If you have heard one behavioural science finding about tiredness and judgment, it is probably this one. It is also one of the most heavily contested findings in the field, and the contest is more interesting than the result.
Key Takeaway
The study found that the probability of a favorable decision drops from about 65% to almost 0% from the first ruling to the last ruling within each session, and that the rate of favorable rulings returns to 65% in a session following a food break[1]. The authors cautiously speculate that the effect might be driven by mental depletion[1]. That mechanism, ego depletion, subsequently failed a 23-laboratory replication, and a published critique argues the pattern could be produced by rational judges using only relevant information[2].
A Note On Scope
Stated first, because this concerns real legal decisions about real people.
This article reviews a research controversy. It is not legal advice, it is not a comment on any court, tribunal or jurisdiction, and it makes no claim about how any decision-maker anywhere behaves.
The study concerned parole boards in one country over a decade ago. Nothing here should be read as describing Canadian institutions or any other.
The findings are contested, and this article's purpose is to set out the contest accurately rather than to endorse either side.
Our Grades For These Claims
Applying the scheme from the first article in this series.
Grade C for the finding as usually reported. It is a single study of one institution, published in a leading journal, with three published critiques and no independent replication we located.
Grade D for the depletion explanation. The mechanism the authors speculated about failed a large multi-laboratory replication, and a further paper argues it is the wrong explanation.
Grade B that some ordering effect exists, because the critiques mostly dispute the size and the cause rather than the existence.
Our position: the pattern is probably real, the explanation is in serious trouble, and the magnitude is genuinely unresolved.
A Note On Method
Everything here is verified to August 2026.
We obtained the study's own abstract and opening pages from two independent hosted copies[3][4], and the critique's abstract describing the original's findings[1].
We did not obtain the full text of the original, nor of any critique, nor of the authors' published reply, and report those through citations and third-party summaries.
The sample details reach us through a blog, which we flag at each use.
We ran our own simulation of the alternative mechanism, and report a result that does not support the conclusion we expected.
This article is not legal advice, and the scope note above governs.
The Study
The paper.
Danziger, Levav and Avnaim-Pesso published Extraneous factors in judicial decisions in the Proceedings of the National Academy of Sciences, 108(17), 6889–6892, in April 2011[3].
Its framing, from the abstract: "Are judicial rulings based solely on laws and facts? Legal formalism holds that judges apply legal reasons to the facts of a case in a rational, mechanical, and deliberative manner. In contrast, legal realists argue that the rational application of legal reasons does not sufficiently explain the decisions of judges and that psychological, political, and social factors influence judicial rulings. We test the common caricature of realism that justice is 'what the judge ate for breakfast' in sequential parole decisions made by experienced judges."[3]
The judges had a mean experience of 22.5 years, with a standard deviation of 2.5[4].
A blog gives sample details we could not verify: 1,121 decisions, 8 Israeli parole boards serving 4 main prisons, about 40 percent of such cases in the country, with committees consisting of a judge, a criminologist and a social worker, and a working day of three sessions divided by meal breaks[5].
One observation, ours. The paper was edited and approved by Daniel Kahneman[4], which is worth recording because this series has twice previously found Kahneman engaged in correcting overstated findings, including his own.
The Finding
The result, as a critical paper's abstract states it, which we prefer to a popular summary.
"They found that the probability of a favorable decision drops from about 65% to almost 0% from the first ruling to the last ruling within each session and that the rate of favorable rulings returns to 65% in a session following a food break."[1]
Two observations, ours.
The recovery after the break is what makes the finding compelling. A simple decline could be many things; a decline that resets is a much more specific pattern.
And note the magnitude. Sixty-five percent to almost zero is not a subtle effect. This series has spent thirty articles reporting effects of a fifth of a standard deviation and smaller. An effect this large in field data should prompt a question rather than only admiration, and the second critique asks it.
The Mechanism They Proposed
What the authors suggested, with their own hedging preserved.
The paper's introduction states: "Prior research suggests that making repeated judgments or decisions depletes individuals' executive function and mental resources, which can, in turn, influence their subsequent decisions."[4]
And the critical paper's characterisation: the authors "cautiously speculate that the effect might be driven by mental depletion"[1].
Two observations, ours.
Cautiously speculate is a fair description and it is a critic's own words about the people he is criticising. The authors did not claim to have demonstrated the mechanism.
