The Identification Process
The real aim of all forensic science is to establish individuality, or to approach it as closely as the present state of the science allows. Criminalistics is the science of individualization.
This citation of Paul Kirk shows how essential the identification process is to criminalistics.
The definition of identification in forensic science may differ largely from the one accepted in science, where the term ‘identification’ is simply used to describe the attribution of an object to a definite class. In criminalistics, the identification process seeks ultimately individualization. For forensic scientists, identifying an object means that it is possible to distinguish this object from all objects considered.
n the forensic literature, the problem of identity of source is often treated by reference to ‘class’ and ‘individual’ characteristics (Table 1). Comparisons that lead to agreement only in class characteristics (without significant differences) will end up with ‘group identification’ conclusions. Only when individual characteristics are present in conjunction with class characteristics can positive identification or individualization conclusions be drawn. The definitions of ‘class’ and ‘individual’ characteristics are only conventional ways of describing selectivity. We will see that the problem of inferring identity of source and its pertinence to criminalistics is more complex than a simple dichotomy between class and individual characteristics.
As illustrated in Table 2, we will distinguish the forensic fields that lead frequently to individualization and those leading rarely (in the present state of the art) to individualization, but more commonly to corroborative evidence of various strength.
Philosophically, identity of source cannot be known with certainty, and therefore must be inferred. As Kwan has demonstrated, the hypothetical-deductive method (assisted by methods of statistical inference) provides a reasonable explanation of how criminalists infer identity of source.
The identification process can be seen as a reduction process, from an initial population to a restricted class or, ultimately, to unity. The initial population constitutes control objects or persons, depending on the type of evidence. We have the combination of two factors:
• A relevant population of control persons or objects defined by its size (and/or other particularities). Put in another way, each member of this population of sources can be seen as a possible source.
• A reduction factor resulting from the combination of concordant characteristics of determined selectivity. In fact, the reduction is proportional to the rarity or random match occurrence of these observed characteristics in that population. As Kwan indicates: ‘this is the sheer rarity of a feature that is important as rarity of that feature with respect to the set of suspected sources being considered. It is important to stress that rarity is relative to the situation at hand.’
Table 2 Classification of forensic evidence types with respect to their identification capabilities
|Fingerprints||Microtraces (glass, paint, hairs,|
|Footwear marks||Biological fluids (now mostly DNA|
|Earmarks||Drugs and toxicology|
|Tool marks and firearms||Explosives and fire residue analysis|
Table 1 Distinction between ‘class’ and ‘individual’ characteristics in some fields
|Field||‘Class’ characteristics||‘ Individual’ characteristics|
|Fingerprint identification||General pattern, ridge count, ridge tracing||Minutiae, pore structure, ridge structure|
|Footwear mark||General pattern, size, manufacturing||Cuts, accidental acquired characteristics, transient|
|Bullet identification||Caliber, number of grooves/lands||Grooves/lands impressions (striae)|
|impressions, angles of grooves/lands|
|impressions, width of grooves/lands|
With respect to the size of the relevant population, an ‘open set’ framework will be distinguished from a ‘closed set’ framework. The open set framework implies that the population at large is considered; meaning, for example, that all living persons on Earth or all produced objects on Earth are taken into consideration as potential sources. The closed set framework corresponds to a situation in which the number of control objects or persons is restricted to a specified set of suspected sources (for example, by taking into account other evidence available describing the putative sources).
To illustrate the identification process graphically, we will assume the following case. A mark is found in association with a crime. Following inquiry, a potential source is submitted to the laboratory for examination. The comparison shows that the recovered evidence and the control share some features. Moreover, there is no significant difference to be accounted for. Given this scenario, the identification process may be illustrated as shown in Fig. 1.
Figure 1 The identification process.
The identification process (in an open set or closed set framework) is a narrowing-down process, reducing the number of possible sources or hypotheses. The hypothesis that a designated suspect or object is the source, is proven by showing that all alternative hypotheses that could explain the phenomenon at hand are excluded. Also, to avoid error, the hypothetical-deductive method makes it imperative that all plausible hypotheses regarding possible sources are taken into account.
