Census, Survey, Experiment, and Observational Study

These four words describe different ways data gets collected, and mixing them up costs you the whole analysis later. A census collects data on every individual in a population — no sampling at all. A survey asks questions of some or all of a population, usually through a sample, to measure opinions or facts people report about themselves. An observational study watches subjects and records variables without interfering — nobody's situation is changed by the researcher. An experiment is the only one of the four that actively assigns subjects to different treatments and then compares the outcomes between those groups.

The test that separates experiment from observational study is simple: did the researcher impose a treatment? If people choose their own behavior — what they eat, which drug they already take, how much they exercise — and the researcher just measures what happens, that's observational, even if it involves two groups being compared. If the researcher flips a coin or uses randomization to decide who gets the new drug and who gets the placebo, that's an experiment. Randomized assignment is what lets an experiment claim cause and effect; an observational study can only show association.

This sets up the next idea: sampling methods. A census is the gold standard for completeness but is often too expensive or slow, which is exactly why random samples exist — they let you estimate a population's traits from a manageable subset, trading perfect accuracy for feasibility.