Foot-in-the-door

The compliance mechanism is close to a dead end, and the evidence for that is old and consistent. Beaman, Cole, Preston, Klentz and Steblay (1983) pooled 120 experimental groups and called the phenomenon replicable but weak, with nearly half producing no effect or an effect in the wrong direction. Burger's (1999) review reads that record as several processes pulling in opposite directions at once — self-perception, reactance, conformity, commitment — so the same manipulation sometimes raises compliance and sometimes lowers it. Gamian-Wilk and Doliński's (2020) direct replication of the original procedure worked in Ukraine and not in Poland in 2003, and was insignificant in both countries in 2013; the effect size Freedman and Fraser reported has never been obtained again. The moderators sit under commitment and consistency. Plan a request sequence as if the compliance gain were zero, and justify it on the grounds below.

The only field test of the sequence delivered as an interface is one paper. Guéguen and Jacob (2001) solicited 1,008 people drawn at random from e-mail lists to visit a humanitarian website, with the manipulation implemented purely as the order of successive HTML pages: a petition to sign and then the donation ask, against the donation ask alone. They report that the sequence raised compliance with the final request, but the published abstract carries no condition-level rates. One study, 25 years old, with no preregistered web replication since — that is the whole direct evidence that progressive sign-up works as a compliance technique.

What does move reliably is the cost the user can see before starting. Galešić and Bošnjak (2009) varied the stated length of a web questionnaire between 10, 20 and 30 minutes: the longer the stated length, the fewer respondents started and completed it, reported as a direction with no percentages. Sandelin (2022) randomised 8,000 Gothenburg residents to a 205-question or a 149-question mixed-mode survey, and the longer instrument cut the response rate to 45.8% from 48.5%. The split by mode is the part to build on: on paper, where the bulk is visible, the longer version lost 2.8 percentage points (34.7% against 37.5%); online, where length is invisible, there was no difference at all (11.1% against 11.0%, not significant). Perceived bulk deters and invisible bulk does not: the argument for a small first step, and the warning that making a long task look short is not the same as making it short.

The claim that chunking an ask into pages improves completion has a null at its centre. Mavletova and Couper's (2015) meta-analysis of breakoff in mobile web surveys — 14 studies, 39 independent samples — found that shorter surveys, mobile optimisation, prerecruitment, reminders and a less complex design all reduced breakoff, while a scrolling design, incentives, stating the expected duration in the invitation and letting respondents skip questions did not. Paging against scrolling is the survey-methodology name for a multi-step form against one long one, and it sits in the no-effect column. Toepoel, Das and van Soest (2009) split the same 40-question instrument across 1, 4, 10 and 40 items per screen: item nonresponse rose as items per screen rose, while more items per screen shortened the interview and worsened how respondents rated the layout. Fewer fields per screen buys answer completeness and costs time; it does not buy completion.

The steps after the first one decide the outcome. Peytchev (2009) found that features visible only after a survey starts predict breakoff, and that respondents who broke off did not look inattentive — they met something specific and stopped. Villar, Callegaro and Yang's (2013) meta-analysis of 32 randomised progress-indicator experiments points the same way: a constant indicator did not significantly reduce drop-off, indicators that moved fast early and slowly later did reduce it, indicators that started slowly increased it, and where a small incentive was promised a constant indicator made drop-off worse. A step counter is not a motivator; it is a statement about how much is left, and it is read as one.

One version of the sequence does raise conversion, and it is the anti-pattern. Cialdini, Cacioppo, Bassett and Miller (1978) named it the low-ball: securing an initial decision and then making performance more costly produced greater final compliance than disclosing the full cost from the outset, and only where the first decision was made with a high degree of choice. Blake, Moshary, Sweeney and Tadelis (2018) measured the same shape at scale on StubHub, splitting several million users over ten days between fee-inclusive prices and base prices with the roughly 15% buyer fee added at checkout. Shrouding the fee raised ten-day revenue per user by 20.64%, the transaction rate by 14.1% and the price of the ticket actually bought by 5.42%; shrouded users were about 19% more likely to select a ticket and then about 45% less likely to buy once the total appeared. The one reliably converting version of this sequence is the deceptive one, and it held for experienced buyers too (15.09% on revenue for people with ten or more prior visits), so it is not a trick users learn their way out of — and in the EU and the US it is now substantially unlawful on the surfaces where it was most common.

The honest grade is weak. The mechanism cited for sequencing asks has a meta-analysis calling it weak, a null direct replication and one unreplicated online study; the interface version — split the form, raise completion — has a null in the one meta-analysis that tests it. What survives is that visible cost deters starting, that what people meet after starting decides finishing, and that deferring the real cost works by surprise. Build a sequence to cut what the user must evaluate at each step and to let them leave with their work saved, and do not expect the first yes to carry the second.