Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Law, Probability and Risk (2009) 8, 85−94 Advance Access publication on July 7, 2009 doi:10.1093/lpr/mgp018 Identification, individualization and uniqueness: What’s the difference?† DAVID H. K AYE * Distinguished Professor and Weiss Family Scholar, Penn State’s Dickinson School of Law, Lewis Katz Bldg, University Park, PA 16802, USA [Received on 8 April 2009; revised on 5 June 2009; accepted on 15 June 2009] Criminalists and many forensic scientists concerned with the identification of trace evidence have distinguished between identification and individualization, but they have not distinguished as precisely between individualization and uniqueness. This paper clarifies these terms and discusses the relationships among identification, individualization and uniqueness in forensic science evidence. Keywords: Identification; individualization; uniqueness; forensic science; trace evidence; DNA; fingerprints. 1. Introduction Problems of identity and uniqueness persist. Heraclitus wondered about stepping into a rushing river. Is the river the same each time we step into it? Plutarch presented the ancient puzzle of the ship of Theseus. Is the ship whose planks have been replaced over the years, one at a time, the original ship or a new one? Forensic scientists ask more mundane questions about the identity of objects, but do they have better answers? According to two distinguished forensic scientists, the field has ‘struggled with the concept of uniqueness, but then so have others—philosophers, logicians, Boolean algebra mathematicians and rare coin dealers. Each of these groups has made an uneasy peace with the concept’.1 In some subfields of forensic science, this peace seems philosophically, logically and mathematically fragile. Too many practitioners and theorists of the arts of comparing toolmarks, bitemarks, fingerprints, fibers, handwriting and manufactured goods toss out terms like ACE-V,2 ridgeology,3 † Presented at a workshop held at George Washington University, August 1st 2009, in honour of the 70th birthday of Joe Gastwirth, one of the founding editors of Law, Probability and Risk. * Email: [email protected] 1 John I. Thornton & Joseph L. Peterson, The General Assumptions and Rationale of Forensic Identification, in 4 M ODERN S CIENTIFIC E VIDENCE : T HE L AW AND S CIENCE OF E XPERT T ESTIMONY § 29:15, at 15 (David L. Faigman et al. eds., 2008–2009). 2 ACE-V stands for ‘Analysis, Comparison, Evaluation, followed by Verification’. See David H. Kaye, The Nonscience of Fingerprinting: United States v. LleraPlaza, 21 Q UINNIPIAC L. R EV. 1073, 1080 (2003). It is ‘a generic protocol used by trained examiners to conduct forensic comparisons using visual recognition’, Christophe Champod, Fingerprint Examination: Towards More Transparency, 7 L., P ROBABILITY & R ISK, 111 (2008), ‘the common sense description of what anyone would do if they were examining a latent and a candidate source print’—in short, ‘an acronym, not a methodology’. Sandy L. Zabell, Fingerprint Evidence, 13 J.L. & P OL’ Y 143, 178 (2005). 3 See, e.g. Kaye, supra n. 2. c The Author [2009]. Published by Oxford University Press. All rights reserved. 86 D. H. KAYE class characteristics4 and, most tellingly, individual characteristics5 as if these represent well-defined methodologies or valid constructs. Rather than conduct the difficult empirical research that would needed to establish that these objects or impressions are traceable to a single common source they resort to such tautologies as ‘[t]he uniqueness of an object may be established by an ensemble of individual [as opposed to class] characteristics’.6 Blithely postulating uniqueness, forensic science textbooks contain advice such as the following: • A positive identification is a conclusion that a particular shoe, and no other shoe, made the crime scene impression. No minimum number of individual identifying characteristics is needed to establish a positive identification.7 • If the suspect cannot be ruled out because of the uniqueness of teeth and biting pattern, the suspect has to be considered the biter.8 • These characteristics, if present in both the questioned impression and the known tire, make that tire unique and allow for positive identification of that tire as having made the questioned impression.9 • By the late teenage years, a person’s writing has matured to the point where his or her writing style is unique . . .. This individualization is a basic principle in document examinations. 