The importance of selecting features judiciously cannot be overemphasized. The criteria for selecting features fall into five areas (without taking cost into account): distinguishability, high intersource to intra-source variance, known variance in time, normalization (standardization) and independence. When individualization is the goal, the object must be defined by a unique set of properties (a set that no other source can share). Of course, each field is focused on different features and it is worth giving a few examples of adequately chosen features in some forensic areas (Table 3).
The Decision Schemes
In practice, for obscure reasons, the identification process leading to individualization is generally operated in an open set framework. This leads to two types of decision schemes: positive identification and corroborative evidence.
Positive identification or individualization
The individualization of an impression is established by finding agreement of corresponding individual characteristics of such number and significance as to preclude the possibility (or probability) of their having occurred by mere coincidence, and establishing that there are no differences that cannot be accounted for.’
Table 3 Features used in identification in some forensic fields
|Field||Features used to characterize (or ultimately to individualize)|
|Fingerprints||General pattern, minutiae, pores and ridge|
|Footwear||Manufacturing characteristics (pattern, size,|
|marks||peculiarities of the manufacturing process)|
|and acquired characteristics (wear features,|
|DNA||Various polymorphic loci on the DNA molecule|
|Microtraces||Optical (color, microscopic features, refractive|
|index), physical (size, length, diameter) and|
|chemical characteristics (Fourier-|
|transformed infrared spectroscopy,|
|elemental composition, pyrolysis coupled|
|with gas chromatography, etc.)|
Following this definition, the size of the control population is systematically set to its maximum (open set framework). This practice is generally used and required in fields like those of fingerprints, footwear marks, tool marks or firearms. For footwear marks, Bodziak stated that: ‘The positive identification means that no other shoe in the world could have made that particular impression’. Analogous definition can be found for fingerprint identification: ‘An ”identification” is the determination that two corresponding areas of friction skin impressions originated from the same person to the exclusion of all others.’
The conclusion of a positive identification is then an opinion, a statement of probability expressing the view that the chance of observing on Earth another object or person presenting the same characteristics is zero. No contrary evidence will ever shake this certainty. According to Stoney, this decision is highly subjective, the identification process for fingerprint identification being described by him as follows:
Beginning with a reference point in one pattern, a corresponding point in a second pattern is sought. From this initial point the examiner then seeks neighbouring details that correspond in their form, position and orientation. These features have an extreme variability, that is readily appreciated intuitively, and which becomes objectively obvious upon detailed study. When more and more corresponding features are found between two patterns, scientist and lay person alike become subjectively certain that the patterns could not possibly be duplicated by chance.
What has happened here is somewhat analogous to a leap of faith. It is a jump, an extrapolation, based on the observation of highly variable traits among a few characteristics, and then considering the case of many characteristics. Duplication is inconceivable to the rational mind and we conclude that there is absolute identity. The leap, or extrapolation, occurs (in fingerprinting) without any statistical foundation, even for the initial process where the first few ridge details are compared.
The schematic description of the identification process can be refined to include the decision step for positive identification (individualization) (Fig. 2). This decision scheme calls for the following comments:
• The threshold, the leap of faith, is in essence a qualification of the acceptable level of reasonable doubt adopted by the expert. Jurists will interpret this threshold as an expression of the criminal standard ‘beyond reasonable doubt’ regarding the issue of the identification. Would jurists accept that the concept of reasonable doubt on the identification of a suspect is outside their province and that the threshold is imposed on to the court by the scientist? The response in the doctrine is negative, as expressed by the members of the Panel on Statistical Assessments as Evidence in Courts:the law may establish different thresholds for what is sufficient evidence in a case from those that statisticians would normally require in drawing conclusions. Clearly, the law must prevail and the statistician must adjust to the law’s standards. Put another way, it is the utility function of the court that is appropriate, not the utility function of the statistician.
Figure 2 The identification process completed with a decision of positive identification.
• It seems illegitimate to set the size of the relevant population at its maximum a priori. Indeed, the number of potential sources from which the mark could originate may be restricted by other evidence available (witness testimonies, other forensic evidence, etc.). Presenting the evidence in an open set framework is too conservative, adopting systematically the extreme defense attorney’s position in trying to make the court believe that all persons or objects on earth could be the origin of the prints.
