Anderson, Bradley wrote:
>>I've worked with some data like these and have observed similar
patterns.
Good point! There's no reason to assume that these items make up a
unidimensional continuum, which is what Cronbach's alpha is designed to
measure.
In the opinion of most in the psychometric community, Cronbach's alpha
is just an approximate analysis of the factor analysis/IRT you really
should have done.
>>In short, measures of internal-consistency reliability imply a
measurement model that is probably untenable in this case.
Yes, I totally agree. A good reference is David Streiner and Geoffrey
Norman. Health Measurement Scales, Third Edition, Oxford. 2003. I plan
to use it for my class in the spring as a supplementary reading for
emphasizing understanding, with M. J. Allen and W. Yen (1979).
Introduction to Measurement Theory for the basic technical results.
JV
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
