topic Re: Random Number Generator in Statistical Procedures

Random Number Generator

ebmancha — Fri, 21 Nov 2014 20:07:47 GMT

Hello,

The change in random number generators between 9.2 and 9.3 is giving me some problems. Specifically, I need to replicate the random number generation process in 9.2 (using ranuni), while using the 9.3 software. I found a link a while back that claimed to have code to do this, but it became inactive before I could test it. Does anyone have experience with this? Specifically, I need to be able to validate that samples generated using 9.2 were done correctly, while having access to only the 9.3 software.

Thanks!

Re: Random Number Generator

Reeza — Fri, 21 Nov 2014 20:29:04 GMT

What does your SAS 9.2 code look like? Did you use the streamint function in 9.2 as well?

Re: Random Number Generator

ebmancha — Fri, 21 Nov 2014 20:33:34 GMT

just your standard ranuni:

proc surveyselect data = frame

out = pj.sample samplesize = 80 seed = 20047

run;

or, ranuni within a datastep.

Emma

Re: Random Number Generator

ebmancha — Fri, 21 Nov 2014 20:34:46 GMT

and no, never used teh streamint function.

Re: Random Number Generator

Rick_SAS — Mon, 24 Nov 2014 18:46:31 GMT

Back in the 9.2 days PROC SURVEYSELECT used the RANUNI algorithm, as documented here:

SAS/STAT(R) 9.2 User's Guide, Second Edition

In your example, the PROC uses the default sampling method, which is SRS (simple random samples without replacement). The doc says that "PROC SURVEYSELECT uses Floyd’s ordered hash table algorithm for simple random sampling.... See Bentley and

Floyd (1987) and Bentley and Knuth (1986) for details." So I'd look up the algorithm and use the RANUNI function in the SAS 9.3 DATA step to try to recreate the 9.2 behavior.