Skip to main content
Category: Effects
Type: A later decision, in a sequence of similar decisions, moving toward the option that ends the question with less work. In the counted file that option is a rejection, and a delay counts as one
Origin: The name is John Tierney, “Do You Suffer From Decision Fatigue?”, New York Times Magazine, online 17 August 2011. The print version is 21 August 2011, page 33, under the headline “To Choose is to Lose.” The Times page says the essay is adapted from a book with Roy F. Baumeister, Willpower: Rediscovering the Greatest Human Strength, then still forthcoming. The essay body past the paywall, and the book, were not opened. The counted case is Shai Danziger, Jonathan Levav, and Liora Avnaim-Pesso, “Extraneous factors in judicial decisions,” Proceedings of the National Academy of Sciences 108(17), 26 April 2011, pages 6889–6892, DOI 10.1073/pnas.1018033108. The journal HTML was read. That paper does not print the words decision fatigue
Also known as: Mental depletion, the interpretation in the 2011 discussion. It is not Choice Overload as one score, and it is not the Status Quo Bias as one score
Quick Answer — Decision Fatigue is the claim that a run of decisions pushes the next one toward the easier option. Tierney used the name in 2011 for a line of work on a limited resource. The counted file is 1,112 rulings by eight Israeli parole judges. Within each sitting the favorable share falls from about 65 percent to nearly zero, then returns to about 65 percent after a food break. The authors did not measure the resource, and a later simulation can draw a large part of a similar curve without it.

​
What is Decision Fatigue?

Decision Fatigue is the claim that after a run of decisions, the next decision slides toward the option that takes less work.
We find that the percentage of favorable rulings drops gradually from ≈65% to nearly zero within each decision session and returns abruptly to ≈65% after a break.
That sentence is Danziger, Levav, and Avnaim-Pesso (2011). A decision session is the run of cases between food breaks. The day has three of them: before a late-morning snack, between that snack and lunch, and after lunch. The drop is inside each run. It is not one slope from the first case of the morning to the last case of the afternoon. In this file the easier ruling is the one that keeps things as they are. The authors call that the status quo, and they count a rejection as including a delay. A delay was a review about a month later, so the prisoner stayed put. A favorable ruling took longer in the room and used more words on the page. Status Quo Bias is a preference for the current state when the current state and a change are both on the table. This page asks whether the share of the easier ruling rises with the case’s place in the run. Choice Overload is the burden of many alternatives at one moment. Decision Fatigue is the place of a decision in a sequence. A menu of fifty jams is the first. The twelfth similar yes-or-no, after eleven others, is the second. Default Effect is what happens when one option is preselected. Here the easier option is the authors’ reading of a rejection. The form did not arrive with a box already ticked. The Hot–Cold Empathy Gap is the difficulty of simulating hunger, fatigue, or craving while that state is quiet. Hunger is one candidate in the parole discussion. It is not a measured state. Mood was not recorded. A useful picture is a stack of similar forms and a stamp that says “not now.” The first forms get a written yes or no. Later forms get the stamp, because the stamp ends the question. The picture fails if the later forms were harder, or longer, or missing an advocate. Those are the objections below. They are why the stamp is not, by itself, a tired mind.

​
Decision Fatigue in 3 Depths

  • Beginner: After several similar decisions, notice whether the next one is the option that ends the question. In this file that option was a rejection, and a promise to look again in about a month counted as a rejection.
  • Practitioner: Put the decision you would regret defaulting into the early slots of a sitting, and put a real stop between sittings. The paper cannot say the food is what changed the next ruling. The break and the meal were the same event.
  • Advanced: A curve that falls and then resets is not a reading of a depleted resource. The authors say they had no such reading. A later simulation can produce a large part of a similar curve because favorable rulings take longer, so a sitting that must stop tends to leave those rulings out of the late slots.

