Date: Sun, 8 Apr 2007 21:47:17 -0700
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: Correlation with adjustments
In-Reply-To: <266757.76898.qm@web31414.mail.mud.yahoo.com>
Content-Type: text/plain; format=flowed
jamiern1us@yahoo.com wote to me personally, instead of to SAS-L as I
intended:
>
>David L Cassell <davidlcassell@MSN.COM> wrote: jamiern1us@YAHOO.COM wrote:
> >
> >I am in interested computing the correlation coefficients of scores on
> >components in a scale (pf01, pf02, pf03) to the summary score for the
>scale
> >(PF). For each correlation, I need to calculate the summary score
>without
> >the ith component and correlate those values with the score on the ith
> >component (ex. PF-pfo1 correlated with pf01). Is there a command I can
>add
> >to the command below for SAS to iteratively do this? If not, is there a
> >simple way to create an adjusted scale score for each component of that
> >scale?
> >
> > proc corr data=physical rank ;
> > var PF;
> > with pf01 pf02 pf03;
> > run;
> >
> > Thank you in advance for any asistance.
> > Jamie
> >
>
>I see that you have already received some very sound advice. But I'm not
>sure that either actually answers your question, since both require some
>assumptions about your data and your real goals.
>
>Could you write back to SAS-L and explain:
>1) just how did you build these scores,
>2) what are the values you built them from,
>3) are the scores categorical, continuous, ordinal, ???
>4) is the summary score really the exact sum of the other variables,
>5) are you really looking to compare PF03 with (PF - PF03),
>6) and why do you want this?
>
>HTH,
>David
>--
>David L. Cassell
>mathematical statistician
>Design Pathways
>3115 NW Norwood Pl.
>Corvallis OR 97330
>
>_
>1) just how did you build these scores,
> I am not sure how to answer this question other than how I have below.
>
> 2) what are the values you built them from,
>The initial values of each answer were 1-5
>
> 3) are the scores categorical, continuous, ordinal, ???
>The scores are ordinal
> 4) is the summary score really the exact sum of the other variables,
> The summary score is the exact sum normalized (0-100) through an
>equation determined by the developers of the survey
> 5) are you really looking to compare PF03 with (PF - PF03),
> I am interested in the correlation of pf03 to the PF scale but
>corrected for overlap. A correlation of pf03 with PF will naturally look
>stronger than a correlation of pf03 with (PF-pf03). Once I obtain that
>correlation I can compare that correlation to the correlations of pf03 to
>other scales used in the survey.
> 6) and why do you want this?
> I am interested in the convergent and divergent validity of my scores
>(pf03 is more closely related to the PF scale than it is to the RF scale)
>before I compare my scores to the norm.
>
>Thank you,
>Jamie
Okay, it does make sense to factor values out of your overall PF before
making comparisons. But you may not be able to if the PF value is
built using some 'secret formula'.
Most likely, it is: a + (sum of PFxx)*b
You can check that by running a quick regression (PROC REG) of the sum of
the PFxx as your X and the final PF as your Y. If I'm right, you'll get the
values for A and B, and you'll have a correlation of roughly 100%. If they
are really scaling the sum to a 0-100 interval, then A and B should be
simple.
(If you have 3 Likert scales going from 1 to 5, then their sum can only
be in the range 3 to 15, so the secret formula would look like
(sum - 3)*(100/12)
and you can then correct by subtracting off (100/12)*PF03 ).
HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330
_________________________________________________________________
MSN is giving away a trip to Vegas to see Elton John. Enter to win today.
http://msnconcertcontest.com?icid-nceltontagline
| 
