What nonresponse bias means
Surveys are often used to learn what a larger group thinks or experiences by asking a smaller group. A well-chosen sample helps, but selecting people is only the beginning. Some invited participants will not answer at all; others will skip particular questions. Nonresponse bias can arise when the answers researchers receive differ in a systematic way from the answers they would have received from those who did not respond.
Imagine a survey about residents’ experiences with a public service. If people who found the service especially frustrating are more motivated to reply, the survey may overstate dissatisfaction. The reverse is also possible: people who have given up on the service may be less likely to respond, leaving their experiences underrepresented. Nonresponse is a problem not simply because some answers are missing, but because the missing answers may differ in ways that matter to the result.
How people become missing from a survey
There are several points at which participation can break down. An invitation may go unnoticed, arrive at an inconvenient time, or appear difficult to complete. A person may open a questionnaire but decide that its wording is confusing or its subject too personal. In an interview, someone may answer most questions while declining one about income or political views. The first case is often called unit nonresponse; the skipped question is item nonresponse.
It helps to distinguish nonresponse from coverage problems. If a survey invites people only by email, residents without usable email addresses may never have a chance to participate. That is a coverage issue, even if everyone who receives an invitation replies. A survey can have both problems at once. Keeping them separate makes it easier to ask whether the sampling method reached the intended population and whether those reached supplied the information needed.
Why the response rate is only a starting point
The response rate describes participation, but it does not, by itself, measure bias. A low rate gives researchers fewer direct observations and more reason to examine who is missing. It does not prove that an estimate is wrong. Likewise, a high rate does not guarantee accuracy if the remaining nonrespondents have unusually different views on the question being studied.
The key relationship is between the reason for nonresponse and the outcome of interest. If people who decline to answer resemble respondents on that outcome, nonresponse may have little effect on a particular estimate. If their views differ substantially, the effect may be large. The same survey could therefore be useful for one question and less dependable for another. A demographic difference among respondents is a clue worth investigating, not automatic proof of the direction or size of the bias.
What researchers can do
Researchers can try to make participation easier by keeping questionnaires clear, offering accessible ways to respond, and following up with people who did not answer an initial invitation. They can also compare respondents with reliable information about the population, when such information is available. If younger residents, for example, appear underrepresented, that comparison identifies a limitation and may guide additional outreach.
Weighting is another common tool. It gives greater influence to responses from groups that are underrepresented in the completed sample and less influence to those that are overrepresented. This can improve estimates when the variables used for weighting are related to both participation and the answers being measured. It cannot recover views that were never observed with certainty. If nonrespondents differ from respondents within the same weighting groups, bias may remain. Researchers may also examine how conclusions change under plausible assumptions about missing answers.
How to read a survey result critically
When a survey informs a civic debate, start with its target population: whose views is it meant to represent? Then look for an explanation of how participants were selected, how they were contacted, how many participated, and whether some questions had substantial missing answers. Consider whether the topic itself might make particular people more or less willing to reply. These questions are more informative than treating a response rate as a simple pass-or-fail grade.
Finally, look for a candid account of adjustments and limitations. Did the researchers compare respondents with known population characteristics? Were results weighted, and on what basis? Do they explain what the adjustment cannot address? Nonresponse bias is often difficult to measure directly because the missing answers are, by definition, unavailable. A careful report makes that uncertainty visible while explaining why its findings remain useful—or why they should be interpreted cautiously.