But the popular version dropped the hedge entirely, which is how a cautious speculation became hungry judges are harsher judges.
The Problem With That Mechanism
The most serious issue, and it arrived five years later. Our analysis.
The mechanism cited is ego depletion: the idea that self-control is a finite resource exhausted by use.
This publication has covered its status elsewhere. A 2016 multi-laboratory replication involving 23 laboratories and over 2,000 participants found no reliable evidence for the classic ego-depletion effect, which is why our article on founder stress and overspending explicitly declined to use decision fatigue as a load-bearing explanation.
Three consequences, ours.
The theoretical foundation the authors reached for is substantially weaker in 2026 than it was in 2011, and they could not have known that.
This does not refute their data. A pattern in parole rulings does not stop existing because a laboratory paradigm failed to replicate.
But it removes the reason to believe the pattern is psychological rather than procedural, which is precisely what the second critique proposes instead.
Critique One: The Ordering
The first published objection.
Weinshall-Margel and Shapard published Overlooked factors in the analysis of parole decisions in PNAS, 108(42), E833, in October 2011[6].
Their objection, as a professional summary describes it: "the parole board tries to cover all cases from one prison in one session, and that prisoners unrepresented by an attorney usually go last and are less likely to be granted parole"[2].
Two observations, ours.
This is the strongest possible form of objection to a field study. It says the ordering is not random with respect to the merits, so late cases differ from early ones in a way that has nothing to do with the judges.
And it is specific and checkable, unlike a general complaint about confounding, which is why it could be answered.
And The Authors' Reply
What happened next, and it matters.
The authors published Reply to Weinshall-Margel and Shapard: Extraneous factors in judicial decisions persist in PNAS, 108(42), E834[7].
A professional summary records the outcome: "In response to this critique, Danziger et al. (2011) reran their analysis, incorporating the comments, but were still able to replicate the original results."[2]
We did not obtain the reply and report only this description.
Two observations, ours.
If accurate, the first critique was answered. We report that plainly because a reader who only encountered the objection would be left with a false impression, and because this series has repeatedly criticised sources for reporting one side of an exchange.
And it is worth noting the speed and venue. Objection and reply appeared in the same issue of the same journal six months after publication, which is the correction process functioning rather than failing.
Critique Two: The Magnitude
The most interesting objection, and it is not about confounding at all.
Glöckner published The irrational hungry judge effect revisited: Simulations reveal that the magnitude of the effect is overestimated in Judgment and Decision Making, 11(6), 601–610, in November 2016[1].
The argument, as a professional summary states it: "the effect could be produced by rational judges using only relevant information. Judges have an approximate time limit for each session, and will avoid starting a new case if they expect that it will overrun. Since favourable rulings take longer than unfavourable ones, it is more likely that the last case in a session will be unfavourable."[2]
Three observations, ours.
This proposes no bias whatsoever. Every judge in the model decides purely on merit and simply manages a clock.
The pattern arises from which cases get heard, not how they are decided. Long favourable rulings cannot fit into a short remaining slot, so the late positions fill selectively with short cases.
And it is structurally identical to the artefacts this series found in the twenty-sixth and thirtieth articles: a selection process producing a pattern that looks like a psychological finding. So we tested it.
We Tested That Mechanism
Our own simulation. The parameters are ours, invented, and come from no paper.
We built sessions with a fixed time budget. Each case has a true favourable rate of 35 percent, constant at every position and never changing. Favourable rulings take longer than unfavourable ones. A judge will not start a case they expect to overrun the session.
There is no depletion in the model at all, no fatigue, and no drift in the merit rate.
Scored by distance from the end of the session, the observed favourable rate came out as follows.
Five cases from the end, 34.7 percent. Three from the end, 35.0. Two from the end, 34.2. One from the end, 33.3. And the last case in the session, 30.8 percent.
A longer session gave the same shape: 35.1, 35.0, 34.1, 32.9, and a final position of 30.9 percent.
What We Actually Found
The honest reading of our own result, including the part that does not suit the argument.
Two findings, and the second is the important one.
The mechanism works, and in the right place. The decline is negligible for most of the session and concentrated on the final case, which is exactly where the censoring bites, because a long favourable ruling is precisely what cannot fit into the time remaining.
But it does not come close to the reported magnitude. Our simulation moves the final slot by roughly four percentage points, from about 35 to about 31. The reported effect is a fall from 65 percent to almost zero.
