When Does New Evidence Matter More? A Bayesian Perspective for Employment Lawyers

Employment Tribunal litigation is, at its core, a contest between competing narratives. The claimant offers one account of events; the respondent offers another. Both sides marshal evidence to support their version. But here's a question that rarely gets explicit attention: does it matter when a party learned about a piece of evidence relative to when they formulated their case theory?

A recent paper in Law, Probability and Risk by Julia Mortera (University of Bristol) and William C. Thompson (UC Irvine) offers a rigorous framework for thinking about this question—and the answer has practical implications for how we evaluate evidence in employment disputes.

Prediction vs Accommodation

Philosophers of science have long debated whether "predicted" evidence—facts that a theory anticipated before they were discovered—should count for more than "accommodated" evidence—facts that were already known when the theory was developed.

The intuition is straightforward: if I tell you my client was dismissed because of her pregnancy, and we later discover contemporaneous emails proving the decision-maker knew about the pregnancy, that feels more compelling than if those emails were the starting point for our case theory. The theory predicted the emails would exist.

But is this intuition reliable? Mortera and Thompson use Bayesian network analysis to show that the answer is: sometimes yes, sometimes no—and the difference depends on the logical structure of the evidence.

The Murder Case Illustration

The authors analyse a hypothetical criminal case (based on a real California matter). A defendant charged with murder claims a third party—"a Samoan"—committed the crime. DNA evidence later emerges showing the unknown perpetrator's profile is 300 times more likely if they were Samoan than if they were from another ethnic group.

The critical question: did the defendant know about this DNA evidence when he formulated his defence?

Using the interactive model below, you can explore the Mortera-Thompson Framework. The key result: setting "KnewDNA" to No yields P(Guilty) ≈ 36%, while setting it to Yes yields P(Guilty) ≈ 91%—the same evidence, radically different conclusions.

The same evidence, the same claim, but radically different probative value depending on the order of discovery.

Application to Employment Tribunals

This framework translates directly to common employment scenarios:

Constructive Dismissal: A claimant resigns and later articulates a "last straw" narrative connecting various incidents. If documents subsequently emerge that fit this narrative perfectly, was the theory predicted or accommodated? Did the claimant identify the pattern before seeing their personnel file, or after?

Victimisation Claims: A respondent asserts they dismissed for capability reasons. If performance reviews emerge showing concerns predating any protected act, does it matter whether the respondent located these documents before or after formulating their defence?

Whistleblowing: The claimant alleges detriment followed disclosure. The respondent produces evidence of a pre-existing restructuring plan. The probative value of that plan depends partly on whether it was known to the respondent's legal team before they committed to their case theory.

The Structural Insight

What makes Mortera and Thompson's analysis powerful is their demonstration that the order of discovery only matters when evidence items are conditionally dependent. If one piece of evidence affects the probability of another, then knowing about evidence A before claiming theory B creates a dependency that undermines B's credibility.

But when evidence items are conditionally independent—each stands or falls on its own merits regardless of the others—the order of discovery is irrelevant.

In the authors' model, witness testimony that "no one else was seen near the crime scene" has the same probative value regardless of when it emerged. But the defendant's claim about the killer's ethnicity, combined with the DNA evidence supporting that claim, creates a dependency that the order of discovery fundamentally alters.

Practical Takeaways

• For Claimants: Document your understanding of events before disclosure. A contemporaneous grievance or resignation letter that accurately predicts what documents will show is powerful evidence that your narrative is genuine, not reverse-engineered.
• For Respondents: Be cautious about case theories that fit disclosed documents too neatly. A Tribunal may reasonably wonder whether the explanation was crafted to accommodate known evidence rather than reflecting genuine recollection.
• For Cross-Examination: "When did you first form this view?" and "What documents had you seen at that point?" are questions that go to the heart of credibility. The Mortera-Thompson framework explains why.
• For Expert Evidence: In complex cases involving statistical evidence (DNA, equal pay comparators, redundancy selection scoring), consider whether the analysis was specified before or after seeing the results. Pre-registered methodologies carry more weight.

The Ethical Dimension

The authors conclude with a provocative observation: defence lawyers often learn about exculpatory evidence through their clients, and may be uniquely positioned to know whether the client's account was predicted or accommodated. This creates tension with the duty to present evidence without misleading the court.

Employment lawyers face analogous situations. When a client's account shifts to accommodate discovered documents, we must consider whether the adjusted narrative can be presented with integrity—or whether the shift itself is a fact the Tribunal should know.

The full paper is open access and rewards careful study. For practitioners interested in probabilistic reasoning and evidence, it offers a masterclass in applying Bayesian methods to legal fact-finding.