Background

Cluster sampling involves sampling units that are groups or clusters, each consisting of one or more subunits. Often, a listing of clusters is available while the complete listing of subunits or observations within clusters is not. Clusters can be sampled, and an enumeration of subunits obtained later for data collection or further subsampling. Even if an enumerated list is available, there could be other constraints on collecting data from units that are selected randomly from among the entire population, and cluster sampling is done instead. (Note that when a listing of all subunits is available, estimates based on a random sample from the entire population are often more precise than those obtained from a cluster sample. This is because of the tendency for units within clusters to be more alike than units between clusters.)

SAS/STAT^® 9.22 in SAS^® 9.2 TS2M3

Beginning with SAS/STAT 9.22 in SAS 9.2 TS2M3, use the SAMPLINGUNIT or CLUSTER statement to name variables that identify the sampling units as groups of observations (clusters).

For example, suppose you have 10 different clusters with one to five people per cluster.

These statements select a simple random sample of three clusters without replacement:

Prior to SAS/STAT 9.22 in SAS 9.2 TS2M3

If a listing of the entire target population is available and you want to carry out a cluster sample, the following shows how PROC SURVEYSELECT can be used in releases prior to SAS 9.2 TS2M3. The steps are to identify the individual clusters, select a random sample of clusters, and then collect all the original observations from each sampled cluster.

Using the same 10 cluster data set above, first identify the individual clusters:

The following statements select a simple random sample without replacement of three of the cluster ID's:

Collect all the observations for each sampled cluster from the original data set to create the final sample:

The IN= data set option creates a new variable that indicates whether the data set contributes to the current observation. Using the MERGE and BY statements above to match-merge the sample of clusters with the original data, and then subsetting using the IF statement causes only those CLUSTERIDs that exist in both the sample of clusters and the original data set to be included in the SAMPLE data set.

Operating System and Release Information