​
Origin

The two words are the magazine’s. Tierney’s essay of 17 August 2011, in print on 21 August, uses Decision Fatigue for the idea that choosing spends the same limited resource as other acts of self-control. The Times page says the essay is adapted from Willpower, the book he wrote with Baumeister. The book was not opened. The essay past the paywall was not opened. This page does not treat a retelling of that essay as a second data set. The laboratory parent, as the parole paper cites it, is older and uses other words. Reference 6 is Mark Muraven and Roy F. Baumeister, “Self-regulation and depletion of limited resources: Does self-control resemble a muscle?”, Psychological Bulletin 126 (2000), pages 247–259. Reference 8 is Kathleen D. Vohs and colleagues, “Making choices impairs subsequent self-control,” Journal of Personality and Social Psychology 94 (2008), pages 883–898. Reference 14 is Martin S. Hagger, Chantelle Wood, Chris Stiff, and Nikos L. D. Chatzisarantis, “Ego depletion and the strength model of self-control: A meta-analysis,” Psychological Bulletin 136 (2010), pages 495–525. None of those three PDFs was opened. The parole paper’s own sentences are the ones used here. It says repeated judgments can deplete executive function, that sequential consumer choices can increase intuitive decision making and reduce pain tolerance, and that German car buyers were more likely to accept a manufacturer’s default later in a sequence of attribute choices. Those three findings are citations. Their designs and their means are not on this page. The parole paper was edited by Daniel Kahneman. It was received on 8 December 2010 and approved on 25 February 2011. The authors’ question is the legal-realist caricature that justice is “what the judge ate for breakfast.” They cite that phrase to Alex Kozinski, Loyola of Los Angeles Law Review 26 (1993), page 993. They say their data cannot test the breakfast directly. They can test whether a break taken in order to eat lines up with a change in the ruling.
However, we cannot unequivocally determine whether simply resting or eating restores the judges’ mental resources because each of the breaks was taken for the purpose of eating a meal.
That sentence is the discussion. The same paragraph says mood was not measured, and that mental resources were not measured, so the authors cannot say those resources changed. They still read the pattern through mental depletion. The reading is theirs. It is not a second test.

​
Key Points

Four distinctions keep the sequence, the easier ruling, and the unmeasured resource from collapsing into one slogan.
1

The easier ruling is the one that keeps the prisoner put

The file has 1,112 rulings, collected on 50 days across 10 months, from eight Jewish-Israeli judges, two of them women. Mean experience on the bench was 22.5 years, SD = 2.5. They presided over two parole boards serving four prisons, and those boards handled about 40 percent of parole requests in the country. Each board also had a criminologist and a social worker, who gave the judge professional advice. Overall, 64.2 percent of requests were rejected. Delay decisions were 48.4 percent of that reject category. A delay meant a review about a month later, which the authors treat as keeping the status quo. Parole requests were 78.2 percent of the file. The rest were changes to parole terms or to the terms of incarceration. A favorable ruling lasted M = 7.37 minutes, SD = 5.11. An unfavorable ruling lasted M = 5.21 minutes, SD = 4.97, t = 6.86, P < 0.01. Written favorable verdicts ran M = 89.61 words, SD = 65.46, against M = 47.36 words, SD = 43.99, for unfavorable verdicts, t = 12.82, P < 0.01. The authors take the shorter, thinner ruling as the easier one.
2

The drop resets inside the day

Judges took two food breaks, and the breaks cut the day into three sessions. The favorable share starts near 65 percent at the first case of a session, falls toward zero, and returns near 65 percent after the break. Figure 1 plots that share by ordinal position and uses the first 95 percent of the cases in each session, because later positions had few cases. The snack was a sandwich and fruit, typically at the bench. Its start ranged from 9:49 to 10:27 a.m. Lunch started between 12:46 and 2:10 p.m. The judge chose the timing and did not have the next prisoner’s file in advance. A clerk gave the file when the prisoner, and the attorney, came in. The authors say the order of cases, with rare exceptions, followed the arrival time of the attorney. A day held 14 to 35 cases, M = 22.58, SD = 4.67. A case took about 6 minutes, M = 5.98, SD = 5.13, with a maximum of 40.
3

Minutes and the case count are not the same test

Cumulative minutes in a session, used as a stand-in for fatigue, also predicted fewer favorable rulings. When minutes and the ordinal-position counter were entered together, only the counter stayed significant. The two predictors correlated, r = 0.72, P < 0.0001. The authors read that as a hint that the act of deciding, rather than the clock, carries the pattern. They flag the correlation, so the hint is not a clean split. Of the legal controls, only the number of prior incarcerations and the presence of a rehabilitation program were consistently significant. Severity of the offense, months served, sex, and ethnicity tended not to be. The authors say the ethnicity result means these judges did not treat prisoners differently by ethnicity. A control for a daily quota of favorable rulings pointed the other way from a filled quota: a higher share of favorable rulings so far made the next favorable ruling more likely, and that coefficient was positive and significant.
4

The resource is an interpretation

The discussion ties the reset to research on rest, positive mood, and glucose. It then says the design cannot separate rest from eating, because every break was a meal, and cannot speak to mood or to the resource, because neither was measured. Andreas Glöckner, Judgment and Decision Making 11(6), November 2016, pages 601–610, read that claim against a simulation. Favorable rulings take longer. A judge who will not start a case that may run past the end of the sitting, or a count that drops the late thin tail, can produce a falling share of favorable rulings with no change in a mental resource. His paper was read. The parole authors’ reply to a 2011 letter, and that letter, were not opened. Glöckner’s summary of both is the one used below.

