|Date: ||Thu, 19 Feb 2004 15:07:08 -0500|
|Reply-To: ||Michael Whitcomb <mwhitcomb@WESLEYAN.EDU>|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Michael Whitcomb <mwhitcomb@WESLEYAN.EDU>|
|Subject: ||Re: weighting survey results|
|Content-Type: ||text/plain; charset="us-ascii"; format=flowed|
I'm no expert on this, but here's a simple approach:
You had 1,025 respondents.
You wanted 42.5% of them to be FT (N=589), but you got 59.7% FT, N= 612 .So
the weight for FT students in you sample is 589/612, or 0.9624
You wanted 57.5% of hem to be PT (N=436) ,but you got 40.3% FT, N= 413 . So
the weight for PT students in you sample is 436/413, or 1.0557
if time = "full' then casewt = 0.9624;
if time = "part' then casewt = 1.0557
then in your PROCs, include the weight statement:
Check your N's - in some PROCs N's get adjusted (FREQ?) and in others they
don't (MEANS?). Not sure about TABULATE.
Hope this is at least a start
As for the lay audience - "Data were weighted so respondent's (enrollment
status) characteristics would mirror that of our student bod." (?)
I'd be interested to hear what others think.
At 02:18 PM 2/19/2004, Pat Moore wrote:
>There is a SAS question further down in this missive, but I need to lay on
>some background first.
>My dataset consists of the responses of 1,025 students to a set of 116
>questions. I'm using proc tabulate to generate the mean responses of the
>subjects to the 100 non-demographic questions.
>I'm also going to have to explain the weighting process to a lay audience
>so they can judge if it is a valid way to look at the findings.
>As always, I appreciate the list and the wonderful help you all give me.
Assistant Director of Institutional Research
237 High Street, 305 North College
Middletown, CT 06459-0280
Fax - 860.685.2011