“Work expands so as to fill the time available for its completion” is one of the most quoted lines in management writing, and it began as a joke. C. Northcote Parkinson, a British naval historian, put it in the opening sentence of an essay in The Economist on Saturday, November 19, 1955. Seventy-one years later, it is repeated as if it were a measured finding. Productivity blogs often back it with one study, Dan Ariely and Klaus Wertenbroch’s 2002 paper on deadlines. That paper was retracted on September 2, 2026.
So does the law hold? The honest answer is more interesting than yes or no. A small set of experiments from 1966 to 1999 does show a Parkinson-like effect in individuals, and a field study with logging crews found one too. But the mechanism is narrower than the slogan, the effect failed at least one replication, and the most famous modern evidence is gone. Here is what the essay actually said, what the experiments measured, and what the mathematics of schedules adds that the psychology alone does not.
A satire about the Admiralty, not a study of people at desks
The essay appeared in the same week as the report of the Royal Commission on the Civil Service, and that timing was the point. Parkinson called the work-expansion line a “commonplace observation” and illustrated it in one paragraph. An elderly lady of leisure spends an entire day sending a single postcard to her niece in Bognor Regis: finding the card, hunting for her spectacles, searching for the address, deciding whether to take an umbrella to the mailbox. A busy man would do the same thing in about three minutes.
The rest of the essay is about something else entirely: why bureaucracies grow whether or not their workload does. Parkinson gave two driving forces. Factor I: an official wants to multiply subordinates, not rivals. Factor II: officials make work for each other. The thing he formally named “Parkinson’s Law” was the growth rule produced by those two factors. The sentence everyone quotes today was the warm-up.
His evidence came from the Royal Navy. Between 1914 and 1928, according to the essay, the fleet shrank by about two-thirds in ships, and the number of seamen fell by 31.5%. Over the same years, dockyard officials rose by about 40%, and Admiralty officials in Whitehall went from 2,000 to 3,569. He then applied a mock formula to the Admiralty and the Colonial Office and “proved” that staff grow by 5.17% to 6.56% a year regardless of workload.
| Admiralty figure (1914 to 1928) | Value in the essay | What the arithmetic gives |
|---|---|---|
| Admiralty officials | 2,000 to 3,569 | +78.45% over 14 years |
| Average yearly growth | 5.6% a year | 78.45 ÷ 14 = 5.6, a simple average |
| Compound yearly rate | Not stated | (3,569 ÷ 2,000)1/14 − 1 ≈ 4.2% a year |
| Predicted band for any department | 5.17% to 6.56% | Two-decimal precision from a joke formula |
The table shows the joke inside the numbers. Parkinson’s 5.6% matches the total growth divided by the number of years exactly, which suggests (our calculation, not his statement) that he used a simple average. Compounded properly, the same headcount gives about 4.2% a year, below the band he presented as universal. Two-decimal precision was part of the parody of official statistics. The essay also made no attempt to separate pointless growth from new administrative work after the First World War. That is our observation, not a flaw Parkinson would have cared about: satire does not control for confounders.
Two laws in one essay, and only one is about your calendar
Researchers now separate a personal version (an individual stretches a task to fit the time allowed) from an organizational version (an institution grows regardless of its workload). They share a name and almost nothing else. The personal version is about pace and goals inside one head. The organizational version is about incentives, hierarchy, and headcount.
The organizational strand has its own quantitative afterlife. In 2008, physicists Peter Klimek, Rudolf Hanel, and Stefan Thurner of the Medical University of Vienna formalized another of Parkinson’s observations: cabinets lose the ability to decide once they grow past about 20 members. Across the cabinets of 197 countries, they found a strong negative linear relationship between cabinet size and indicators of governance and human development, and most countries with above-average scores had fewer than 20 cabinet members. This is correlational. Large cabinets may be a symptom of weak governance rather than its cause, and the model explains the threshold rather than proving it. It says nothing about whether you will spend all Friday on a report that needs two hours.
The rest of this article is about the personal version, because that is the claim people apply to their own deadlines.
