Strong designs use probability-based selection or another carefully validated method, work to reach people the sampling frame might miss, and report how interviews were collected and weighted. Question wording, field dates, sample size, response patterns, and sponsor all matter. A very large voluntary website poll is not scientific merely because many people answered; self-selection can make the respondents unlike the population.

Treat every estimate as evidence with limits. Sampling uncertainty is only one concern. Coverage error, nonresponse, wording, order, interviewer effects, and turnout assumptions can also alter results. The right exam habit is simple: identify the population, inspect how the sample was reached, and make no claim broader than the design can support.

Video lesson: Science Behind the News: Opinion Polls & Random Sampling