Recruitment channels shape sample bias
Aliases: recruitment source bias · coverage error · channel selection
What it is
Recruitment channel bias arises because entering a sample depends on coverage by a channel, exposure to an invitation, willingness to respond, and passing screening. An in-product prompt reaches current active users; support records concentrate people who encountered problems and sought help; social media favors brand-aware people willing to speak publicly. A channel is not merely where an invitation appears. It creates a sequence of selection from target population to completed participant.
Why it happens
Selection bias occurs when factors affecting participation are also related to the outcome. People with lower digital capability may encounter more difficulty yet be less likely to see online recruitment or finish a long screener, inflating observed success. Increasing numbers within one channel estimates its covered population more precisely without recovering people it never reaches. Multiple channels do not automatically cancel bias when they repeatedly cover the same active group.
Studying it
Map the recruitment funnel: target population, contactable frame, invitation receipt, screening start, eligibility, consent, and completion. Record counts and available characteristics at each step, then compare enrollment, no-show, and outcome patterns across channels. When population benchmarks exist, coverage differences and weighting can be examined under explicit assumptions; otherwise channel composition describes the sample but does not establish representativeness. Limited nonresponse evidence can be probed by varying invitation mode, timing, or offline access.
Where it stops holding
Channel difference does not make a source unusable. Qualitative exploration may intentionally select channels rich in relevant experience as long as frequency is not treated as population prevalence. Rare populations may require community organizations or referral sampling, making network structure part of the interpretation. Demographic resemblance to a population also does not prove absence of bias because unobserved motives, access, and attitudes may differ.