← Evaluate statistically based reports
3 Sampling & error
Watch: Sampling & error Opens a YouTube search: pick a video that clicks for youSampling methods
- Simple random, everyone equally likely; the gold standard.
- Stratified, split into groups (strata), sample each proportionally; improves representativeness.
- Cluster / systematic. Practical but can introduce bias.
- Self-selected / convenience, cheap but usually biased and unrepresentative.
Two kinds of error
- Sampling error: natural variation because we only measured a sample: shrinks as sample size grows.
- Non-sampling error: bias from method, poor questions, non-response, measurement error, is not fixed by a bigger sample.
- A bigger sample reduces sampling error only; it can’t rescue a biased method.