Data Mining The Decision-Maker: Can Adjudicator Statistics Prove Bias?

In an era of increasing data analytics, it was perhaps only a matter of time before litigants attempted to "moneyball" the justice system. A recent decision¹ from the Human Rights Tribunal of Ontario (the "Tribunal") addressed a novel attempt by an unsuccessful Applicant to overturn a decision based not on legal errors, but on a statistical analysis of the adjudicator's past rulings.

In Safavi-Naini v. Gauthier, the Tribunal found that the Applicant's "win rate" table, which relied on outdated and incomplete data, was insufficient to establish a reasonable apprehension of bias. The decision provides reassurance regarding the finality of decisions and confirms that allegations of bias require specific, evidentiary grounds rather than general statistical trends.

The Background

The dispute arose in the context of a medical residency program. The Applicant, a medical resident, filed an application alleging sexual harassment and discrimination based on race, ethnic origin, and sex. She made serious allegations against a supervising physician in her residency program and his brother, a lawyer who served on the board of the fundraising arm of the hospital where she was based. The allegations included claims of sexual harassment during training rotations and threats of institutional reprisal.

On September 19, 2025, the Tribunal dismissed the Application on its merits.

Dissatisfied with the result, the Applicant sought a reconsideration of the decision. However, she did not argue that the Tribunal had made an error of law or fact in the traditional sense. Instead, she argued that the decision should be set aside because of a "reasonable apprehension of bias" based on a statistical analysis of the adjudicator's previous decisions.

The "Statistical Method" Argument

The Applicant submitted two tables to the Tribunal in support of her request. The first table purported to represent a "statistical method for assessing judicial bias," claiming to show the "win rate" for applicants before various adjudicators. Based on this data, the Applicant argued that the adjudicator in her case displayed a "significant favouring of respondents."

The second table served as the "raw data" for these conclusions, listing decisions from 2011 to 2025. The Applicant argued that these statistics demonstrated that the adjudicator was predisposed to rule against applicants, thereby justifying a reconsideration of the dismissal.

What Did the Tribunal Decide?

The Tribunal dismissed the request for reconsideration, dismantling the statistical argument on both evidentiary and legal grounds.

First, the Tribunal noted significant flaws in the data itself. The "raw data" was incomplete and outdated, listing adjudicators who had left the Tribunal years prior while omitting many current decisions. Regarding the specific adjudicator in question, the table accounted for only 16 of nearly 300 issued decisions. The Tribunal found that such incomplete data could not possibly support the sweeping conclusions suggested by the Applicant.

More importantly, the Tribunal rejected the premise that a statistical analysis of past decisions is sufficient to establish bias, finding that such data does not substitute for the legal requirement to prove, through a decision-maker's words and conduct, the presence of a closed mind. Relying on established Supreme Court of Canada jurisprudence², the Tribunal reiterated that the test for reasonable apprehension of bias is whether an informed, right-minded person, viewing the matter realistically, would think it more likely than not that the decision-maker would not decide fairly.

To meet this threshold, a party must point to specific words or conduct demonstrating that the decision-maker did not have an open mind. A general assertion based on past outcomes does not satisfy this requirement. The Tribunal found that the Applicant's speculative statistical evidence was not a factor that outweighed the public interest in the finality of decisions.

Key Takeaways

This decision serves as a critical reminder that the threshold for establishing adjudicative bias remains high. Litigants cannot rely solely on data scraping or "judicial analytics" to circumvent a Tribunal decision they dislike. The Tribunal reiterated that the legal test for bias focuses on the adjudicator's specific words and conduct, rejecting the Applicant's attempt to satisfy this threshold through an aggregation of past outcomes.

Ultimately, this ruling underscores that abstract statistical arguments are unlikely to constitute the "compelling and extraordinary circumstances" required to outweigh the strong public interest in the finality of Tribunal decisions. It confirms that Tribunal decisions will not be easily upended by creative but legally unsound theories regarding an adjudicator's track record.

Footnotes

1. Safavi-Naini v. Gauthier, 2025 HRTO 2671

2. Committee for Justice and Liberty et al. v. National Energy Board et al., 1976 CanLII 2 (SCC)