Home Statistical Dictionary Sampling Unit
Sampling Unit
The individual element or group of elements that is directly selected at a given stage of a sampling design -- may be the ultimate unit of analysis (a person) or an intermediate unit (a household, school, or cluster) in a multistage design.
In Plain English
The sampling unit is whatever the sampling design actually picks at a given step, which isn't always the same thing you're ultimately measuring. In a simple one-stage survey of individuals, the sampling unit and the unit of analysis are the same person. But in a multistage design, the sampling unit at the first stage might be an entire school district, at the second stage an individual school, and only at the final stage an individual student, each of those is a sampling unit at its own stage, even though only the student is the unit whose test scores you're actually analyzing.
Definition
A sampling unit is the entity that is directly and explicitly selected at a particular stage of a sampling design; in single-stage designs the sampling unit coincides with the observation or analysis unit, but in multistage designs each stage has its own sampling unit -- the primary sampling unit (PSU) at the first stage, secondary sampling units nested within selected PSUs at the second stage, and so on down to the ultimate sampling unit, the smallest unit actually observed and measured. Distinguishing the sampling unit from the population element (the entity a study's substantive conclusions are about) matters because selection probabilities, and therefore the appropriate weighting and variance-estimation methods, are defined stage by stage in terms of the sampling unit at each level, not directly in terms of the final population element.
Properties
- The choice of sampling unit at each stage of a design directly determines the design effect, larger, more heterogeneous sampling units (like entire cities as PSUs) generally preserve more precision than smaller, more homogeneous ones (like individual city blocks), because within-unit homogeneity is what drives the intraclass correlation ρ in the design effect formula.
- Distinguishing sampling units from population elements matters for variance estimation, techniques like Taylor series linearization, the jackknife, or balanced repeated replication all operate at the level of the primary sampling unit, since the PSU, not the individual final observation, is what carries the design's actual randomization.
- A cluster and a sampling unit are related but not identical concepts, a cluster refers to the natural or administrative grouping used in the design (a school, a neighborhood), while sampling unit refers more specifically to whichever level of that grouping is directly selected at a given stage of the sampling process.
At a Glance