Read on Twitter
Doing some ARC modelling. If you factor in a 5% bias on women, it leads to around a two-fold drop in success for projects not in the very top ranking. 1/n
I've assumed here that all projects have an 80% mean real score with a 20% standard deviation (based on pruning at university level). The top 15% get funded, after being ranked. 2/n
The assessment has a 20% noise for both men and women - only difference is a systematic 5% lower ranking for women. Note, in student evaluations women get a _20%_ lower ranking. 3/n
Which is to say that our ARC system doesn't need much bias against women to see dramatic changes to outcomes. And, of course, similar biases apply for experience - since it is blatantly obvious that "relative to opportunity" is not done well. 4/n
Next step will be to try "lottery" systems with an explicit stochastic element designed. But these results were bad enough to share as is. 5/5
