Nonresponse bias is the systematic error that occurs when the people who answer a survey differ in important ways from the people who do not, so the sample no longer represents the target population. If the 80% who ignored your questionnaire hold different views from the 20% who replied, your findings are skewed, often without the researcher ever noticing. It is one of the most common and most underestimated threats to survey validity in academic research.
This guide covers exactly what nonresponse bias is, what causes it, worked and real-world examples, how it differs from related biases (response, voluntary-response and selection bias), and a practical, evidence-based checklist for reducing it in your own dissertation or study.
What Is Nonresponse Bias?
In research, we tend to obsess over the data we collect. But what about the data we do not have? Nonresponse bias occurs when there is a meaningful, systematic difference between the people who respond to a study and those who do not. Because these “non-responders” often have different opinions, lifestyles, circumstances or experiences from the “responders”, the final dataset becomes a distorted version of reality rather than a faithful picture of the population. It is the invisible weight that can tip the scales of a study without the researcher ever realising it.
Crucially, a low response rate is not, on its own, proof of bias. A survey can have a 30% response rate and still be unbiased if the non-respondents are essentially identical to the respondents on the variables you care about. The bias appears only when nonresponse is related to the thing you are measuring. A poll about voting that loses busy professionals, or a health survey that loses the seriously ill, is biased precisely because the missing people would have changed the answer. This is why nonresponse bias is a recognised form of research bias rather than just a sampling inconvenience.
Nonresponse Bias vs a Low Response Rate
Think of response rate as a warning light, not the fault itself. The fault is differential nonresponse: when the probability of responding correlates with the answer. Two surveys with identical 25% response rates can have wildly different levels of bias depending on who the missing 75% are.
A Worked Example: How the Bias Distorts the Number
It helps to see nonresponse bias as a calculation rather than an abstract idea. Imagine a university surveys 1,000 students about satisfaction with campus mental-health services.
- 700 satisfied students (true mean 7.0) respond at a rate of 40% → 280 responses.
- 300 dissatisfied students (true mean 3.7) respond at a rate of only 15% → 45 responses (they have disengaged and are less likely to reply).
Reported result: (280 × 7.0 + 45 × 3.7) ÷ 325 = (1960 + 166.5) ÷ 325 = 6.54 out of 10. The survey reports 6.54, but the true population value is 6.0. Nonresponse bias has inflated satisfaction by more than half a point, and a manager reading the report would wrongly conclude the service is performing better than it is. Notice the distortion comes entirely from the difference in response rates between happy and unhappy students, not from the sample size.
This is also why weighting can repair the damage. If we knew the true population split (70% satisfied / 30% dissatisfied) we could re-weight the 280 and 45 responses back to that ratio and recover an estimate much closer to 6.0. We return to weighting and other fixes in the reduction section below.
How Nonresponse Bias Happens (At a Glance)
What Causes Nonresponse Bias?
Nonresponse is rarely random. It clusters around predictable causes, and recognising them early in your research design is the first line of defence. The main drivers are:
- Topic sensitivity. People skip surveys on income, health, immigration status or mental health because the subject feels private or stigmatising, so the most affected respondents drop out.
- Survey burden. Long, complex or repetitive instruments exhaust respondents; drop-off rises sharply after roughly 10–12 minutes.
- Access and mode mismatch. An online-only survey silently excludes those with poor internet access; a phone survey excludes those who screen unknown numbers.
- Salience and interest. People with no stake in the topic ignore the invitation, while those with strong views over-participate.
- Timing and reminders. A single email sent at a bad time reaches only the most diligent; busy or shift-working groups are systematically lost.
- Trust. Where the sender, sponsor or potential sources of the survey are unfamiliar or feel untrustworthy, suspicious groups decline at higher rates.
Two distinct mechanisms sit underneath these causes: unit nonresponse (a sampled person provides no data at all) and item nonresponse (a respondent answers the survey but skips particular questions). Both can bias results; item nonresponse is especially common on sensitive items and is easy to miss because the respondent technically “took part”.
Nonresponse Bias vs Related Biases
Nonresponse bias is frequently confused with several neighbouring concepts. The distinctions matter because each one calls for a different fix. The tables below set them side by side. For the broader family of distortions and how they interrelate, see our guide to research bias.
Nonresponse Bias vs Response Bias
| Response Bias | Nonresponse Bias | |
|---|---|---|
| Definition | People take part but answer in a way that does not reflect their true views — through misunderstanding, wanting to please the researcher, or giving socially acceptable answers. | Selected individuals do not respond at all, so the people who are present differ systematically from the people who are missing. |
| Cause | Poor question wording, social desirability, leading questions or misunderstanding. | Inaccessibility, lack of motivation, survey burden or time constraints among non-respondents. |
| Impact | Skews answers in a particular direction depending on how respondents alter their replies. | Under-represents certain groups, reducing the validity and generalisability of findings. |
| How to reduce | Use neutral, clear wording, pilot-test the survey and allow anonymous responses. | Raise response rates with reminders, incentives, shorter surveys and confidentiality assurances. |
| Example | Respondents under-report alcohol consumption because of social stigma. | A phone survey on internet usage excludes those without phones, biasing the conclusions. |
Nonresponse Bias vs Voluntary Response Bias
| Nonresponse Bias | Voluntary Response Bias | |
|---|---|---|
| Definition | People chosen for the sample do not respond, so respondents may differ systematically from non-respondents. | People self-select into the study (common in online or phone polls), so the sample reflects who chose to opt in rather than the wider population. |
| Source of bias | Sampling friction — surveys that are too long, complex or demanding of time and motivation. | Self-selection — people with strong opinions or high interest are far more likely to take part. |
| Effect on data | Over- or under-estimates true population values when non-respondents differ from respondents. | Biases results toward extreme opinions or specific demographics, reducing generalisability. |
| How to minimise | Follow-ups, simpler surveys, higher engagement, or adjusting the sampling method. | Use probability (random) sampling, or weight responses to correct demographic imbalances. |
Voluntary-response bias is closely tied to self-selection bias: in both, the participant, not the researcher, decides who ends up in the data.
