>I can assure you that this is not easy.  The SEL has been collecting human
>effort data, program size data, error rates, and many other metrics for over
>two decades.
>...
>What you're left with is a situation where only many experiments over time
>can be used to arrive at statistically valid results.
  We'll take results at the 50% confidence level.  ;)
Seriously, it may take hundreds of years to get enough data to prove,
at the 99% level and without a shadow of a doubt, that X is true.  But
even knowing that X is more likely true than not, is *useful*
information.  While we wait for certainty, people *are* making
decisions that are more likely bad than good, based on their
conclusions, statistically valid at *very* low confidence, from single
circumstance anecdotes.