Writing Evaluative Statements in Forensic Reports

Forensic propositions operate at three hierarchical levels, and choosing the wrong level is the most consequential structural error a scientist can make in an evaluative report. The level must match the legal question the court is actually deciding.

In most contested cases, the parties do not dispute that the complainant's DNA was found on the knife: they dispute how it got there. A source-level opinion that says 'the DNA profile is 10,000 times more likely if it came from the complainant than from a random person' does not help the court resolve the actual dispute. An activity-level opinion that says 'the findings are 50 times more likely if the suspect used the knife to stab the complainant than if someone else did while the suspect was never in contact with the knife' is directly responsive to the legal issue.

Activity-level propositions are harder to evaluate because they require the scientist to estimate or model the probability of the evidence under a scenario involving human behaviour, not just a laboratory measurement. The scientist must incorporate assumptions about transfer rates (how likely is it that stabbing would deposit DNA of the observed amount?), persistence (how much DNA would survive the intervening time?), and background (how likely is it that the complainant's DNA would be on the knife for an innocent reason?). These assumptions must be stated explicitly in the report. Omitting them makes the opinion unverifiable.

A pair of propositions is correctly constructed when it satisfies three conditions: both propositions are at the same inferential level, they are mutually exclusive (both cannot be true simultaneously), and together they cover the realistic alternative explanations in the case context. The scientist does not invent propositions. They are derived from instructions received from the instructing party, informed by the likely defence case as disclosed.

A common error is to pair a source-level prosecution proposition with an activity-level defence proposition. For example: 'H1: The fibres on the seat originated from the suspect's jacket' versus 'H2: The suspect was never in the car.' H1 is a source proposition; H2 is an activity proposition. Evaluating evidence under this mismatched pair produces an LR whose denominator includes outcomes that are irrelevant to the prosecution proposition, making the number uninterpretable.

The defence proposition must also be realistic. A proposition framed as 'the fibres came from any other blue wool jacket in the world' is technically correct but so broad as to make the LR trivially large. The accepted practice, recommended in ENFSI Guideline for Evaluative Reporting (2015) and reflected in guidance from the UK Forensic Science Regulator, is to anchor the defence proposition to the realistic alternative advanced by the defence or, where no defence case has been given, to the most plausible alternative given the case circumstances.

Likelihood ratios span many orders of magnitude. A DNA LR might be 10 billion; a voice comparison LR might be 8. Presenting raw numbers to a jury or to a lay judge risks either inducing false precision or causing incomprehension. Verbal equivalent scales translate LR ranges into standardised descriptive phrases. The goal is to communicate the direction and approximate magnitude of evidential support in language accessible to the court, without abandoning the logical structure.

The table above follows the ENFSI recommended scale (2015). Other scales exist: the Association of Forensic Science Providers (AFSP) in the UK published a similar six-tier scale; the International Association for Identification (IAI) uses different terminology for friction-ridge evidence. Individual laboratories may use in-house scales, provided those scales are disclosed in the report and validated for the discipline.

The verbal equivalent must correspond to the actual LR range. Selecting a phrase that overstates the LR is a drafting error with direct legal consequences. It effectively converts a moderate finding into a very strong one in the mind of the jury. Courts have cited such overstating as grounds for appeal in several jurisdictions. Where the LR has been estimated rather than calculated, the uncertainty in the estimate must be reflected in the choice of verbal equivalent: use the more conservative phrase when the LR spans more than one tier.

A well-structured evaluative section follows a fixed sequence. Deviating from this sequence invites the reader to confuse the proposition with the conclusion, or the LR with the posterior probability. The sequence below reflects the structure recommended in ENFSI Guideline for Evaluative Reporting and is consistent with Forensic Science Regulator guidance in England and Wales, as well as guidance issued by forensic science institutes in Australia and the Netherlands.

