LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous messageNext messagePrevious in topicNext in topicPrevious by same authorNext by same authorPrevious page (September 2005)Back to main SPSSX-L pageJoin or leave SPSSX-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Thu, 29 Sep 2005 18:02:51 -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: Weighted Correlation
Comments: To: Siraj Ur-rehman <Siraj.Ur-rehman@Ipsos-ASI.com>
In-Reply-To:  <5101F2BA382BB245A7033EC1E5B8E35501C7B4B7@namail2.na.ipsos>
Content-Type: text/plain; charset="us-ascii"

1. You are not doing right. In my opinion your procedure makes no sense. Correlation should not get greater or smaller depending on the number of cases. What may increase with the number of cases is their statistical significance: a coefficient of 0.30 may not be significant with a few cases, but significant with more cases. However, even this would be improper in your cases, because for variables other than X1 you did actually have fewer cases, and just telling the computer to pretend it has a larger number of cases would not increase the intrinsic significance of your correlations, though it will make them APPEAR to be more significant to uncautious readers.

2. There MIGHT be a way out if you could ESTIMATE the value of other variables (other than X1) based on information actually given by those subjects. The SPSS MISSING VALUES module (and other softwares) allows you to estimate the expected value of, say, X2 for a subject lacking that information, based on a regression of X2 on other variables where that subject gave valid responses. Estimating is one thing and observing is another. However, these estimated or imputed values for the other variables would give you some indication of the correlation of X2 with X1 including all cases for whom the estimation can be carried out. Advice: if these estimated cases are many, abstain. Do it only if they are few, and only if the regression estimates have a small margin of error (i.e. very high significance).

3. Your subject line reads "Weighted correlation", but that's a misnomer: what you proposed to do is not weighting, but rigging the correlation to make them appear larger.

Hector

> -----Original Message----- > From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] > On Behalf Of Siraj Ur-rehman > Sent: Thursday, September 29, 2005 5:46 PM > To: SPSSX-L@LISTSERV.UGA.EDU > Subject: Weighted Correlation > > Hi Guys > I have ten variables and except the first variable (X1) all > have so many missing observations. I calculated the > correlations between X1 and rest of the nine variables. So I > have nine correlation values with nine different bases > depending on X2 to X10. I want to calculate the correlation > values with the base of X1. I tried to do as > > (Correlation of X1 and X2)*(# of non missing values in X1 AND > X2)/(# of values in X1) (Correlation of X1 and X3)*(# of non > missing values in X1 AND X3)/(# of values in X1) > > And so on. My question am I doing right or there is another > way? Thanks Siraj > > __________ Informacisn de NOD32 1.1237 (20050929) __________ > > Este mensaje ha sido analizado con NOD32 Antivirus System > http://www.nod32.com > >


Back to: Top of message | Previous page | Main SPSSX-L page