Date: Thu, 28 May 1998 11:11:47 -0500
Reply-To: "Tverdek, Edward" <etverdek@SPSS.COM>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Tverdek, Edward" <etverdek@SPSS.COM>
Subject: Re: how to use table cell info in syntax
Gonzalo,
The CROSSTABS procedure has a WRITE subcommand available in
command syntax which, in combination with the PROCEDURE OUTPUT
command, can write the cell counts to a text file. This file can
be read back into SPSS and merged with your original data through
a table look-up MATCH for further processing.
In your case, however, I think it might be easier to avoid producing
contingency tables altogether and to determine the joint frequency
counts for each combination of age, sex, and region through the
AGGREGATE procedure. If you BREAK on each of these variables and
ask for the number of cases within each break group, SPSS will return
a file with as many cases are there are unique combinations of the
values (i.e. the number of cells you would have across your layered
tables). You can then use MATCH FILES (Data...Merge Files...Add
Variables on the menus), assigning age, sex, and region as your key
variables and, again, treating the aggregated outfile as a TABLE to
assign the "cell counts" back to each case on your original file.
Thus for each case, you'll have a new variable which represents the
count of whichever cell it would fall into had you actually done
Crosstabs. You can then divide this variable by the total number
of cases to generate the weighting variable you describe.
I hope I've understood your question and that these suggestions are
useful,
Ed Tverdek
SPSS Technical Support
>-----Original Message-----
>From: Gonzalo Kmaid [SMTP:gkmaid@INTERNET.COM.UY]
>Sent: Thursday, May 28, 1998 9:59 AM
>To: SPSSX-L@UGA.CC.UGA.EDU
>Subject: how to use table cell info in syntax
>
>Hi All,
>
>I would like to use the particular values obtained in a crosstab table to
>create a new var to be used to weight a particular sample. The weighting
>variable is always defined the same way. So life would be less tedious if
>that can be done automatically. The situation is as follows:
>
>Usually I produce a crosstab between age and sex controlling by region
>(three, two, and two groups respectively).
>
>Then,
>1. I compute the actual proportion of cases in each cell with respect to
>the total sample size (by hand).
>
>2. Elswhere, I have the real value for each cell combination (population
>distribution value for region equal 0, sex equal 0 and age equal 0 and so
>on). Therefore, each factor weight is equal to the population cell value
>divided the sample proportion calculated in 1.
>
>Through a series of if statements I create the weighting variable
>
>if ((age eq 0 and sex eq 1) and region eq 0) peso1=x.xxxx
>if ((age eq 1 and sex eq 0) and region eq 1) peso1=z.zzzz
>etc.
>etc.
>
>Where x.xxx is the result of 2 (same for the rest of statements)
>
>What I would like to do is to tell spss to divide each cell value over the
>sample size, and then use these numbers to calculate each factor weight
>score, and then write them into the series of if statements. Any hint on
>how to acomplish this? Any reference to check? TIA
>
>
>Gonzalo Kmaid
>gkmaid@internet.com.uy
>
>Gonzalo Kmaid
>gkmaid@internet.com.uy
>Carlos Maria Morales 944/801
>Montevideo-URUGUAY
| 
