Date:         Thu, 26 Feb 2009 14:08:15 -0200
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: comparing weighted data to unweighted data
Comments: To: J P <jp7837@yahoo.com>
In-Reply-To:  <47927.62340.qm@web59902.mail.ac4.yahoo.com>
Content-Type: multipart/alternative;

Several consequences. First, if the weights were “inflationary”, i.e. they
amplify sample size to the size of the population, SPSS would consider the
expanded size as the size of the sample, and compute significance
accordingly (thus overstating the significance of results). Besides, if
different cases have different weights (i.e. weights are correcting
proportions in the sample to proportions in the population), using the
unweighted set gives some cases more weight than they deserve, and other
cases less, again distorting the results. 

I do not understand your remark about weights being no Langer achúrate
because the data have been subsetted. If you have N cases, each with a
weight, and you take one subset of M cases (M<N), you can still use the
weights. For instance, súpose you extract a sample of 1000 from a
population, with a complex sampling design mandating that each case has a
specific weight in the sample; your weighted sample would yield results
representative of the referente population. Then suppose you select one
subset consisting of about one half of the samle (say, just all the males):
if you keep the weights, your subset would represent the characteristics of
the male population, just as the entire sample represented the entire
population of both sexes.

 

Hector.

 

  _____  

From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On Behalf Of J P
Sent: 26 February 2009 13:46
To: SPSSX-L@LISTSERV.UGA.EDU
Subject: comparing weighted data to unweighted data

 

Dear List,

 

I appreciate your patience with this non-SPSS question.

 

From a practical standpoint, what is the consequence of conducting a
statistical test comparing means or frequencies when one data set is
weighted and the other is not (data has been subsetted so weights are no
longer accurate). 

 

Thank you,

John  

 


[text/html]