What is Attrition Rate?

Attrition rate is the proportion of participants who leave a study before completing it: the share of survey starters who don't finish, of diary-study recruits who stop logging, of longitudinal panel members who don't return for the next wave, of experiment participants whose outcomes never get measured. The number itself is a cost problem (sample shrinks, power falls, budgets stretch). The deeper problem is differential attrition: when leaving correlates with the thing being studied. If the frustrated abandon the usability survey, the completers report a happier product than the recruits would have; if the treatment arm's dropouts are the ones the treatment failed, the survivors flatter the treatment. Attrition then stops being a sample-size footnote and becomes a selection bias that arrived after recruitment, the close relative of non-response bias, and analysis of completers alone answers a question about completers.

Where It Bites

Surveys. Drop-off climbs with length, grid questions, and open-text demands; the abandonment curve is fatigue made visible, and question-level exit points are diagnostic data about the instrument.

Longitudinal and diary designs. The longitudinal researcher's central enemy: every wave loses members, and the stayers skew engaged, stable, and available, so trends measured on stayers understate the churners' reality by construction.

Experiments. Unequal dropout between arms breaks the balance randomisation created; a variant that quietly repels a segment can "win" among those who remained. Sample-ratio checks (do the arms still match the intended split?) are the cheap alarm.

Multi-session usability studies. Participants who found the first session hard are the ones least likely to return for the second, thinning the difficulty signal precisely where it mattered.

Designing Against It

1. Shorten and front-load.
Put the questions the study cannot live without early, cut what's merely nice to know, and make progress and remaining effort visible. Short instruments and single-session formats (an unmoderated Ballpark task study completes in one sitting, which structurally removes between-session attrition) prevent more dropout than any incentive recovers.

2. Pay for continuation, fairly.
Staged incentives for multi-wave designs, reminders that respect the participant, and completion bonuses that don't turn into coercion for sensitive studies, all within the bounds of consent that keeps withdrawal a genuine right.

3. Over-recruit against the expected loss.
Historical attrition rates for each format set the recruitment buffer; a diary study planned at the exact n needed will end below it.

4. Collect early, keep the partial.
Save responses progressively so abandoners' completed answers survive, and analyse partials where the design permits, with the missingness documented.

Analysing and Reporting It

Report the rate, per arm and per wave, alongside every result; compare leavers to stayers on baseline data (the test of whether attrition was differential); and, in experiments, prefer analysis that respects original assignment (the intention-to-treat principle from clinical trials) over completers-only comparisons that quietly re-select the sample. Where attrition is large and differential, say plainly that findings describe the persistent rather than the recruited, and mine the drop-off itself: where people left is a finding about the instrument, and why they left is a follow-up interview worth running.

The Takeaway

Attrition rate counts the leaving; the analysis has to ask who left. Shorten, front-load, pay fairly, over-recruit, save progressively, report per arm and wave, compare leavers to stayers, and analyse by assignment rather than by survival. A study that only hears from the people who stuck around has been re-sampled by the participants themselves, and the honest readout says so.