10 These theories of uniqueness and individualization are more than lingering curiosities. They determine the course of people’s lives. Day in and day out, in case after case, the testimony of criminalists reflects this paradigm of positive, uniquely specific identification. Commentators from other disciplines have complained bitterly. Some have called for a ‘paradigm shift’11 that would replace talk of individualization with statements of probabilities12 or have demanded the exclusion of certain testimony pending better research on the ability of analysts to perform as claimed.13 Indeed, 4 ‘Class characteristics are general characteristics that separate a group of objects from a universe of diverse objects . . .. Class objects do not, and cannot, establish uniqueness’. Thornton & Peterson, supra n. 1, at § 29:7, at 8. 5 ‘Individual characteristics, on the other hand, are those exceptional characteristics that may establish the uniqueness of an object.’ Ibid. 6 Ibid. 7 William J. Bodziak, Forensic Footwear Evidence, in F ORENSIC S CIENCE : A N I NTRODUCTION TO S CIENTIFIC AND I NVESTIGATIVE T ECHNIQUES 297, 309 (Stuart H. James & Jon J. Nordby eds., 2003); see also ibid. at 298. 8 R. Tom Glass, Forensic Odontology, in James & Nordby, supra n. 7, at 61, 73. 9 William J. Bodziak, Forensic Tire Impression and Tire Track Evidence, in James & Nordby, supra n. 7, at 313, 325. 10 Frank H. Norwich & Howard Seiden, Questioned Documents, in James & Nordby, supra n. 7, at 357, 359. Some authors are slightly more reserved. For example, Thomas A. Kubic & Nickolas Petraco, Microanalysis and Examination of Trace Evidence, in Ibid. at 251, 252, 264, 273. 11 Michael J. Saks & Jonathan J. Koehler, The Coming Paradigm Shift in Forensic Identification Science, 309 S CIENCE 892 (2005). 12 Michael J. Saks & Jonathan J. Koehler, The Individualization Fallacy in Forensic Science Evidence, 61 VAND . L. R EV. 199, 217, 218 (2008) (advocating ‘a frank reliance on probability’ in the future, but recognizing that ‘[i]n some areas . . . producing sound probability estimates may be particularly difficult . . .’.). 13 For example, Simon A. Cole, Does ‘Yes’ Really Mean Yes? The Attempt to Close Debate on the Admissibility of Fingerprint Testimony, 45 J URIMETRICS J. 449 (2005); Lyn Haber & Ralph N. Haber, Experiential or Scientific Expertise, L., P ROBABILITY & R ISK 143 (2008); D. Michael Risinger & Michael J. Saks, Science and Nonscience in the Courts: Daubert Meets Handwriting Identification Expertise, 82 I OWA L. R EV. 21 (1996); Saks & Koehler, supra n. 12, at 218 (concluding that ‘claims [of individualization that] exaggerate the state of their science . . . would be a prime target for exclusion’). IDENTIFICATION, INDIVIDUALIZATION, UNIQUENESS: WHAT’S THE DIFFERENCE 87 some have written or implied that because assertions of uniqueness of physical objects are metaphysical or not directly provable, all claims of individualization are fallacious.14 This paper re-examines the related concepts of identification, uniqueness and individualization. Section 2 explains that individualization is one type of identification and that it can take two forms— ‘universal’ and ‘local’. It also suggests that the prominent distinction between class characteristics and individual characteristics is not easily drawn in practice. Section 3 unpacks the meaning of ‘uniqueness’. It argues that uniqueness must be defined relative to a particular set of items and emphasizes the distinction between the uniqueness of a single item (‘special uniqueness’) and the uniqueness of all items with respect to the set (‘general uniqueness’). Section 4 applies these ideas to DNA evidence and fingerprints. It argues that because special uniqueness in a proper subset of all objects in existence is sufficient for individualization within that subset, testimony that a specific individual probably is the source of a DNA sample or a fingerprint may be justified in at least some cases. 