Gender Bias in Particle Physics: A Statistical Analysis

UPDATE: After posting this entry, I found out that the paper I discussed here is not actually slated at this time to be published in a peer-reviewed journal; it is merely available as a preprint. Nevertheless, I hear that the folks at Nature have picked up on this and have interviewed the author; we may see something next week there about it.
Remember that famous line about how women need to be twice as good as men to be considered half as good? A new statistical study by Sherry Towers available on ArXiv.org shows just how true this is in the world of particle physics.
Here’s the scoop:

…the females in our cohort had to be on average 3 times more productive than their male peers in order to be awarded a conference presentation…if conference presentations were allocated by the administration of the experiment in a gender-blind fashion, we would expect that around 50% more of the females in our cohort would have moved on to faculty positions. The gender-biased allocation of conference presentations…appears to be an effective gate-keeping mechanism that chooses which females can move on to faculty positions and which cannot.

Towers presents a statistical analysis very similar to that of Wenneras and Wold
in their landmark paper in Science. (Wenneras and Wold, 1997, “Nepotism and Sexism in
Peer Review
” Nature vol 337 pp 341) Wenneras and Wold, you’ll remember, found that females had to be 2.5 times more productive than males to receive a postdoctoral fellowship.
In particle physics experiments, conference presentations are a chance for postdocs to make themselves known to potential future employers. It’s a way to stand out in a world where publications can have as many as 700 coauthors. But you have to have permission from the experiment administrators to give a conference presentation, and this is where the bias enters in. Towers looked at productivity defined as number of internal papers (physics analysis papers and service papers) produced by a postdoc and found that the ratio of conference presentations to internal physics papers for females was about half that of males.
The analysis is more complicated than the brief description I’ve given here, but the paper reads very clearly and is easy to follow.
Disclosure: Towers thanks me at the end of the paper for “insightful discussions”. I don’t really know that I was more than a sounding board for her as she worked on this paper, so that was very kind of her.
I have also blogged previously (on the earlier version of this blog) about Towers’ personal experiences at Fermilab, in response to a Chronicle of Higher Education article about her. (subscription needed for the Chronicle piece.)

  1. April 23, 2008 at 4:55 pm

    My reading of her paper is conference presentations is proposed as a plausible explanation for the fact that women landed fewer jobs than those to which they would have been entitled, based on their productivity.
    I question whether her data really show that, though, namely that women were hired in significantly smaller number than men. I really don’t think that’s true, or in any case i don’t think that the data presented in the paper really show that. It is possible that women may have been invited to give talks less often than men (although I would contend that her data do not show that either), but if in the end that does not have a measurable impact on hires, is it really that important ?

  2. April 24, 2008 at 4:02 pm

    She doesn’t say women were hired in significantly smaller numbers than men.
    Nor have I claimed that she does. What she does say (on page 14, for example) is that
    “If the experiment allocated physics conference presentations based on physics productivity rather than gender, we predict that around 50% more females in our cohort would have moved on to faculty positions”
    Is it inaccurate to rephrase the above as “fewer women were hired than productivity considerations would have dictated” ? If it is correct, then conference presentations were merely the mechanism to actuate the bias.
    My difficulty accepting her thesis stems from the following:
    1) I do not buy her argument of “around 50% more females [should have been hired]” (which means “around 2”). This predictions comes from Towers’ own parametric model (page 13), whose reliability is not obvious, especially given the small sample size. A more objective statement could be made if six of the nine women ranked by productivity in the top twenty (something that Towers does not say), in which case one could indeed claim that discrimination took place (assuming that productivity were all that mattered in a faculty hire, which it is not). However, even in that case it would not be specifically discrimination against women, for just as many men, proportionally, could argue that their position was unfairly given to a less productive colleague.
    2) I do not believe that a simple statistical fluctuation can be ruled out, when talking a deviation of two from a target of six in a sample of nine.
    I am also not convinced by her “conference reward ratio” measure, which I regard as misleading because it gives most of the weight to presentations given by “unproductive” individuals. I think it would be easier to assess to what extent men were awarded significantly more conference slots if the raw numbers were available, as opposed to ratios.

  3. April 25, 2008 at 12:52 pm

    Nobody said it could be completely ruled out!
    Towers seems to think that it is sufficiently unlikely that one may appropriately speak of “gender bias”. I think that that is a pretty serious accusation, and I think that the statistical evidence required to make it is much stronger than the one she presents in her paper. Just my opinion, of course.
    You […] don’t seem to disagree that there was a clear effect related to the presentations, which I thought was the main take-away message of the paper.
    Actually I do disagree on both counts. First off, I do not think that Towers’ data clearly show that conference invitations were allocated in significantly greater numbers to men. That is suggested by her “conference reward ratio” measure, which I think is misleading for the reasons that I stated above.
    Secondly, even accepting the above premise, I disagree that there was any effect whatsoever, let alone a “clear” one. Towers’ point is that if conference time allocation had been fair, it would have had to reflect productivity. This means, in turn, that hires should also ultimately mirror productivity. In order for Towers’ case to be valid, observed hiring frequencies should deviate significantly from what one would expect based on productivity. In my opinion, they do not (if you are interested in details, I have a blog entry on this subject), which is why I do not buy her claim of gender bias (which does not mean that I claim that it does not exist).

  4. April 25, 2008 at 3:09 pm

    I still don’t understand why you are only interested in the hiring.
    Because the whole point of the paper is that career prospects of female researchers were significantly hurt by the alleged infrequency of their invited talks. I respectfully submit that that is what makes this paper potentially so important.
    I think very few people, including Towers herself, would care about speaking invitations at all, if they had little or no measurable effect on career prospects. How many of those female researchers, given a choice, would rather give a talk than get a job, in your view ?
    My point is, even if allocation of invited talks did not reflect productivity, it does not seem to have affected careers.
    Of course, things should always be done fairly, including giving every one a chance to speak at conferences.

  5. April 26, 2008 at 10:48 am

    Dude, Gerard has no good faith interest in seeing the underlying data. The data being “unavailable” is just a fake-ass bullshit diversionary tactic that lets him sound like he is all about “scientific method”, when what he is really about is “fearful woman-hating apologetics”.

  6. Schlupp
    April 28, 2008 at 3:15 pm

    Gerard, in which way does this contradict my statement? Ask the guy at the statistics department, because the data are there. Fine that you finally agree with me that the data are not lost.

  7. April 29, 2008 at 8:26 am

    Evgeny, as much as I like The Kids in the Hall (and thanks for the link to that…I didn’t know there were a bunch of TKH skits on YouTube), I think sarcasm is actually a relatively innocuous response.
    I think you underestimate the zealousness and viciousness of people like Harbison and the people he apparently associates with; nothing gets your attention like getting an email from an anonymous hotmail account (you know the kind…mybigasstruck357@hotmail.com) from someone who claims to be a “friend” of Harbison that lists my home address and the names of my husband and kids.

  8. absinthe
    April 30, 2008 at 12:29 pm

    Damn, I just came across the original comment thread. It’s here http://forum.darwincentral.org/viewtopic.php?f=9&t=12900&hilit=sherry+towers

  1. No trackbacks yet.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: