Nonresponse bias occurs when individuals chosen for a sample can't be reached or refuse to respond, and those nonrespondents differ systematically from respondents, making the sample unrepresentative of the population (AP Stats Topic 3.4, LO 3.4.A).
Nonresponse bias is what happens when the people who don't answer your survey are different in some meaningful way from the people who do. The CED puts it plainly in Topic 3.4. Individuals chosen for the sample for whom data cannot be obtained, or who refuse to respond, may differ from those for whom data can be obtained. The key word is chosen. Your sampling method might be perfectly random, but if 85% of the selected people toss the survey in the trash, randomness can't save you.
Here's the trap. Nonresponse isn't a problem just because your sample got smaller. It's a problem because who responds is often connected to what you're measuring. A survey about income gets ignored by people uncomfortable sharing their income. A survey about free time gets answered mostly by people with free time. That's systematic, not random, so collecting more data the same way won't fix it. Bias doesn't shrink as n grows.
Nonresponse bias lives in Topic 3.4 (Potential Problems with Sampling) under learning objective 3.4.A, which asks you to identify potential sources of bias in sampling methods. It's one of the named bias types in the essential knowledge, alongside voluntary response bias and undercoverage bias, so you need to tell them apart on sight.
It also reaches forward into Unit 5. Topic 5.4 (LO 5.4.A) defines an unbiased estimator as one whose average value equals the population parameter. The sampling distribution facts in Topics 5.5 and 5.6, like μp̂ = p, only hold when the sample actually behaves like a random sample from the population. Nonresponse bias breaks that assumption, so your point estimate is systematically off-target no matter how big your sample is. That's the through-line the exam loves: a bad collection method in Unit 3 poisons every calculation you do in Units 5 and beyond.
Keep studying AP Statistics Unit 5
Voluntary Response Bias (Unit 3)
These are mirror images. In voluntary response bias, people insert themselves into the sample. In nonresponse bias, properly selected people remove themselves. Same CED list in Topic 3.4, opposite direction of the problem.
Undercoverage Bias (Unit 3)
Undercoverage means part of the population had a reduced chance of being selected in the first place, like a phone survey missing people without phones. Nonresponse means they were selected but didn't answer. Ask yourself whether the failure happened before or after selection.
Biased and Unbiased Point Estimates (Unit 5)
Nonresponse bias is how a 'bias in sampling' becomes a 'biased estimate.' If affluent households answer an income survey at higher rates, the average of p̂ across repeated samples no longer equals p, which is exactly what LO 5.4.A means by a biased estimator.
Sampling Distributions for Sample Proportions (Unit 5)
The formulas μp̂ = p and σp̂ = √(p(1-p)/n) assume the sample is random and representative. Nonresponse quietly violates that, and notice that increasing n shrinks σp̂ but does nothing to the bias. Bigger sample, same wrong center.
Nonresponse bias shows up most often in multiple-choice scenario questions where you have to name the bias type from a short description. Classic stems include a mail survey about income with a 15% response rate where most responses come from affluent neighborhoods, or a finding that higher-income individuals are less likely to disclose their income. You may also be asked which technique would be least effective at reducing nonresponse in a mail survey, so know the real fixes (follow-up contacts, incentives, shorter surveys, multiple contact methods). No released FRQ has used the phrase verbatim, but Unit 3 study-design FRQs routinely ask you to identify a flaw in a sampling method and explain its effect, and nonresponse is a go-to answer. To earn credit you must do three things in context: name the bias, explain the mechanism (nonrespondents likely differ from respondents on the variable being measured), and state the direction of the bias (will the estimate be too high or too low, and why).
Both involve people self-selecting, so they feel identical, but the sampling setup is different. Voluntary response bias starts with an open invitation (an online poll, a call-in survey) where anyone who feels strongly opts IN, and there was never a random selection. Nonresponse bias starts with a legitimate sampling method that selects specific people, but some of those chosen people opt OUT. Quick test: if the researcher picked the individuals and they didn't answer, it's nonresponse. If the individuals picked themselves, it's voluntary response.
Nonresponse bias occurs when people selected for a sample cannot be reached or refuse to respond, and those nonrespondents differ from respondents in ways related to what's being measured.
The sample selection can be perfectly random and you can still have nonresponse bias, because the damage happens after selection, not during it.
Increasing the sample size does not fix nonresponse bias; it just gives you a more precise estimate of the wrong value.
On the exam, distinguish it from voluntary response bias (people opt in) and undercoverage bias (people never had a fair chance of being selected).
A strong exam answer names the bias, explains why nonrespondents likely differ from respondents in context, and states whether the estimate will be too high or too low.
Nonresponse bias connects Unit 3 to Unit 5 because it makes p̂ a biased estimator of p, breaking the assumption that μp̂ = p.
It's the bias that occurs when individuals chosen for a sample can't be contacted or refuse to respond, and those nonrespondents differ systematically from respondents. It's one of the named bias types in Topic 3.4 under learning objective 3.4.A.
Not automatically, but it's a major red flag. Bias only occurs if nonrespondents differ from respondents on the variable being studied. In practice, exam scenarios pair a low response rate (like 15%) with a clear pattern, such as mostly affluent households responding to an income survey, and that combination is nonresponse bias.
Direction of self-selection. In nonresponse bias, the researcher randomly selects people and some opt out. In voluntary response bias, there's an open invitation and people with strong opinions opt in. If specific individuals were chosen and didn't answer, it's nonresponse.
No. A bigger sample reduces variability (σp̂ gets smaller), but the center of the sampling distribution stays shifted away from the true parameter. The fix is improving the response rate through follow-ups, incentives, or easier survey formats, not adding more of the same flawed data.
No. Nonresponse bias is about people not answering at all. Response bias is about people answering but giving inaccurate answers, often because of question wording or social pressure. If higher-income people skip the income question entirely, that's nonresponse; if they answer but understate their income, that's response bias.