On 29/04/2013 05:05, Michael StJohns wrote:
At 08:53 PM 4/28/2013, Margaret Wasserman wrote:
The question that people are asking is why the diversity of the IETF leadership doesn't reflect the diversity of _the IETF_.
Let's consider for a moment that this may not actually be the correct question. Instead, consider "Why the diversity of the IETF leadership doesn't reflect the diversity of the set of the IETF WG chairs"? I believe this is a more representative candidate population for the IAB and IESG.
By my count (using the WG chairs picture page), there are 202 current working group chairs. Of these 15 are female - or 7.4% of the population [It would be more reliable to do this for any WG chair in the last 5-10 years, but the above was readily available and I think provides at least the basis for discussion. Anticipating the argument, I would assume for the sake of discussion a fairly similar percentage of ex-working group chairs per gender unless there is evidence to the contrary]
There are 14 (current area directors plus the chair) members of the IESG, of which none are currently female.
There are 12 current IAB members of which 1 member is female.
Assuming perfect distribution, that would suggest that 14 * (15/202) or 1.03 IESG members should be female.
Assuming perfect distribution, that would suggest that 12 * (15/202) or .89 IAB members should be female.
Assuming perfect distribution, that would suggest that 26 * (15/202) or 1.93 IAB + IESG members should be female.
And pretending for a moment that picks for the IAB and IESG are completely random from the candidate set of Working group chairs, the binomial distribution for 7.4% for 27 positions is:
0 - 12.5%, 1 - 27.0%, 2 - 28.1%, 3 or more - 32.5%. (e.g. about 40% of the time, the IAB and IESG combined will have 0 or 1 female members).
for 7.4% for 15 positions (IESG) is:
0 - 31.4%, 1 - 37.8%, 2 - 21.2%, 3 or more - 9.5%
for 7.4% for 12 positions (IAB) is:
0 - 39.6%, 1 - 38.1%, 2 - 16.8%, 3 or more - 5.4%
But the actual one you should consider is 7.4% for 14 positions (annual replacement):
0 - 34%, 1 - 38.1%, 2 - 19.9%, 3 or more - 8%.
This last one says that for any given nomcom selection, assuming strict random selection, 72% of the time 0 or 1 females will be selected across both the IAB and IESG. You should use this one as the actual compositions of the IAB/IESG are the sum of all the nomcom actions that have happened before.
There are statistical tests to determine whether there is a statistically significant difference in populations, but my admittedly ancient memories of statistics suggest that the population size of the IAB/IESG is too small for a statistically valid comparison with either the WG chair population or the IETF population.
Of course, the nomcom doesn't select and the confirming bodies do not confirm based on a roll of the dice.
But looking at this analysis, it's unclear - for this one axis of gender - that the question "why the diversity of the IETF leadership does not reflect the diversity of the set of IETF WG chairs" has a more correct answer than "the luck of the draw".
My base premise may be incorrect: That you need to have been a WG chair prior to service as an IAB or IESG member. I hope it isn't as I think this level of expertise is useful for success in these bodies.
Assuming it is correct, then the next question is whether or not there is a significant difference in percentage of female attendees vs percentage of female working group chairs and is there a root cause for that difference that the IETF can address in a useful manner.
Mike
This is in line with my own estimate based on an approximation of 1:10
which with random selection gives an error approximation of sqrt(1)=1
The other thing to remember is that whilst your proportional estimates
are likely to be correct, in a random process you will get long runs of
"bias" that only average out in the long run. So you will get long runs
of 0. Very infrequently you will also get long runs of 27. In both cases
it is in human nature to would assume something is wrong, when it is an
artifact of random numbers. Humans have considerable difficulty
discriminating between systematic and statistical problems, and taking
the long view rather than the short view.
For that reason, as I noted in my previous post, we need a rigorous
statistical analysis with proper confidence intervals, rather than
simple averages on spot years.
- Stewart