In some forensic fields (fingerprints, tool marks, footwear marks), practitioners have voluntarily excluded probability statements – other than exclusion and positive identification – from their conclusions. All items of evidence between these extremes are classified as ‘inconclusive’. There is no logical reason for avoiding probability statements; the refusal of qualified opinions is a policy decision, even if the distinction of the arguments (policy or scientific argumentation) is not so clear in the literature.
(Appreciate the dogmatic statement proposed recently by a North American working group on fingerprint identification: ‘Friction ridge identifications are absolute conclusions. Probable, possible, or likely identification are outside the acceptable limits of the science of friction ridge identification’.)
Indeed, the attention of jurists or scientists is too easily concentrated upon the ability of a technique to provide an absolute certainty. If the technique can positively conclude an identification, it is greeted as a panacea; if it cannot, it is damned as unreliable. This ignores the vital point that any technique will only function to a high degree of precision and accuracy under controlled conditions, and the conditions under which forensic scientists work are far from ideal. It follows that, in many cases, a forensic scientist will not be able to provide a definitive answer but only a probabilistic figure or opinion. If the ultimate set of specific features is not present or not detected in the evidence, then the examiner will not provide an identification but will express a probability statement, verbally or numerically, which attempts to assess the value of the evidence.
Each piece of evidence is relevant if it tends to make the matter that requires proof more or less probable than otherwise. Hence, a piece of evidence that only approaches the absolute identification constitutes relevant evidence that should not be ignored.
The identification process remains, in essence, a statistical process based on objective empirical data and/or on subjective evaluations related to the examiner’s experience. When the evidence is only corroborative, it is necessary to form a conclusion in a way that reflects the statistical uncertainty.
In various fields which accept corroborative evidence, examiners have expressed the meanings of conclusion terms pragmatically. Their respective meaning relates to their power of reduction of the initial population. An agreement on the following terms seems to have been achieved: identification, very probable, probable, possible, inconclusive (could not be determined), appears not, negative, not suitable for analysis.
The identification process scheme can be refined as illustrated in Fig. 3. This decision scheme calls for two comments:
• The conversion between the objective statistical value and the verbal statement is never declared or explained. This naturally leads to obvious variations between examiners when assessing the same case.
• The blind and dogmatic allegiance to the open set framework is never questioned as for the positive identification.
Figure 3 The identification (ID) process completed with corroborative evidence.
Identification and Statistics
Positive identification or individualization
From a statistical point of view, to conclude an individualization means that the probability (Pr) of the event, here the event ‘identification’, after examining the evidence is equal to 1. Because of the probabilistic nature of the identification process, the probability of 1 is an absolute that cannot ever be reached numerically. Hence, the examiner expresses an opinion of a positive identification when this probability is so near 1 that it can be reasonably set as 1. This probability of identification, given the evidence, can be represented in the form Pr(IDlE), where ID denotes the event identification, which is uncertain, and E denotes the information available, the evidence E, which has been taken into account. In this way, the vertical line can be seen as shorthand for the word ‘given’. But we have seen that the identification process is related to the specificity of the print under examination, its random match occurrence in a population and the size of the suspect population being considered (N). The probability we are interested in is then conditioned on both E and N and becomes Pr(IDlE, N). Probability theory provides a tool to calculate Pr(IDlE, N). Let us denote the frequency of a random occurrence of the mark by f:
If the population is set to N =5 billion (including the control) and f = 1 in 5 billion, then, using Equation 1, Pr(IDlE, N) is close to 0.5. That means that if you have a control object or person in hand with such a frequency, the probability that you have the wrong one is about 0.5! This expression is the correct statistical value even if it is counterintuitive. It has been largely demonstrated that intuition is a bad substitute for the laws of probability. If we accept that Pr(IDlE, N) must be above a certain threshold value in order to declare an identification, then it is possible to calculate the frequency that must be considered to achieve such a preset value. If Pr(IDlE, N) is fixed at 0.9998, N =5 billion, using Equation 2 we see that f must be equal to 4.0 x10-14, which represents 1 in about 5000 times the size of the initial population of 5 billion. A conclusion of individualization (according to this threshold and in an open set framework) must be at least the expression of such a small random match probability.
