Re: [Gendispatch] New Version Notification for draft-knodel-nomcom-gender-representation-00.txt

[Brian E Carpenter <[e-mail]>]

On 23-Nov-23 10:18, Eric Rescorla wrote:
> 
> 
> On Wed, Nov 22, 2023 at 1:09 PM Brian E Carpenter <[e-mail] <mailto:[e-mail]>> wrote:
> 
>     On 23-Nov-23 09:05, Eric Rescorla wrote:
>      >
>      >
>      > On Wed, Nov 22, 2023 at 11:57 AM Christian Huitema <[e-mail] <mailto:[e-mail]> <mailto:[e-mail] <mailto:[e-mail]>>> wrote:
>      >
>      >
>      >      > On Nov 21, 2023, at 8:37 PM, Martin Thomson <[e-mail] <mailto:[e-mail]> <mailto:[e-mail] <mailto:[e-mail]>>> wrote:
>      >      >
>      >      > On Wed, Nov 22, 2023, at 15:25, Brian E Carpenter wrote:
>      >      >> Yes, but if (hypothetically) we had four possible responses: Male,
>      >      >> Female, Other, Prefer-not-to-say,
>      >      >> I really don't know what to do with Prefer-not-to-say.
>      >      >
>      >      > I have no strong preference here, but it could be reasonable to treat other and pnts as distinct categories.  Or both together.  Based on my understanding of demographics, it is unlikely that either (or both) would be selected for more than 8 or 9 out of 10, so while we need to write up firm rules, that would be the only reason we'd need that level of precision; it's unlikely to be a problem in practice.
>      >
>      >     Prefer-not-to-say amounts to opting out of gender based selection. That means the candidate would be selected based on random ranking only, and should never be selected based on "reserved for minority" preferences.
>      >
>      >     There are many reasons to prefer not to say anything. One is a general aversion for keeping personal preferences in databases, based for example of the history of repurposing such databases for other purposes, like the Gestapo did 60 some years ago. Another is because the participant would prefer to keep private details hidden. Yet anther is to introduce enough noise in the system to provide cover for those participants who want their private details hidden. These are all very good reasons, but they should not provide any priority in nomcom selection.
>      >
>      >
>      > Just so I understand what you are proposing in practice. Suppose we have four
>      > values: A, B, C, and PNTS, then the selection algorithm would be:
>      >
>      > exclude = None
>      > while selected_total < 10 {
>      >    candidate = draw_random()
>      >    if selected(gender(x)) == 8 {
>      >       exclude = gender(x)
>      >    }
>      >    if exclude != None {
>      >      if exclude == gender(x)  {
>      >         skip;
>      >      }
>      >      // This avoids giving pnts a preference
>      >      if gender(x) == pnts {
>      >         skip;
>      >      }
>      >
>      >     select x;
>      > }
> 
>     I think that gives PNTS a normal chance of being selected for the first 8 slots
> 
> 
> Once you have 8 majority, right?
> 
>     , but zero chance of being selected for the 9th or 10th slot.
> 
> 
> Yes.
> 
>     If so, a member of the majority gender gains no advantage by selecting PNTS, which was the tactic I was concerned about. If a member of a minority gender selects PNTS, they do worsen their chance of selection, however.
> 
> 
> They make it the same as majority gender, right? I may need more coffee, but it's not clear to me how to not have that affect, if we were to think it's undesirable

I think you're right. I'm OK with it, but it does need to be made explicit in the draft, I think, so that people are fully informed.

On the question of 8 vs 9 as the trigger, I'm open-minded on that choice.

     Brian