Match Probabilities Beyond DNA

DNA random match probabilities rest on three features that are not replicated in most other forensic disciplines. First, the population genetics of allele frequencies is well understood, and allele frequencies at the relevant STR loci have been measured in large, diverse reference populations. Second, the assumption of independence between loci is grounded in linkage disequilibrium studies and supported by validation work across different population groups. Third, the multiplication rule, which combines per-locus probabilities across a multi-locus profile, has been tested and challenged in courts in multiple jurisdictions and has withstood that scrutiny.

None of these three features transfer directly to fingerprints, glass, fibres, or footwear. For those categories, the relevant features are not alleles at defined loci with known population frequencies. They are configural properties, meaning the combination and spatial arrangement of features matters, but no consensus method exists for how to count or weight them. The relevant population is also harder to define: for DNA the population is the human gene pool; for footwear the population is all shoes of the same make and model that have been worn comparably.

This does not mean non-DNA evidence is scientifically worthless. It means the inferential pathway from observation to probability statement is longer and the uncertainty at each step is larger. The discipline of forensic statistics exists, in part, to make those steps explicit and to develop the tools needed to quantify them more rigorously. Progress has been made in glass analysis, modest progress in fibres, and substantial ongoing work in fingerprints. Footwear remains the least developed.

Fingerprint examiners compare latent marks against reference prints by examining ridge detail: ridge endings, bifurcations, enclosures, and other minutiae at specific spatial positions. The traditional conclusion in many jurisdictions was a categorical identification or exclusion, with no probability statement attached. Courts in the United Kingdom, Australia, and elsewhere have moved away from the categorical twelve-point standard and toward conclusions expressed as levels of support, but the underlying frequency data needed for a quantified LR have been slow to arrive.

Several research programmes have attempted to address this. The most systematic is the work arising from the US National Institute of Standards and Technology (NIST), which has published fingerprint image databases and error rate studies. Simon Cole's historical analysis of fingerprint misidentification cases identified a non-zero false positive rate. The 2016 PCAST report concluded that latent fingerprint analysis was a foundational-validity method but that the false positive rate in black-box studies was higher than practitioners had claimed, in the range of 1 in 18 to 1 in 306 depending on the study, not the 1 in a billion implied by some expert testimony.

An LR for fingerprint evidence requires: a model of how many minutiae would be expected to match in marks from the same finger (the numerator probability), and how often a mark from a different finger would produce that many matching minutiae (the denominator probability). Research groups including Champod, Evett, and colleagues at the Forensic Science Service developed probabilistic models using minutiae frequency data from small samples. These models are used in some jurisdictions to support LR-based fingerprint testimony, but critics note that the underlying databases are small and the models have not been independently validated at scale.

Glass comparison in forensic science relies primarily on refractive index, measured by the temperature of matching oils in a hot-stage microscopy technique or by automated instruments. When glass from a suspect's clothing matches glass from a broken window in RI, the examiner asks: how often would a randomly selected piece of glass from the relevant population have this RI value? The answer requires a reference database of RI measurements from glass in circulation.

The UK Forensic Science Service maintained such a database, and similar collections exist in Australia and some other jurisdictions. These databases record the RI of glass from different sources, including window glass, container glass, and vehicle glass. When a crime-scene RI value is very common in the database, the evidential value of a match is lower; when the RI is rare, the value is higher. This is directly analogous to the logic of DNA frequency estimation but the databases are far smaller, covering thousands rather than millions of samples.

Additional discriminating characteristics include elemental composition, measured by techniques such as laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS). Elemental composition adds discriminating power beyond RI and is increasingly used in high-value cases. However, reference databases for elemental composition are even smaller than RI databases, which constrains the precision of match probability statements. Forensic glass scientists in the UK have published LR models combining RI and elemental data, and these are used in some high-profile cases, but the approach is not yet routine.

Textile fibre evidence involves comparing fibres found on a victim or crime scene to fibres from a suspect's garment. Comparison criteria include fibre polymer type (cotton, polyester, nylon, acrylic), colour, diameter, cross-sectional shape, and, for dyed fibres, the specific dye combination identified by high-performance liquid chromatography or microspectrophotometry. A match on all these characteristics increases the evidential value; the question is how to quantify that value.