What is the statistical meaning, in the open set framework, of conclusions like: ‘it is possible (or probable or highly likely) that this print/mark has been made by this particular item’? Such verbal scales can be translated into numbers expressing Pr(IDlE, N). It has been demonstrated that these terms are understood quite uniformly (in a numerical conversion) by experts or jurists. The numerical conversion is given in Table 4 (columns l and 2). From these probabilities, their corresponding frequencies (f) can be derived, using Equation 2, considering an open set framework (N = 5 billion) as shown in Table 4 (column 3).
The analysis of Table 4 leads to a paradox. In the open set framework (the one claimed to be adopted in major forensic fields like footwear marks, firearms or tool marks), the rarity of the shared characteristics which corresponds to the most negative verbal statement is quite small (2.0 x 10-7). For the court, every piece of evidence that would lead to such small frequencies will be considered as highly relevant (by analogy with DNA evidence, for example). Hence, the verbal statement proposed here does not make this evidential value very clear.
To escape this paradox, forensic scientists could be tempted to adjust the size of the relevant population; hence, to pass from an open set framework to a closed set framework. To understand if it is legitimate to do so, we will briefly introduce a proposal for an alternative inferential system. This proposal advocates the use of a bayesian model to assess the value of identification evidence.
The Bayesian Framework
In the bayesian framework, it is not assumed that the size of the relevant population is at its maximum (the world population of control objects). It seems that this relaxation constitutes the major difference between the two approaches (strict open set versus bayesian approach). Both closed set and open set situations can be handled in a bayesian framework; the open set constitutes a specific situation.
From information gathered through normal investigation or through the use of a database, a limited number of people or objects, or a single person or object, can be pinpointed in a general or limited population. There is no way to go back along the time axis and deduce the identification. The only way to evaluate the strength of the evidence is to consider this evidence from two perspectives: that the designated person or object is, or is not, the source.
The bayesian approach permits the revision of a measure of uncertainty about the truth or otherwise of an issue (here the identification ID) based on new information. It shows how data can be combined with prior or background information to give posterior probabilities for particular issues.
Table 4 Relationship between verbal statements in qualified opinions and the probability (Pr) of identification, along with the corresponding frequencies f (assuming a relevant population of 5 billion)
|Verbal statement||Pr(/D|E, A/)(%)||Frequency f|
|Likelihood bordering to certainty||99.9||2.0×10~13|
|Highly (very) likely||98||4.1 x10~12|
|Very well possible (plausible)||75||6.5X10-11|
|Possible (evens) not||40||3.0×10~10|
|Very well possible (plausible) not||25||6.0×10-10|
|Not likely/unlikely||15||1.1 x10~9|
|Highly (very) unlikely||2||9.8×10~9|
|Likelihood bordering certainty not||0.1||2.0×10~7|
The following events can be defined from the previous example:
I Some background information has been collected before the forensic examination. For example, data from police investigation, eyewitness statements or data from the criminal historical record of the suspect will contribute to I. Typically this information will reduce the number of potential suspects or objects that could be at the origin of the mark.
E The evidence:features agreement (without significant differences) has been reported between the print left at a scene and a person or object under examination.
ID The mark has been produced by this person or object in examination.
ID The mark has not been produced by this object or person, and another unknown person or object is at the origin of the mark.
The definition of the two exclusive hypotheses (ID and ID) requires consideration of the context of the case (the defense’s strategy, for example); they are not always as straightforward or exhaustive as could be deduced from our example.
The Bayes’ formula (Equation 3) shows how prior odds on ID are modified by the evidence E to obtain posterior odds on the issue:
between the mark and the control, given that the mark has not been left by this person or object (ID is true).
Equation 1 can also be derived in a bayesian framework, where the posterior odds on the identification considering the evidence E,O(IDlE,I), are obtained by multiplying the prior odds of 1/N— 1 with a likelihood ratio of 1/f.
