Tuesday, May 01, 2007
Update: In the comments, Arosko has collected data contradicting the study's claims.
Via Omni-Brain, a press release is circulating wildly about a new study by David Figlio (an economist), which suggests that girls with more feminine-sounding names are less likely to pursue math and the sciences compared to girls with less feminine names.* First, let me admit that out of laziness I didn't read the press release in full and conjectured at Omni-Brain's blog that it might be an effect of genes (i.e., masculine parents giving their daughters masculine names plus masculine genes), but it's a twin study. Whenever I hear an argument of the form, "group X does / doesn't tend to do well in activity Y because of social expectations," my toes curl.**
There are several puzzling, unanswered questions, such as how exactly is the femininity of a girl's name measured? In languages with grammatical gender, there are regularities that allow you to predict better than chance if a word is masculine or feminine, but English doesn't have that. Then again, some sound sequences might be more frequent in male than female names, or something like that. But is there independent evidence that this something that ordinary people are sensitive to, enough to behave differently toward Abigail than Isabella? I don't think these are that hard to answer, and the metric may be perfectly sound.
The problem I want to focus on is more dull but potentially more damning: it may be that the "result" is not due to anything other than sampling error. The reason is that non-feminine names are not that popular, and therefore the variance in outcomes will be greater among Abigails than Elizabeths, due only to the much smaller sample size of the former. Below the fold is a graph showing the relationship between femininity of a name and its popularity.
The 15 names and their femininity scores are as reported at the end of the press release linked to at the beginning. To judge popularity, I entered the names into the HowManyOfMe search engine and rounded the figure to the nearest thousand. I couldn't search by sex, so for "Alex," this returned male as well as female names. Note that "Alex" is the full name, not the nickname, since "Elizabeth" is not reported as "Liz," for example. I went with the figure for "Alexandra" (that of "Alexis" was about the same), which is probably an upper-bound, again since it's unlikely to give a girl the full name Alex. As the graph shows, feminity is moderately correlated with its popularity, with r = +0.39 (r^2 = 0.15). That's not surprising if there really is something to the idea that some names are more feminine than others, and if parents in general don't want to give their daughters manly names.
Thus, the statements about girls with less-feminine names must be based on very small sample sizes, which we know would increase the variance in their outcomes. The same is true for the one outlier at the very-feminine end -- "Isabella" is the most feminine but least frequent, so pronouncements about Isabellas are premature. To be concrete, Figlio's study used 1000 twin pairs, or 2000 girls total. Based on the figures reported at HowManyOfMe, together with that site's estimate of the US population as 301,734,581 (the female half of which is ~150,867,000), I multiplied the population frequency of each name times the sample size of 2000 to see how many individuals of each name we might expect to see. True, there are probably age effects for frequency of names, but this at least gives a rough idea. We would expect to see 18.7 Elizabeths, 9.8 Jessicas, and 8.8 Annas (more feminine), but only 0.8 Alexs, 0.5 Abigails, and 3.8 Graces (less feminine). For the outlier of Isabella (more feminine), we'd expect 0.2.
Even if we partitioned the names into equivalence classes of femininity, it's still clear that the least-feminine group is going to be small, while the most-feminine group will be pretty big. Concretely, dividing the number of girls with a given name by the total number of girls with any name that was covered in the sample (rather than out of all girls), and multiplying this relative frequency by 2000 still gives us just 35 Alexs and Abigails combined, and 4 Isabellas, compared to 774 Elizabeths and Jessicas combined. Maybe girls with names as non-feminine as Abigail are more likely to pursue math because, just by chance, the handful of girls with such names in the sample happened to prefer science. The name Isabella is getting a lot of attention (it gets 4 mentions in the brief press release, apart from its entry in the list of names and their scores), but these conclusions must be based on 1 or 2 individuals at most. For what it's worth, in my 26 years, I've met one Isabella (and I lived in Barcelona for a year, where I should've met many more!), and she was struggling in and hated math. Why my anecdote isn't being given a megaphone, I don't know.
So, there are three take-home messages here. First, smaller sample sizes produce greater variance, which despite being taught in any statistics class is still not widely appreciated, as Howard Wainer illustrates in his essay "The Most Dangerous Equation". Second, since journalists in search of a juicy scoop and academics in search of recognition make a perfect pair to collude against public understanding, don't believe any popular press release about an unpublished study; delay judgment until the study becomes publicly available and you can inspect it yourself. And third, extra caution is probably wise when the academic is an economist writing about anything other than economics -- this isn't a credentialist turf-war, but just an observation that economics must be a hard way to earn fame, since economists are increasingly trying to invade other fields. That may or may not work out fine: physicists who invade typically do better than you'd expect for career-changers, but the track record for economists who invade is less impressive.
* This is another one of those studies that gets a lot of PR but hasn't yet been published (though it is at least in press at Journal of Human Resources). These episodes drive me nuts because while everyone is busy gossiping about them, we have to take the PR department's word for it that the interpretation is sound, but once the data are finally made public, everyone has stopped caring. This makes an easily broadcastable critique impossible. Obviously my critique is subject to this as well: maybe my complaints will prove baseless -- but I want to know. It's almost as obnoxious as the increasing trend toward publishing the data in online supplements, which often tell a more watered-down story than the headline and "Discussion" section of the article, but where most lay readers aren't going to bother to look. So, we'll have to really see what's up once the study comes out.
** Consider a trivial example: very physically attractive girls project femininity at least as much as do girls with very feminine names. And at least to judge anecdotally, they tend to go into more feminine careers like PR, advertising, and so on, rather than mathematics, sports, etc. Is this because they have succumbed to social expectations about what feminine women ought to do? Or did they just figure out that some niches are more tailored to feminine women than others, and rationally decided to play their strong suit?
Labels: unreadable studies