Is nonresponse the same as coverage error?

No. Coverage concerns which population units are represented by the collection frame; nonresponse concerns missing information from units intended to contribute.

Can a high response rate still be misleading?

Yes. An important subgroup or outcome may remain poorly observed, and measurement problems can also persist.

Does nonresponse necessarily invalidate a study?

No. Its implications require investigation, suitable analysis, and transparent limitations.

Can one response rate summarize everything?

Usually not. Unit-level participation and item-specific availability answer different questions.

AI Glossary / Statistics and probability / Beginner

Nonresponse

Definition

Nonresponse occurs when information intended to be collected from an eligible unit is not obtained or is incomplete. In surveys, unit nonresponse means that the unit provides little or none of the requested information, while item nonresponse concerns unanswered individual questions. Similar missing-result problems can arise in experiments and observational studies. Nonresponse describes missing information; nonresponse error describes its effect on a statistical result.

What it means

A person may be unavailable, decline participation, stop midway, or skip a question. A measurement may also fail to be recorded. These events can leave the analyst with fewer observations than planned and with an observed group that differs from the intended group.

The cause and pattern matter. Missing answers are not automatically equivalent to zero, disagreement, dissatisfaction, or absence of an event. Nor does every instance of nonresponse create the same amount of bias. Its consequences depend on which information is missing, how it relates to the outcome, and how the analysis handles it.

How it works

Define eligibility and record what happened to every selected unit. Distinguish completed responses, partial responses, refusals, noncontacts, and other relevant outcomes. Keep legitimate “not applicable” responses separate from unanswered questions when their meanings differ.

Calculate participation summaries using clearly specified denominators. Item response can vary even among people counted as survey participants, so a single overall response rate may hide missingness in an important outcome. Examine patterns across groups and collection stages.

Improve collection where feasible through clear questions, accessible formats, reasonable burden, and appropriate follow-up. For analysis, consider methods such as weighting or imputation only with stated assumptions and diagnostics. Preserve a record of which values were observed and which were supplied analytically. Sensitivity analyses can show how conclusions change under plausible assumptions about the missing information.

Three practical examples

An employee survey receives no completed questionnaire from some selected employees, producing unit nonresponse.
A customer answers most questions but leaves the AI-use question blank, producing item nonresponse for that variable.
A planned experiment lacks a final measurement because a recording process fails, leaving the intended outcome unobserved.

Think of it this way

Think of a class register with some blank marks. A blank does not tell you whether the student was absent, present but overlooked, or excluded from that activity. Understanding the missing mark is part of interpreting the record, not merely a formatting task.

Worked example

Consider an original survey example with 200 known eligible customers, all invited under a simple stated design. There are 140 returned questionnaires and 60 customers who provide no questionnaire. Under a definition that counts all 140 returns as responding units, the unit response rate is 140/200 = 70%, and unit nonresponse is 30%.

Of the 140 returns, only 120 answer the key satisfaction item. There are 20 item nonresponses among returned questionnaires, giving item nonresponse of 20/140, approximately 14.3%. The key outcome is observed for 120/200 = 60% of all invited eligible customers.

Suppose 90 of those 120 answers are favorable. The observed favorable proportion is 90/120 = 75%. Without additional information, the favorable proportion among all 200 could range from 90/200 = 45%, if all 80 missing outcomes were unfavorable, to 170/200 = 85%, if all were favorable. These are extreme logical bounds for this finite invited group, not a confidence interval or a claim about what the missing customers actually think. They show why returning a questionnaire and answering the key item are different milestones.

Illustrative case study

Fictional case study: Cedar Feedback surveys users after an AI-assisted support interaction. Its dashboard counts every opened questionnaire as a response, even when the main outcome question is unanswered. Management interprets that headline count as evidence of comprehensive customer feedback.

An analyst separates invitations, questionnaire starts, partial returns, and completed outcome items. The revised report shows exactly which denominator supports each percentage. Cedar also tests whether the final question is difficult to understand. The fictional case illustrates how better collection records can reveal a missing-information problem that an attractive headline response rate concealed.

Common misconceptions and mistakes

“A blank answer means no.” That interpretation must follow the question's design and data coding, not convenience.

“Every survey participant answered every item.” Partial response can produce different analytic sample sizes for different outcomes.

“Nonresponse always means refusal.” Noncontact, inability to respond, technical failure, and other causes can also matter.

“Imputed values are observed answers.” They are analytically supplied values whose validity depends on the method and assumptions.

Business use

Businesses should track nonresponse because feedback and evaluation systems can systematically miss the people or situations most relevant to a decision. Users with failed sessions may never reach a feedback form, and customers with limited access may be harder to contact.

Make response definitions visible in reports and preserve the collection history. Compare observed characteristics of responders and nonresponders when legitimate information is available, while recognizing that similarity on recorded variables does not prove similarity on unobserved outcomes. Choose follow-up and analysis methods based on the decision's needs rather than simply maximizing the headline number of completed forms.

Check your understanding

In the example, why is the key-outcome availability 60% rather than 70%?

Answer: Only 120 of the 200 eligible invited customers provide the key answer. Twenty additional customers returned questionnaires but omitted that item, so the unit response count overstates outcome availability.

Related terms

Nonresponse error
Missing data
Coverage error
Unit nonresponse
item nonresponse
response rate
imputation
survey weighting

Learn next

Map a collection process from eligibility to each recorded outcome. Then distinguish missingness mechanisms from observable response patterns. Study weighting, imputation, and sensitivity analysis, paying particular attention to the assumptions each method makes about unobserved information.

Sensitivity analysis

Sources

NIST/SEMATECH e-Handbook glossary, nonresponse: https://www.itl.nist.gov/div898/handbook/glossary.htm

Statistics Canada, Non-sampling error, unit and partial nonresponse: https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch6/nse-endae/5214806-eng.htm

The counts and Cedar Feedback case are original illustrations. Response-rate definitions are explicitly simplified for the example; formal survey standards may use additional disposition categories.

Previous: Nonprime attribute Next: Nonresponse error