nonresponsebias

Nonresponse bias occurs in survey research when individuals who do not respond differ in meaningful ways from those who do respond, causing estimates to deviate from the true values for the target population. This bias can affect any statistic derived from survey data, including means, proportions, and regression relationships.

Nonresponse is often categorized as unit nonresponse, where no data are obtained from a sampled unit, and item nonresponse, where some questions remain unanswered by respondents. The potential for bias depends on the mechanism of nonresponse: if data are missing completely at random (MCAR), nonresponse may not bias results; if missing at random (MAR), biases can be mitigated with appropriate adjustments using observed data; if missing not at random (MNAR), bias may persist even after adjustments because nonresponse is related to unobserved information.

Causes of nonresponse include survey design features (length, complexity, or sensitivity of questions), mode of data collection (web, phone, in-person), timing and incentives, household accessibility, language barriers, and privacy concerns. Practical indicators of potential bias include low response rates and differential response across subgroups.

Mitigation strategies encompass follow-up contacts, mixed-mode data collection, and efforts to reduce respondent burden. Statistical adjustments include weighting adjustments (post-stratification, raking), modeling response propensity, calibration, and imputation methods such as single or multiple imputation. Researchers may perform nonresponse analyses to compare respondents with known population benchmarks or with late-responding cases as proxies for nonrespondents.

Nonresponse bias is a major concern because it threatens the representativeness of survey findings and their generalizability to the population of interest. Transparent reporting of response rates, weighting schemes, and sensitivity analyses is essential to convey the potential impact of nonresponse on study conclusions.