Small Pushes, Large Effects

AEvery choice reaches us already framed by somebody's decision about how to present it, and the recognition that these frames alter behaviour has proved one of the most exportable ideas in modern social science. A nudge, in the strict sense, is an alteration to the presentation of a choice which changes what people do without forbidding anything and without changing what anything costs. Placing fruit at eye level is a nudge; taxing chocolate is not. The idea travelled quickly because it promised results at a fraction of the price of legislation, and because the early demonstrations were startling: switching a form from opt-in to opt-out, or adding a single sentence to a tax letter, appeared to move whole populations. By 2015 the approach had been written into the working practice of dozens of finance ministries, and a claim that began as a modest empirical observation had hardened into an expectation.

BThe first serious complication came from within. Miriam Dallow, an economist at the Redbridge Institute of Public Policy, was asked to compile the effects reported by government trial units, and discovered that a great many trials had never been written up at all. When she pooled the 97 published studies in her field, the average effect on take-up was 8.7 percentage points; when the 214 unpublished trials held in departmental archives were added, the same average fell to 1.4. Dallow argues that nothing dishonest need be assumed to explain the gap, since a unit that runs a trial producing no detectable change has little incentive to write it up and no journal is waiting for the result. What the literature records, on her account, is not the average nudge but the memorable one, and the difference between those two quantities is large enough to overturn most published cost calculations.

COwen Traskin, who directs a government trials unit, treats the whole exercise as beside the point. An average taken across interventions with nothing in common, he insists, is a number without a referent: a reminder letter, a redesigned form and a rearranged canteen are not three doses of the same drug, and averaging them tells a minister nothing about what to do next. The relevant test, in his view, is the cost of a unit of outcome. A letter that costs 97 pence to send and shifts behaviour by a single percentage point remains extraordinarily good value beside a campaign costing millions, and Traskin's unit reports reaching 1.4 million households in a year on a budget smaller than that of a regional advertising round. His critics note, however, that these figures omit the salaries of the staff who design the letters, and that most of his trials measure whether a form was opened rather than whether a debt was eventually paid, often within a fortnight of the letter arriving.

DA third line of criticism leaves the size of the effects untouched and questions their consequences. Hana Sorby, a political scientist, contends that publicising a behavioural remedy can lower public appetite for the structural one, so that a cheap intervention crowds out an expensive but more powerful alternative. In a survey experiment conducted between 2017 and 2022, respondents shown a description of a recycling nudge subsequently supported a packaging levy at a rate 11 points lower than a control group shown nothing. The effect appeared only among respondents who had been told that the nudge was expected to solve the problem; where the same intervention was described as a partial measure, support for the levy was unchanged. Sorby is careful about what this licenses. Her design measures stated preference in an online panel rather than votes, and she concedes that no field study has yet detected the substitution she describes.

EThe conditions under which nudges perform well are by now reasonably clear, and they are narrower than the early enthusiasm implied. Interventions that exploit inertia succeed only where the obstacle is the effort of acting rather than the expense of the outcome, which is why defaults transform enrolment in schemes that cost the participant nothing and barely register where a substantial payment is involved. Timing matters as well: a prompt delivered at the moment of decision outperforms the same message sent a week earlier by a factor of roughly three. Repetition, by contrast, erodes the effect. A message that alters behaviour on first sight typically loses most of its power by the fourth exposure, a decay that trials lasting six weeks are too short to observe and that longer evaluations published since 2023 have consistently found. The distinction is not merely academic, since a department that reports a first-year result and then repeats the same mailing annually will be budgeting for an effect it no longer has.

FNone of this returns the argument to where it began. The evidence that presentation matters is not in doubt, and Dallow herself works on nudges rather than against them; what her audit removes is the licence to quote the largest published figure as though it were typical. Traskin is right that cost per outcome is the decisive comparison, but his own accounting is incomplete in ways that would narrow the advantage he claims. Sorby's worry is the hardest to settle, because a displaced policy leaves no trace in any trial register. The safest summary is that a nudge is a dependable instrument for removing friction and an unreliable one for shifting anything that people have a strong reason to avoid, and that its political cost, unlike its financial cost, has not yet been measured.