The MCLC is one of the largest in the developing world, with around 25,000 active cases and 5,000 new cases filed each year. By law, the court is meant to resolve cases within 100 days of filing, but typically takes several years, with almost one-third of cases filed in 2012 unresolved after four years.

One reason for the long case durations is that parties are over-optimistic about the likelihood they will prevail. When the researchers asked plaintiffs and defendants how likely it was they would win the case, the two probabilities summed to almost 150 per cent.

Over the past two years, the team have worked with the MCLC to test innovations to make the court more efficient. The first provided parties to a suit with information on average settlement amounts from previously settled cases with characteristics that matched their own. The researchers tested whether this information helped to overcome bargaining impasses by giving parties a realistic starting point for negotiations. They used machine-learning techniques to develop personalised predictions based on the outcomes of 5,000 cases filed in 2012 and resolved before the end of 2016.

They compared the effect of the information on predicted outcomes with two alternatives in a randomised control trial. One alternative was a meeting with the court’s conciliator; each of the MCLC’s 20 subcourts is assigned a conciliator who, if the parties to a case desire, attempts to intermediate a settlement between them. They randomly assigned some hearings to receive encouragement to sit with the conciliator. The second alternative was simply the status quo. That is, some hearings were assigned to proceed exactly as they normally would.

The researchers found that the parties reached a settlement in about 6 per cent of the hearings where they did not intervene. Either of the interventions almost doubled this rate. Parties were also about five times more likely to reach a settlement when the plaintiff was present in person. (In many hearings, only the two lawyers are present.) It turns out that the researchers’ two interventions were also more effective when the plaintiff was present at the hearing. Indeed, they found that providing the two lawyers with predictions for settlement amounts had no effect whatsoever, either on the day or – using administrative records for the cases – for up to six months after the information was provided. This raised questions about whether the lawyers had the plaintiff’s best interest in mind, an issue the team are exploring in ongoing work with the court.

The research is being used by both the court and lawmakers in Mexico to design new rules governing labour litigation. In 2017, Mexico passed a constitutional reform mandating conciliation hearings prior to the filing of a lawsuit. The research was the first evidence of the effectiveness of these hearings, albeit with cases that had already been filed. In ongoing work, the team are testing the effect of arranging hearings before cases are filed, as the new law mandates.