Home Statistical Dictionary Domain Estimation
Domain Estimation
The estimation of a population quantity for a subpopulation (domain), such as a demographic group or geographic region, defined within a survey whose overall sample design was not necessarily planned around that specific subgroup.
In Plain English
Domain estimation is about producing estimates for a particular slice of the population, say, women aged 25-34, or residents of a specific state, using data from a survey whose overall sample wasn't necessarily built with that specific slice in mind. Because the survey's stratification and sample-size targets were set for the full population (or for different planned subgroups), a domain defined after the fact might end up with a smaller, less carefully controlled sample size than a subgroup the survey was actually designed around, which is exactly why domain-specific estimates are often noticeably less precise than the survey's headline, national-level figures.
Definition
Domain estimation is the estimation of population quantities, means, totals, proportions, for a domain (a subpopulation defined by demographic, geographic, or other characteristics) using data collected under a survey design whose sample allocation was not necessarily planned specifically around that domain; domains are distinguished from the strata of a stratified design, strata are explicit, design-time divisions with controlled sample allocations, while domains can be defined after the fact, cutting across strata in ways the original design did not anticipate, and their effective sample sizes are simply whatever the design happened to produce for that particular subgroup. Standard design-based estimators (Horvitz-Thompson, ratio, or regression estimators) remain valid for domain estimation, applied to the subset of sample units belonging to the domain, but because domain sample sizes are typically smaller, and sometimes much smaller, than the survey's overall sample size, domain estimates generally carry higher variance than headline population-level estimates, and very small domains may require small area estimation techniques instead of straightforward direct domain estimation.
Properties
- Domain sample sizes are a consequence of the overall design, not a design choice in themselves, unless a domain happens to coincide with an explicit stratum, its sample size falls out of however the original stratification and sampling allocation happened to distribute units across that particular subgroup, which is why some domains end up well-represented and others thin, purely by the luck of the underlying design.
- Survey reports routinely flag domain-level estimates with especially small sample sizes as unreliable or suppress them entirely, a common and important practical convention, since a domain estimate built on a handful of sample observations can carry enormous, sometimes uselessly large, sampling variance even when the overall survey is well powered.
- When a domain's direct sample size is too small to support acceptable precision even with standard design-based estimators, the domain becomes a candidate for small area estimation methods instead, which borrow strength from related domains or auxiliary models rather than relying solely on the domain's own thin direct sample.
At a Glance