What Small Commitments Can and Cannot Do

Advice about change often says to start small. The idea is old and appealing, and parts of it have good support. Other parts are weaker than the popular version suggests. We separate what the research shows from what it does not.

Two different ideas under one name

A small commitment can mean two things. The first is a small first request that makes a larger one more likely to be accepted, usually called the foot-in-the-door technique. The second is a small, specific plan that makes a person more likely to act on a goal they already hold. They rest on different evidence and should not be mixed.

The first idea comes from Freedman and Fraser, who published two experiments in the Journal of Personality and Social Psychology in 1966. One summary reports that about 17% of homeowners asked cold agreed to display a large sign, against 76% of those who had agreed to a small sign first. That is a striking gap, and it is the reason the idea is repeated so often.

The first idea has not held up well

A later direct replication repeated the original design in Poland and Ukraine in 2003 and again in 2013. The authors note that meta-analyses report small effect sizes and that the large original effect has never been reproduced. In 2003 the technique worked in Ukraine but not in Poland, and in 2013 the effect was not significant in either country. We read this as a caution rather than a refutation: the effect may exist, but it is smaller and less reliable than the 1966 figures imply.

The second idea has stronger support

An implementation intention is an if-then plan. The National Cancer Institute describes the format as "If situation Y occurs, then I will initiate goal-directed response Z." The same page cites a meta-analysis of 94 studies by Gollwitzer and Sheeran from 2006, with a medium to large effect on goal attainment (d = .65), and values of .61 for getting started and .77 for preventing derailment. The page notes that it is no longer maintained, so we treat it as a summary of the 2006 work and not as new evidence.

A larger and more recent meta-analysis covered 642 tests. Effects ranged from d = .27 to d = .66 across cognitive, affective and behavioral outcomes, and were larger when plans used an if-then format, when people were highly motivated, and when plans were rehearsed. The last two points matter. A plan helps most when the person already wants the outcome.

A worked example

The numbers in this example are illustrative, not real data. Expected effort is the product of three things.

Expected minutes per week = Planned days x Completion rate x Minutes per session

Plan A asks for 30 minutes on 5 days, and suppose 15% of those sessions happen: 5 x 0.15 x 30 = 22.5 minutes. Plan B asks for 5 minutes on 7 days, and suppose 70% happen: 7 x 0.70 x 5 = 24.5 minutes.

The totals are close, so minutes alone do not favour the small plan. What differs is repetition: Plan B yields 4.9 sessions a week against 0.75 for Plan A. Whether that matters depends on how much repetition builds a habit, which brings us to the next point.

What the evidence does not show

Intention is not behavior. A meta-analysis by Webb and Sheeran of 47 experimental tests found that a medium to large change in intention (d = 0.66) produced only a small to medium change in behavior (d = 0.36). Wanting to act, or promising to act, gets less done than people expect.

Habit timing is also uncertain. A study by Lally and colleagues, summarised by the British Psychological Society, followed 96 undergraduates over 84 days, and 82 completed it. Average time to the plateau of automaticity was 66 days, with a range from 18 days to a predicted 254 days, extrapolated beyond the study. A single missed day had little effect, though many misses reduced the level reached. One small sample of students cannot tell us how long any particular person will take.

None of these studies shows that a small commitment is better than a large one for a given person. They show conditions under which specific plans help.

Going deeper

Effect sizes in this article are given as d, the standardised mean difference. OpenStax defines/10%3AHypothesisTestingwithTwoSamples/10.02%3ACohen'sStandardsforSmallMediumandLargeEffectSizes) it as Cohen's d = (Mean of group 1 - Mean of group 2) / Pooled standard deviation, and gives Cohen's standards of 0.2 as small, 0.5 as medium and 0.8 as large, so 0.65 sits between medium and large and 0.36 between small and medium.

Three checks are worth applying to any behavior claim. Ask whether the outcome was behavior or only stated intention. Ask whether the sample was a general population or a convenience group. Ask whether anyone has replicated it independently. The Poland and Ukraine replication shows why the third matters.

For further reading, the meta-analyses linked above report their moderators in full. We invite scrutiny of every figure here; each is linked to the page where we read it.