Sampling

Before sampling begins, primary sampling units (PSUs) are defined and grouped into strata. PSUs consist of one or more contiguous counties. The largest PSUs, as of the 1990 census, are designated as self-representing PSUs. The remaining PSUs are grouped to form nonself-representing strata of approximately equal size within the state using a clustering algorithm. During the first stage of sample selection, one PSU is selected from each nonself-representing strata based on population size. There are currently 428 self-representing and 326 nonself-representing PSUs in the CPS sample.

Within a selected PSU, multiple sampling frames are used in the next stage of selection. The determination of which frames should be used is based on the the type of enumeration during the last decennial census and the availability of housing permits. If complete addresses are available for a particular block from the decennial address files and the block is in an area requiring building permits, the sample is selected from the decennial files and from a frame of addresses created by sampling building permits issued for post-April 1990 construction. Otherwise, a sample of areas (one or more contiguous blocks) is selected; a field representative of the Census Bureau lists all the housing units in the selected areas; and a sample is selected from the listed units.

At this stage of sampling, ultimate sampling units (USUs) are selected using systematic sampling. USUs are clusters of four housing units. The clusters from the decennial address lists are usually geographically compact. The clusters from the area frame, while in expectation should consist of four units, may contain more or less if there has been major building or demolition within the area since the census. (Group quarters are sampled separately from housing units on the decennial census list.)

Multiple samples are selected all at once, each sample being a 'neighbor' or geographically close to the previous sample. Each sample is divided into 8 subsamples (referred to as panels or rotation groups.) Each month a new rotation group is introduced for its first month of interviewing; the rotation group which had its eighth interview in the preceding month is retired permanently. Thus, over the decade until the results of the next census are available, the specific sample of units changes, but their general characteristics do not. Each rotation group is interviewed for 4 consecutive months, out of sample for the next 8 months, and then interviewed for 4 months. This 4-8-4 rotation pattern implies a 75 percent overlap month-to-month and a 50 percent overlap year-to-year.

The additional Hispanic sample is from the previous November's basic CPS sample. If a person is identified as being of Hispanic origin from the November interview and is still residing at the same address in March, that housing unit is eligible for the March survey. This amounts to a near doubling of the Hispanic sample since there is no overlap of housing units between the basic CPS samples in November and March.

Annual Demographic Survey (March 1996 CPS) Sampling and Weighting Page

CPS Main Page