​
Applications

These four moves use splits the 2011 paper and the 2016 simulation actually make. They are not a claim that a break, a snack, or a reordered agenda changes a court’s grant rate, a clinic’s prescriptions, or a household’s evening choices by a measured amount.

Park the costly yes or no early

The favorable share in this file is highest at the start of a session, not at a magical hour. If a decision is the one you would regret answering with the easier option, give it an early slot in a sitting, and start a new sitting after a stop. Do not copy the sandwich into a rule about glucose. The paper cannot separate the food from the pause.

Count a delay as a decision

A promise to look again kept the prisoner in place for about a month, and the authors put that promise in the reject category. In a meeting, “let’s revisit” can be the stamp that ends the question. Write it down as the choice it was. A later review is a new case, not evidence that the first answer was still open.

Do not use the clock as the cause

Minutes and the number of decisions moved together, r = 0.72. The joint test left the position counter, not the minutes. A rule that says “nothing important after 4 p.m.” adds a clock the joint test did not keep. A rule that says “nothing important after many similar decisions, unless the sequence stops” is closer to the contrast they printed.

Ask what the late cases were

Before treating a falling grant rate as fatigue, ask whether late cases were longer when granted, missing an advocate, or left out of the figure because the tail was thin. The 2016 simulation is that question. A briefing that skips it has turned the authors’ interpretation into a finding they say they could not measure.

​
Case Study

Eight judges on two Israeli parole boards decided 1,112 requests on 50 days in a 10-month period. The prisons held people convicted of offenses from embezzlement and theft to assault, murder, and rape. A judge saw the day’s cases in a row and chose when to break for a snack and for lunch. The next file was not in the judge’s hands until the prisoner appeared. The authors’ question was whether the place of a case in that row, and the food break, lined up with a favorable ruling once the legal facts were taken into account. The legal facts in the regression were severity of the offense, months served, previous incarcerations, and whether a rehabilitation program was in place. Prisoner sex and nationality were included. So was the proportion of favorable rulings already made that day. The order variables were dummies for the first three cases in a session, dummies for which of the three sessions the case was in, and a counter for position in the session or in the day. Judge-specific fixed effects absorbed each judge’s general tendency. The authors report that the first-three dummies stayed positive and significant with those controls, that the downward trend between breaks stayed significant, and that every judge was more likely to rule favorably at the start of a session than at the end. Nested tests fit better once the order variables were added. Restricting the file to parole requests, or dropping the two most frequent judges, or dropping each judge in turn, left the pattern nearly the same. Those checks are reported in the supplement. The supplement’s coefficients were not copied here. Table 1’s coefficients are not in the HTML prose and are not on this page. The result they emphasize is the reset. About 65 percent favorable at the opening of a session, nearly zero late in that session, about 65 percent again after the break. They also note what did not explain it. Severity and months served tended not to predict the ruling. A slight correlation between recidivism and position across the whole day was not significant inside a session, so it cannot explain the spike after a break. The type of case just decided did not significantly predict the decision to break. Attorneys waited where they could not see the rulings or the break, and a later survey, described but not tabulated in the text, found them unaware of an advantage to appearing after the break. A matching survey of board members is described the same way. The lesson the authors draw is that repeated rulings increased the tendency to keep the status quo, and that a meal break overcame that tendency. The boundary is in the same discussion, and it is widened by Glöckner (2016). The resource was not measured. Rest was not separated from food. Glöckner shows that because a favorable ruling runs longer, a rational stop before a case that may not fit, plus the choice to plot only the first 95 percent of each session, can manufacture a steep drop. In his example, with 5 minutes left, 12 percent of favorable cases would still fit and 46 percent of unfavorable cases would. For a long remainder the favorable share sits near the 36 percent base rate. For a short remainder it approaches zero. Dropping the last 5 percent made his simulated drop steeper. Without any foresight about the next case, censoring plus a small autocorrelation, r = .10, in a simulated N of 50,000, still produced a drop of roughly 15 percent. He is explicit that this does not rule depletion out. The simulated curve started lower than the parole curve, near 45 percent rather than 65 percent, and was somewhat smaller. He writes that planning and selective dropout can account for a drop on the order of 15 to 45 percent, not for the whole pattern, and that the data were not available for a finer split. The percentages in the parole paper belong to these eight judges, these 50 days, and this coding of delay as rejection. They are not a grant rate for another court or another year.

