**Date:** Mon, 12 Sep 2005 09:45:37 -0300
**Reply-To:** Hector Maletta <hmaletta@fibertel.com.ar>
**Sender:** "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
**From:** Hector Maletta <hmaletta@fibertel.com.ar>
**Subject:** Re: cox regression with data weight
**In-Reply-To:** <20050912084741.14743.qmail@web50914.mail.yahoo.com>
**Content-Type:** text/plain; charset="us-ascii"
Simon,
All SPSS procedures can be done with or without weighting. To be precise,
they can only be done WITH weighting, because SPSS always multiplies the
given data by the value of a hidden variable called $WEIGHT, but this
variable is by default set to 1 in all cases. When you issue the command
WEIGHT BY [somevariable], $WEIGHT is given the values of that variable.

SPSS computes significance levels and related values such as CI based on the
sum of weighted cases, or more precisely the sum of weights. By default this
equals the number of cases.

The distortion of significance and CI measures occurs when you use so-called
inflationary weights, i.e. weights that expand the total number of cases to
population size, or more generally, where the sum of weights differs from
the sum of cases. But you may choose a set of weights lacking this effect,
so-called non-inflationary proportional weighting, when each case is
augmented or reduced in weight but the sum of weighted cases is always n,
the original sample size. In this list's archives there are some
contributions of mine dealing in detail with this.

Notice that if your sampling probabilities are not equal for all cases, i.e.
if yours is not a simple random sample, then obtaining your CI or
significance probabilities from unweighted data would also produce distorted
results.

Differential sampling ratios for different cases may arise from two main
features in sample design: stratification and clustering. Using reverse
sampling ratios (N/n) as weights corrects for the effect of stratification
but cannot correct for clustering. Even if you use non-inflationary weights,
your results would be distorted from failing to account for clustering.

However, whenever you need differential weighting you are in the presence of
complex samples, and in that case you should use the Complex Samples module
of SPSS which gives the right estimates. The weighting facility in SPSS was
originally intended only to expand frequencies to population scale, not to
deal with inferential estimates.

Hector

> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU]
> On Behalf Of LAI Man Kin
> Sent: Monday, September 12, 2005 5:48 AM
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: cox regression with data weight
>
> May I ask if there is a way to perform cox regression
> analysis with data weight applying to the dataset in SPSS?
> Besides, my colleague told me the using data weight in SPSS
> may produce erroreous p-value and CI. I am aware of the
> re-base issue but are there other issues I should be aware of
> in using data weight in SPSS? Thank you for your attention.
>
> Simon.
>
>
>
> __________________________________
> Yahoo! Mail - PC Magazine Editors' Choice 2005 http://mail.yahoo.com
>
> __________ Informacisn de NOD32 1.1213 (20050909) __________
>
> Este mensaje ha sido analizado con NOD32 Antivirus System
> http://www.nod32.com
>
>