Three things we conclude, all ours.
Our version is crude, uses invented parameters, and is not Glöckner's model. A published simulation calibrated to the real distributions could plausibly produce far more than ours does, and we have not obtained it.
So our test establishes that the mechanism is real and directional and does not establish that it accounts for the finding.
And we are reporting a result that cuts against the critique we found most persuasive. Had our simulation produced a 65 point swing we would have said so, and it did not.
Critique Three: The Explanation
A third objection, which separates the finding from its interpretation.
A legal scholar's chapter records a preprint by Daljord, Urminsky and Ureta (2017), The Status Quo Theory of Depletion Does not Explain the Israeli Parole Decisions, University of Chicago, and notes that the authors "suggest that the status quo depletion theory is the wrong explanation, but do not call the general finding of the study into question"[8].
We did not obtain this preprint and report only that characterisation.
Two observations, ours.
That second clause is the one to keep: the mechanism is disputed and the phenomenon is not, at least by these authors.
Which produces the position we set out at the top. The pattern survives three critiques in some form. What has not survived is the explanation that made it famous.
And One In The Opposite Direction
A finding that points the other way, recorded for completeness.
A reference list identifies Plonsky, O., Chen, D. L., Netzer, L., Steiner, T., and Feldman, Y. (2019), Best to Be Last: Serial Position Effects in Legal Decisions in the Field and in the Lab, Bar Ilan University Faculty of Law Research Paper[9].
We did not obtain this paper and report only its title.
One observation, ours: Best to Be Last asserts the opposite of the popular reading of the hungry judges study, in which last is worst. We do not know what the paper found, in what context, or whether it concerns the same kind of decision. We record it because a reader should know the serial position literature in law is not one-directional.
Where That Leaves It
Our own summary of a genuinely unresolved position.
Four statements we think the evidence supports.
Something happens across a session. Three critiques dispute the size or cause; we did not find one arguing the pattern is absent.
The first critique was answered by a reanalysis the authors published, on the summary we obtained.
The depletion explanation is in serious trouble, both because its laboratory basis failed a large replication and because a further paper says it is the wrong explanation.
And the magnitude is unresolved. A published simulation says it is overestimated; our own crude version of that mechanism produced only a small part of it.
What we would not say is that the study has been debunked, and what we would not say is that hungry judges are harsher judges. Both statements are more confident than anything we found.
The Use It Is Being Put To
Why the accuracy of this matters beyond an academic dispute. Ours.
The study appears in reference lists on algorithmic decision-making, alongside work on algorithm aversion, in papers concerned with moderating inconsistency between human decision-makers[10].
Three observations.
The argument it supports is powerful and simple: if human decisions turn on the time of day, a machine would be fairer. That is a serious argument and this article does not dismiss it.
But it rests on a single contested study of one institution whose proposed mechanism has since lost its laboratory support and whose magnitude is disputed in print.
And a legal scholar has written a chapter specifically on the legal implications of this study, tracing what he calls the historical development of the hungry judge slogan[8], which suggests the gap between the finding and its slogan is itself now a subject of study.
Our position: a finding used to justify replacing human judgment should be held to a higher evidentiary standard than one used to justify taking a lunch break, and this one is currently being used for both.
What This Means For Professional Work
The honest translation to an accounting or advisory practice. Ours, and modest.
Three things we would not say.
We would not tell you that your afternoon decisions are worse than your morning ones, because the evidence for that in professional work is not in this study, which concerned parole boards.
We would not attribute anything to blood sugar, which no source we obtained claims.
And we would not use this to sell a scheduling product or a policy, which is what most business writing about it does.
Two things we would say.
The censoring mechanism is worth understanding in its own right, and it does not depend on the parole study being right. If your firm reviews a queue of items against a time budget, and thorough outcomes take longer than quick ones, then the items at the end of the queue are being treated differently by the clock rather than by you.
And that is checkable in your own records, which beats any published study for your purposes.
The Usable Part
The one operational point that survives everything above. Ours.
Whether or not judges deplete, the structural fact in the alternative explanation is real: a decision that takes longer is less likely to be reached when time is short.
Three consequences.
If the thorough answer takes longer than the quick answer, then time pressure systematically favours the quick answer, with no bias and no fatigue required.
Which means the remedy is about the schedule, not about the person. Do not put the item that requires the most work at the end of the day's list.