​
Boundaries and Failure Modes

Decision Fatigue does not say that a person who has decided many things will choose badly, or that a meal repairs judgment. The 2011 discussion says the opposite of a settled mechanism. Every break was a meal, so rest and food are one event. Mood was not measured. The resource was not measured. Using the reset as evidence that glucose was restored adds a blood reading the study does not contain. Two other readings overshoot the table. One is a single slide from morning to evening. The figure has three sessions and two returns. A briefing that keeps only the late-day zero has deleted the reset. The other is a filled daily quota of favorable rulings. The coefficient on the share of favorable rulings so far was positive. Judges who had already said yes more often were more likely, not less likely, to say yes again. Glöckner (2016) is the boundary on magnitude. He converts a drop from 65 percent to 5 percent into an odds ratio of 35 and, by the formula in Chinn (2000), into d = 1.96. The parole prose prints “nearly zero,” not 5 percent. He then notes that a publication-bias-corrected meta-analysis he cites, Carter and McCullough (2013), puts mental depletion around d = −0.10 to 0.25, and that a 23-laboratory replication he cites, Hagger and Chatzisarantis (2016), found d = 0.04, N = 2,142, not distinguishable from zero. Those two papers were not opened. Their numbers are his citations, not a re-analysis of the parole file. A courtroom, a clinic, a hiring meeting, or a household that imports 65 percent, zero, d = 1.96, or d = 0.04 has left the sample those figures belong to. The rehabilitation contrast is thin in a different way. The authors say prisoners without a planned program were less likely to receive a favorable ruling, and they also say 98.3 percent of prisoners had such a program. The control matters in the regression. It is not a large untreated group. Prior incarcerations are the other legal control that consistently mattered. Severity and time served tended not to.

​
Common Misconceptions

Three readings add a mechanism, a clock, or a merge that the parole paper does not print.
The authors set out to test that caricature and then say the data cannot test it. They do not record what was eaten, and they do not compare a rest break with a meal. What they record is the place of the case and whether a break for a meal had just occurred. A sentence that names the menu, the calories, or the blood glucose is not in the paper. The Hot–Cold Empathy Gap is a different question, about judging a state you are not in. Hunger here is an unmeasured candidate, not a score.
Choice Overload is how many alternatives sit in one decision. A parole request is not a shelf of jams. Status Quo Bias is a preference for the current state. The authors use “status quo” for the reject-or-delay outcome, and their question is whether that outcome becomes more common as the session goes on. Default Effect needs a preselected option. This file does not have one. The three pages were not edited. The names stay apart because the questions differ.
The discussion says there is no direct measure of the judges’ mental resources. Glöckner (2016) shows a way for the curve to appear because yeses take longer and sittings stop. He also says his account does not cover the whole curve, especially the high share at the start of the day, and that depletion remains possible. The authors’ 2011 reply, as he summarizes it, still finds case order and the meal break after they add attorney representation. He notes that they do not say whether the size of the effect stayed the same. The reply and the letter it answers were not opened, so this page does not print the letter’s rates. A claim that the drop has been cleared of every alternative, or that it has been explained away, is past both papers.
These six pages sit beside Decision Fatigue. Each one asks a different question.

Choice Overload

Too many alternatives at one moment. Decision Fatigue is the place of a decision after other decisions. A long menu and a long docket can both stall a person. They are not one count.

Status Quo Bias

Preferring the current state when a change is available. The parole paper uses the status quo as the name of a rejection or a delay. The added claim is that this share rises inside a session and resets after a break.

Default Effect

Taking the option that is already selected. The easier parole ruling was easier in time and in words. It was not printed on the form as the preselected box.

Hot–Cold Empathy Gap

Missing the force of hunger, fatigue, or craving while that state is quiet. A meal break is a candidate explanation in the parole discussion. The state itself was not scored.

Planning Fallacy

Underestimating how long a task will take. Glöckner’s judge who avoids starting a long case before a break is making a time judgment. That judgment is his alternative to depletion, not a finding that the judges misforecast.

Overconfidence Effect

Too much trust that a judgment is accurate. Decision Fatigue is not a confidence score. The parole judges were not asked how sure they were. The recorded outcome is the ruling.

​
One-Line Takeaway

When the next decision would be costly to answer with the option that merely ends the question, give it an early slot in a sitting that has just stopped, and do not call the snack the cause.