The excess-time experiments, 1966 to 1999
The first lab test came from social psychologist Elliot Aronson and his coauthor Gerard, published in the Journal of Personality and Social Psychology in 1966 under the title “Beyond Parkinson’s Law.” TIME described Aronson’s experiments in 1969 with a simple design. One group of students had five minutes to prepare a talk on smoking, and another had 15. Each group then received a new, similar task with no time limit. The five-minute group finished it within their original deadline. The 15-minute group took an average of eight minutes.
That second step is the important part. Using all 15 minutes when 15 are offered could be politeness or caution. Taking longer on a fresh task with no limit at all suggests the generous allowance had changed people’s sense of how long this kind of job takes. The effect became known as the “excess time effect”: surplus time does not just get spent, it carries over to the next task. Aronson was candid about why he cared. He had spent three summer weeks on a lecture that, by his own estimate, could have been written in about three hours.
The follow-ups are more mixed than most summaries admit:
- Aronson and Landy (1967), Journal of Experimental Social Psychology: a replication and extension of the excess time effect.
- Bryan and Locke (1967), Organizational Behavior and Human Performance: two experiments that recast the effect as goal setting. Their hypothesis was that different time limits change work pace only to the degree that people set different goals under them. A long deadline works as an easy goal.
- Landy, McCue, and Aronson (1969), Journal of Applied Psychology: tested how distractions that stretch or shorten a task affect time wasted on later tasks.
- Orpen and Riese (1973), Journal of Social Psychology: published under the blunt title “Beyond Parkinson’s Law: A Failure to Replicate.”
- Brannon, Hershberger, and Brock (1999), Psychonomic Bulletin & Review: argued the law had never had an artifact-free demonstration at the individual level. Undergraduates judged four sets of face photos. Those told before the third set that the fourth had been canceled dawdled on the third set compared with controls. The authors named this the “cancellation-dalliance effect” and reproduced it in two exact replications and one conceptual replication.
| Study | Setting | What changed | Result |
|---|---|---|---|
| Aronson and Gerard, 1966 | Lab, students | Generous vs short preparation time | Surplus time carried over to a later task |
| Bryan and Locke, 1967 | Lab, two experiments | Different time limits | Shorter limits, faster pace; attributed to the goals people set |
| Orpen and Riese, 1973 | Replication attempt | Excess time effect | Not replicated |
| Latham and Locke, 1975 | Field, logging crews | Mill buying limited to 1 or 2 days a week | Higher output per working day under restriction |
| Brannon et al., 1999 | Lab, four studies | Next task “canceled” | Slower work on the current task |
Taken together, these studies support a narrow claim: when people have more time than a short, well-defined task requires, many of them slow down, and the slower pace can persist. They do not show that every task expands to fill any deadline, and they were run on lab tasks measured in minutes, not on projects measured in weeks.
The logging crews: Parkinson’s Law with a paycheck attached
The strongest real-world test is Gary Latham and Edwin Locke’s 1975 paper in the Journal of Applied Psychology, titled “Increasing Productivity with Decreasing Time Limits.” Pulp and paper mills sometimes had more wood than they could process, so they restricted the days on which they would buy it. Independent logging crews showed a significantly higher rate of output when mills accepted deliveries on only one or two days a week than when no restriction was in effect.
It is a clean natural setting with one detail blogs leave out: crews and owners were paid on a piece-rate basis. A crew that could sell only on limited days had a direct financial reason to cut as much as possible before the window closed. That is a time limit working as a goal with money behind it, which fits the Bryan and Locke reading. It does not show that a salaried employee given a looser deadline for a slide deck produces less per hour. Carrying the result over to office work is an extrapolation and should be labeled as one.
The deadline study that was supposed to settle it, and was retracted in 2026
For two decades, the modern evidence cited for “set tighter deadlines” was Ariely and Wertenbroch’s “Procrastination, Deadlines, and Performance: Self-Control by Precommitment,” published in Psychological Science in May 2002. It asked three questions: Will people impose costly deadlines on themselves? Do those deadlines improve performance? Do people set them optimally? The paper’s answers were yes, yes, and no, with externally imposed, evenly spaced deadlines working best.