And it is testable in any practice that records both what was decided and how long it took, which most do.
What To Do
Do not repeat the slogan. Hungry judges are harsher judges is more confident than the evidence, and the authors themselves only cautiously speculated about the mechanism.
Do not treat it as debunked either. The first critique was answered by a published reanalysis, and no source we found argues the pattern is absent.
Separate the finding from the explanation. One paper's own title says the depletion theory does not explain the result while not questioning the result.
Note that the mechanism failed elsewhere. Ego depletion, which the paper cited, failed a 23-laboratory replication in 2016.
Understand the censoring mechanism on its own merits. If thorough outcomes take longer, time pressure favours quick ones with no psychology required.
Schedule the heavy item early. That follows from the structural point and needs no contested study to support it.
Be careful what you build on it. A single contested study of one institution is thin ground for replacing human judgment with an algorithm.
Check your own queue. Whether outcomes differ by position is answerable from records most practices already keep.
The Limits Of This Analysis
Several caveats matter. This article reviews a research controversy. It is not legal advice, is not a comment on any court, tribunal or jurisdiction, and makes no claim about how any decision-maker behaves; the study concerned parole boards in one country over a decade ago. Everything is verified to August 2026. We did not obtain the full text of the original study, only its abstract and opening pages from two hosted copies. We did not obtain any of the three critiques, the authors' published reply, or the paper pointing in the opposite direction, and report all of them through citations, abstracts and third-party summaries. The sample details come from a blog and could not be verified. The account of the authors' successful reanalysis comes from a professional advisory firm's summary, not from the reply itself. Our simulation is entirely ours, uses invented parameters, is not the published model, and produced only about four percentage points against a reported effect of sixty-five; it establishes that the mechanism is real and directional and does not establish that it accounts for the finding. The ego depletion replication is described from this publication's own earlier coverage rather than from the replication paper. Every application to professional practice is our own extension to a setting this research did not examine.
Frequently Asked Questions
Is the hungry judges finding true?
What is wrong with the explanation?
What is the alternative explanation?
Does that explain the whole effect?
Should this affect how I schedule work?
Why does the accuracy matter?
References
- Glöckner, A. (2016). The irrational hungry judge effect revisited: Simulations reveal that the magnitude of the effect is overestimated. Judgment and Decision Making, 11(6), 601–610, published abstract, on Danziger, Levav and Avnaim-Pesso (2011) having analyzed legal rulings of Israeli parole boards concerning the effect of serial order in which cases are presented within ruling sessions; on their finding that the probability of a favorable decision drops from about 65% to almost 0% from the first ruling to the last ruling within each session and that the rate of favorable rulings returns to 65% in a session following a food break; and on the authors arguing that these findings provide support for extraneous factors influencing judicial decisions and cautiously speculating that the effect might be driven by mental depletion. Note: we obtained the published abstract only, not the paper or its simulations. doaj.org
- Professional economics advisory firm's article on behavioural biases in the judiciary, on one challenge to the study having been that the parole board tries to cover all cases from one prison in one session and that prisoners unrepresented by an attorney usually go last and are less likely to be granted parole; on Danziger and colleagues having reran their analysis in response, incorporating the comments, and still being able to replicate the original results; and on a further critique arguing the effect could be produced by rational judges using only relevant information, since judges have an approximate time limit for each session and will avoid starting a new case if they expect it to overrun, and since favourable rulings take longer than unfavourable ones it is more likely that the last case in a session will be unfavourable; with its reference list confirming the citations for Danziger and colleagues (2011), Weinshall-Margel and Shapard (2011), Glöckner (2016), and Chen and Philippe (2020). Note: a professional advisory firm's article, not peer-reviewed. This is our sole source for the outcome of the authors' reanalysis and for the summary of the second critique's argument; we obtained neither the reply nor the critique. oxera.com