2. Identification and individualization Forensic scientists are sensitive to the breadth of the word ‘identification’. It can mean classification, as in ‘I identified the specimen as an Ursus horribilis’ or ‘an automobile as a red Buick’.15 It also can mean ‘individualization’, as when a fingerprint analyst testifies that ‘I identified the latent fingerprint as having been made from the right ring finger of the defendant’.16 In the latter situation, ‘the term “individualized” would be more felicitous’.17 But the term ‘individualization’ admits further distinctions. Does the individualization mean that, no matter how many people could be considered as possible suspects, the analyst confidently can state that the defendant is the source of the latent print? Or does it mean that with respect to a smaller set of possible perpetrators, the defendant is far and away the most probable source? To highlight this distinction, we can call the former identification an instance of universal individualization and the latter an assertion of local individualization.18 As noted above, forensic scientists distinguish between ‘class characteristics’ and ‘individual characteristics’,19 but all identifications are classifications. Some of the classes are simply larger than others. The larger the class, the less discriminating the identification, but all such associations provide relevant information. A Bayesian perspective brings these two ideas together. Whether the individual is the source of the trace evidence depends on the prior odds and the likelihood ratio. The prior odds are influenced by the size of the plausible population of suspects, while the likelihood ratio depends, at least in 14 Saks & Koehler, supra n. 12. 15 Thornton & Peterson, supra n. 1, § 29:10, at 11. 16 Ibid. 17 Ibid. 18 Forensic scientists recognize a related, but slightly different distinction when they speak of ‘open’ and ‘closed sets’. See, e.g., John Buckleton et al., Disaster Victim Identification, Identification of Missing Persons, and Immigration Cases, in F ORENSIC DNA E VIDENCE I NTERPRETATION 395, 431 (John S. Buckleton et al. eds., 2005); Philip Rose, F ORENSIC S PEAKER I DENTIFICATION 84 (2002). With an ‘open set’, the number of possible suspects is unknown, and an assertion of identity is an instance of universal individualization. With a ‘closed set’, the number and features of all possible suspects are known, and an assertion of identity is an instance of local individualization. The murder-on-the-yacht example presented below straddles this distinction, for it involves a known set of possible suspects with an unknown feature for one of them. It produces an assertion of local individualization even though the set of suspects is ‘open’ to the extent that one remains to be examined. 19 See, e.g. Ibid., § 29:7, at 8 (describing the distinction and its use in the process of comparison). 88 D. H. KAYE part, on the size of the class that shares the identifying characteristics. For instance, about 4% of the population in the UK20 and the USA21 has type AB blood. Finding a type AB bloodstain at the scene of a violent crime when the victim is type O and the suspect is type AB therefore is 25 times more probable under the hypothesis S that the suspect is the source of the stain than it is under the hypothesis S C that someone unrelated to the suspect is. Hence, the likelihood ratio involving the match M is L = P(M|S )/P(M|S C ) = 25, (1) and the identification has increased the prior odds by this factor of 25. Although this likelihood ratio means that M is relevant and has definite probative value,22 M hardly amounts to universal individualization. That type of individualization occurs only in the limit, as 1/L approaches 0 (L → ∞). It is an idealization that, strictly speaking, is never realized. DNA typing with a substantial number of short tandem repeat (STR) loci produces a very large value for the likelihood ratio for the hypothesis that the suspect is the source of the bloodstain, but the number is finite. Nonetheless, as a practical matter (when the probability that an identical twin was responsible or that laboratory or handling error has occurred is close to 0), it becomes reasonable to believe that the extensive multilocus match indicates that the suspect is the source of the stain.23 The likelihood ratio is so large that it swamps any realistic prior probability. This pragmatic form of universal individualization may occur with fingerprints as well. Local individualization might be attainable with the ‘class characteristic’ of the ABO blood system if the population of plausible suspects were small. Suppose that a dead body (with type O blood) is found in a cabin of a yacht. Ten other unrelated people were on board. There are signs of a struggle and a violent death. Blood drops, determined to be of type AB, lead towards the