Response Bias and Nonresponse Bias
You'll learn how the way questions are asked and who chooses to respond creates systematic deviations from true population values.
What's the Problem?
Even with a perfectly random sample, you can still end up with wrong conclusions if your data collection process introduces bias. Two critical culprits are response bias and nonresponse bias.
Response Bias
Response bias occurs when the way you ask questions influences how people answer, causing responses to systematically differ from their true values.
Examples:
- Leading questions: "Don't you agree that our amazing product is excellent?" pushes toward positive answers
- Social desirability: People underreport embarrassing behaviors (alcohol consumption, income) or overreport "good" behaviors (exercise, charitable giving)
- Question order effects: Asking about crime rates before asking about neighborhood satisfaction affects responses
The analogy: Imagine asking "How much do you love chocolate?" versus "How do you feel about chocolate?" The first phrasing nudges people toward positive responses.
Nonresponse Bias
Nonresponse bias happens when people who choose not to participate differ systematically from those who do participate.
Examples:
- Survey about work-life balance sent during business hours—busy, overworked people won't respond
- Health studies where sicker people are less likely to participate
- Political polls where one side feels more motivated to respond
The key issue: Your sample no longer represents the population because a specific group self-selects out, and their absence skews results.
The Combined Impact
These biases can completely invalidate your estimates, even with proper sampling methods. You might calculate a sample mean or proportion perfectly, but if your data collection introduced systematic errors, your statistics don't reflect the true population parameters.
Key Takeaway: Response bias distorts how people answer, while nonresponse bias distorts who answers—both create systematic deviations from true population values that random sampling alone cannot fix.