proc_surveyselect

Alternative sampling of a given dataset with same purpose as PROC SURVEYSELECT but a machine independent implementation.

%proc_surveyselect(idsn, odsn, method, sampsize=, rep=, strata=, seed=0, ilib=WORK, olib=WORK);

Arguments

• idsn : a reference input dataset;
• method : sampling method; only SRS (i.e. simple random sampling) and URS (i.e. unrestricted random sampling, with replacement) are currently implemented;
• sampsize, rep, seed, strata : (option) same arguments as in PROC SURVEYSELECT; default: 1, 1 (no repetition), 0 and '' respectively;
• ilib : (option) name of the input library; by default: empty, i.e. WORK is used;
• olib : (option) name of the output library; by default: empty, and the value of ilib is used.

Returns

odsn : name of the output table where the sampled data (of size sampsize, see above) will be stored.

Examples

Run macro %_example_proc_surveyselect for examples.

Notes

1. The current version of SAS (9.2 TS Level 2M3) on Solaris machines does not ensure the same output as SAS 9.3 and upper for the PROC SURVEYSELECT. The issue has been documented here. This macro provides with an alternative approach/algorithm that circumvents this problem for SRS and URS methods. In practice it enables the user to produce the same output whatever the SAS version used (see note 4 below). When running on different SAS versions, the sampled outputs generated from identical datasets with the same seed will also be identical.
2. In short, this is somehow equivalent to:

PROC SURVEYSELECT DATA=&ilib..&dsn OUT=&olib..&odsn 
    METHOD = &method rep = &rep 
    SAMPSIZE = &sampsize
    SEED = &seed
    ID &var;
    STRATA &strata ;
run;

with the parameters defined above.

1. Following the implementation of PROC SURVEYSELECT, in case of stratified sampling, the algorithm will select by default 1 unit in each stratum. In case sampsize is set to one given number *n*, the algorithm will select *n* units in each stratum. Finally, in case sampsize designates the name of a dataset, this dataset is supposed to provide the stratum sample size in a variable named _NSIZE_.
2. The SYS method should also be implemented in a later version.
3. This approach was implemented by P.BBES.Lamarche (pierr.nosp@m.e.la.nosp@m.march.nosp@m.e@ec.nosp@m..euro.nosp@m.pa.e.nosp@m.u).

Reference

Fan, C.T., Muller, M.E., and Rezucha, I. (1962): "Development of sampling plans by using sequential (item by item) selection techniques and digital computers", JASA, 57(298):387-402, DOI: 10.2307/2281647.

See also

%ds_sample, SURVEYSELECT.