A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Compares probability and non-probability sampling approaches with attention to representativeness and generalizability. Covers random, stratified, and cluster sampling for surveys; theoretical, purposive, and snowball sampling for qualitative research; and how sampling strategy affects validity and inference.
Design sampling frames for different populations, calculate power and sample size requirements, evaluate sampling in published studies for potential biases.
From probability and statistics you know that random sampling is what allows inference from a sample to a population — and why. When every member of a population has an equal, known chance of selection, the sample's properties mirror the population's within calculable margins of error. This is the mathematical foundation for polling, clinical trials, and national surveys. Sampling strategy in social research is the practical translation of this principle: how do you actually draw a sample that allows the inferences your research question demands?
Probability sampling methods preserve the mathematical guarantees. Simple random sampling gives every individual an equal chance — workable with a complete population list but often impractical at scale. Stratified random sampling first divides the population into meaningful subgroups (strata) — by age, region, income — and then samples randomly within each stratum. This improves precision when the variable of interest varies across strata, and ensures rare groups appear in sufficient numbers to analyze. Cluster sampling is a practical compromise: instead of sampling individuals directly, you randomly sample naturally occurring clusters (schools, neighborhoods, households) and then survey everyone within the selected clusters. It reduces fieldwork costs dramatically at the price of statistical efficiency — clustering introduces correlation within groups that must be corrected for.
Non-probability sampling does not offer statistical representativeness, but this is not a failure — it is a different research logic. Purposive sampling selects cases for their theoretical relevance: if you are studying how hedge fund managers make decisions, you don't want a random sample of Americans, you want hedge fund managers. Snowball sampling starts with accessible contacts who then refer additional participants — invaluable for studying populations with no sampling frame (undocumented immigrants, illicit drug users, underground musicians). Theoretical sampling in grounded theory selects new cases to probe emerging conceptual categories rather than to represent a population. The standard of quality shifts from statistical representativeness to theoretical saturation — the point at which new cases add no new conceptual variation.
The deepest error is applying the wrong standard to the wrong type of sampling. Criticizing a purposive sample for not being representative misunderstands why it was chosen; criticizing a random sample for not capturing lived experience misunderstands what survey data is for. Sampling strategy follows from research question. If you want to know what percentage of Americans support a policy, you need probability sampling. If you want to understand how people who have lost a child to gun violence construct meaning and mobilize politically, snowball sampling into that community may be your only option — and the depth you gain compensates for what you sacrifice in breadth. Good researchers state their sampling strategy, justify it in terms of their question, and are honest about what it allows them to claim and what it does not.
Topics in reflective domains aren't scored by quiz answers. Read, reflect, and mark when you've thought it through.