Subject: Version of Case Control Program

Here's a somewhat novel approach which *some* may find interesting. <g>

Crude outline:

1.) Segregate male, female datasets as necessary.

2.) Make a "short pass" thru the female dataset. Create a var (catcount) for the # of obs/category. Create a var (rec) which points to the last rec/category in female dataset. Create an index on categories.

3.) To select the matched sample, get a male rec, use the male category (indexed access) to get the female metadata, use such to randomize the pointer to the actual female data within the appropriate categories.

Tested W2k 8.1:

*** test data: ***; data male (drop = i) female; do agecat = 20 to 55 by 5; *** <--- age category ***; do inccat = 20 to 100 by 20;** <--- income category ***; do i = 1 to 5; output female; end; if ranuni (6) <.4 then output male; end; end; run;

*** female metadata(sounds sexist, don't it?):-) ***; data femalept (index = (ageinc = (agecat inccat) ) ); do catcount = 1 by 1 until (last.inccat); set female (keep = agecat inccat); by agecat inccat; rec+1; end; run;

*** matched sample: ***; data match nomatch; set male; set femalept key = ageinc; If _iorc_ then output nomatch; else do; pt = rec - floor (catcount * ranuni(8)); set female point = pt; output match; end; run;

This approach _might_ present some efficiencies if the data were voluminous ...

Na Zdrowie, Puddin'

****************************************************** *** Puddin' Man *** Pudding_Man@lycos.com ******** ******************************************************;

-----Original Message----- From: John Gerstle [mailto:JGerstle@SW.UA.EDU] Sent: Tuesday, February 26, 2002 2:48 PM To: SAS-L@LISTSERV.UGA.EDU Subject: Version of Case Control Program

SAS-Lers, I'm looking for some sample Case Control code. Presently, I have a dataset where I would like to match the 20 males with 20 females that match them along some parameters out of a total of 200 females. In other words, I would like to run some simple analyses based on gender, but only have 20 males versus 200 females. And let's say, that the parameters I would like to use are age (in five year increments) and income level (in $20K increments). So for every male, there is some subset of females that match in terms of age level and income level, and I would like to randomly select one of these females to 'pair' with the males. In this way, the two genders will be matched on age and income. I have been successful in manually creating matches (printing out the data, highlighting matches, hard coding ids to pull, etc...). But I would like to have something that would generalize over to a larger dataset with many more levels (say, 100 males and 2000 females). Does anyone have something that would do this? (I haven't checked SCONSIG or the SAS site yet...thought I'd question the List first in case others out there might need something similar.)

Thanks, John

John Gerstle, M.S. Program Analyst, Sr. Applied Gerontology Program University of Alabama

--- ****************************************************** *** Puddin' Man *** Pudding_Man@lycos.com ******** ******************************************************;

2,000,000,000 Web Pages--you only need 1. Save time with My Lycos. http://my.lycos.com