Home Statistical Dictionary Nonresponse Bias

Nonresponse Bias

The bias introduced when sampled units that do not respond differ systematically, on the variable of interest, from those that do respond, so the achieved sample no longer represents the full sample originally selected.

Sampling Theory & Survey MethodsIntermediateFree

In Plain English

Nonresponse bias shows up whenever the people who don't answer a survey are systematically different, on the very thing you're trying to measure, from the people who do answer. A survey about political engagement, for instance, is likely to be answered disproportionately by people who are already politically engaged, precisely the people less likely to be too busy or uninterested to bother responding, which skews the results toward showing higher engagement than the true population average. The mere fact of a low response rate doesn't automatically mean nonresponse bias is large (if nonrespondents are similar to respondents on the outcome, a low response rate can still yield accurate estimates), but a low response rate is exactly the condition that makes large nonresponse bias possible, which is why response rates are watched so closely as a survey quality signal.

Definition

Nonresponse bias is the bias in a survey estimate that results when the subset of sampled units that actually respond differs systematically, on the outcome variable being measured, from the subset that does not respond, so the responding sample no longer represents the original probability sample drawn. Formally, if $\overset{y}{ˉ}_{r}$ is the mean among respondents and $\overset{y}{ˉ}_{n r}$ is the (unobserved) mean among nonrespondents, the bias in using $\overset{y}{ˉ}_{r}$ to estimate the full-sample mean is approximately $(1 - R) (\overset{y}{ˉ}_{r} - \overset{y}{ˉ}_{n r})$ , where $R$ is the response rate, showing that nonresponse bias depends on the product of the nonresponse rate and the gap between respondents and nonrespondents, not on the response rate alone; a survey can have a low response rate yet negligible nonresponse bias if nonrespondents are similar to respondents, and conversely a high response rate does not guarantee low bias if the small group of nonrespondents happens to be extremely different. Common remedies include intensive follow-up of initial nonrespondents, weighting adjustments (post-stratification, raking) using auxiliary variables correlated with both response propensity and the outcome, and, where feasible, comparing respondent characteristics against known population benchmarks to detect and correct for nonresponse patterns.

Formula

Bias \approx (1 - R) (\overset{y}{ˉ}_{r} - \overset{y}{ˉ}_{n r})

Notation

R

—Response rate (proportion of sampled units that respond)

\overset{y}{ˉ}_{r}

—Mean of the outcome among respondents

\overset{y}{ˉ}_{n r}

—Mean of the outcome among nonrespondents (typically unobserved)

Properties

Nonresponse bias depends on the product of the nonresponse rate and the respondent-nonrespondent gap, not on the response rate alone, which is why response rate by itself is only a rough proxy for nonresponse bias risk, not a direct measure of it.
Nonresponse bias is a specific mechanism within the broader category of selection bias, the units that end up observed (respondents) differ systematically from the units that were sampled but not observed (nonrespondents), on exactly the outcome the survey is trying to measure.
Weighting adjustments like post-stratification and raking can partially correct for nonresponse bias when the auxiliary variables used for weighting are correlated with both response propensity and the outcome of interest, but they cannot fully correct for nonresponse driven by factors unrelated to the weighting variables.

At a Glance

Driversboth the nonresponse rate and how different nonrespondents are from respondents

Low response rate alonedoes not guarantee large bias, if nonrespondents resemble respondents

Common remediesfollow-up efforts, weighting adjustments, benchmark comparisons

Type of selection biasa specific mechanism by which selection bias enters survey data

Last updated September 8, 2026← Back to the Dictionary