ItemNonresponse

Item nonresponse refers to missing data at the item level in survey research and related data collection contexts, where a respondent provides a response for some questions but not for others. It is distinguished from unit nonresponse, where the respondent does not participate at all. Item nonresponse is common across surveys and administrative datasets and can affect any item, including sensitive or complex questions.

Causes of item nonresponse include respondent-related factors such as privacy concerns, recall difficulty, or perceived sensitivity, as well as instrument-related factors like unclear wording, ambiguous response options, or the placement of questions (e.g., near the end of a long questionnaire). Missingness is often discussed in terms of mechanisms: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR). In practice, MAR—where the probability of a missing response depends on observed data—is a common working assumption, though MNAR cannot be ruled out.

Implications of item nonresponse include reduced effective sample size and increased variance, with potential bias if the missingness is related to the unobserved values or to the outcomes of interest. Prevention strategies focus on questionnaire design and administration, such as clear wording, appropriate response options, placing sensitive items later, and offering options like “prefer not to answer.” Analysis typically involves methods that use all available data or account for missingness, including complete-case analysis, weighting adjustments, and imputation approaches (single or multiple), as well as model-based methods like maximum likelihood or the EM algorithm under MAR assumptions. Sensitivity analyses help assess robustness to different missing-data assumptions.

See also: Missing data, Nonresponse bias, Imputation, Multiple imputation, EM algorithm.