the perfidy of the barista
Tim Harford has a piece on Slate reporting on a new study claiming that women are discriminated against by male workers at coffee shops: on average, “Men get their coffee 20 seconds earlier than do women.” And this remains true even when you account for the fact that women tend to order more complicated drinks than men.
Harford comments, “It is not clear whether women were held up by male staff because the men viewed them with contempt or because the male staff members were flirting furiously.” These are the only two possible explanations he, following the authors of the study, considers, which means that he apparently does not imagine any others. But there’s a way of accounting for the discrepancy that, given my extensive experience waiting in lines at coffee shops, strikes me as far more likely than either of the ones mentioned by Harford: men are, on average, considerably more impatient than women.
In the coffee-shop lines that I know, women tend to be more relaxed, less in a hurry, more likely to be with friends and therefore in conversation. They’re more likely to have to be told twice that their drinks are ready.
Men, on the other hand, are more likely to give off every possible signal that they’re in a hurry. They stand closer to the people in line in front of them, they have their payment ready before it’s asked for, they plant themselves as near as possible to the barista and in some cases stare down the poor coffee-craftsperson until their drinks are ready, at which point they snatch up the cups and bolt from the store.
It seems to me that baristas, then, might be responding to these signals by hustling to serve the obviously impatient men, and relaxing a bit when they’re making drinks for women. And it also seems likely that male baristas would be more sensitive to the signals given off by their fellow men, more eager to show them that they share their emphasis on speed.
Maybe I have misobserved; maybe coffee-shop behavior is different in Chicagoland than in the Boston area, where the study was conducted. We can’t know, because the research seems to have paid attention to the behavior only of the employees, not the customers (whose gender alone was considered relevant).
The title of the study is “Ladies First? A Field Study of Discrimination in Coffee Shops.” The assumption that difference always equals discrimination seems in this case to have limited the investigation’s terms unnecessarily, and to exclude certain relevant factors from its view.
When I worked at Hunter’s Bookstore in La Jolla, CA, any customer who showed signs of impatience would get the worst possible treatment from my co-worker, a dark-haired, scowling, brooding—published—science fiction author. Arguments and insults free of charge, along with your hardcover bestseller (or hard-to-find science fiction novel) and be grateful he stopped short of his worst. I don’t suppose that behavior would go over all that well in a coffee shop.
— Joules · Nov 11, 04:20 AM · #
This is almost a perfect example of researchers asserting unfounded conclusions.
There are both technical and conceptual problems with the assertions that have been made here (to glorify some pretty common-sense issues).
Alan, you put forward a sensible example of an alternative explanation for why women might have slower service times than men for reasons other than discrimination. Another example would be that women have more complicated orders, on average, than men. The researchers obviously thought about this, and coded anything other than a coffee or tea or something like that as “fancy” (a 0/1 variable). They didn’t distinguish between kinds of “fancy” orders (e.g., “tall americano” vs. “decaf two pump grande iced caramel macchiato with extra ice in a venti cup”). The researchers observed that women do place more fancy orders than men. It is a non-trivial difference: ~75% women’s orders were fancy vs. ~55% for men.
They note, as you do, that after they control for “drink complexity” (meaning fancy vs. non-fancy), that there is still a service time difference between men and women, but as they also note, there is a non-trivial possibility that the level of complexity of fancy drinks for women may be higher than that for men. This is a non-academic possibility, since it is very common for a behavior or characteristic that is more prevalent in sub-population A than sub-population B to have a higher absolute value for the maximum coded range for A. Consider, for example, if you had all adult males vs. females in the US divided into wage bands, the highest of which was $100K or more per year. A higher proportion of males than females would be in the maximum wage band, and the average income of those males in the 100K+ wage band would also be higher than the average income of those females in the 100K+ wage band. In fact, for this reason the researchers in this study informally spoke with (“surveyed” in their language) four coffee shop workers, three of whom indicated that women were more likely to make additional special requests when placing fancy orders than men. The default assumption ought to be that we would have to consider, and either prove or disprove, this effect, as we have ample a priori reason believe it might matter.
How did the researchers deal with this issue? Indirectly and not persuasively. They did a separate interaction analysis that “showed” the extra wait time for females was higher when there were more male employees in the store and lower hen there were more female employees. They hang their hat on this, but as they note, it is not statistically significant. That’s putting it mildly: the p-value is 0.31. Said differently, there is a 31% chance that even this correlation is random variation in the data, never mind the question of whether it represents a causal relationship. As a point of reference, in a normal analysis variables are accepted only if p<0.05. And remember that even if they had demonstrated such an interaction, it still wouldn’t directly address the issue of whether there is a complexity difference between male and female fancy orders.
This is an example of the problem of omitted variable bias, which is a jargonish way of saying that actual causal variables might be correlated with male vs. female, and therefore when the researchers find “discrimination” that are really missing the causal relationship. Alan, as I said, you put forward one such alternative explanation. The researchers identify (and substantiate) the additional potential issue of differential fancy order complexity, and fail to eliminate it. A third (if my experiences at Starbucks are representative) is that when the shop gets busy, there is a lot of jockeying to get your order into the actual barista, rather than the person at the register, which actually drives a major component of wait time, and males could be more aggressive at doing this. In fact, if you look at Model 6 in their paper, this would be consistent with the “customer female * line length” interaction term (of course, as with most of the findings in this paper, it is not statistically significant). A fourth possibility is that, as the authors note, women may be worse tippers than men, especially for small-check transactions such as a coffee shop orders. Differential treatment may be a result of this effect. Even if you think it is wrong to treat people based on “stereotypes” (even if, as in this hypothetical, they are based on correct data analysis), many coffee shops have a very large number of known, repeat customers. It may be that the servers are treating only the known customers differently based on historical tipping behavior, and this difference creates a significant difference for male vs. female treatment across the whole population. It’s pretty easy to generate fifth, sixth and seventh possibilities. These are not “what if the moon is really made of green cheese and that makes people do what they do” possibilities, either. Human behavior is complex, and usually what causes some effect in retail interactions across a population turns out to be obvious only in retrospect.
This isn’t to dump on these researchers in isolation. They at least seem to highlight a number of the causes for skepticism of their findings. This is an example of a structural problem with a lot of what passes for “social science.”
— Jim Manzi · Nov 11, 11:44 PM · #
That’s an extremely helpful response, Jim.
— Alan Jacobs · Nov 12, 09:16 PM · #