| |
 |
|
|
Science Forum Index » Statistics - Math Forum » 95%CI for Cronbach's alpha coefficient with SAS
Page 1 of 1
|
| Author |
Message |
| cat.. |
 Posted: Wed Apr 23, 2008 11:55 pm |
|
|
|
Guest
|
Dear SAS-Lers,
I've been asked to compute the 95%CI for Cronbach alpha coefficient
based on a specific reference. ok, no problem excepted that SAS does
not support it and this is the only program I can use.
The reference is: JM Van Zyl et al., "On the distribution of the
maximum-likelihood estimator of conbach's alpha", biometrika, vol 65
(3):271-80, 2000.
This paper is really tricky and I cannot figure out a SAS code for
computing it. I guess it may require SAS/IML langage, which I don't
know at all.
An on-line SAS sample code does exist that computes several 95%CI,
based on several references (but not "mine"). The code is complex and
written in SAS/IML.
Has anyone ever developed such a code ? Any clue for me ?
Another question I am asking to myself is: What if the distributional
validity assumptions are not met ? Is it worth it to calculate this
CI ? Eg: If you had to calculate a 95%CI of a mean on any continuous
variable for which you know that it is not normal. What value could
have this 95% CI ?
Thanks.
Catherine. |
|
|
| Back to top |
|
| Guest |
| Posted: Thu Apr 24, 2008 11:17 pm |
|
|
|
|
Hi,Catherine,
Would it be easier to use simulation study? Say bootstrap. |
|
|
| Back to top |
|
| Ryan |
| Posted: Fri Apr 25, 2008 5:20 pm |
|
|
|
Guest
|
On Apr 24, 5:55 am, "cat.." <cat....@gmail.com> wrote:
Quote: Dear SAS-Lers,
I've been asked to compute the 95%CI for Cronbach alpha coefficient
based on a specific reference. ok, no problem excepted that SAS does
not support it and this is the only program I can use.
The reference is: JM Van Zyl et al., "On the distribution of the
maximum-likelihood estimator of conbach's alpha", biometrika, vol 65
(3):271-80, 2000.
This paper is really tricky and I cannot figure out a SAS code for
computing it. I guess it may require SAS/IML langage, which I don't
know at all.
An on-line SAS sample code does exist that computes several 95%CI,
based on several references (but not "mine"). The code is complex and
written in SAS/IML.
Has anyone ever developed such a code ? Any clue for me ?
Another question I am asking to myself is: What if the distributional
validity assumptions are not met ? Is it worth it to calculate this
CI ? Eg: If you had to calculate a 95%CI of a mean on any continuous
variable for which you know that it is not normal. What value could
have this 95% CI ?
Thanks.
Catherine.
These might help:
http://pages.infinit.net/rlevesqu/Syntax/Bootstrap/BootstrapCIforCronbachAlpha.txt
http://www2.sas.com/proceedings/forum2008/230-2008.pdf
Ryan |
|
|
| Back to top |
|
| Richard Ulrich |
| Posted: Sat Apr 26, 2008 8:52 pm |
|
|
|
Guest
|
[cross-posted to sci.stat.consult where the same
question was posted separately.]
I can't help with the SAS programming, but I have a
couple of comments.
On Thu, 24 Apr 2008 02:55:52 -0700 (PDT), "cat.." <cat.b41@gmail.com>
wrote:
Quote: Dear SAS-Lers,
I've been asked to compute the 95%CI for Cronbach alpha coefficient
based on a specific reference. ok, no problem excepted that SAS does
not support it and this is the only program I can use.
The reference is: JM Van Zyl et al., "On the distribution of the
maximum-likelihood estimator of conbach's alpha", biometrika, vol 65
(3):271-80, 2000.
What I've read about the Cronbach alpha described it
as a statistic based on and created out of variances.
That makes me wonder what is gained (or lost) by using
its MLE.
Quote:
This paper is really tricky and I cannot figure out a SAS code for
computing it. I guess it may require SAS/IML langage, which I don't
know at all.
An on-line SAS sample code does exist that computes several 95%CI,
based on several references (but not "mine"). The code is complex and
written in SAS/IML.
Has anyone ever developed such a code ? Any clue for me ?
Another question I am asking to myself is: What if the distributional
validity assumptions are not met ? Is it worth it to calculate this
CI ? Eg: If you had to calculate a 95%CI of a mean on any continuous
variable for which you know that it is not normal. What value could
have this 95% CI ?
One version of Cronbach's alpha computes an alpha solely
from average correlations among items, and scale length.
Based on that, I would say that alpha is pretty robust
against underlying deviations on the separate items, and
the loss would largely be mediated by the size of those
correlations. However, I think that one buried assumption
is that the set of correlations are pretty homogeneous.
- If the correlations are *not* homogeneous, then I
expect that the formula-CI on the alpha will be too small.
Do keep in mind that alpha, based on correlations, reflects
correlations in another regard -- The reliability is a
description of the scale in *this* sampled population.
In more general use, the CI would not necessarily hold.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Tue Oct 07, 2008 5:01 pm
|
|