I could not agree more with Dave. I have had some experience and success
using a "critical incident" approach to measuring success and failure with
error management in the military. Originally validated by John Flanagan,
a WWII researcher, it has been used extensively in social science over the
past 50 years. The methodology is simple, but labor intensive due to the
need for qualitative analysis and coding by experts. I've heard some new
qualitative analysis software exists, but have not seen it used yet in
this way. Critical incidents are more usable than accident data because
of a larger N, as well as detailing positive and negative aspects of
training.
Tony Kern