Date: Sun, 11 Sep 2005 15:00:33 -0400
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Art Kendall <Art@DrKendall.org>
Organization: Social Research Consultants
Subject: Re: improving reliability
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
see interspersed comments
gokhan kaya wrote:
>Thanks a lot
>These itemms were selected as a theoritical reasons
What language were the items administered in?
What language were the items originally developed in?
>tehse items under the same theotical concept(actually we discussed
>the method of construction as before )
>There are seven items in each scale, they show similar behaviżor..
If they showed similar behavior they would have small to moderate
positive iteritem correlations and therefore a high Cronbach's alpha.
Or do you intend to say that they are intended to be measures of the
>Yo are right, but regrading with
>if item deleted values cause number of items as a
Are you sure you are putting the items in the correct sets (i.e.,
scales)? Did you have splitters in the factor analysis?
> three. Also, I do not know whether number of items
> is larger than 4 or more alpha value is important?
If the items are intended to be measures of the same construct and they
behave similarly (as above) then this will be shown as a sufficiently
high alpha. Sometimes it is necessary to live with alphas as low as .70
but that should be rare if the scale is well constructed.
We seem to be having a communication problem. Where are you located?
It is often easier to overcome language barriers face-to-face. Perhaps
someone on this list could meet with you.
Social Research Consultants
University Park, MD USA Inside the Washington, DC beltway.
> Art Kendall <Art@DrKendall.org> wrote:
> Were the items designed to measure particular constructs?
> What are they? How many items did you have for each potential scale?
> How did you get the items? Did you use pre-existing scales?
> starting with all of the items, do a PA2 factor analysis, see if it
> suggests that the number of factors to extract is about the number
> you intended to have scales for.
> Be sure that items are reflected as necessary to to assure that a high
> item score indicates the direction of a high scale score.
> Then do reliability with the set of items that are indicated to go
> Do you have at least 4 and preferably more items on each proposed
> Social Research Consultants
> University Park, MD USA
> (301) 864-5570
> gokhan kaya wrote:
> >Dear All
> >To imprrove reliability (Conchran alpha) what should I do?
> >I started with removing some items becuse of larger values in the
> column of alpha if item is deleted
> >and use factor analysiss to reduce data
> >Finally, these results make me to remove some items but this
> efforts still not neough since conchran alpha is increased from
> 0.32 to 0.47 but it is under the level of 0.6.8 (common use)
> >and I've also another problem:
> >My questionnaire is applied in two similar qualified(similar
> socio economic status) provinces but one of them reveal much
> smaller alpha values than another one for 3 scales.
> >Are there alternative way to improve alpha values?
> >What is the source of this difference between two porvinces and
> howcan I overcome it?,
> >If I can
> >Any helps and recommendetions are appreciated
> >Thanks Advance
> >Gokhan KAYA
> >Yahoo! for Good
> > Watch the Hurricane Katrina Shelter From The Storm concert
> Do You Yahoo!?
> Tired of spam? Yahoo! Mail has the best spam protection around