FRAME

A frame is used to identify elements in the population. Elements are the fundamental unit of observation in the survey. A frame may look very different depending on how the population of interest is defined and how its elements are defined. A well-defined appropriate frame is essential to the sampling process, the development of weights for use in analyses of survey data, the minimization of coverage error, and the understanding of what coverage error may exist. This entry describes the basic concept of a frame; the impact it has on sampling, weighting, and coverage; and how it is developed in relation to the survey population and the survey sample. It also discusses several commonly used frames and their specific issues.

A major goal of most surveys is to describe a specific population. For example, the U.S. government conducts two surveys specifically to estimate the rate of unemployment in the country each month: the Current Employment Statistics program (a survey of business establishments) and the Current Population Survey (CPS; a survey of people). Each month, the U.S. Census Bureau interviews a sample of people for the CPS. However, selecting that sample is difficult as there is no accurate, up-to-date list of people in the United States with contact information. Without such materials, it is difficult to draw a sample. But the U.S. Census Bureau can construct a frame of housing units in the country using various sources (the decennial census, building permits, etc.). Therefore, the U.S. Census Bureau defines the survey population as people living in housing units. This revised definition of the survey population is important because it allows for a better frame to be constructed. Of course, a disadvantage is that the homeless are not included, but this is judged to be acceptable to meet the goal of this survey.

Among the domains of research that use statistical techniques, survey research is unique in assigning so much importance to the source of sample units. Whereas most statisticians view sample units as a way to describe a process of interest, survey statisticians view sample units as a way to describe a population of interest. Other statisticians would only be interested in [Page 291]elements that are missing from the frame if the missing elements were informative of the process under study. For example, in sampling to study the effect of a certain drug, if the sample had no women, this would be a concern only if women reacted differently from men to the drug under study. The survey statistician may not be interested in whether women reacted differently to the drug but would want women proportionally represented in the sample frame so their role in the population could be described.