The obvious way to check is to repeat the experiment and see if the statistics hold up or whether they are just noise. Such a repeat failed to reproduce any claim of discrimination. So either you can conclude that something has radically changed or that the original conclusion was flawed.

On the basis of the evidence provided, there is no way of telling which is the case.

And once again let me reiterate that I think it highly likely that sex discrimination exists in many places including particle physics: my point is only that those who “prove” it by such dubious means are actually doing more harm than good in making the case.

In such situations, the candidate’s “parental lineage” becomes a more important factor than the candidates actual abilities in terms of research and teaching. The reasons such tendencies are unhealthy should be obvious to anyone who wants to maintain a high-quality research and teaching institution.

In Spain, this is known as the “endogamy problem” (endogamy = marriage within the tribe or extended family). It occurs because, as I understand it, university hiring committees usually include several prominent citizens from the local city, and these people will tend to favor applicants who are locals over those from other regions (or foreigners). The result is that people who aren’t from that locality can sometimes have a harder time getting faculty or equivalent positions.

(This isn’t nepotism in the strict sense — the people being favored are not necessarily directly related to the hiring committee; they’re just “locals”.)

Notable anecdotal stories of this process include faculty members who load the short list with obvious non-starters in order to make their favored candidate look better, as well as “job listings” that, after inquiry, turn out to be in-house promotions of junior members that have to be presented to the public as “job searches”.

Does this really produce the best scientific research teams, whether big or small? In other words, “why is nepotism bad”?

Nepotism is the showing of favoritism toward relatives and friends, based upon that relationship, rather than on an objective evaluation of ability or suitability. For instance, offering employment to a relative, despite the fact that there are others who are better qualified and willing to perform the job, would be considered nepotism. The word nepotism is from the Latin word ‘nepos’, meaning “nephew” or “grandchild”.

In such situations, the candidate’s “parental lineage” becomes a more important factor than the candidates actual abilities in terms of research and teaching. The reasons such tendencies are unhealthy should be obvious to anyone who wants to maintain a high-quality research and teaching institution.

Right now, it should be noted, academics have a pretty cushy position - but imagine a situation in which external political considerations control all university activities - as was the case in Lysenko’s Soviet science “community”, and as was the case at all the German scientific institutions during the 1930s and 40s. The main external influence over science institutions in the U.S. today is the growth of secretive public-private relationships and the new emphasis on proprietary corporate research within all of the U.S. public universities. Extrapolate the trends of the past two decades forward 20 years, and what do you get?

This is why people are calling for more openness and transparency in all aspects of academic life - but especially in hiring and funding decisions.

Hmm… looks like countries with higher fractions of women physicists also have higher fractions in engineering. I’d say your “fairly obvious” explanation isn’t.

Except that we weren’t talking about undergraduate degrees, but rather the numbers further along in career progression. The figures that you report reflect the higher priority of STEM subjects in some countries. The more salient question is how these numbers translate into PhD students, postdocs and faculty positions in physics compared to engineering. Clearly, the relatively small difference between the fraction of female undergraduates studying physics in the UK (21%) and Spain (27%) is not sufficient to explain the much larger differences further down the pipe.

Precisely the conclusion at which I arrived after reading the paper. I really do not believe that her case is strong enough to make the claims that she makes, even if we accept all of her premises (some of which are dubious).

Is this actually true? (I should point out that I’ve heard a somewhat similar argument from a female Turkish friend about the roles of physics and engineering in Turkey — that is, Turkish men who might be interested in physics are encouraged to study engineering instead, on the grounds that they’ll be the ones supporting a family and so need a better-paying career — so I’m inclined to take the argument seriously. But I have this weird habit of wanting to look for actual evidence, so…)

Here are some numbers for the fraction of undergraduate physics degrees going to women in countries you’ve mentioned, for 2004:
US: 21%
UK: 21%
Italy: 36% [*]
Spain: 27%

According to your argument, we should expect lower female fractions in engineering in Italy and Spain (where engineering has higher “esteem”), and higher fractions for the US and UK.

Fraction of undergraduate degrees going to women in engineering:
US: 21%
UK: 16%
Italy: 28.%
Spain: 31%

Hmm… looks like countries with higher fractions of women physicists also have higher fractions in engineering. I’d say your “fairly obvious” explanation isn’t.

(Data mostly from the NSF’s “Science and Engineering Indicators 2008″ report, plus a few other reports found here and there.)

[*] This is actually the number for 1998; I haven’t been able to find any newer numbers for Italy. Tentatively, I would assume that the 2004 value is, if anything, slightly higher.

However, a one-dimensional statistical test along the gender axis misses important “hidden variables” and results in systematic biases, the ever-present issue in any statistical analysis. You would need to do a two-dimensional statistical analysis along the axis of gender and the axis of nepotism, at least. Nepotism is often gender-independent, for example.

It’s high time that some independent sociologist types - ideally a external investigation, not an in-house review - took a look at this issue in the United States. It’s been done elsewhere with far more statistical rigor:

But the question is, what should we do if we find evidence for nepotism? It seems to me that at some level it is unavoidable. Given the choice between employing someone you know personally and therefore already know that you can have a useful working relationship and an academically-equivalent but otherwise-unknown applicant, which do you go for? When reading two superlative references, one from someone you know and trust and the other from someone you have never met, to which do you give more weight?

