Sampling Methods

The one idea: How you pick your sample decides what your data can honestly say before you compute a single statistic. Probability methods — simple random, stratified, cluster, systematic — let you calculate how far a sample is likely to be from the truth. Convenience sampling can't, no matter how large it gets, because "whoever showed up" is not a description you can do math with.

⚠️ The trap: "We surveyed 2,000 users through an in-app popup — that's a huge sample, the numbers are solid." That feels right and it's wrong. A big convenience sample is still a convenience sample: size shrinks the wobble around whatever the popup-clickers represent, but it does nothing to close the gap between popup-clickers and your actual user base — a biased sample just becomes a very precisely wrong one.

Every statistic you'll ever compute — a mean, a proportion, a lift, a correlation — is standing on top of a sampling method, and most people never look down. They ask "is n=2,000 big enough?" when the load-bearing question is "who had a chance of ending up in this 2,000, and who didn't?" Get the sampling method wrong and no amount of downstream statistical sophistication rescues you. Get it right and even a modest sample supports real inference.

This lesson is about the menu of sampling methods, when each earns its keep, and the one method — convenience sampling — that looks free and isn't.

Why the method matters more than the size

A sample exists to stand in for a population you can't fully measure. It can only do that job if you can say something precise…