This study has so many problems, it's hard to know where to start:

"...five studets and an instructor visited eight coffee shops in the central Boston area over the course of two weekdays in January 2007."

Aside from missed type (studets), there's selection bias in both the city, the date, and the coffee shops choosen.

"In order to control for possible effects of appearance... enumerators ranked each customer’s appearance on a scale of 1 to 10."

Translation: We controlled for appearance by adding observational bias. Wait, what?

The statistical analysis is even worse. This study looks like an activity meant for undergraduates, not something to be cited or taken very seriously.

You're right, this study was terribly designed and executed. Claiming significance for a term with a p-value of .31 is just embarrassing.

I would say the primary flaw in this study was an inability to tie the responses variables back to the order. Categorizing the order as fancy or not-fancy collapses way too many categories to be meaningful. Particularly when that variable, fancy or not-fancy, turns out to be the primary term in all of the models for which it was included.

Looking at the distributions given in figure 1, I am at a loss to see how the reported results were calculated. The distributions of serving time by sex and drink type look almost exactly the same. If anything, women who ordered plain drinks were served faster.

As a side note, i'ts really cool that the wait times for non-fancy drinks appear to be exponentially distributed, while the wait times for fancy drinks appears to be normally distributed.

That is exactly what one would expect if the non-fancy drinks required a single step performed when the server got arond to it, while the fancy drinks required several steps, with each step having a duration independent of that of the other steps.

The .31 value is REPORTED AS NOT SIGNIFICANT.

However, in a coffee shop with all female employees a female customer’s wait is estimated to be 7 seconds longer than that of a male, a differential which is not statistically significantly different from zero. Although the coefficient on the interaction term is not significant (with a p-value of 0.31), the result is suggestive that it is not order type but rather some action on the part of employees that is driving the result

"This study looks like an activity meant for undergraduates, not something to be cited or taken very seriously."

It actually was an activity for undergrads. I've read some of Myers personal response to the "hoopla" and she's pretty horrified at the way the media has glommed on to the more "salacious" parts of the paper (her words). She has also said time and again that although the preliminary results were interesting, much further research is needed. Fun for her students, but for her, not so much. Alnother example of how our media dumbs down material for 60-second sound bite consumption. (Tastes like McDonalds french fries. Or how about over-roasted big-box coffee-shop swill.)