Date: Mon, 18 Jun 2001 11:49:04 -0500
Reply-To: "Pietrzak, Dale" <dpietrza@USD.EDU>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Pietrzak, Dale" <dpietrza@USD.EDU>
Subject: Re: RELIABILITY ANALYSIS
Content-Type: text/plain; charset="iso-8859-1"
It means what you indicate. This means that people did not responding in a
consistenct manner to the 10 items. Thus if thy answered true to one, there
is little relationship to this and whether they ansered true or fale to any
other item on the intventory. As the scale construciton process you are
using carries the assumption of indvidual differences (that each person has
the target trait to some degree) this does not bode well for the scale in
its present form. Assuming the scale is intended to measure a single
something, this suggests several possible things (others are possible as
well but it is early here). One is that you have a restricted range in your
sample. Nearly all the subjects coming from the low end (don't express much
of the target condition) would cause this for example. Second, the scale is
to "hard," meaning the item difficulty is such that it takes an extreemly
high amount of the target condition (compared to what your sample has) for
it to be captured by the scale (trigger an item to be responded to in such a
manner as to indicate the condition). Third, while internal consistency is
not an indicator of multidimentionality as such, it is often lower in scales
that are multidimentional. Thus your scale is really two or more scales
which are tapping different constructs in some way.
In all cases unless you are able to get the reliability higher the utility
of the scale would be questionable in most circles with which I am familiar.
The minimum lower limit for reliability being .70 (Nunnlley & Bernstein,
1994, Psychometric Theory 3rd Ed; Janda, 1999, Psychological Testing; and
similar) for most applied purposes. And, before someone asks, no one form of
reliability (stability for example) does not make up for this difficulty.
If it were me, and I wanted to know what was happening, I would perform an
item analysis (unless my sample is large a classical item analysis) to check
item difficulties. Second, I would test for multidimentionality, at minimum
running some sort of factor analysis to check this out.
Hope that was of some help.
From: c dommeyer [mailto:vcmkt001@CSUN.EDU]
Sent: Sunday, June 17, 2001 7:34 PM
Subject: RELIABILITY ANALYSIS