Randomization in Surveys, Experiments, and Observational Studies

Random selection means every member of a population has a known chance of landing in the sample. That one feature is what makes a few hundred people a fair stand-in for millions: because nobody self-sorted into the sample, the people who ended up in it look, on average, like the people who didn't. Break that randomness and the sample drifts away from the population in whatever direction the selection process favored.

Volunteers break it hardest. People who respond to an ad, call in to a poll, or fill out a survey on a website are not a random slice of anyone. They tend to care more, have stronger opinions, or have more free time than people who ignore the request. That gap is systematic, meaning it points the same direction for everyone affected by it, so it does not average out and no sample size shrinks it. Five thousand self-selected replies carry less trustworthy information about a population than five hundred randomly chosen people, because the five hundred were not filtered by anyone's willingness to answer.

Random assignment is a separate tool, used once subjects are already in a study, and it answers a different question. An observational study just watches people who already chose their own treatment, diet, or habit, so any group difference might come from that choice rather than the treatment itself. An experiment that randomly assigns subjects to treatment and control spreads every other difference, age, health, motivation, evenly across groups. That balancing is what lets the researcher credit a group difference to the treatment and call it a cause, not just a correlation.