Reliability problem

[Thomas M. Guterbock]

Chronbach's alpha will come out very low, or weird if you don't have a set
of items that are all positively intercorrelated.  (In contrast, you can do
factor analysis on variables that include positive and negative items). So,
if some of your measures are coded positively and others negatively, you
need to recode the negs to positives before you run reliability.  That is,
all measures should be in the same direction in relation to the underlying
construct.  Hope this helps.
                                                        Tom Guterbock


Thomas M. Guterbock                        Voice: [phone]
Director                         CSR Main Number: [phone]
Center for Survey Research                   FAX: [phone]
University of Virginia     EXPRESS DELIVERY:  2400 Old Ivy Road
P. O. Box 400767                                      Suite 223
Charlottesville, VA 22904-4767        Charlottesville, VA 22903
                e-mail: [hidden email]

[Humphrey Paulie]

Mark, Thomas and Stephen,
  Thanks alot for your input.
  I have no reversed coded items.
  in fact the entire data set has 160 cases and 14 dichotomously scored items.
  The alpha for the entire data is 0.81.
  I split the sample to three ability cohorts, N=70, 52 and 38.
  When I compute the alpha for each cohort separately then I get the wierd figuers.
  For cohort 1 it is 0.11 for the 2nd one it's -1.71 and for the third it's -.17.
  Could it be the result of small variances?
  I couldn't access the paper. It requires a password.
Cheers
  Humphrey
"Thomas M. Guterbock" <[hidden email]> wrote:
  Chronbach's alpha will come out very low, or weird if you don't have a set
of items that are all positively intercorrelated. (In contrast, you can do
factor analysis on variables that include positive and negative items). So,
if some of your measures are coded positively and others negatively, you
need to recode the negs to positives before you run reliability. That is,
all measures should be in the same direction in relation to the underlying
construct. Hope this helps.
Tom Guterbock


Thomas M. Guterbock Voice: [phone]
Director CSR Main Number: [phone]
Center for Survey Research FAX: [phone]
University of Virginia EXPRESS DELIVERY: 2400 Old Ivy Road
P. O. Box 400767 Suite 223
Charlottesville, VA 22904-4767 Charlottesville, VA 22903
e-mail: [hidden email]



---------------------------------
Do you Yahoo!?
 Everyone is raving about the  all-new Yahoo! Mail.

[Swank, Paul R]

You should not break the group up into ability levels if you are using
Cronbach's alpha. It necessarily will be sensitive to a lack of
heterob=geneity as is any measure based on correlations. That's another
advantage of Rasch scaling. It is relatively independnet of the ability
level of the group.


Paul R. Swank, Ph.D.
Professor, Developmental Pediatrics
Director of Research,


University of Texas Health Science Center at Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Humphrey
Sent: Thursday, October 26, 2006 10:37 AM
To: [hidden email]
Subject: Re: Reliability problem

Mark, Thomas and Stephen,
  Thanks alot for your input.
  I have no reversed coded items.
  in fact the entire data set has 160 cases and 14 dichotomously scored
items.
  The alpha for the entire data is 0.81.
  I split the sample to three ability cohorts, N=70, 52 and 38.
  When I compute the alpha for each cohort separately then I get the
wierd figuers.
  For cohort 1 it is 0.11 for the 2nd one it's -1.71 and for the third
it's -.17.
  Could it be the result of small variances?
  I couldn't access the paper. It requires a password.
Cheers
  Humphrey
"Thomas M. Guterbock" <[hidden email]> wrote:
  Chronbach's alpha will come out very low, or weird if you don't have a
set of items that are all positively intercorrelated. (In contrast, you
can do factor analysis on variables that include positive and negative
items). So, if some of your measures are coded positively and others
negatively, you need to recode the negs to positives before you run
reliability. That is, all measures should be in the same direction in
relation to the underlying construct. Hope this helps.
Tom Guterbock


Thomas M. Guterbock Voice: [phone] Director CSR Main Number:
[phone] Center for Survey Research FAX: [phone] University
of Virginia EXPRESS DELIVERY: 2400 Old Ivy Road P. O. Box 400767 Suite
223 Charlottesville, VA 22904-4767 Charlottesville, VA 22903
e-mail: [hidden email]



---------------------------------
Do you Yahoo!?
 Everyone is raving about the  all-new Yahoo! Mail.