Date:         Tue, 6 Mar 2001 17:19:27 -0800
Reply-To:     Dale McLerran <dmclerra@MY-DEJA.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Dale McLerran <dmclerra@MY-DEJA.COM>
Subject:      Re: square matrix for crosstabs
Comments: To: smiller@wvmi.org
Content-Type: text/plain

Scott,

It is not possible to force the square matrix with zero values
for all cells in the missing row/column.  However, what I have
done in such circumstances is to add an observation which forces
the matrix to be square.  At the same time, I construct a variable
(call it WEIGHT) with value 1 for every real observation, and
value 1E-10 for every dummy observation.  Now, if you run the
proc freq using the WEIGHT statement to apply your constructed
weight, you can compute KAPPA.  By playing with the weight value
for the dummy observations, you can see the sensitivity of KAPPA
to the WEIGHT value.  Even if you used dummy weights of 0.001,
there would be little change to the value of KAPPA.  At 1E-10,
you have really added essentially nothing to the data, but
tricked SAS into believing that there is a square matrix.

Dale


>Date:         Tue, 6 Mar 2001 11:39:24 -0500
>Reply-To: "Miller, Scott" <smiller@WVMI.ORG>
>From: "Miller, Scott" <smiller@WVMI.ORG>
>Subject:      square matrix for crosstabs
>To: SAS-L@LISTSERV.UGA.EDU
>
>i have run into this delima before and decided to pose this question.  when
>trying to get kappa's for interrater analysis, many times one of the raters
>will have no responses for one level of a variable, while the other rater
>does.  so when doing the table a*b a non-square matrix is produced and none
>of the statistics are generated for the agree option.
>
>a/b     1       2       3
>-----------------------------------
>1       5       10      15
>3       15      10      5
>
>
>is there a way to force a square matrix to be produced?  in essence to
>impose a format that includes all the levels for each variable?
>
>
>a/b     1       2       3
>1       5       10      15
>2       0       0       0
>3       15      10      5
>
>
>thanks
>
>Scott D. Miller, MA
>
>Evaluation and Informatics Analyst
>West Virginia Medical Institute
>(304) 346-9864 ext 2240
>
>Every good scientist should be one part P.T. Barnum and one part B.F.
>Skinner.
>
>The goal of science is to build better mousetraps.  The goal of nature is to
>build better mice.
>
>Sumus quid sumus




---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@fhcrc.org
Ph:  (206) 667-2926
Fax: (206) 667-5977
---------------------------------------

------------------------------------------------------------
--== Sent via Deja.com ==--
http://www.deja.com/