Home Statistical Dictionary Coverage Error
Coverage Error
The error introduced when a sampling frame does not perfectly match the target population, either by omitting eligible units (undercoverage) or including ineligible or duplicate ones (overcoverage).
In Plain English
Coverage error is what happens when the list you're sampling from doesn't quite match the group you actually want to study. If your sampling frame is a list of landline phone numbers but your target population is all adults, you've got undercoverage, plenty of adults who only use cell phones or have no phone at all are simply invisible to your frame, no matter how carefully you randomize within it. Coverage error is fundamentally a frame problem, not a sample-size problem, and it's one of the reasons survey researchers spend so much effort trying to build and maintain frames, like address-based sampling, that track the target population as closely as possible.
Definition
Coverage error arises from a mismatch between the sampling frame and the target population, undercoverage occurs when eligible population units are entirely absent from the frame, giving them zero probability of selection regardless of the sampling design used within the frame; overcoverage occurs when the frame includes ineligible units, duplicate entries, or units that no longer exist. Coverage error is distinct from, and not correctable by, increasing sample size, since units missing from the frame remain missing no matter how large a sample is drawn from what the frame does contain; it is instead addressed at the design stage, by improving the frame itself (for example, address-based sampling to reduce landline-only undercoverage) or after data collection, through post-stratification or raking adjustments that reweight the achieved sample to match known population totals on key characteristics, though such reweighting can only correct for coverage gaps correlated with the variables used in the adjustment, not for gaps on unmeasured dimensions.
Properties
- Coverage error is a design-time, frame-level problem, not a sampling problem, this is the key conceptual distinction from sampling error, no amount of randomization within an incomplete frame can recover the units that frame never included in the first place.
- The classic modern example is telephone survey undercoverage as landline-only sampling frames became increasingly unrepresentative of adults with only cell phones, survey organizations had to fundamentally redesign their frames (dual-frame landline-and-cell designs, then eventually address-based sampling) to keep pace with population coverage.
- Post-survey adjustments like post-stratification and raking can partially compensate for known coverage gaps by reweighting the achieved sample to match population totals on variables like age, sex, and region, but they cannot correct for coverage error on unmeasured or unmodeled characteristics that happen to be correlated with the outcome of interest.
At a Glance