Can an honest respondent give an incorrect answer?

Yes; misunderstanding and recall problems can be unintentional.

Does agreement with another survey prove accuracy?

No; both may share the same error.

Can an AI assistant eliminate respondent error?

No; it may help clarify questions but can also influence answers or misinterpret them.

AI Glossary / Statistics and probability / Intermediate

Respondent error

Definition

Respondent error is a component of survey measurement error arising when a respondent gives an incorrect answer, deliberately or inadvertently. It can result from misunderstanding, imperfect recall, estimation difficulties, sensitivity, fatigue, or mistakes in selecting or entering a response. The term identifies a source of discrepancy; it does not imply that the respondent is solely responsible for the survey's measurement problems.

What it means

A completed answer can differ from the quantity or fact the survey intends to measure. Someone may remember the wrong period, interpret a term differently, round a difficult number, or give an answer they believe is more acceptable.
Questionnaire design, collection mode, interviewer behavior, and respondent circumstances interact. A confusing question can create an incorrect response from a conscientious person. Improving measurement therefore often means improving the process, rather than simply asking people to be more accurate.

How it works

Define the target information precisely. A question about technology use needs a time period, a unit of reporting, and a clear meaning of use. Without those definitions, a discrepancy may reflect incompatible interpretations rather than a simple factual mistake.
Consider the response process: the person must understand the question, retrieve relevant information, form a judgment, and map that judgment to an answer format. Problems at any stage can affect the recorded response.
Recall error can omit events, duplicate them, or place them in the wrong reference period. Rounding and rough estimation may create heaping at convenient numbers. Social desirability or perceived consequences can encourage selective reporting, although intent should not be inferred without evidence.
Reduce avoidable burden with clear language, suitable reference periods, relevant answer options, and appropriate instructions. Where feasible, allow respondents to consult records. Cognitive testing can reveal how people interpret terms and arrive at answers before a large survey begins.
Use checks carefully. Plausibility or consistency rules can identify answers needing clarification, but an unusual answer is not necessarily wrong. Forced validation rules can themselves cause people to supply inaccurate values merely to continue.
Assess error using a credible reference when available, such as verified records, a carefully designed reinterview, or another measurement process. The reference can also be imperfect. Agreement between two answers does not prove truth, and disagreement does not by itself identify which answer is wrong.
Separate respondent error from nonresponse, sampling error, coverage error, and processing mistakes. These sources can coexist, but increasing sample size mainly reduces sampling variability and does not automatically remove systematic misreporting.
Report how potential error was investigated and what assumptions support corrections. If AI is used to clarify or code responses, preserve the original answer and validate the added processing. An automated interpretation can introduce new error rather than repair the original measurement.

Three practical examples

AI adoption research: a respondent counts ordinary automation as AI because the questionnaire does not define the term clearly.
Consumer research: a participant estimates last month's purchases and forgets several small transactions.
Workforce surveys: an employee reports training hours using a different reference period from the one requested.

Think of it this way

Think of measuring a journey by asking someone to reconstruct it from memory. A sincere answer can still miss a stop or use the wrong dates. Better prompts and records can help, but the resulting measurement must still be evaluated against the intended journey.

Worked example

Suppose five respondents report monthly training hours of 12, 8, 15, 9, and 6. For this invented illustration, assume verified records give the target values 10, 8, 12, 10, and 5.
Reported minus reference errors are 2, 0, 3, -1, and 1 hours. The mean signed error is (2 + 0 + 3 minus 1 + 1)/5 = 1 hour. The reported mean is 10 hours, compared with the reference mean of 9 hours, confirming the same one-hour difference.
The mean absolute error is (2 + 0 + 3 + 1 + 1)/5 = 1.4 hours. The mean squared error is (4 + 0 + 9 + 1 + 1)/5 = 3 squared hours.
Signed errors can cancel, while absolute and squared measures reflect discrepancy size. These five records do not establish the error rate in a wider population or reveal whether the discrepancies came from memory, interpretation, deliberate reporting, or imperfect reference records.

Illustrative case study

Fictional case: Hazel Research surveys employees about AI-assisted work. Interviews reveal that some respondents include built-in spelling correction while others count only generative tools. The team revises the definition, tests examples with new participants, and reports the earlier results with their measurement limitation instead of silently relabeling every inconsistent response as dishonest.

Common misconceptions and mistakes

Do not equate respondent error with intentional deception. Many errors arise from ordinary memory limits or unclear questions.
A larger sample can estimate a systematically wrong response pattern more precisely without correcting it. Do not present a small sampling margin of error as a bound on all measurement error.
Missing answers are not automatically respondent measurement errors; they also raise nonresponse questions.
Avoid overwriting plausible but unexpected values solely because they disagree with an analyst's expectations. Keep an audit trail for verified corrections.

Business use

Businesses rely on survey answers to size markets, evaluate training, and understand customers. Respondent-error analysis improves those decisions by identifying which questions need redesign, where records can support recall, and how much confidence to place in reported measures. The objective is a better measurement system, not assigning blame.

Check your understanding

Why is mean absolute error larger than the mean signed error in the example?

Answer: Taking absolute values prevents positive and negative discrepancies from canceling.

What is assumed about the reference records?

Answer: They represent the intended target values for the arithmetic illustration.

Related terms

Measurement error
Recall bias
Nonresponse error
Response Bias
social desirability bias
questionnaire design
cognitive interviewing
total survey error

Learn next

Learn next how cognitive testing separates comprehension and recall problems, how validation studies assess reporting discrepancies, and how total survey error combines multiple sources of uncertainty and bias.

Sources

Sources: NIST/SEMATECH glossary, respondent error and survey measurement error, https://www.itl.nist.gov/div898/handbook/glossary.htm ; Statistics Canada, questionnaire development and testing guidance, https://www150.statcan.gc.ca/n1/pub/12-539-x/steps-etapes/4147792-eng.htm . These primary sources establish the terminology and practical testing context. The five-record comparison and fictional case are original.

Previous: Resource constraint Next: Response