What you are missing is that it is an apples and oranges comparison. PADI programs, by design, and by policy, can be idealized by a rather steep bell curve (small standard deviation) if you measure "quality" on the X-axis and frequency on the Y-axis. Other agencies (again idealized) have no left hand tail, that is cut off by minimum standards, but have a right hand tail that extends way out since the only limit (with minor exception) to what may be added is the instructors' creativity and judgment. So, while the standard deviation is much higher, the average "quality" is also slightly higher, both because of the right shift due to the tail length as well as the requirement that for certain items like rescue.
If you were to apply an ANOVA approach to the problem I suspect that what you would see significant variation due to agency. In the old days, when most non-PADI instructors greatly exceeded minimum requirements to properly meet local needs, I suspect that the between agency variance greatly exceeded the between instructor variance, however, today, with many non-PADI instructors teaching what is basically a PADI course with a few additional items tossed in the between instructor variance appears more important, and I'd posit that the between agency variance, while still significant, has dropped but only as a proportion of the whole with concomitant growth in the instructor*agency interactive term.