door. The 10 people disappear after making port. Medical records of nine of them are located. These show that only one has blood group AB. He and the individual with the unknown ABO type now are the only viable suspects. If we assume that, before considering the blood group evidence, the two suspects were equally likely to be the source of the bloodstain, then the prior odds are 1:1.24 The blood group evidence increases the odds as follows: Odds(S|M ) = L × Odds(S ) = 25:1. (2) If a posterior probability of 25/26 = 96.15% is good enough for an attribution to an individual, then the serological data on a class characteristic have produced local individualization. Seen in this light, individualization is a statement that, like any other empirical claim, could be wrong. The epistemological question is whether the probability that the statement is correct is high 20 National Blood Service for England & North Wales, All about Your Blood Types, <http://www.blood.co.uk/pages/ all about.html> (Mar. 8, 2009). 21 Stanford Medical School Blood Center, Blood Types in the U.S., <http://bloodcenter.stanford.edu/about blood/blood types.html> (Mar. 10, 2009). 22 For example, DAVID H. K AYE ET AL ., T HE N EW W IGMORE : A T REATISE ON E VIDENCE : E XPERT E VIDENCE (2004); David H. Kaye & J.J. Koehler, The Misquantification of Probative Value, 27 L. & H UMAN B EHAVIOR 645 (2003). 23 See I AN W. E VETT & B RUCE S.W EIR , I NTERPRETING DNA E VIDENCE : S TATISTICAL G ENETICS FOR F ORENSIC S CIENTISTS 241 (1998) (suggesting that because the DNA analyst can supply the value of the likelihood ratio, it is not necessary to give ‘individualization’ testimony for DNA—the jury can round off 1/L to 0 on its own). 24 On the selection of prior odds, see, e.g. A. Biedermann et al., Equal Prior Probabilities: Can One Do Any Better? 172 F ORENSIC S CI . I NT ’ L 85 (2007); Kaye et al., supra n. 22; David H. Kaye, Rounding Up the Usual Suspects: A Legal and Logical Analysis of DNA Database Trawls, 87 N. C AR . L. R EV. 425 (2009). IDENTIFICATION, INDIVIDUALIZATION, UNIQUENESS: WHAT’S THE DIFFERENCE 89 enough to warrant the claim. This probability will turn on how discriminating the feature set is and the size of the population of objects. A provocative hypothetical case devised by Michael Saks and Jay Koehler to show the difficulty of demonstrating that a feature set takes on unique values in a large population also illustrates the finite chance of error in an individualization. They posit a case in which: [E]xactly 100 pairs of firearms out of an estimated 100,000 guns in a Texas town share indistinguishable gun barrel markings. If each of 100 firearms experts examined 10 pairs of guns from the town’s gun population every day for 10 years (n = 3,650,000 gun pairs), there is about a 93% chance that none of the indistinguishable pairs will have come under examination. That is, despite 1,000 “collective years” of forensic science experience (100 experts multiplied by 10 years), the failure to find even a single pair of guns with indistinguishable markings would offer little basis for drawing conclusions about whether gun barrel markings, even in this single town, are unique.25 The example treats the firearms examiners as drawing a simple random sample of n = 3 650 000 pairs of guns out of a population of all pairs of the N = 100 000 guns in the town during the 10-year period.26 This population consists of N (N − 1)/2, or about 1010 /2, possible gun pairings. The examiners have observed X = 0 matches in n trials. The binomial probability θ is the chance of stumbling on 1 of the 100 matching pairs in any one comparison: θ = 100/(1010 /2) = 2×10−8 . The probability of finding none of the special pairings of matching guns (X = 0) is therefore (1 − θ)n , which is approximately 1 − nθ = 92.6%, as asserted. Now let’s look at this as a problem in statistics. The observed number of matches, namely, X = 0 (for n = 3 650 000 tries), summarizes the sample data. Under the null hypothesis that θ = 0, the probability of data at least as surprising as the observed value X = 0 is 1. We cannot reject the null hypothesis of uniqueness. Sak’s and Koehler’s point is that we also have ‘little basis’ for accepting the hypothesis. Since a single match suffices to reject the null hypothesis that θ = 0, the critical region is X > 1. If the true number of matching pairs in the population of all pairs is not 0, but 100, then using the calculation of the