The most useful caveat, from the paper:

  Since the average professional software engineer has more experience than our average subject,
  the effect sizes reported may be overstated for some populations