Track 6 · Sampling and the Central Limit Theorem · Lesson 2 of 8
Sampling bias and nonresponse
Recognise coverage, nonresponse and voluntary-response bias, quantify how nonresponse distorts an estimate, and correct a survey mean by weighting.
An engagement survey reports that staff are fairly proud of their company. But only 45% of Operations, one of the least engaged departments, answered, against 80% of Human Resources. The respondents are not a random sample of staff — and if the people who stayed silent are the less engaged ones, the headline number is too rosy, no matter how many people replied.
A sampling method is biased when it tends to give estimates that are off in the same direction, sample after sample. Random sampling error averages out over repeated samples; bias does not. The main sources:
- Coverage (undercoverage) bias: the sampling frame leaves out part of the population (a phone survey using landline numbers misses people with only mobiles).
- Nonresponse bias: some selected units do not answer, and the non-responders differ from the responders on what is being measured.
- Voluntary-response bias: people choose themselves (online polls, comment cards), so strong opinions are over-represented.
- Response (measurement) bias: people answer but not accurately — leading questions, sensitive topics, an interviewer's presence.
Included with a Formula Quest membership
It has 7 graded exercises with feedback, hints and worked solutions. After it you will be able to:
- Identify coverage, nonresponse, voluntary-response and response bias in a survey design
- Explain why a larger sample reduces random error but not bias
- Compute response rates and post-stratification weights
- Correct a survey mean by weighting, and state what weighting cannot fix