preceding paragraph, the chance of a rejection is 1 − 0.926 = 0.073. The 10-year effort has very little power (about 7%) to reject the null hypothesis even when it is false (in the way they postulate, namely, θ = 2 × 10−8 ). What does this example tell us about the chance of a false individualization? The hypothetical does not state that 100 guns out of the 100000 in town always will be indistinguishable. It refers to 100 pairs, and the analysis basically establishes that it is hard to distinguish the small binomial probability of θ = 2 × 10−8 from θ = 0 even with a sample of nearly 4 million pairs. The number of guns (not pairs of guns) whose bullets could be confused with another gun is between 21 and 200.27 Therefore, a firearms expert who declares individualization when a bullet matches in the Texas town 25 Saks & Koehler, supra n. 12, at 212–13. 26 Presumably, the examiners put the pair of guns that they test back into the pool for future testing; otherwise, they could have exhausted the town’s supply of guns before doing millions of tests. Saks and Koehler do not specify whether the examiners also keep track of which particular pairs that the examiners have sampled so that they do not do test the same pair twice. Sampling without replacement in this fashion introduces a dependence in the outcomes, but the effect is negligible when n is much smaller than 1/θ . See, e.g. Frederick Mosteller, Understanding the Birthday Problem, 55 M ATH . T EACHER 322 (1962). 27 The maximum number of nonunique guns, 200, occurs when the all 100 pairs are distinct. The minimum of 21 nonunique guns occurs when the separately indistinguishable clusters are of size 14 (producing 91 pairs) + 4 (for 6 pairs) + 3 (3 pairs). I am grateful to Jay Koehler for this analysis. 90 D. H. KAYE will be correct with a probability in the range of 1 − 200/100 000 = 99.50% to 1 − 21/100 000 = 99.98%. Even though general uniqueness does not exist, an individualization is very likely to be correct. 3. Special and general uniqueness One might think that a claim of individualization is equivalent to an assertion of uniqueness, but we have just seen that it is not. If an object is discernibly unique, it can be individualized; however, the converse is not necessarily true. Individualization—in the sense of an almost certain source attribution—does not imply uniqueness—in the sense of a set of features that are different in every member of a set of objects.28 In the Texas gun example, the vast bulk of the hypothetical 100 000 guns leave unique impressions on bullets, but a small fraction—no more than 1 in 500—do not. Because the set of guns can be partitioned into subsets of unique and nonunique items, and because the unique items vastly the nonunique ones, individualization exists in the Texas gun case. To generalize this idea, and to be clear about why testimony that a particular item is unique is a much weaker claim than testimony that all items are unique, some definitions will be useful. Let γ refer to an item (such as a human finger) that leaves traces with distinctive features (such as fingerprints). How distinctive the traces are remains to be seen. Let Γ be a set of items of this sort (such as all human fingers). Let yi be a vector that describes the variable features on the traces left by γi , the ith such item. A particular feature vector yi marks a unique element γi (with respect to Γ ) if, for all γ j in Γ, yi 6= y j (when i 6= j). We can call this situation special uniqueness. The element, γi , is the only one in Γ that produces the mark yi . If Γ is the set of all γ in the universe, then γi is unique in the universe. This is ‘universal special uniqueness’. Other γs also might be specially unique in Γ , but they need not be. General uniqueness is special uniqueness for all the γs in Γ . It means that every feature vector yi can be mapped back to the one and only one γi capable of generating it. In other words, the elements of Γ are ‘generally unique’ if for all γi and γ j in Γ, yi 6= y j (when i 6= j). If Γ is the set of all γs in the universe, then these items are ‘universally generally unique’. There are no duplicates in the universe. Fingerprint examiners traditionally insist that fingerprints from the same finger are unique in this very strong sense, but the claim is remarkably hard to prove.29 The Section 4, which discusses DNA evidence, indicates why. 4. The uniqueness and individualization of DNA profiles Biologists accept and expound the proposition that the full sequence of more than 6 billion base pairs in the human diploid genome is unique to each individual (and any clones). For example, in 1992, a committee of the National Academy of Sciences recommended that ‘[c]ourts should take judicial notice of [the] scientific underpinnings of DNA typing’—including the fact that ‘[e]ach person’s DNA is unique (except that of identical twins) . . .’