What is sampling bias, and what are common types?
Sampling bias happens when the sample systematically differs from the population, so conclusions do not generalise. Common types are selection bias, survivorship bias, non-response bias and convenience sampling.
Explain it simply
If you want to know how Indians feel about online payments but you only ask people at a tech park, you will hear from the most digital crowd. A big sample of the wrong people is still wrong.
Common types:
- Convenience / selection bias: you pick whoever is easy to reach, such as friends or a WhatsApp group.
- Non-response bias: the people who answer differ from those who do not, as with long surveys that only enthusiasts finish.
- Survivorship bias: you only see what lasted, such as studying successful startups and ignoring the many that closed.
- Undercoverage: part of the population is not in your list, like an online-only survey missing offline users.
Example
An app asks for ratings after a successful ride only. The average rating of 4.7 ignores riders who cancelled or gave up. The data looks great but it only measures happy completions.
A fix is to survey a random sample of all booking attempts, not just completed trips.
What to say in the interview
"Sampling bias means my sample is not representative because of how it was collected. A bigger sample does not cure it. I reduce it with random sampling, a good sampling frame, follow-ups for non-responders, and by asking what data is missing and why before trusting a dataset."
Common mistakes
- Believing that more rows automatically means better evidence.