Date: Fri, 12 Aug 2005 17:41:32 +0200
Reply-To: Frank Thomas <firstname.lastname@example.org>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Frank Thomas <email@example.com>
Subject: Re: weighting with two weight variables
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Jan and Mr N Forcheh,
I intend to do regression analysis for all European countries together.
So, if I understand you well, I should only use the design weight (which
I did in my previous analyses). However, this provides a queer result:
the 80 mill. of Germans will have the same weight as the several
millions of, say, Austrians. This is why there is the population weight.
So, ... ?
Spousta Jan wrote:
>A remark: If weghting with the compound weight, all confidence intervals
>and statistical tests are biased. So use this method only for computing
>the mean scores (e.g. a mean trust of an European respondent in the
>European Parliament). If comparing the countries (ANOVA-like), it is
>appropriate to use the design weight only.
>From: Spousta Jan
>Sent: Friday, August 12, 2005 4:35 PM
>Subject: RE: weighting with two weight variables
>compute wgt = w_design * w_country.
>and weight by the compound weight wgt.
>From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On Behalf Of
>Sent: Friday, August 12, 2005 4:24 PM
>Subject: weighting with two weight variables
>I am doing regression analysis with the European Social Survey data set.
>The data need weighting. There are two weights:
>- a design weight to correct for within-country deviation from the
>- a population weight for correcting for country size when you do an
>analysis for all countries together. Thus, the population weight
>neutralizes the fact that you have a similar sample size in all
>countries. When doing an analysis with all countries together the
>population sizes have to be used to take the different sizes of the
>population into account.
>If doing analysis on the complete data set both weights have to be
>combined. How to do this ? The weight command only allows for a single
>Thanks for your help.
>Dr. Frank Thomas
>FTR Internet Research
Dr. Frank Thomas
FTR Internet Research