Thank you all for your remarks. The basic dependant variable is recidivism. Do the respondents re-engage in criminal acts within half a year (as self reported) and/or after a year (as reported by police officials). Also: we are not a purely scientific organisation (though obviously we don't want to be unscientific!!!). The results should be scientificaly right but also intuitively understandable. This was the reason to balance the design by weighting .

Thanks Antoon Smulders

----- Original Message ----- From: "Richard Ristow" <wrristow@mindspring.com> Newsgroups: bit.listserv.spssx-l To: <SPSSX-L@LISTSERV.UGA.EDU> Sent: Thursday, February 17, 2005 4:37 AM Subject: Re: weighting

> At 09:04 AM 2/15/2005, Antoon wrote: > >>I have a question about weighting. I work with a file of young >>respondents that have committed minor delinquency. Half of the group >>got a learning and or work penalty, the other half (control group) got >>no (official) punishment. >>Regretfully the groups are not completely comparable as to ethnicity >>(6 categories) and whether the act was committed alone, or with others >>(4 categories). I would like to create weights to adjust for these >>differences. > > That is, I take it, you want to assign weights such that, weighted, > this is a balanced design. > >>I have two questions: >>1 Is my reasoning conceptually valid? > > I think it's doubtful. There are better mathematical statisticians here > than I am (Marta? Hector? you there?), but my understanding is that you > do NOT try to correct an unbalanced design by weighting. > > Weighting, when it goes beyond treating one observation as several when > it really represents several with the same observed values, is tricky, > and I don't pretend to understand the subtleties. The legitimate use, I > believe, is to match a stratified sample to the population > distribution. > > You may want to post more about the analysis you intend: What dependent > variable or variables are you observing? Continuous, so this is an > ANOVA problem, or what? >