. L. REV. 509, 550-54 (1974)CrossRefGoogle Scholar. But see Jasanoff, & Nelkin, , Science, Technology and the Limits of Judicial Competence, 214 SCIENCE 1211 (1981)CrossRefGoogle Scholar.

50 Weiss, , The Diffusion of Social Science Research to Policymakers: An Overview, in REFORMING THE LAW: IMPACT OF CHILD DEVELOPMENT RESEARCH 63, 80 (Melton, G.B. ed. 1987)Google Scholar.

51 Melton, Judicial Notice of “Facts“ About Child Development, in id. at 232, 244-45 (suggesting consultation with experts may be sanctioned by the Code of Judicial Conduct, Canon 3(A)(4)); Leventhal, supra note 49, at 549-50 (noting Fed. R. Civ. P. 53 permits appointment of masters who may serve expert consultative function); Sperlich, supra note 43, at 288 (pointing to Fed. R. Evid. 706 as allowing courts to appoint their own expert witnesses).