Selection Bias vs Nonresponse Bias
| Selection Bias | Nonresponse Bias | |
|---|---|---|
| Definition | The participants are not representative of the population because of how the sample itself was chosen. | The sample is drawn fairly, but the people who actually respond differ systematically from those who do not. |
| Source of bias | The way the sample is selected, often inadvertently — e.g. an online survey excludes non-internet users. | An issue at the response stage — the survey is too long, or people lack the time or motivation to reply. |
| Effect on data | Distorts findings because the sample never matched the intended population, harming generalisability. | Over- or under-estimates true values when non-respondents differ markedly from respondents. |
| How to minimise | Random, stratified or other probability sampling so everyone has a known chance of selection. | Follow-ups with non-respondents, making the survey easier and more appealing, or adjusting the design. |
A useful rule of thumb: selection bias is a problem with who you invite; nonresponse bias is a problem with who actually answers. Both are distinct from the ecological fallacy, which is a reasoning error (drawing individual-level conclusions from group-level data) rather than a sampling problem.
Real-World Examples of Nonresponse Bias
Nonresponse bias is easy to spot once you know the pattern. Here are five everyday situations where it routinely creeps in.
Healthcare Satisfaction Surveys
In a survey on healthcare satisfaction, patients who had negative experiences may be more likely to refuse participation (or to have left the service entirely), so the results overestimate overall satisfaction. The sickest or most dissatisfied patients are precisely the ones least likely to fill in the form.
Customer Satisfaction Surveys
When businesses send out customer satisfaction surveys, those with extremely positive or negative experiences are more motivated to respond than customers with moderate experiences. The resulting sample exaggerates the extremes and may not represent overall customer sentiment.
Online Product Reviews
Review platforms let anyone leave feedback, but people with strongly positive or negative experiences are far more likely to write a review than those with an average experience. This produces the familiar J-shaped distribution of ratings and rarely reflects the typical customer.
Opinion Polls
Phone and online opinion polls suffer nonresponse bias when people with particular views or strong opinions are more willing to respond. This over- or under-represents certain perspectives and undermines the accuracy of the headline figure — a well-documented cause of high-profile polling misses.
Job Application Feedback
Employers often request feedback from job applicants about the recruitment process. Applicants with exceptionally good or bad experiences are more inclined to reply, so the feedback may not capture the overall candidate experience.
How to Reduce Nonresponse Bias
Reducing nonresponse bias means tackling it on two fronts: preventing nonresponse before it happens (design and fieldwork) and correcting for it afterwards (analysis). The strongest studies do both. Build these safeguards into your chosen methods of data collection from the outset rather than bolting them on at the end.
1. Clear and Concise Communication
Explain the purpose, importance and benefits of the study in plain language. When people understand why their answer matters and how it will be used, they are far more likely to respond.
2. Personalised Invitations
Address participants by name and explain why their specific participation is valuable. A personal, relevant invitation reliably lifts response rates over a generic mass email.
3. Multiple Modes of Data Collection
Offer several ways to take part — online, telephone, postal or in person — so that no group is excluded by the channel. Mixed-mode designs are one of the most effective ways to recover the hard-to-reach. Sound data collection in statistics begins with making participation easy across every channel.
4. Keep the Survey Short and Simple
Cut every non-essential question, use a clear layout and show a progress bar. Because survey burden is a leading cause of drop-off, a shorter questionnaire directly reduces nonresponse. If you reference the instrument in your write-up, remember to cite the survey correctly so others can scrutinise its design.
5. Offer Incentives
A modest, fair incentive — a prize draw, voucher or small token — signals that you value people’s time and raises response rates. Keep incentives proportionate so they do not themselves distort who responds.
6. Follow Up With Non-Respondents
Plan two or three polite reminders. Follow-ups are among the most cost-effective tools available, and a final push often reaches the busiest, most different respondents — exactly the group whose absence causes the bias.
7. Guarantee Anonymity and Confidentiality
For sensitive topics, credible assurances of anonymity and confidentiality reassure reluctant participants and reduce both unit and item nonresponse. State clearly how data will be stored and who can see it.
8. Adjust Statistically for Nonresponse
After fieldwork, use post-hoc methods to correct what prevention could not. Where you know the population profile, statistical techniques such as post-stratification weighting can re-balance the sample, while a wave (or follow-up) analysis compares early and late responders to estimate the likely direction of any remaining bias. Always report your response rate and any weighting transparently.
“Nonresponse is a problem not because it reduces the sample size, but because it can reduce the sample’s resemblance to the population.” — Robert M. Groves, Survey Methodology
Why Nonresponse Bias Threatens Validity
Nonresponse bias is ultimately a threat to the reliability and validity of your conclusions. Even a large, carefully analysed dataset gives the wrong answer if the missing respondents would have changed the result. For a dissertation, examiners look for evidence that you anticipated nonresponse, measured your response rate, and reflected honestly on its likely effect on your findings. Acknowledging the limitation and explaining how you mitigated it is a mark of methodological maturity, not weakness.
Worried your survey results won’t stand up?
Our methodologists help you design representative surveys, lift response rates and defend your findings — and you can Learn More about our research support.
