How many would you expect in a random selection of people, assuming they constitute % of available speakers?
This selection has:
Over-representation is therefore about times as likely as no representation.
Tech conference speaker line-ups frequently contain few or no women at all. I believe, as many others do, that most conference selection processes are biased towards dominant demographics (you know: men, white, cis, able-bodied) and that addressing and removing this bias is an important part of the battle to increase diversity in the industry at large.
I sometimes encounter the argument that speaker line-ups that fail to adequately represent women are not the product of systemic discrimination, but rather an inevitably frequent occurrence in an industry as male-dominated as ours. On the face of it, this makes intuitive sense.
Human beings, however, are notoriously bad with probabilities.
This calculator was inspired by comments from Dave Wilkinson and Paul Battley, who modelled the probability distribution for male/female speaker balance based on a Poisson distribution and found that the likelihood of an unbiased selection process yielding a line-up with no women at all is far lower than intuition might suggest, and—depending on the numbers you plug in—can often be far lower than the likelihood of their over-representation. That is to say: in an unbiased selection, you’re significantly more likely to see more than the expected number of women than none at all.
A caveat: there’s no obvious value to use for the percentage of “women in tech”. Wilkinson used 20%, “based off of Taubee studies and corroborated by other conferences that achieve speaker lineups with this distribution”. Battley used 9.1%, based off the number of self-identified female developers in A List Apart’s annual survey. 10% is used here as an arbitrary, conservative-feeling number. Feel free to plug in anything from Wikipedia’s page on women in computing, if you want to play.
The statistical modelling technique employed in this calculator has also been used in articles tackling the same biases in physics (see this article in the American Physical Society newspaper) and mathematics (see this article in The Verge).