Forensic likelihood ratios are meant to bring discipline to the courtroom: a clear, logical way of saying how strongly evidence supports one explanation over another. In theory, the number does the job neatly. In practice, it often lands with a thud. Presiding Officers, lawyers, and even experts may hear "30 times more likely" and quietly translate it into something it does not mean.

Recent work by Thompson et al. and Morrison et al. shows the problem is not simply whether likelihood ratios are scientifically defensible. The harder question is whether they are being communicated in a way that non-specialists can actually understand and use without falling into statistical traps. The two papers are close relatives, but they do different jobs. Thompson et al. 2025 is an empirical study: it asks whether explaining likelihood ratios in videoed expert testimony improves lay understanding. Morrison et al. 2025 is a review paper: it surveys prior empirical work and asks whether a particular presentation format, if any, is best for legal decision-makers.

Both documents align in several important ways in their treatment of likelihood ratios.

First, they endorse the likelihood-ratio framework as the logically correct framework for evaluating forensic evidence. They also use and prioritise numerical likelihood ratios, treating them as central to forensic reporting rather than relying primarily on vague verbal expressions. Both documents emphasise the need to distinguish source-level claims from guilt- or offence-level claims, recognising that a likelihood ratio should address the strength of the evidence under competing hypotheses, rather than directly stating the probability of guilt. They also use the effective likelihood ratio as a diagnostic tool to assess how people actually interpret and apply the evidence. In addition, both evaluate understanding through the related concepts of sensitivity, orthodoxy, and coherence. Finally, both documents suggest that while people may often recognise that a larger likelihood ratio represents stronger evidence, this basic sensitivity is easier to achieve than full Bayesian understanding.

The main point of divergence is not a direct contradiction, but a difference in emphasis.

Thompson et al. test the hope that "explain the number better" is the solution. Morrison et al. conclude that neither explanation nor alternative formats have yet solved the communication problem. That is the main tension. Thompson is experimentally narrow and explanation-focused; Morrison is review-based and format-sceptical. The former asks, "Does explanation help?" The latter effectively answers, "Not enough evidence and no format gets a clean win."

The real challenge with forensic likelihood ratios is not whether the numbers are scientifically defensible; it is whether they survive translation into the courtroom.

Thompson et al. show that even a careful explanation may do little, while Morrison et al. show that alternative formats have not yet produced a clear solution either.

The safest path, for now, is not to abandon likelihood ratios, but to present them with discipline: use the number, define the competing hypotheses, keep the claim at the source level, explain that the ratio is not a probability of guilt, and avoid dressing uncertainty in vague verbal labels. In short, the number may be right, but unless the audience understands what it means and what it does not mean, it risks becoming just another impressive-looking statistic lost in translation.

The overlap is substantial, but the editorial punchline is stark:

Neither paper gives a silver-bullet answer. Anyone pretending otherwise is selling courtroom numeracy snake oil.

Read the articles via the link in the source.

Sources: 

Morrison, G. S., Bali, A. S., Martire, K. A., Grady, R. H., & Thompson, W. C. (2025). What is the best way to present likelihood ratios? A review of past research and recommendations for future research. Science & Justice, 101342. https://doi.org/10.1016/j.scijus.2025.101342

Thompson, W. C., Grady, R. H., & Morrison, G. S. (2025). Does explaining the meaning of likelihood ratios improve lay understanding?. Science & Justice, 101352. https://doi.org/10.1016/j.scijus.2025.101352

Image: OpenAI. (2026). Forensic likelihood ratios in court: An AI-generated illustration of evidence, uncertainty, and lay understanding [AI-generated image]. ChatGPT.