.30 Like the fingerprint community, geneticists believe in universal general uniqueness. 28 Contra Thornton & Peterson, supra n. 1, § 29:10, at 11 (‘Individualization implies uniqueness . . .’.). 29 See, e.g. David H. Kaye, From Snowflakes to Fingerprints: A Dubious Courtroom Proof of the Uniqueness of Finger- prints, 71 I NT ’ L S TAT. R EV. 521 (2003). 30 NATIONAL R ESEARCH C OUNCIL C OMMITTEE , DNA T ECHNOLOGY IN F ORENSIC S CIENCE 23 (1992). IDENTIFICATION, INDIVIDUALIZATION, UNIQUENESS: WHAT’S THE DIFFERENCE 91 Of course, sequencing the entire genome is not feasible. The immediate issue for DNA identification is whether the fewer than 20 loci used in STR typing—involving on the order of a 1000 base pairs—are sufficient for individualization. This section suggests that (1) they might produce universal general uniqueness; (2) they probably produce universal special uniqueness; and (3) they almost certainly produce special uniqueness in the populations that are relevant in some criminal cases. 4.1 Universal general uniqueness In 1996, a committee of the U.S. National Academy of Sciences suggested that ‘[w]ith an increasing number of loci available for forensic analysis, we are approaching the time when each person’s profile will be unique (except for identical twins and other close relatives)’.31 Its report on forensic DNA technology distinguished between special and general claims of uniqueness for VNTR loci.32 A specific profile might be unique: ‘Suppose that in a population of N unrelated persons, a given DNA profile has probability P. The probability (before a suspect has been profiled) that the particular profile observed in the evidence sample is not unique is at most NP’.33 A small probability NP indicates that the one profile under consideration is likely to be unique within a population that contains as many as N unrelated people. Being conditioned on a given genotype, this is special uniqueness. General uniqueness, as we have defined it, refers to all the profiles in the population. In this regard, the committee discussed the famous Birthday Problem34 and wrote that ‘[a] lower bound on the probability that every person is unique depends on the population size, the number of loci, and the heterozygosity of the individual loci’.35 With some simplifying assumptions, the probability of this event also can be estimated. ‘Neglecting population structure and close relatives, 10 loci with a geometric mean heterozygosity of 95% give a probability greater than about 0.999 that no two unrelated people in the world have the same profile’.36 4.2 Universal special uniqueness Even if STR profiling at 13 or so STR loci (currently the norm in forensic work) does not establish a high enough probability of universal general uniqueness, it still might be possible to state that particular genotypes are very likely to be unique in the world. Probabilities on the order of 10−23 31 C OMMITTEE ON DNA F ORENSIC S CIENCE : A N U PDATE , NATIONAL R ESEARCH C OUNCIL , T HE E VALUATION OF DNA E VIDENCE 161 (1996) [hereinafter NRC 1996]. 32 VNTR stands for ‘variable number of tandem repeats’. VNTR alleles come in a wide range of sizes due to the variations in the number of core sequences of base pairs that are stacked one after the other, like freight trains with different numbers of boxcars. No single locus is unique in the human population, but allelic combinations from several loci can be very rare (depending on which alleles are present). 33 NRC 1996, supra n. 31, at 161. 34 Ibid. at 165. In its simplest form, the Birthday Problem assumes that equal numbers of people are born every day of the year. The problem is to determine the minimum number of people in a room such that the odds favour there being at least two of them who were born on the same day of the same month. Here, the problem is to determine, in a population of size n, the probability that no one’s DNA type matches anyone else’s. If all DNA profiles have the same probability θ , and if profiles are independent, then the probability that there are not two or more instances of any profile in a set of n profiles is approximately exp(–n 2 θ/2). 