st: Alternative to Cronbach's Alpha for Messy Data

[Lloyd Dumont <lloyddumont@yahoo.com>]

Hello.

We have about 96 survey items that we are trying to partition into meaningful constructs.  The problem is that the data come from about 8 years of worker surveys, and most of the survey items come and go (and often come back again) over the 8 year period.  That precludes the use of PCA or even calculating a typical alpha measure.  (I think there needs to be at least one observation that has values for all the items in a particular construct/index, and that can never happen given these data.)  In short, since we cannot run the –alpha- command, we can mimic what it does, but end up without a reliability coefficient!