The centerpiece, Study 2, had 60 participants proofread three texts over three weeks, earning $0.10 per error found and losing $1 for each day late. Twenty had evenly spaced deadlines, 20 chose their own, and 20 had a single deadline at the end. The reported gap was enormous: the evenly spaced group averaged 136.1 corrections against 71.1 for the end-deadline group, nearly double.
Two events in 2026 ended that result. First, economists Kyle Hyndman and Alberto Bisin published a preregistered replication in Psychological Science (online July 15, 2026) with 124 participants, at least 40 per condition, and doubled incentives ($0.20 per error, $2 per day late). The first finding held: 93.33% of participants allowed to set their own deadlines chose binding ones for the first two tasks. The headline finding did not. Errors found, days late, and earnings showed no significant differences between conditions. What survived, in the authors’ account, was modest: intermediate deadlines make people finish earlier.
Second, on August 31, 2026, the research blog Data Colada published an analysis of what appear to be the original data files, which Hyndman had received from Ariely’s MIT email account in 2006. The reported effect size was Cohen’s d = 2.5, larger than the gender gap in adult height (about 1.8). Eighteen of the 20 end-deadline participants had a “twin” with identical correction counts on all three texts, with participant IDs exactly 10 apart. Only 11.7% of self-reported work times were round numbers, against 85% in the replication. Data Colada concluded the data “were severely tampered with or fabricated.” Ariely said the data contained serious anomalies his records and memory could not explain. Wertenbroch said he never had access to the data and asked for retraction. The journal retracted the paper on Wednesday, September 2, 2026.
The deadline evidence that survived 2026 is narrower than the advice built on it: people want deadlines and finish sooner with them, but there is no reliable evidence that spacing them out makes the work better. worldtimedata
Any guide that says Ariely proved deadlines improve performance is now citing a retracted paper. The willingness to precommit replicated; the performance benefit did not.
Why a looser limit slows people down
The experiments point to mechanisms, though none has been isolated cleanly enough to call settled. These are the best-supported readings of the evidence above, not established laws:
- The limit works as a goal. This is the Bryan and Locke account. People treat the allotted time as information about how demanding the task is meant to be and pace themselves accordingly. A generous limit reads as an easy target.
- The pace gets anchored. Aronson’s carryover result suggests the first allowance becomes the reference for “how long this kind of task takes,” shaping the next attempt even when no limit exists.
- The task gets elaborated. Extra time often goes into polishing, rechecking, and widening scope. Whether that is waste depends on whether the polish has value, which these lab studies rarely measured.
- Start time slips, finish time does not. Project managers call this “student syndrome,” a term popularized by Eliyahu Goldratt’s 1997 book Critical Chain: effort begins late because the deadline is far away. Strictly, this is procrastination, not expansion. The work does not grow, it just starts later, and the two are often confused.
The distinction decides the fix. Late starts respond to checkpoints. A slow pace or creeping scope does not; it needs a clear definition of “done” and a limit sized to the real task.
The schedule math: early finishes vanish, late ones add up
The most practical consequence of Parkinson’s Law shows up in project schedules, and it does not depend on any single psychology study. In 1991, Genaro Gutierrez and Panagiotis Kouvelis argued in Management Science that critical path methods such as PERT and CPM ignore workforce behavior, and they built a stochastic model of task duration that formally includes Parkinson’s Law. The core idea fits in one line: if a task that finishes early is reported as finishing on time while a late task is reported late, the recorded duration is the larger of the actual time and the allotted time.
Average slip per task = E[max(X, d)] − d, which is always ≥ 0
Example: X uniform between 7 and 13 days, d = 10 days
Slip per task = (3² ÷ 2) ÷ 6 = 0.75 day
10 sequential tasks: 100 days planned, 107.5 days delivered on average
In that example every estimate is perfectly unbiased: the average task really takes 10 days. The schedule still slips 7.5%, because time saved on early tasks is quietly absorbed while overruns pass down the chain. A Monte Carlo simulation of 200,000 such projects lands on the same 107.5 days. Nobody has to be lazy for this to happen. It only takes a team in which finishing early earns nothing.