35 Ibid. 36 Ibid. The computation, ‘an application of the ”Birthday Problem” with unequal probabilities’, can be found in appendix 5C of the report. More STR loci (the type currently used in DNA identification) would be required to achieve the same probability of uniqueness. See Ibid. at 165; see also Kaye et al., supra n. 22, § 12.5.3. 92 D. H. KAYE for random matches to unrelated individuals have been quoted in some court cases.37 If one could be confident that the random match probability of a particular profile really is this small, then it would be fair to conclude that the profile is unique among everyone in the world who is not a close relative of the individual who is known to have this profile. This is universal special uniqueness. Because a figure like 10−23 exceeds the precision that can be justified by existing data and because suspect populations include relatives, however, this claim of universal special uniqueness for STR loci remains theoretical. 4.3 Local uniqueness as the basis for local individualization Federal Bureau of Investigation (FBI) examiners, focusing on special uniqueness, have testified to source identifications in cases for which they consider the duplication probability for a particular profile in the U.S. population to be quite small.38 National Academy of Science report, issued this year, seems comfortable with such testimony.39 The criteria for individualization have been described as follows: Rather than ask whether a profile probably is unique in the world’s population, the examiner focuses on smaller populations that might be the source of the evidentiary DNA. When the surrounding evidence does not point to any particular ethnic group, the analyst takes the random-match probability and multiplies it by ten (to account for any uncertainty due to population structure). The analyst then asks what the probability of generating a population of unrelated people as large as that of the entire U.S. (290 million people) that contains no duplicate of the evidentiary profile would be. If that “no-duplication” probability is 1% or less, the examiner must report that the suspect “is the source of the DNA obtained from [the evidentiary] specimen . . ..” Similarly, the FBI computes the no-duplication probability in each ethnic or racial subgroup that may be of interest. If that probability is 1% or less, the examiner must report that the suspect is the source of the DNA. Finally, if the examiner thinks that a close relative could be the source, and these individuals cannot be tested, standard genetic formulae are used to find the probability of the same profile in a close relative, that probability is multiplied by ten, and the resulting no-duplication probability for a small family (generally ten or fewer individuals) is computed. Once again, if the no-duplication probability is no more than one percent, the examiner reports that the suspect is the source.40 This approach, although debatable in some particulars, illustrates how individualization is possible even with feature sets that do not uniquely identify all objects in existence. If one defines the relevant population more narrowly (‘everyone in Boston’ or ‘everyone in New York’, for example), 37 People v. Nelson, 185 P.3d 49, 52 (Cal. 2008); David H. Kaye, The Role of Race in DNA Evidence: What Experts Say, What California Courts Allow, 34 S W. L. R EV. 303 (2008). 38 Roberto Suro, DNA Now Used To Make Specific Identification; FBI Calls Lab Match ‘Major Breakthrough’, Wash. Post, Nov. 13, 1997, at A4. 39 NATIONAL R ESEARCH C OUNCIL C OMMITTEE ON I DENTIFYING THE N EEDS OF THE F ORENSIC S CIENCE C OM MUNITY, S TRENGTHENING F ORENSIC S CIENCE IN THE U NITED S TATES : A PATH F ORWARD 3-2 (2009) (prepublication copy) (‘no forensic method other than nuclear DNA analysis has been rigorously shown to have the capacity to consistently and with a high degree of certainty support conclusions about “individualization”. . .’). 40 David H. Kaye & George Sensabaugh, Reference Guide on DNA Evidence, in R EFERENCE M ANUAL ON S CIENTIFIC E VIDENCE 485, 548 n. 282 (2d ed., 2000). IDENTIFICATION, INDIVIDUALIZATION, UNIQUENESS: WHAT’S THE DIFFERENCE 93 then larger random match probabilities warrant inferences of uniqueness, and hence, individualization, relative to that population.41 Such local special uniqueness is comparable to the local individualization for the people on the yacht in Section 2. 