Articles

Six self-control claims that don't hold up

By Michael Fuchs · Published 2026-09-29

In short The classic study on self-set deadlines was retracted in September 2026. The willpower battery has not held up after short efforts. Once publication bias is corrected, the average nudge shows no reliable effect, and reminders and prompts to commit show evidence against one. Cheat days are barely studied, and two numbers quoted about streaks and blockers lead back to no study I could find. What survives is smaller: many people bind themselves when it costs them little, and a step in front of an app turns some attempts away, though that need not mean less time on the phone.

I am building an app that puts a price on more screen time, so I have read a lot of research about self-control. In September 2026 one of its classics, a study about deadlines, was retracted. That made me go back over everything I had cited in my own design notes. The numbers in them were mostly right. The claims I had built on them were often stronger than the papers.

Here are six claims you meet again and again in articles about habits and screen time, and what I found when I followed each one to its source. Where I give an effect size as d, 0.2 is conventionally called small, 0.5 medium and 0.8 large [1].

“Deadlines you set yourself beat procrastination”

This one no longer stands in its famous form. In 2002, a study reported that students who set their own costly deadlines did better than those who had only one final deadline for everything, though not as well as those given evenly spaced deadlines [2, retracted]. By Web of Science’s count, it was cited nearly 1,000 times [3].

On 2 September 2026 the journal retracted it [4]. One of its authors, Klaus Wertenbroch, had asked for the retraction, and both agreed with it [3; 4].

Two things had come before. A preregistered replication of its proofreading study, with 124 students at a US university, found that deadlines made a negligible difference to performance [5]. That prompted analysts to examine data files from 2006, and they concluded that the data of that same study were, in their words, “severely tampered with or fabricated” [6; 3]. The notice says the replication “raised questions about the underlying data in the original study” and the analyses “called into question the veracity of the overall findings” [4]. In the other study, run in 2 sections of a course, 13 final grades had been altered. Without those changes, the students who spread out their own deadlines did not do about as well as those given evenly spaced ones, but significantly worse [7].

What is left is narrower: many people bind themselves when it costs them little. In the replication, between 89% and 93% chose a binding deadline, depending on the task, [5]. At a data-entry firm in India, 102 workers were offered, on random days, to set themselves a daily target that cut their pay rate by 50% if they missed it. They took one on 36% of those days, and their output was about 2% higher on days the option was offered [8]. In a 7-week experiment, 47 of 80 students chose a commitment when it was free. It was worth little to them: at a price of 25 cents, 91% kept their options open, and when paid 25 cents, 90% committed [9, pp. 1098, 1104].

“Willpower is a battery that runs down”

The idea is called ego depletion: acts of self-control draw on a limited resource, so one act leaves less for the next [10].

Three large, preregistered tests across many labs, each after at most 15 minutes of effort, found little or nothing [11; 12; 13]. The first, in 23 labs with 2,141 people, found d = 0.04, too small to tell from zero [11]. The second tested 3,531 people in 36 labs. Its preregistered analysis, which kept 2,461 of them under rules set in advance, found d = 0.06, again not distinguishable from zero; with everyone included, the effect was d = 0.08 and just distinguishable [12]. The third found d = 0.10 in 12 labs with 1,775 people: small, and only just distinguishable from zero [13].

An earlier meta-analysis of 116 experiments, unpublished ones among them, corrected for small studies showing larger effects, a sign of publication bias, and found “very little evidence that the depletion effect is a real phenomenon, at least when assessed with the methods most frequently used in the laboratory” [14].

Longer effort may be different. A 2025 study with a task lasting 30 to 40 minutes found d = 0.35 across 14 samples with 2,078 people [15]. Most of those samples were not preregistered, and the task comes from the authors’ own group.

“Nudges change what people do”

A large meta-analysis in 2022 reported an average effect of d = 0.43 across 447 effect sizes from 212 publications, in its corrected version [16]. At least two groups then corrected the same data for publication bias: small studies with big effects get published, small studies with none stay in a drawer. After that correction, one group found that “no evidence remains that nudges are effective as tools for behaviour change” [17]; the other’s corrected averages range from d = −0.01 to 0.08 [18].

That is the average, and effects vary widely [17; 18]. By kind, the first group found evidence against an average effect for nudges that give information, and for nudges meant to help self-control, such as reminders and prompts to commit to something. For nudges that change how the options are set up, such as defaults or making an option more or less effort, the evidence was undecided [17]. The second group names defaults as one of a few exceptions to its conclusion [18].