• Propositions: State both propositions explicitly and label them (e.g., H1 and H2 or Hp and Hd). This tells the reader exactly what question is being answered.
• Key assumptions: List the conditioning assumptions the opinion depends on, particularly for activity-level opinions. These include assumed transfer rates, persistence rates, and the relevant reference population for background frequency estimates.
• Findings: Describe the observed evidence briefly. The full analytical results belong in an earlier section; here, state only what is relevant to the evaluation.
• Likelihood ratio: State the LR value or range, and whether it was calculated, estimated, or derived from a model. If it was estimated, state the basis for the estimate.
• Verbal equivalent: State the verbal equivalent and identify the scale being used. This is the sentence the court reads. It must translate the LR accurately.
• Limitations: State what the opinion does not cover. If only source-level evidence is available but the legal issue is at activity level, say so explicitly.

The order matters. Stating the verbal equivalent before the propositions allows the reader to attach the conclusion to an implicit question rather than the explicit one, which is exactly the error these guidelines are designed to prevent. Stating the LR before the assumptions invites the reader to treat the number as a raw fact rather than a conditional probability.

The errors below recur across disciplines and jurisdictions. Each has appeared in published appellate judgments or in peer-reviewed audits of forensic reporting practices.

• Proposition-level mismatch: The legal issue is activity-level but the report evaluates source-level propositions. Correction: re-draft propositions to match the court's question; if activity-level evaluation is not possible, state that explicitly.
• The prosecutor's fallacy: Reporting 'there is a one in ten million chance this profile came from someone other than the suspect' instead of 'the evidence is ten million times more likely if the profile originated from the suspect than from an unrelated individual.' Correction: always frame the opinion as the LR, never as a posterior probability.
• The defence attorney's fallacy: Arguing that because thousands of people share the profile, the evidence has little value. This ignores all other case evidence and misapplies probability. Forensic scientists should be alert to this framing in cross-examination and correct it.
• Verbal overstating: Choosing 'very strong support' for an LR that falls in the 'moderate support' range, because the scientist believes the suspect is guilty and wants the jury to reach the same conclusion. Correction: the verbal equivalent must follow from the LR, not from the scientist's private opinion about the case.
• Omitting conditioning assumptions: Reporting an activity-level LR without stating the transfer and persistence rates assumed in its calculation. If those assumptions are wrong, the LR is wrong. Correction: list all key assumptions in the evaluative section and, where possible, give the sensitivity of the LR to changes in those assumptions.
• Ambiguous language: Using phrases such as 'consistent with', 'cannot be excluded', or 'supports the view'. These phrases carry no defined probabilistic meaning. Juror perception studies in the UK, Australia, and the Netherlands have shown that different jurors interpret them across the full range of probabilities. Correction: replace with a verbal equivalent tied to an explicit LR.

Peer review of evaluative sections before a report is finalised catches most of these errors. Several forensic science regulators, including the UK Forensic Science Regulator and the Netherlands Register of Court Experts (NRGD), now require that evaluative reports be reviewed by a second scientist before submission to a court. This requirement is not yet universal, but the practice has measurably reduced reporting errors in organisations that apply it.

Reading examples from real or realistic casework is the most direct way to internalise the difference between correct and flawed evaluative statements. The examples below are illustrative; the analytical details are typical of published casework reports in the UK and Australia.

Example 1: Correct activity-level report. 'H1: The suspect fired the firearm. H2: The suspect was present in the room where the firearm was fired but did not fire it. Key assumptions: I have assumed that firing a weapon deposits residue of this composition on the shooter's hands in approximately 80 to 95% of cases (based on published studies), and that secondary transfer from a bystander is possible but substantially less likely. Findings: Gunshot residue particles consistent with primer composition were recovered from the right hand of the suspect. LR: I estimate the LR at between 100 and 1,000. Verbal equivalent: The findings provide moderately strong support for H1 compared to H2, using the ENFSI recommended scale. Limitation: This evaluation does not address whether the suspect loaded or unloaded the weapon without firing it; additional data would be needed to evaluate that alternative.'

Example 2: Flawed source-level report for an activity-level issue. 'The glass fragments recovered from the suspect's jacket are indistinguishable in refractive index from glass from the broken window at the scene. There is therefore strong evidence that the glass came from the scene window.' The legal issue is whether the suspect broke the window. The report answers a source question (where did the glass come from?) rather than an activity question (did the suspect break the window?). A bystander near the scene, a secondary transfer from clothing contact, or contamination could all explain the same finding without implicating the suspect in the breaking. The report overstates the significance of the evidence for the legal question.