Overcoverage

Overcoverage occurs in survey sample frames when the frame contains more than enough sample records. This primarily results from two situations. In one case, there are records in the sample frame that do not contain respondents or members of the target population. In other cases, the same respondent is targeted by duplicate or multiple records in the sample frame. In either case, the sample frame contains sample records that should be interviewed.

Different types of overcoverage are commonly referred to as "ineligible units" or "multiple records." Different researchers use the term overcoverage inconsistently, so it is important to consider whether over-coverage in a given sample frame is caused by ineligible units, multiple records, or both.

Sample frames ideally contain a perfect one-to-one correspondence between sample records and members of the target population for a survey. In some cases, multiple records refer back to a single member of the target population. This type of overcoverage is sometimes called a "multiplicity of elements." In other cases, sample records fail to lead to members of the target population. These cases are sometimes referred to as "blanks" or "foreign elements."

Multiple sample records that refer to a single member of the target population are common in sample frames. In cases in which directories or lists are used as sampling frames, respondents can be included more than once if lists are compiled from multiple sources. More commonly, multiple records lead back to a single respondent when the sample frame and target populations are measured (covered) inconsistently. For example, if telephone numbers are sampled for a survey of households, a household with multiple telephones will be included multiple times. If sales records are used as a source for a consumer survey, then consumers who have made multiple purchases might be included in a sample multiple times.

Overcoverage caused by duplicate or multiple records can be adjusted for either by cleaning the sample frame or by providing sample weights to adjust for different probabilities that a respondent is included in the sample frame. Frame cleaning can be accomplished either before or during the survey field process. Cleaning before involves checking the sample frame for duplicate or multiple records and eliminating them. This "de-duping" is a basic part of constructing and [Page 560]checking a sample frame and is made enormously easier and more practicable by increased computer power.

The second type of overcoverage, in which sample records do not contain valid members of the target population, is present in almost all sample frames. This occurs for a variety of reasons. In some cases, sample records do not correspond to anything similar to the target population. For example, telephone samples often contain disconnected telephone numbers or numbers that have not been assigned. Household surveys may send an interviewer to an empty lot. A business directory might contain mailing addresses for establishments that went out of business many years ago. These listings are referred to as "blanks," "empty records," "empty listings," or more colloquially as "bad records" or "duds." In other cases, the sample record reaches a unit that can be screened for eligibility, but the record turns out to not be a member of the target population for the survey. For example, a survey of eligible voters may reach nonvoters, a survey that targets telephone households in one city instead may reach some households in a neighboring town, or a survey of college students may reach some recent graduates. These records are called "foreign elements," "out-of-scope units," or, colloquially, "screen-outs."

...