Sampling

Sampling is drawing a subset from a larger population to make inferences about the whole. The power of sampling is efficiency — you don't need to survey every customer or test every product unit to reach reliable conclusions. The risk is bias: if the sample isn't representative, every conclusion drawn from it is systematically wrong. Selection bias, convenience sampling, non-response bias, and survivorship bias all corrupt samples in predictable ways. In business, sampling errors appear constantly: customer feedback skewed toward vocal complainers, NPS surveys with low response rates, competitor analysis limited to visible successes. The model teaches that how you select your sample matters more than how large it is — a biased sample of a million is worse than a representative sample of a hundred.

For every data-driven decision, ask: who's in this sample and who's missing? Actively design samples to include populations that don't self-select in — churned customers, failed experiments, silent employees. Increase sample size to reduce random error, but focus first on reducing systematic bias, which sample size alone can't fix. When you can't get a representative sample, acknowledge the limitation explicitly in your conclusions.

Sampling draws a subset to understand the whole. Representative samples produce reliable conclusions; biased samples produce confident errors. Focus on who's missing from your sample before worrying about its size, and acknowledge sampling limitations in every data-driven conclusion.