Who Actually Gets Studied
A population is the entire group a researcher wants to draw conclusions about; a sample is the subset actually studied. Because studying an entire population is almost always impractical, how the sample is drawn matters enormously for whether conclusions generalize back to the population.
Random sampling gives every member of the population an equal chance of being selected, minimizing systematic bias and supporting strong generalizability — though it can still leave small subgroups underrepresented by chance. Stratified sampling improves on this by first dividing the population into meaningful subgroups (strata) — by age, gender, or region, for instance — and then sampling proportionally from each, guaranteeing that important subgroups are represented in their true proportion. Convenience sampling, by contrast, simply recruits whoever is easiest to reach (college students in an introductory psychology course are the classic, heavily criticized example) — fast and cheap, but prone to serious sampling bias, since the people easiest to reach are rarely a random cross-section of the broader population.
| Method | How it selects | Generalizability |
|---|---|---|
| Random sampling | Equal chance for everyone | Strong |
| Stratified sampling | Proportional by subgroup | Strong, subgroup-protected |
| Convenience sampling | Whoever is easiest to reach | Weak |
The consequence of poor sampling is a threat to external validity: findings from a biased sample may simply not hold true for the broader population the researcher actually wants to say something about. This is precisely the concern behind the WEIRD-samples critique discussed earlier in this course — a discipline built heavily on convenience samples of undergraduates risks systematically over-generalizing from a narrow slice of humanity.
Common pitfall: assuming a LARGE sample automatically means a REPRESENTATIVE sample. Sample size and sampling method are separate issues — a huge convenience sample of, say, 50,000 self-selected online volunteers can still be badly biased, while a smaller, carefully drawn random sample can generalize far better.