The core difficulty with fibres is that rarity depends on the garment, not just the fibre. A polyester fibre of a common red dye is unremarkable. A polyester fibre of an unusual dye combination from a limited production run of a specific garment is potentially very significant. The examiner must estimate how many garments with that exact specification exist and how many people who could have committed the offence own one. These estimates are often qualitative rather than quantitative, relying on the examiner's experience and on garment manufacturer or retailer records in the rarest cases.

No general-purpose population database for fibre frequencies exists comparable to the glass RI databases. Some academic research has characterised colour distributions in commercial textiles, and the UK Forensic Science Service developed internal reference data before its closure in 2012. In practice, fibre match probability claims are usually expressed qualitatively, for example as evidence that the fibres are consistent with transfer from the suspect's garment and that fibres of this combination are uncommon, rather than as a numerical LR. Courts have generally accepted this approach while acknowledging its limitations.

Footwear impression evidence involves three analytically distinct comparison layers. The first is sole pattern: the tread design that identifies the manufacturer and model of shoe. The second is wear pattern: the specific areas of the sole that have worn down through use, which reflects the wearer's gait and the surfaces they habitually walk on. The third is manufacturing characteristics: random marks introduced during the production process that may be unique to one shoe. The evidential value of a footwear match depends on which layers contribute to the comparison and how rare the observed combination is.

Pattern databases exist and are reasonably well developed. The UK's Footwear Intelligence Technology (FIT) database catalogues sole patterns from many manufacturers and is used to identify the make and model of a shoe from a crime-scene impression. Similar databases are maintained by national law enforcement agencies in the US, Canada, and elsewhere. These databases support statements such as: this impression is consistent with a size 10 Nike Air Max 90, and that model sold approximately 200,000 pairs in the UK in the relevant period. That is a frequency statement, though an imprecise one.

Wear pattern and manufacturing characteristics are much harder to quantify. No large-scale study has established how often two shoes of the same model, worn for comparable periods, produce impressions as similar as two impressions from one shoe. Without such data, the claim that a wear pattern match significantly increases the probability that the crime-scene impression came from the suspect's shoe is expert opinion rather than a validated statistical claim. PCAST and equivalent bodies in the UK have identified footwear analysis as needing empirical studies to establish foundational validity at the wear and manufacturing-mark level.

Courts in England and Wales, the US, and India under the Bharatiya Sakshya Adhiniyam 2023 (which, like the US Federal Rules of Evidence Rule 702, requires expert testimony to be based on sufficient facts and reliable principles) have received footwear evidence at varying levels of quantification. The general judicial approach is to admit the evidence while directing the jury that a categorical identification claim unsupported by validated error rates should be given less weight than the expert's language might suggest.

Validation in forensic science means demonstrating, through empirical study, that a method produces accurate results at a known error rate when applied by qualified examiners to casework-quality samples. For a match probability claim to be credible, the method used to generate it must be validated. DNA profiling is validated in this sense: proficiency testing programmes, blind verification procedures, and large-scale error rate studies exist across multiple countries. Most other forensic comparison disciplines are not validated to the same degree.

The PCAST report distinguished foundational validity, which is whether the method works at all under controlled conditions, from validity as applied, which is whether a specific examiner's application of the method in a specific case meets the required standard. Both matter for a court. A method can be foundationally valid but applied badly in a particular case, or it can lack foundational validity entirely. For fingerprint and footwear analysis, PCAST found foundational validity at a basic level but identified that error rates were higher than practitioners had claimed and that validity as applied was often impossible to assess because examiners did not document their reasoning in ways that could be independently checked.

The path toward quantified match probabilities for non-DNA evidence requires three parallel developments. First, larger and more representative reference databases must be built and maintained: glass RI databases must be updated as glass manufacturing changes; fingerprint minutiae frequency data must be collected at scale. Second, statistical models must be developed that correctly handle feature dependencies, such as spatial correlations between adjacent minutiae. Third, the models must be validated in black-box studies where examiners use the model output to reach conclusions on test sets with known ground truth, and the error rates measured.

Progress is uneven. Glass analysis using both RI and elemental composition, expressed as an LR with explicit database citation, is now routine in UK courts and increasingly used in Australia and New Zealand. Probabilistic genotyping software for DNA mixtures has set a precedent for using likelihood ratio software in court that may accelerate acceptance of similar tools for fingerprints. Fibre and footwear remain qualitative in most jurisdictions, and the forensic science community continues to debate whether quantified probabilities for those evidence types are achievable in the near term or whether a well-structured qualitative LR is the more honest approach given current data.