- Danziger, S., Levav, J., & Avnaim-Pesso, L. (2011). Extraneous factors in judicial decisions. Proceedings of the National Academy of Sciences, 108(17), 6889–6892. DOI 10.1073/pnas.1018033108, PMID 21482790, published abstract, on whether judicial rulings are based solely on laws and facts; on legal formalism holding that judges apply legal reasons to the facts of a case in a rational, mechanical and deliberative manner while legal realists argue that the rational application of legal reasons does not sufficiently explain judicial decisions and that psychological, political and social factors influence rulings; and on the authors testing the common caricature of realism that justice is what the judge ate for breakfast in sequential parole decisions made by experienced judges. Note: we obtained the abstract and opening only, not the full paper, and report no statistics in the authors' own voice. pmc.ncbi.nlm.nih.gov
- Hosted copy of Danziger, Levav and Avnaim-Pesso (2011), on the authors' affiliations at Ben Gurion University of the Negev and Columbia Business School; on the paper having been edited and approved by Daniel Kahneman of Princeton University on 25 February 2011, having been received for review on 8 December 2010; on the judges having a mean experience of 22.5 years with a standard deviation of 2.5; on the realist view being commonly caricaturized by the trope that justice is what the judge ate for breakfast; and on prior research suggesting that making repeated judgments or decisions depletes individuals' executive function and mental resources, which can in turn influence their subsequent decisions. Note: a hosted teaching copy; we obtained the opening pages only. pages.stat.wisc.edu
- Neuroscience blog, on the study having analyzed 1,121 decisions of 8 Israeli parole boards serving 4 main prisons, corresponding to about 40 percent of the cases of such committees in the country; on the committees consisting of a judge, a criminologist and a social worker; on the working day consisting of 3 sessions divided by meal breaks; and on the percentage of positive decisions being about 65 percent at the beginning of each session and falling almost to zero at the end. Note: a blog, not peer-reviewed. These sample details could not be verified against the paper. mozgowiec.com
- Reference lists identifying Weinshall-Margel, K., & Shapard, J. (2011), Overlooked factors in the analysis of parole decisions, Proceedings of the National Academy of Sciences, 108(42), E833, DOI 10.1073/pnas.1110910108. Note: a citation only. We did not obtain this letter and report its argument only as the professional summary at reference 2 describes it. cambridge.org
- Reference list identifying Danziger, S., Levav, J., & Avnaim-Pesso, L. (2011), Reply to Weinshall-Margel and Shapard: Extraneous Factors in Judicial Decisions Persist, Proceedings of the National Academy of Sciences, 108(42), E834, DOI 10.1073/pnas.1112190108. Note: a citation only. We did not obtain the reply; its outcome is reported in this article solely on the summary at reference 2. mozgowiec.com
- Legal scholar's draft chapter on the legal implications of the hungry judge study, tracing the historical development of the hungry judge slogan, and recording in its footnotes Glöckner (2016), Judgment and Decision Making, 11, 601–610; Daljord, O., Urminsky, G., & Ureta, J. M., The Status Quo Theory of Depletion Does not Explain the Israeli Parole Decisions, preprint, University of Chicago, December 2017, whose authors suggest that the status quo depletion theory is the wrong explanation but do not call the general finding of the study into question; and Weinshall-Margel and Shapard (2011), PNAS, 108, E833. Note: a draft chapter hosted on an academic's personal site; we obtained portions and did not obtain the preprint it describes. chrisbublitz.de
- Reference list identifying Plonsky, O., Chen, D. L., Netzer, L., Steiner, T., & Feldman, Y. (2019), Best to Be Last: Serial Position Effects in Legal Decisions in the Field and in the Lab, Bar Ilan University Faculty of Law Research Paper. Note: a citation only. We did not obtain this paper and report its title alone, which points in the opposite direction to the popular reading of the study discussed here; we do not know what it found or in what context. mozgowiec.com
- Academic preprint on moderating inconsistency between human decision-makers, whose reference list cites Danziger, Levav and Avnaim-Pesso (2011) alongside work on algorithm aversion, including Dietvorst, Simmons and Massey (2015), Algorithm aversion: People erroneously avoid algorithms after seeing them err, Journal of Experimental Psychology: General, 144(1), 114, and Dietvorst, Simmons and Massey (2018), Overcoming algorithm aversion, Management Science, 64(3), 1155–1170. Note: a preprint; cited here only to evidence that the parole study appears in reference lists on algorithmic decision-making. We obtained none of the works listed. arxiv.org
This article reviews a research controversy. It is not legal advice, is not a comment on any court, tribunal or jurisdiction, and makes no claim about how any decision-maker behaves. No paper discussed was obtained in full. The account of the authors' successful reanalysis rests on a single professional advisory firm's summary. The simulation reported is the authors' own, uses invented parameters, is not the published model, and produced about four percentage points against a reported effect of sixty-five.