If DNA Evidence Excludes 99.9% of People, Why Is a Defendant Not 99.9% Likely to Be Guilty?

Geoffrey Chen

When Numbers Start Making Decisions · Season Five, “Things That Have Not Happened Yet” · Article 9

1. One statistic, two very different impressions

In 2012, the High Court of Australia decided Aytugrul v The Queen. A hair found under the deceased’s fingernail had a mitochondrial DNA type matching the accused. The expert evidence included two descriptions: about one person in 1,600 might share the type, and about 99.9% of people could be excluded. High Court of Australia: Aytugrul v The Queen

The statements convey similar population frequency information but have different psychological force. “One in 1,600” keeps a relevant population in view. “Excludes 99.9%” can be completed in the listener’s mind as “it is almost certainly the accused” and then as “the accused is almost certainly guilty”.

The High Court dismissed the appeal. The majority considered that the relationship between the frequency and exclusion percentage had been explained at trial and did not, in the full evidentiary context, create unfair prejudice requiring exclusion. But the judgment did not declare that 99.9% was a probability of guilt. It demonstrated the need to manage the distance between a laboratory proposition and a legal conclusion.

2. DNA statistics first answer whether a trace is consistent with a source

Forensic comparison commonly evaluates hypotheses such as: how likely is the observed DNA result if the material came from the accused, and how likely is it if it came from an unrelated person? Modern reports may express relative support using a likelihood ratio. The mitochondrial evidence in Aytugrul was presented through population frequency and exclusion percentage.

Those quantities primarily address source level: who could be the origin of biological material. A criminal trial must eventually address activity and offence: how the trace arrived, what the accused did and whether the prosecution has excluded reasonable doubt.

A person can be the source without depositing the trace during the offence. Material may remain from innocent contact, be transferred secondarily or predate the event. Mixed samples, degradation, contamination and interpretation methods also affect strength. Even very strong source evidence needs a timeline, location, witnesses, alternative explanations and other evidence to connect it to criminal conduct.

This does not weaken DNA science. It keeps the science within the question it can answer.

3. How the prosecutor’s fallacy occurs

The prosecutor’s fallacy reverses conditional probabilities. A low probability that a random unrelated person would match the trace is not the probability that the accused is innocent given the match. Nor does subtracting that frequency from one produce the probability of guilt.

Selection of the suspect also matters. If police identify a person through independent evidence and then compare DNA, the statistical context differs from a database search that generates a candidate. Database size, relatedness, population frequency and the search process affect interpretation.

The denominator was important in Aytugrul. A broad population estimate may differ from the frequency in a relevant subpopulation. The case also contained evidence that the type was more common in the accused’s ethnic group. An exclusion percentage without its database and reference population encourages the listener to substitute an imagined national population for the real candidate group.

DNA also cannot supply prior case information. It does not say why the accused became a suspect, whether the trace is contemporaneous with the crime or who else had a reasonable opportunity for contact.

4. Evidence strength and decision strength are separate

DNA can properly support an investigation, exclude a suspect, justify further evidence collection and contribute strongly at trial. Its capacity to correct wrongful convictions is profound. Difficulty with probability is not a reason to deny those uses.

But consequences intensify from investigation to arrest, bail and conviction, and the evidentiary structure must thicken with them. A database hit can open an investigation; it should not complete it. A strong source likelihood ratio can be an important part of a case; it cannot replace analysis of transfer and activity.

Beyond reasonable doubt is a legal standard applied to the evidence as a whole, not a laboratory percentage. Even if the ratio is vastly greater than 1,600, the logical boundary remains: it changes support for a proposition and does not directly compute a posterior probability of guilt.

The direction is asymmetric. A valid non-match can strongly exclude a person as the source. A match retains the person among possible sources and increases support according to statistical strength. Exclusion and identification are not mirror-image decisions.

5. Presentation changes judgement

One in 1,000 and 99.9% exclusion can be mathematically related while directing attention differently. The first keeps possible matching people visible; the second emphasises a vast excluded group. The High Court recognised that frequency and exclusion expressions may communicate differently even when they derive from the same analysis.

Statistical communication is therefore part of evidence law. Expert reports should state the propositions, database, assumptions, error and limits. Judicial directions should guard against reversing conditional probabilities. Both parties require sufficient material for independent review.

Equivalent statistics should not be stacked as if they were independent confirmations. Frequency, exclusion rate and a likelihood ratio derived from the same test are not three separate pieces of DNA evidence. Repetition can amplify psychological weight without adding information.

6. Error can enter anywhere in the chain

Forensic error is not limited to a laboratory calculation. Collection can contaminate or mislabel material. Chain of custody can fail. A sample can contain mixtures. A database may be unsuitable. A statistical proposition may be misaligned with the legal issue. Courtroom language can then introduce another transformation.

Review must trace the whole chain: who collected what and where, how it was packaged, laboratory method and version, quality controls, preserved electropherograms and interpretation, and whether the defence can repeat the work.

If systemic contamination or a software defect is discovered, correction cannot apply only to future cases. Institutions must identify affected convictions, notify parties, preserve material for retesting and provide genuine appeal and compensation pathways. When liberty was wrongly removed, better wording in the next report is not a sufficient remedy.

7. Powerful knowledge remains limited knowledge

DNA evidence forms among bodily traces, collection procedures, laboratory systems, population data, statistical assumptions and courtroom propositions. No one person possesses the whole process, yet knowledge can travel across institutions.

That distribution creates a duty to preserve role boundaries. A laboratory knows about samples, not the complete activity. Police know an investigation, not guilt. A jury judges the case but cannot rerun the analysis. Each actor receives limited knowing and limited authority.

The more impressive the number, the more tempting it becomes to abandon judgement. The court retains the human commitment to reason across the evidence. It cannot say that 99.9% has already decided.

Conclusion: DNA can strongly support source, but cannot calculate guilt alone

DNA statistics should be presented through clearly stated propositions and likelihood ratios or frequencies. Exclusion percentages may also be used when the denominator, database, assumptions and distance from the guilt question are explained. Aytugrul should not be read as permission to substitute 99.9% for proof as a whole; its outcome depended on the context and explanation at trial.

Courts should maintain independent review for database searches, mixtures and software analysis and separate source, activity and offence levels. However enormous a ratio becomes, it can carry only the decision authority its measurement supports.

DNA’s power lies partly in reliably excluding many people. Criminal justice remains responsible for not transforming “one of those who remain” into “the person who committed the crime” before the rest of the evidence is proved.

Primary sources

Discover more from Geoffrey Chen

Subscribe to get the latest posts sent to your email.