If we agree that, in a given case, it is too conservative, or even misleading, to state that the number of potential objects or persons is maximal, then each case bears its own specificity in this regard. Hence, the estimation of the number of potential objects or persons will be based on the background information of the case (I) gathered by the police (investigation, eyewitness statements or data from the criminal historical record of the suspect). But generally these data are not known to the forensic examiner and are outside his or her province. Hence, the assessment of the relevant population remains the confines of the courtroom (judges, members of the jury, etc.). Consequently, before the presentation of the expert evidence, the judge will assess the size of the relevant population that could be involved in the case; this will lead to the so-called ‘prior odds’on the identification. In the bayesian perspective, the open set framework covers the situation where the prior odds are estimated in reference to the total population possible (I represents no prior knowledge at all); any other situation with an estimate of the prior odds can be viewed as a closed set framework.
The statement (numerical or subjective) made to the court by the forensic scientist should only be the expression of the reduction factor to be applied to the prior odds (opinion of the court without the knowledge derived from the scientific evidence). Without the information about the prior odds, it is not possible for the scientist to state the fact itself (the probability that this particular person or object has produced this mark): he can only state the degree of support given to this hypothesis (or version) versus the alternative. In the bayesian framework, the strength of the evidence is given by the probability of observing the evidence under two chosen propositions. This ratio (called the ‘likelihood ratio’) has a numerator close to 1 and a denominator equal to the frequency of the shared features in the relevant population. The statement given by the scientist is not alone sufficient to decide on the issue (the identification or probable identification). The scientist’s statement is then completely different from assessing the fact itself. The concept of evidence is therefore relative: it shows how observations should be interpreted as evidence for ID vis-a-vis ID, but it makes no mention of how those observations should be interpreted as evidence in relation to ID alone. Therefore, the same piece of evidence may be insufficient proof for the court in one context, but may be the factor essential in clinching the case in another.
The odds in favour of the identification itself, that the person or object in examination has produced the mark given the circumstances of the case (background I) and the observations made by the forensic scientist (the evidence E), is a judgement based on posterior odds combined with the evidence. The essence of the decision on the identification remains the province of the fact finder.
This logical tool permits an understanding of how events interact mathematically. It clarifies the position of the scientist as well as that of the judge, and defines their relationship. The scientist is concerned solely with the likelihood ratio, whereas judges deal with the odds on the identification (Fig. 4). This impossibility of assessing (even by degrees) the issue in itself has already been identified in some forensic identification fields, such as handwriting examination, voice identification and tool marks.
This solves the philosophical dilemma encountered by scientists asked to express a view on the identification process. Identification is obtained through a logical process that requires a judgment at the end. Such a conclusion can be drawn only by combining prior knowledge and forensic evidence. In a strictly bayesian point of view, the examiner is never in position to identify definitively. But, it may be (for fingerprints, footwear marks or DNA evidence, for example) that the likelihood ratio in favor of the identification is so strong that identification is declared, whatever prior odds might be. In such instances, a conclusion of individualization is made.
Figure 4 The Bayesian framework positioning the actors of the judicial system.
The bayesian model helps us to understand the respective duties of the court and of the forensic scientists (Fig. 4). Our schematic representation of the identification process can then be modified, taking into account the bayesian perspective (Fig. 5).
Figure 5 The value of evidence in the identification (ID) process.
Various decision schemes for the question of identification of sources have been presented. Two of them lead to decisions of positive identification or of qualified opinions (possible, probable, very probable identification, etc.). Both of these schemes are defined with reference to a relevant population size set to its maximum in an open set framework. The use of an open set framework forces the scientist to deal -sometimes without being aware – with thresholds to take decisions and with prior or posterior probabilities on the issue. It appears legitimate to question if these practices (or accepted assumptions) are reasonable for each particular case. Moreover, the statistical relationship between verbal opinions and frequencies of occurrence of the shared features has been shown, with all its counterintuitive and paradoxical consequences.
Another scheme, the bayesian interpretation framework, overcomes most of these difficulties, in particular by relaxing the necessity of adopting an open set framework. It provides a coherent way of assessing and presenting identification evidence. From a logical point of view, the concept of evidence is essentially relative to the case and its value is best expressed using a likelihood ratio. The question of the size of the relevant population (open set versus a closed set) depends on prior probabilities, which are outside the province of the scientists but belong to the court.