Home Statistical Dictionary Non-Probability Sampling
Non-Probability Sampling
Any sampling design in which units are selected by a mechanism that does not assign every population unit a known probability of selection, so formal design-based sampling error cannot be rigorously calculated from the design alone.
In Plain English
Non-probability sampling covers every method that doesn't guarantee each person a known, calculable chance of being included, convenience sampling (whoever's easiest to reach), quota sampling (fill target numbers per group without randomizing within them), snowball sampling (ask each participant to refer others), and so on. These methods are often faster, cheaper, or the only realistic option for hard-to-reach populations, but they come with a real statistical cost: you can't legitimately compute a margin of error from the design itself, since there's no known selection probability to base it on, any 'error bars' reported alongside a non-probability sample rely on model-based assumptions rather than the sampling design.
Definition
Non-probability sampling encompasses any selection mechanism, convenience sampling, judgment (purposive) sampling, quota sampling, snowball sampling, and respondent-driven sampling among them, in which population units' inclusion probabilities are unknown, undefined, or not controlled by the design, precluding the direct application of design-based inference. Valid statistical inference from a non-probability sample, when attempted at all, must instead rely on model-based assumptions, for example assuming the sample is representative conditional on observed covariates, and adjusting via weighting or calibration against known population totals, but such inference carries the added risk that the model's assumptions, rather than the physical randomization of a probability design, are what's actually doing the work, and those assumptions are inherently unverifiable from the sample data alone.
Properties
- The absence of known selection probabilities is a fundamental, not merely practical, limitation, no amount of post-hoc statistical adjustment can fully substitute for the physical randomization that a probability design provides, which is why non-probability samples are generally considered a weaker evidentiary basis for population-level claims.
- Non-probability sampling is often the only practical choice for populations without any usable sampling frame, hidden or hard-to-reach groups like people experiencing homelessness, undocumented populations, or members of stigmatized communities are frequently studied via non-probability methods like respondent-driven sampling precisely because no complete list of them could ever be constructed.
- Modern survey practice increasingly blends the two categories, some large-scale opt-in web panels use non-probability recruitment but then apply sophisticated statistical calibration against known population benchmarks (age, region, education) to try to approximate the properties of a probability sample, with mixed and actively debated success in the survey methodology literature.
At a Glance