5. Conclusions Forensic scientists often strive for individual identification. They call the identification of a trace or mark as having originated from an individual person or object ‘individualization’. Individualization can be ‘local’ (relative to a proper subset) or ‘universal’ (relative to the entire world). In principle, it can be accomplished (to a high probability) without examining every member of the population. Uniqueness can be ‘general’ or ‘special’. General uniqueness implies individualization for every member of the set of unique elements. Claims of general uniqueness for the set of all people in the world have been made for fingerprints and genomes (‘universal general uniqueness’). Less grandly, claims of special uniqueness have been made (for a particular forensic DNA profile) in local populations (less than the entire world). From the statistical standpoint, universal general uniqueness is not a necessary condition for individualization. A statement of individual identification can be almost certainly correct even when the identifying features are not unique throughout the world. In some circumstances, an expert can reason that a given individual is likely to be the only one in a smaller population of plausible suspects who would match the trace evidence. ‘The issue for the forensic scientist is not “Is this profile unique?”’ in the world.42 A fortiori, is it not whether all profiles are unique. ‘Rather, the issue is “Is there sufficient evidence to demonstrate that they originate from the identical source?”’43 Or is it? Ultimately, the last issue is left to the jury, and how the forensic scientist should convey the match to the jury is open to debate.44 The conclusion that the defendant is the source can be framed as a decision (as in acceptance sampling), and it is often presented without any recognition of the uncertainty inherent in the identification.45 Yet, if the expert’s task is to describe the extent of the match and to inform the judge or jury of the strength of this evidence of association, not to decide what the factfinder should believe, then the expert should present likelihoods, not ultimate conclusions.46 Unfortunately, for most pattern matching by human analysts, this ideal is difficult to attain. When the process is subjective, the underlying likelihoods are not readily quantified. A pragmatic and traditional procedure is for the expert to present a personal belief about the source of the trace and leave it to the legal system to explore the basis of this belief. Probabilists find this strategy oversimplified and misleading, but pragmatists defend it as a kind of ‘rounding off’. This essay does not 41 In a Bayesian framework, the size of the relevant population affects the prior odds. See supra § 2. 42 Evett & Weir, supra n. 23, at 239. 43 Ibid. 44 See, e.g. Kaye et al., supra n. 22; supra n. 23. 45 A. Biedermann et al., Decision Theoretic Properties of Forensic Identification: Underlying Logic and Argumentative Implications, 177 F ORENSIC S CI . I NT ’ L 120 (2008) (describing the elements of statistical decision theory as they apply to a decision to act on the hypothesis that the defendant is the source, complaining that ‘it is the very fact of deliberately suppressing uncertainty—which the established schemes need to achieve individualization—that breaks with a logical approach’, and appealing ‘to key players that engage in decision processes to assume their responsibility to explicate their probabilistic beliefs along with their preferential matrix’); cf. I SAAC L EVI, G AMBLING WITH T RUTH A N E SSAY ON I NDUCTION AND THE A IMS OF S CIENCE (1967) (using Bayesian decision theory to describe belief formation in science). 46 For a discussion of this premise in the context of presenting the results of statistical hypothesis tests, see David H. Kaye, Is Proof of Statistical Significance Relevant? 61 Wash. L. Rev. 1333 (1986). 94 D. H. KAYE resolve this debate. It simply seeks to sharpen the issues by clarifying the meaning of ‘identification’, ‘individualization’ and ‘uniqueness’ and thus untangling the relationships among these concepts. Acknowledgements I am grateful to Jay Koehler, Stephen Stigler and an anonymous referee for comments on a draft of this paper. Funding This work was supported by a summer research grant from Penn State’s Dickinson School of Law.