A recent very clear example was highlighted in The Times. The article points out that in the recent prioritization of UK astronomy and particle physics facilities required by funding cuts, those in which the panel members had a vested interest seemed to be preferentially at high enough priority to avoid being shut down. I have the greatest respect for the members of the panel who were confronted with an impossible job and did the absolute best they could under appalling conditions. I am sure they took all the appropriate steps to ensure that their interests were declared and mitigated, but in any such exercise a number of things have to be taken on trust because no-one is an expert on everything, and it is a lot easier to have that trust when you have some personal knowledge.

This kind of bias seems to me unavoidable, and if we try to deny its existence and strive for some completely unachievable gold standard of objectivity, surely all we are doing is deluding ourselves.

From what the author says, I gather that the productivity is not normally distributed, which is not surprising. Many things aren’t; you can have an asymmetric distribution with a small number of highly productive people, for example. A Kolmogorov-Smirnov test to compare the distributions of productivity is then more appropriate than a t-test.

Indeed, given that the metric for productivity is fairly arbitrary, it can have pretty much any distribution you want. I suppose you could use a KS test to see whether male and female productivity differs, but I wouldn’t recommend it since the numbers are sufficiently small that you would have to MC the significance, plus a positive result doesn’t tell you how they differ, which would seem to be the salient issue. Comparing the means of the distributions., on the other hand, is relatively robust even with the small numbers here, and directly addresses the interesting question “are female postdocs on average more productive than their male counterparts,” to which the answer is “no.” So, the facts remain that on average the productivity of male and female postdocs is indistinguishable, and on average the fraction who went on to faculty positions is indistinguishable. Faced with those very basic results, you would be hard pressed to find evidence of sex discrimination in the data.

If only you were true. Unfortunately one does not control things merely because one is aware of things.

Gravity rules me despites my understand of its mechanism.

You point out that Success is social, even in science. OK. Now, what can we do about this? How can we advance science *despite* the social obstacle?

Here is another anecdote that also indicates a good degree of gender discrimination at top U.S. universities: Lawrence Summers, while President at Harvard

However, a one-dimensional statistical test along the gender axis misses important “hidden variables” and results in systematic biases, the ever-present issue in any statistical analysis. You would need to do a two-dimensional statistical analysis along the axis of gender and the axis of nepotism, at least. Nepotism is often gender-independent, for example.

It’s high time that some independent sociologist types - ideally a external investigation, not an in-house review - took a look at this issue in the United States. It’s been done elsewhere with far more statistical rigor:

Wenneras & Wold (1997) Nepotism and sexism in peer-review Nature v387

In the multiple-regression analysis, we assumed that the competence scores given to applicants are linearly related to their scientific productivity. We constructed six different multiple-regression models, one for each of the productivity variables outlined above. In each of these models, we determined the influence of the following factors on the competence scores: the applicant’s gender; nationality (Swedish/non-Swedish); basic education (medical, science or nursing school); scientific field; university affiliation; the evaluation committee to which the applicant was assigned; whether the applicant had had postdoctoral experience abroad; whether a letter of recommendation accompanied the application; and whether the applicant was affiliated with any of the members of the evaluation committee.

About women in physics: I don’t know about experimenters, they seem often like large corporations where things work differently. But in Theory, I have known one occasion when a very smart female theorist did extremely well for herself, two instances where pretty smart female theorists sank after postdocs like many others in the field, and two cases where not-so-smart female theorists did *extremely* well for themselves. I cannot say that I have met many guys who can fit into this last category in string theory. Of course I am going to be attacked on my personal views on “who is smart” and on my relying on anecdotal evidence,…

In any case, in the (obviously biased) view of some of their male colleagues, the last category had easy entry into male dominated collaborations, their mentors were more generous on them with their times, they often were able to associate themselves with the “superstars”, etc. because they were not unattractive.

Society has a way of trying to correct past wrongs by doing compensating wrongs now. I am certainly not saying there isn’t a problem (women are so extremely rare in theory, afterall!), but that the problem is more complicated than how it is often discussed.

Of course it isn’t completely bogus: it addresses the question of whether, on average, the population of males and the population of females are treated differently, which is the sensible first question to ask. As I am sure you know, the central limit theorem tends to make statistics based on means a lot more robust than other statistics, since there is some hope of knowing the underlying distribution of the statistic. Any higher-order statistics may be more powerful, but they suffer greatly in robustness, as the fact that someone else can reach completely the opposite conclusion based on the same statistics but a different test demonstrates.

Mike, this is not good statistical procedure for several reasons. The “productivity” here is the independent variable; we might wish to measure conference presentations as a function of productivity. From what the author says, I gather that the productivity is not normally distributed, which is not surprising. Many things aren’t; you can have an asymmetric distribution with a small number of highly productive people, for example. A Kolmogorov-Smirnov test to compare the distributions of productivity is then more appropriate than a t-test.

As a hypothetical example, it’s possible to construct two populations with the same distribution of x, one in which y is correlated with x with slope +1, and one in which y is correlated with x with slope -1. (Let us say, x is hair color, y is success as a Hall impersonator or an Oates impersonator, respectively). These populations could have the same mean and and pass the respective t-tests. However, the relation of y(x) is directly opposite, which would be apparent by measuring a correlation coefficient.

#23 Mike M: What makes you think that physics is held in high esteem (compared to say, engineering, which is your example) in the US? By whom does it have to be held in high esteem for gender imbalance in numbers to be correlated with esteem?

Mainly I am reporting what I have been told by people I know who have worked in southern European countries and in the US as their explanation for the difference in fraction of women doing physics. My own experience is that physics in the US is held in relatively high esteem compared to, say, the UK, but I have less direct experience of Italy and Spain.