ds_sample

Perform the sampling of a given dataset following PROC SURVEYSELECT method.

%ds_sample(idsn, odsn, sampsize=1, method=SRS, nreps=1, var=, seed=, strata= , ilib=WORK, olib=WORK, debug=no);

Arguments

• idsn : a dataset;
• sampsize, method, nreps, seed, strata : (option) arguments of the PROC SURVEYSELECT; default: 1, SRS (i.e. simple random sampling), 1 (no repetition), '' (not specified/used) and '' respectively;
• var : (option) list of (unquoted and blank-separated) strings that store the name of the variables/fields (which must exist in the dataset) to be returned in odsn;
• ilib : (option) name of the input library; by default: empty, i.e. WORK is used.

Returns

• odsn : name of the output table where the sampled data (of size sampsize, see above) will be stored;
• olib : (option) name of the output library; by default: empty, and the value of ilib is used.

Example

Let us consider the table _dstest31 as follows:

geo	value	unit
BE	0	EUR
AT	0.1	EUR
BG	0.2	NAC
LU	0.3	EUR
FR	0.4	NAC
IT	0.5	EUR

we then shall run the sampling of geo and value fields only:

%ds_sample(_dstest31, dsn, sampsize=2, var=geo unit, method=SRS, nreps=1);

which returns (seed not set) into the dataset dsn the following table:

geo	unit
BG	NAC
FR	NAC

Run macro %_example_ds_sample for more examples.

Notes

1. In short, this macro runs, in case strata is not passed:

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

or in case strata is specified:

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

with the parameters defined above. Check the online documentation of the PROC SURVEYSELECT procedure for more details.

1. No consideration on the SIZE (sampling unit size measure) statement is made, which implicitly means that you cannot perform unequal probability sampling with this macro.
2. For SRS and URS methods (simple sampling with or without replacement), an alternative algorithm is available so as to produce the same output whatever the machine used (see note 4 below). These algorithms have been 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).
3. This macro runs on different machine, but with the same seed on the same dataset will produce the exact same samples. Please note however that the current version of SAS (9.2 TS Level 2M3) on Solaris machines does not ensure the same output as SAS installed on a Windows machine for the PROC SURVEYSELECT. Therefore alternative macros have been implemented for SRS and URS methods; the SYS method should also be implemented in a later version.

Reference

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

See also

%var_check, %ds_count, %ds_delete, SURVEYSELECT.