News for hackers. Stuff that matters.

A comment on Do women face a reputational bias when they co-author?. Parent comment

Re:Do women face a reputational bias when they co-author? by jordigh on Friday November 13, 2015 @09:44AM (#10560053)
A bias is not bigfoot. You can clearly prove that there is fairness ("no bias", but we can phrase your negative as a positive) by simply showing that things are equal. All you have to do is count males and females and show that they are equal.

If things are not equal, you can still prove that there is no bias by asking the disadvantaged group for their opinion. It's not like polling this group is impossible.

[ Parent | on HN ]
Re:Do women face a reputational bias when they co-author? by omginternets on Friday November 13, 2015 @10:03AM (#10560156)
> by simply showing that things are equal.

Statistics do not work like this. You either positively reject the null hypothesis (i.e.: "these things are different") or you fail to reject the null hypothesis (i.e.: "I cannot assert these are different").

Formally speaking, I am doing the latter.

Informally speaking, everybody else understood my point as a rebuttal to OP's. More explicitly: "claims of X seem grossly exaggerated" should be taken as "I haven't seen compelling evidence for X" -- it may be awkwardly phrased, but I suspect you know exactly what I meant.

[ Parent | on HN ]
Re:Do women face a reputational bias when they co-author? by jordigh on Friday November 13, 2015 @10:18AM (#10560246)
There are more statistics than a simple frequentist-style hypothesis testing. You can totally make a probability statement about equality if you allow Bayesian-style statistics. Frequentist-style hypothesis testing was designed for a time when computations were done by hand and data was scarce, but we can do better now.

You can also just outright swap around your null and alternative hypotheses for a frequentist-style test. For example, your null hypothesis could be "proportion of women to men is 0.1" and your alternative hypothesis is that it isn't, for a standard two-tailed test.

[ Parent | on HN ]