Suppose you are playing Dungeons & Dragons (or Bunkers & Badasses). Suppose you throw 20 20-sided dice. What are the chances of getting a least a single dice with a 1?
Well, the easier way to formulate that question is to ask the dual question: “what is the chance that we get no 1’s at all?” This is 19/20 on the first dice, 19/20 on the second dice and so on. Since each throw is independent, we can simply multiply those events together to get a result. This number is (19/20)²⁰. Now, since we want the opposite result, we use 1-(19/20)²⁰ which is about 64%.
- At 15 dice thrown it is 53%
- At 20 dice thrown it is 64%
- At 25 dice thrown it is about 72%
- At 40 dice thrown it is about 87%
That is, as the number of dice thrown goes up, the chance of getting at least a single dice as a “critical fail” of 1 goes up as well. But at the lower end of dice thrown, there is a fair chance we are not going to see a single 1 turning up.
Github recently abandoned their “Electronconf” conference because it “was not diverse enough.” You may agree on the merit of this decision or not, but I claim Github doesn’t know enough about dice statistics to make an informed guess. What Github did was that they ran a blind review process in which the reviewers doesn’t know the identity of the submitters of abstracts. This is a generally good move as it removes any bias people might have on the process: well-known speakers tend to get an advantage because people know their history of being great speakers, even though their material may be fairly weak. Also — it has been claimed — women tend to have a disadvantage due to their gender. A blind process removes that bias.
But a bias-free fair process will invariably create a situation in which the organizers will run into trouble. A fair process gives equal place on the podium to a neo-nazi as to a 12 year old girl who has written her first program.
Another way organizers run into trouble is now apparent: the D&D dice throwing game is against them. We have — roughly — 5% women in tech who does programming as their major job. I’m not counting finance spreadsheet witches here, who often are better programmers than many in tech. Nor am I counting UI, UX, and Design — important areas, but they are not likely to speak at a conference such as Electronconf. It is very close to a “critial fail” of a 1 in the example above. In short:
- Draw 25 speakers at random. Chances of getting at least one woman: 74%
- Draw 20 speakers at random. Chances of getting at least one woman: 64%
Here we make an assumption that women and men are equally skilled in tech among the candidates. This assumption favors the women a lot. So it is about 1/4 or 1/3 chance that you will have 0 women among the speakers in this case.
In short: you have to drop one out of four or three conferences depending on the speaker count. If you have multiple years with no female speakers, you may want to adjust something, but a single year of 0 isn’t a problem because of the relatively few women tech has.
The “fix” Github has proposed is to drop the blind application process. This decision isn’t without some risk. Namely that your process now is governed by a bias. This bias may lower the quality of your conference in the longer run as you want to include everyone. So you end up with some “filler token talks” from the higher end of the Fitzpatrick scale (Types IV/V/VI) laced with XX chromosomes, and as a result people don’t think they get enough for their money.
My personal counter-proposal is that once we stop caring for soft subjects such as “gender” or “diversity”, we are far more likely to succeed in the long run. There is now a full cottage industry running on the subject of “diversity” and none are the wiser. If we stopped that game, we would immediately start gaining ground where it matters. And once we have the talks back in focus, we will easily find the rights speakers.
Perhaps we would even learn how to throw dice.
 It is a degenerate corner case in binomial probability where one side is at the extreme which makes this particular case so easy. More advanced binomial experiments require more elaborate formulæ.