Temptation bundling

The direct evidence is two field experiments, both with Katherine Milkman as an author, both on gym attendance, both using audiobooks as the want, and nobody outside that group has replicated either. The first is small. Milkman, Minson and Volpp (2014) assigned 226 university gym members who wanted to exercise more to one of three arms: four self-chosen page-turner audio novels on an iPod kept in a monitored locker at the gym and reachable only there; the same novels loaded onto the participant's own iPod with a recommended rule to only let yourself enjoy them while exercising; or a $25 bookshop gift card, pretested as equal in value. In the first week the locked-iPod arm made 0.48 more gym visits a week than control, a 51% increase over control's 0.94 visits, and the self-imposed-rule arm 0.27 more, a 29% increase. Both decayed, by 0.07 and 0.05 visits a week across the nine-week study. Over the seven weeks before the university gym closed for Thanksgiving the locked arm was worth 0.31 visits a week and the self-rule arm 0.14; averaged across the whole study, 0.21. The closure ended the effect, which is the one documented way to kill it.

The large test dropped the restriction entirely. Kirgios, Mandel, Park, Milkman, Gromet, Kay and Duckworth (2020) added audiobooks to StepUp, a four-week digital exercise programme for 24 Hour Fitness members. Every arm earned points worth about $0.22 a visit, planned workouts and pre-workout texts; one arm also received a code for a free Audible audiobook, and one received the code plus a cartoon, a short video and a customised recommendation teaching the bundling rule, which 78.6% of them pledged to follow. Against a control with the same incentives, planning and reminders and no audiobook (N = 6,792), the taught arm was 6.7 percentage points more likely to produce a gym visit in a given week during the programme, and still 5.1 points and 0.131 visits a week ahead ten weeks after it ended — a 10–14% lift in the likelihood of a weekly workout and 10–12% in weekly workouts, holding for up to seventeen weeks. Nobody's access was restricted; the audiobook was theirs to use anywhere. The preregistered comparison that separates the lesson from the gift gave the lesson 2.2 to 3.2 points, all of it on whether people showed up at all rather than how often.

Two things temper that. Only 14% and 13% of the two audiobook arms ever downloaded the free audiobook, so most of the effect came from people who never obtained the want; the authors could not verify that anyone bundled, and without a cash-equivalent control they could not rule out gratitude toward the gym. And 29% of enrollees in those arms swam or took classes — activities you cannot listen through — and were excluded from the preregistered analysis.

StepUp was one arm of a megastudy, which is the best available check on durability. Milkman and colleagues (2021) compared 53 four-week interventions across 61,293 members of the same chain: 45% significantly raised weekly gym visits, by 9% to 27%, and only 8% produced a significant increase in the four weeks after the programme ended, against 2.5% expected by chance. The taught-bundling arm was the third largest of the 53 on the likelihood of a weekly visit. But in the megastudy's own post-intervention analysis of visit counts, the arm that cleared significance was the audiobook with no explanation at all (0.213 visits a week) and not the taught version (0.130). Scenario studies say why the bare gift travels so far: shown only the plain audiobook screens, 80% of 75 gym-goers said the audiobook was meant to be used at the gym. The gift leaks the instruction, so the screen that explains the rule is worth less than the screen that hands over the content.

The first link in the mechanism — enjoyable media makes effort feel better — is heavily replicated, just not on return rates. Terry, Karageorghis, Curran, Martin and Parsons-Smith's (2020) meta-analysis of 139 studies found music listening associated with more positive affective valence (g = 0.48), better physical performance (g = 0.31) and lower perceived exertion (g = 0.22), with no effect on heart rate. Woolley and Fishbach (2016) show the same logic inside an activity: having people choose the more enjoyable exercise rather than the more useful one, or adding music to a high-school classroom, increased persistence, and immediate rewards predicted persistence more strongly than delayed ones; in two field studies (2017), enjoyment predicted sticking with New Year's resolutions and with single sessions of studying and exercising while the delayed benefits did not. That body of work supports making the minutes better, not the claim that users come back.

The version usually sold under this name — do the chore, unlock the content — was never tested. In both experiments the want was available while the should was happening, either because it was kept at the gym or because the user adopted a rule about when to enjoy it; in neither was it handed over for finishing. A payout on completion is a different object with its own literature: Deci, Koestner and Ryan (1999) found engagement-contingent and completion-contingent rewards undermined free-choice persistence at d = −0.40 and −0.36 across 128 studies, so an unlockable puts you inside those cells. The honest shape of the principle is a small, twice-positive effect in one domain from one research group and a well-supported story about why the moment feels better: build the want into the task, measure after the content stops, and do not expect a habit.