Hungry Judge Effect

What it is

The hungry judge effect comes from a paper by Shai Danziger, Jonathan Levav and Liora Avnaim-Pesso, "Extraneous factors in judicial decisions", published in the Proceedings of the National Academy of Sciences in 2011. The authors analysed 1,112 parole rulings by eight judges, drawn from two parole boards and 50 days of sessions over ten months. Each day had three sessions, separated by a late-morning snack and by lunch. Within a session, the share of favourable rulings stood at about 65 per cent at the start or just after a break and fell to nearly zero by the end. Over the whole sample, 64.2 per cent of requests were rejected.

The authors read the pattern as a case of mental depletion. A tired judge, on this account, falls back on the easy option, which in parole decisions is to leave things as they are and refuse. The break restores the judge, and the cycle starts again. The paper frames the question as a test of the legal realist caricature that justice is what the judge ate for breakfast. The label hungry judge effect is later shorthand, and it puts weight on food that the data cannot separate from the passage of time within a session.

The authors also addressed the obvious worry about case order. They reported that order was set mostly by when each prisoner's lawyer arrived, that it appeared to be exogenously determined, and that the judges set the timing of breaks without knowing which cases came next. They did not describe the order as random.

In effect

In research: the critique from case order

In the same journal, Keren Weinshall-Margel and John Shapard argued that case order within a session was not random. From interviews with people in the process, they reported that prisoners without a lawyer tend to be heard last and are less likely to be granted parole. If so, part of the late-session decline reflects who is heard last and not how tired the judge is. As Andreas Glöckner later summarised the exchange, Danziger and colleagues replied that the downward trend survived when representation was controlled for.

In statistics: the simulation

In 2016 Glöckner published simulations in Judgment and Decision Making. Favourable cases took longer than unfavourable ones, on average 7.37 minutes against 5.21. A judge who ends a session rather than start a case likely to overrun would therefore see favourable rulings thin out toward the end, with no fatigue involved. His simulated judge produced a decline comparable to the reported one, though somewhat smaller. He noted that the reported effect was more than twice the size usually counted as large, and he allowed that some of the original pattern remains unexplained.

In commerce and management: the borrowed lesson

The finding is still quoted in management training. Turning it into a rule for when to schedule important decisions would rest on a disputed dataset, and a dispute over case order and session planning is a poor foundation for a rule about when to make a sales call or approve a budget.

The figures are worth stating exactly, because the critique turns on them. Andreas Glöckner's 2016 paper puts the reported drop at 65 per cent to about 5 per cent and calculates that it corresponds to an odds ratio of 35, or a standardised effect of about 1.96, more than twice the conventional limit for a large effect. His simulation of a judge who plans the session found that favourable rulings averaged 7.37 minutes (standard deviation 5.11) against 5.21 minutes (standard deviation 4.97) for unfavourable ones, and that the simulated decline came out somewhat smaller than the original. He concluded that rational time management and selective dropout explain the pattern only in part, so the remainder is unexplained and not refuted. The original authors' reply, as Glöckner reports it, was that the downward trend also holds when representation by an attorney is controlled.

What it does not say

It does not say that the 2011 data are fraudulent or that the paper was carelessly done. The authors reported their sampling and their view of case order openly, and the critics worked from the same published figures. It does not say that hunger explains the pattern: the paper shows a decline within sessions and a recovery after breaks, and the data cannot separate food from rest, from the passage of time or from the order of cases. It does not say that the critique removes the whole effect. Glöckner's simulated decline was somewhat smaller than the reported one, and he allowed that some of the original pattern remains unexplained. The honest position is that the finding is contested, and that it should be cited with the critique beside it.

---