Fresh start effect

The mechanism Dai, Milkman and Riis propose is mental accounting applied to time. A landmark closes one accounting period and opens another, so the record of missed workouts and abandoned diets stays filed under the period that just ended. The carrier they measured was dissociation from the past, imperfect self: participants shown a date framed as a new beginning expected to feel more disconnected from their past selves, and that rating mediated the increase in goal motivation (indirect effect b = 0.07, 95% bias-corrected CI [0.003, 0.22], and [0.01, 0.29] in a supplementary replication). Both intervals clear zero only narrowly, so the mechanism is indicated rather than settled. A landmark also pushes people out of execution and into big-picture thinking, which favours desirability over feasibility, so the costs of starting look smaller. And a dated landmark is more salient than an interval, which is why “after your next birthday” and “in two months” behave differently on the same button even when they name the same day.

The founding paper is three archival field studies. Daily United States Google search volume for “diet” across 3,104 days from January 2004 to June 2012 rose 14.4% at the start of a new week, 3.7% at a new month, 82.1% at a new year and 10.2% after a federal holiday; placebo terms (news, gardening, laundry, weather) showed no comparable pattern. Card-reader records covering every undergraduate member of a university gym (11,912 members, 442 days, 5,265,104 person-days) put the probability of a visit 33.4% higher at the start of a week, 14.4% at a month, 11.6% at a year, 47.1% at a semester and 24.3% on the first day back from a school break. Commitment contracts created by 43,012 stickK.com users rose 62.9% at the start of a week, 23.6% at a month, 145.3% at a year, 55.1% after a federal holiday and 2.6% after a birthday. Of those, only the week start recurs often enough to build a repeating job around.

The weekly pattern replicates in independent data. Ayers and colleagues applied a wavelet decomposition to Google queries about quitting smoking in six languages from 2008 to 2012 and found Monday volumes 25% (95% CI 24–26%) above the Tuesday-to-Sunday mean, with Monday significantly higher in 33 of 36 day-pair comparisons; the paper was funded through a cooperative agreement with The Monday Campaigns and one author worked there. Fast and colleagues triangulated five Winnipeg datasets collected between 2012 and 2019, predicting that cycling to work would not follow the effect, and found that it did: the three trails with commuter-shaped daily profiles carried 22% less traffic on Friday than Monday, 147 fewer cyclists a day (95% CI 95 to 199 fewer), while the two leisure trails did not change significantly (22 fewer, 95% CI 59 fewer to 15 more). Accesses to a large employer's protected bike park (56,307 over seven years) fell about 20% from Tuesday to Friday, Tuesday being that site's landmark because the company closed every second Monday. Their positive control, fitness-centre attendance, fell 21% across the week, 592 fewer visits a day (95% CI 425 to 759 fewer), and their negative controls moved the other way: car counts rose about 7% and coffee sales about 10%. It is observational, one city, no individual-level data, and two co-authors work for the firms that supplied the coffee and parking records. If you send one prompt a week, the first working day is the slot the data supports, and the relevant first day is the user's week, not the calendar's.

All of that is correlational. The causal evidence is five laboratory experiments in which the only thing manipulated was the label on a date. Offered a list of dates for a goal reminder, 25.61% of participants chose 20 March as “the first day of spring” against 7.23% as “the third Thursday in March” (N = 165, p = .001); in a student sample, 28.57% chose 14 May as “the first day of Penn's summer break” against 4.35% as “Penn's Administrative Day” (N = 278, p < .0001). The behaviour behind those choices is thin: in another experiment 82% of the landmark group and 90% of the control group clicked no goal-related links at all, a mean of 0.62 against 0.21 clicks (N = 216, p = .01). Only three of the five measured an act at all: the other two collected motivation ratings, one about a hypothetical stranger and one about an imagined apartment move, and the mediation result rests on those ratings rather than on behaviour. The label on the date is the lever, and the lever is light.

One randomised field experiment, by Beshears and colleagues, tests the framing on consequential behaviour. Mailings to 6,082 university employees offered an immediate retirement contribution increase or a scheduled future one. Framing the future date as a fresh start (next birthday, New Year's, first day of spring) rather than as an interval (“in 2 months”) raised take-up of the delayed option by 1.3–1.4 percentage points on a control base of 2.6%, a 50–54% relative increase; take-up of either option rose 2.2–2.3 points on a base of 7.6%, and average contribution rates over the next eight months rose 17–19 basis points against 73, a 23–26% increase. Placebo landmarks that pre-testing showed do not feel like new beginnings (Thanksgiving, Martin Luther King Day, Valentine's Day) produced no positive effect, and fresh-start framing beat placebo framing by 1.7–1.8 points on delayed take-up. Take-up of the immediate option did not fall, but that estimate was not significant and Bayes factors modestly favoured the null, so treat “it does not cannibalise acting now” as untested rather than shown.

The same programme contains the strongest warning. In the companion working paper covering all 8,682 randomised employees at four universities, offering a delayed option at all, without landmark framing, lowered average contribution rates by 14 basis points for the targeted plan and 27 basis points across all plans compared with offering only an immediate increase, apparently because “later” leaks the message that saving is not urgent. Fresh-start framing recovered about 10 basis points, concentrated in birthday framing (22 basis points on the targeted plan, 33 across all plans) while holiday framing did not reach significance; the published paper reports the same contrast. A landmark label is a repair for a delayed option, not an improvement on asking people to start now.

Nulls exist and are not obscure. In a preregistered study by Milyavskaya and colleagues, 362 people rated three goals each on 16 characteristics in January or in March, and their ratings came out very similar in the two months; the differences that appeared were tied to goal age, not the calendar. A Center for Advanced Hindsight pilot mailed welcome boxes to 1,310 new movers in three North Carolina cities, testing the move itself as the landmark, and found no significant difference in drive-alone mileage (71.4 against 79.2 miles) or drive-alone rate (60.7% against 64.0%). The effect has also never been through a megastudy: the 61,293-member gym megastudy that tested 54 four-week programmes carries no fresh-start arm, so nothing puts this framing against planning prompts, reminders or micro-incentives at scale. That megastudy is the best available estimate of how long such boosts last: 45% of its interventions raised weekly visits by 9–27% during the programme, but only 8% produced change still measurable afterwards. Build the landmark as a scheduling trick for initiation, and put something else in the product for week two.