| Task | Allotted (working days) | Actual work | Recorded if early finishes are absorbed |
|---|---|---|---|
| 1 | 10 | 7 | 10 |
| 2 | 10 | 12 | 12 |
| 3 | 10 | 8 | 10 |
| 4 | 10 | 14 | 14 |
| 5 | 10 | 9 | 10 |
| Total | 50 | 50 | 56 |
Put that on a real calendar. A 50-working-day plan starting Monday, January 4, 2027, skipping Martin Luther King Jr. Day (January 18) and Presidents’ Day (February 15), ends on Tuesday, March 16, 2027. The 56-day version ends on Wednesday, March 24. The work came in exactly on budget, and the project is still eight calendar days late.
Now add a realistic bias to the estimates. In Roger Buehler and colleagues’ 1994 study, 37 students predicted they would finish their theses in 33.9 days on average and actually took 55.5. The planning fallacy pushes estimates low, and Parkinson-style absorption stops early finishes from compensating. The two errors do not offset each other; they stack in the same direction. The model also implies (an inference, not a tested result) why padding every task “just in case” often fails: each pad belongs to one task, so it gets used up there instead of protecting the project.
When a tighter limit backfires
If generous limits slow people down, it is tempting to conclude that tight ones always help. The evidence says otherwise, and the type of evidence matters.
Teresa Amabile, Constance Hadley, and Steven Kramer analyzed daily diaries from 177 employees at seven US companies in chemicals, high tech, and consumer products, reporting the results in Harvard Business Review in August 2002. Creative thinking under extreme time pressure was unlikely when people felt they were on a treadmill. The exception, which the authors called being “on a mission,” came with strong focus on a single problem. These are observational diaries from knowledge workers, so they show association rather than proof that pressure causes the drop. Still, they cut against the idea that squeezing time always sharpens output.
A field experiment gives cleaner causal evidence. Maria De Paola and Francesca Gioia (Journal of Economic Psychology, 2016) randomly assigned Italian university students to take exam tests with or without time pressure, then swapped conditions on the next test. Time pressure had a negative, statistically significant effect on both verbal and numerical tasks, driven mainly by female students, with no significant effect for male students. Exams are not office work, but this is a randomized comparison, which most deadline advice cannot claim.
Short windows also distort priorities. Meng Zhu, Yang Yang, and Christopher Hsee (Journal of Consumer Research, 2018) ran five experiments in which people chose tasks with short expiration windows over tasks that paid objectively more. They named it the “mere urgency effect.” It was stronger in people who saw themselves as busy and weaker when payoffs were made salient at the moment of choice. A tight limit on one task can pull time away from more valuable work that has no limit at all.
Using the law without trusting the slogan
What survives the evidence points to specific practices rather than a blanket rule to give yourself less time:
- Size limits from history, not from hope. The excess-time studies show people calibrate to the limit they are handed. A limit far below the real task size does not create efficiency; it creates corner-cutting or a missed date. Your own records of how long similar tasks took are a better anchor than any rule of thumb.
- Use checkpoints against late starts, not as a quality tool. The one deadline effect that held up in the 2026 replication was earlier completion.
- Define “done” before the clock starts. Surplus time expands scope when the finish line is vague. A fixed time limit paired with a fixed definition of done is the core of timeboxing, a term James Martin popularized in his 1991 book Rapid Application Development.
Calendar defaults are a quiet version of the same problem. A meeting booked for 60 minutes because that is the standard block has its time allowance set before anyone states its purpose, a pattern examined in detail in why meetings default to 30 and 60 minutes.
For short solo tasks, the Aronson design suggests a simple self-test: set a deliberately modest limit on a countdown timer, then record how long the next similar task takes with no limit. If the untimed version runs much longer, the excess time effect is likely at work. If both take about the same time, the task was already sized correctly and the timer adds nothing.
For teams, the schedule math matters more than any personal habit. Make early finishes visible and hand them to the next task at once, instead of letting each task own a buffer it will use up. Hold shared contingency at the end of the project, where only genuine overruns can draw on it. Parkinson’s postcard writer cannot be fixed with a stricter deadline, but a schedule that rewards finishing early can stop a team from repeating her day across every task in a project.









