Expert Evidence and Statistical Interpretation in Court

The Daubert decision (1993) changed the relationship between science and US courts. Before Daubert, the Frye test required only that a technique be generally accepted in the relevant scientific community. Daubert replaced general acceptance with a structured, multi-factor inquiry: Has the theory been tested? Has it been subjected to peer review? Is there a known or potential error rate? Is it generally accepted? A trial judge must run through these questions as a gatekeeper before scientific testimony reaches a jury.

For wildlife forensics, these questions carry real bite. DNA barcoding using COI sequences satisfies all Daubert factors in well-established taxa: the method is published, tested, has a known success rate by taxonomic group, and is accepted by the systematic biology community. But barcoding of recently described species, or species where the reference database is thin, sits in a grayer zone. A knowledgeable defence attorney can challenge the database coverage and force the scientist to concede that the match probability is conditional on an incomplete reference.

Outside the United States, the rules vary. England and Wales now use Criminal Procedure Rules Part 19, which require experts to help the court, state the limits of their expertise, and indicate where their opinion is disputed. Australia follows similar principles under the Evidence Act uniform provisions. The practical effect across jurisdictions is the same: an expert who overstates certainty or conceals methodological limitations faces serious credibility damage in cross-examination, and the wildlife forensic community's own reporting standards, through SWFS, align with these legal expectations.

The logical framework that underpins modern forensic DNA interpretation is Bayesian. Two hypotheses compete: under the prosecution's hypothesis (Hp), the evidence DNA comes from the species, individual, or population in question; under the defence hypothesis (Hd), it comes from some other source. The likelihood ratio is the ratio of P(evidence | Hp) to P(evidence | Hd). When that ratio is high, the evidence strongly favours the prosecution account; when it is near 1, the evidence is neutral.

In human forensics, the denominator is typically the random match probability from a large population database. Wildlife forensics faces a harder problem. For many traded species, population genetic reference data is patchy. The SWFS standards address this: the expert must state the reference database size and composition, and if the database is incomplete, the opinion should be bounded accordingly. A match to African elephant using a well-sampled database of hundreds of loci and thousands of individuals is far more defensible than a match using twelve microsatellites from fifty samples.

Bayesian barcoding takes the LR concept and combines it with a prior probability drawn from what is known about the composition of trade in a given market or seizure context. If 80% of ivory confiscated on a particular trade route is from African forest elephant rather than savanna elephant, that prior changes the posterior probability of species assignment from a barcoding match. Courts that understand Bayesian reasoning accept this framing; courts that do not can find it confusing, which places an obligation on the expert to explain it in terms the judge and jury can follow.

Wildlife trafficking cases often involve fragmentary material: hundreds of bear paws, thousands of turtle shells, dozens of dried seahorses. The charge almost always depends on a count of individual animals, not just a weight of material. But the same animal contributes multiple parts, and duplicate counting inflates the number. The minimum number of individuals calculation provides the floor.

1. Inventory and sort
   Separate all biological material into element types: left forelimb, right forelimb, skull, pelvis. A single animal contributes one of each paired element on each side.
2. Identify the most frequent single-side element
   Count the most numerous element from one side (left or right). Forty-five left front bear paws means at least 45 individual bears, regardless of how many right paws, spines, or skulls are present.
3. Apply DNA typing to refine
   Where budget permits, genotyping distinguishes individuals even from the same element type. Two samples of the same element with different genotypes confirm two individuals. This raises MNI above the morphological floor.
4. Report as a floor, not an estimate
   MNI is conservative. The actual number of animals in the shipment may be higher. Courts should be told explicitly that MNI understates, not overestimates, the scale of the crime.

Two cases in the English-language wildlife prosecution literature illustrate how courts have tested forensic expert evidence in practice.

R v Bailey (UK) involved raptor persecution, specifically peregrine falcons and their eggs. The prosecution relied on forensic identification of eggshell fragments recovered from a gamekeeper's property, chemical analysis of residues, and feather morphology. The case was significant because it required the court to assess whether ornithological and ecotoxicological expertise met the threshold for admissibility under English rules. The defence challenged the chain of custody and the uniqueness of the shell fragments. Expert witnesses for the prosecution had to demonstrate that their identification methods, drawn from published comparative collections, were reliable and not merely opinion. The case established a precedent for treating specialist natural-history identification as properly scientific expert evidence rather than mere conjecture.

United States v. Kapp involved the trafficking of tigers and leopards and their meat, hides, and parts in violation of the Endangered Species Act and the Lacey Act. The defence contested aspects of the forensic identification evidence. The court accepted the testimony, relying on the laboratory's accreditation and the analyst's documented training. The case illustrates that accreditation of the laboratory and the analyst's qualifications carry independent weight alongside the methodological factors that Daubert requires.

The Society for Wildlife Forensic Science developed its minimum reporting standards precisely because wildlife forensic scientists work across many laboratory and agency contexts without the centralised accreditation structure that governs human forensic science in most countries. The standards require that a report answer the question the investigator actually asked, describe the evidence examined, state the methods clearly enough for a competent scientist to evaluate them, acknowledge limitations, and express the conclusion as a degree of support for the hypothesis rather than a categorical declaration.

• State the question: what was the forensic question? Species identity? Geographic origin? Individuality? The answer to a narrower question than what the prosecution needs can be a problem if the gap is not stated.
• Describe the evidence: item number, condition received, any continuity concerns, and the type of biological material examined.
• State the method and its limitations: which markers, which reference database, what the database coverage is for the species group, and the method's published error rate in comparable taxa.
• Express the conclusion proportionately: use a verbal scale tied to the likelihood ratio, or state the posterior probability explicitly. Avoid 'the sample is definitely from species X' when what the data supports is 'the evidence is very strongly consistent with species X given current reference data.'
• Note any competing explanations: if another species could produce the same barcode due to reference database gaps, say so. Courts expect the expert to address the defence hypothesis, not only the prosecution's.

Wildlife cases are often heard by judges and juries with no scientific background. The forensic scientist's job is to translate statistical conclusions into language that is accurate and comprehensible without being misleading. Three persistent problems arise in practice.

The first is the transposition fallacy, also called the prosecutor's fallacy. The probability of the evidence given innocence is not the same as the probability of innocence given the evidence. A scientist who says 'there is only a one in ten thousand chance this is not from the endangered species' has stated the LR in reverse and may be giving the court a misleading impression about the probability of guilt. The correct statement is 'the evidence is ten thousand times more probable if it came from the species than if it came from a non-target source.'

The second is overconfidence from a matching result when the reference database is small. A unique genotype match to a reference population of thirty animals is statistically interesting but far from conclusive. The third is treating class-level evidence as individual-level: a species-level identification does not prove a specific individual animal was taken from a specific location, and conflating the two inflates the strength of the evidence.