Closer to what my app does is effort in the way, though the evidence for it is of a weaker kind. When 40 Android users recruited through a university community had to type digits before a target app opened, the share of launch attempts they abandoned rose with the effort: 13% when they only had to tap OK, 27% for 10 digits, 48% for 30. The size of the step was drawn at random and kept for an hour at a time, so the rise is a fair comparison [19]. An app that shows a short breathing pause, a message and a button to leave before a chosen app opens saw 36% of attempts closed again, among the 280 of its 719 new users who kept it for 6 weeks. In a separate randomised experiment with videos, the button to leave did most of the work, and the waiting added nothing to it [20].

“A cheat day helps you stick to the plan”

The evidence is thin and points both ways. A 2025 review found 8 studies of cheat meals in total: 4 on the body, with 125 participants over 4 to 6 weeks, and 4 on the mind, 3 of them cross-sectional snapshots [21]. The reviewers write that allowing such free eating “appears to be a potentially effective strategy for maintaining diet adherence”. That rests on 4 small studies. In 2 of them nobody dropped out, but both were of bodybuilders, and neither compared cheat meals with ordinary dieting. Of the 2 randomised trials that did, the one where the cheat-meal group lost fewer people is the one the reviewers rate at high risk of bias; in the other, the same number dropped out of each group. The same review warns that treating cheat meals as a reward may go with disordered eating [21]. None of the three studies it names for that link compared people who see cheat meals that way with people who don’t. One coded Instagram posts rather than asking anyone [22], and the two surveys below compared people by whether and how often they ate cheat meals, not by how they thought of them [23; 24].

A survey of 2,717 young Canadians found those who ate cheat meals more likely to report binge eating, in every gender group it looked at [23]; a smaller one of 248 students, sharing two of its authors, reports a link for men only, though its own table puts it under women [24]. Both measured at a single moment, so neither can say which comes first, and the smaller survey’s authors could not rule out that a cheat meal was reported a second time as a binge.

What looks somewhat better is a small reserve of exceptions agreed in advance, which the researchers call an emergency reserve. In a 4-week field experiment, 273 students and staff of one US university were randomly given daily step targets under different weekly goals. Those asked to reach the target every day, but allowed 2 emergency skips a week, reached it on 4.00 days a week. On paper, that goal is the same as 5 days out of 7. Those given the 5-day goal reached it on 3.11 days, and those asked for all 7 without skips on 2.83 [25]. Online experiments by the same researchers, mostly with paid crowd workers doing short tasks, point the same way [26].

“Track a streak, and one missed day makes you 63% more likely to quit”

Even with extensive help from AI tools, I could not find where this number comes from. One blog attributes it to a 2020 paper in the Journal of Personality and Social Psychology [28]; another, in a nearly identical sentence, links it to a 2020 CHI study [29]. Neither names an author or links a paper, and a search of both venues’ records found no such paper.

The research that does exist finds smaller, shorter-lived effects. In a study of how habits form, with 82 volunteers over 12 weeks, missing a single opportunity “did not materially affect the habit formation process” [30]. In 7 studies, mostly online experiments with streaks of 3 or 4 steps, people shown a broken streak were less likely to carry on with the activity right away than people shown an intact one, even when both had done exactly the same; the gap shrank, but did not close, when the streak could be repaired [31]. In interviews, 17 runners who had ended a streak of at least 100 days described sadness, anger, disappointment, the most common, and relief, and some a grief that lasted weeks; some had kept running injured to save the streak. All who could went on running, most of them on a new streak, and none reported lasting harm [32].

Streaks also do something well: the runners’ study concludes that streaking can support lasting change like few other techniques, with a small risk of backfiring for some [32].

“60–70% of users disable or bypass blockers within a month”

Even with extensive help from AI tools, I could not find this figure anywhere except one app maker’s blog, which gives no source [33]. The nearest real measurements count something else. Of 719 new users of the pause app above, 280 were still using it after 6 weeks [20]. And among 1,240 long-term users of a browser tool that nudges people away from time-wasting sites, 73% had chosen to be shown no intervention by their 200th visit [34].

What this means for the app I am building

I make I Have No Self Control, an iPhone app that gives the apps you choose a daily limit and puts a price on more time. Some of what it does rests on the research above, some on my own preference, and some on things the research does not reach.

How I checked

I checked these claims with the help of AI research tools that read each paper in full, methods and results, not only its abstract; for one of the two cheat-meal surveys, only the abstract was open to us, and it is cited for what the abstract says. Every number from a study is traced to the study itself, not to a summary of it. The account of the retraction rests on the journal’s notice, the Data Colada analyses and Retraction Watch’s report, and the blogs were read on 25 September 2026. Each source below says what kind of evidence it is: a meta-analysis weighs more than a lab study, and a vendor blog counts as a place where a claim appears, not as evidence for it. Where a result has been corrected or retracted, the list says so. If you find something I got wrong, the imprint has my address.

This site belongs to an iPhone app: a daily limit on the apps that pull you in, and a price on more time, both set by you earlier. See how it works