Here are some specifics. Ten percent of the 60 proposers had one or more letter with inappropriate content for the purpose of the letter and all such cases were in letters for the women. Prof. McNutt cites examples of mentioning that the candidate was "so good to her elderly mother", "spending time in nature with her husband and her animals friends". Another discussed the candidate's balancing being a scientists and a mother. Also, the language was on-average different between men and women in a detrimental way for women. In some cases, the women got adjectives such as "friendly", "kind", "pleasant", "humble", and frequently "nice". Typical language for the male candidates, and also many of the females candidates, included "brilliant", "creative", "hard-working", insightful" and "showing leadership".
Graphic from Science May 2015 issue (McNutt)
Interestingly, the same bias was noted in letters from male and female writers. This may seem surprising at first, but is actually quite common in the world of unconscious bias. For example, in my blog last month (April 20, 2015) it was noted that female and male professors show about the same bias when it comes to inviting candidates of different genders to see them during campus visits.
What can be done to combat such pervasive, but subtle, bias in the way we treat young people entering science? One way is to be aware of it. High profile articles such as this one in Science help a lot. Next time I write a recommendation letter, I will pay more attention to the words and style.
My wife suggested an easy way to check if the language has biases when writing a letter for a female student or colleague. Replace all the she's with he's and all the name occurrences with a male name, then read the letter. Even a single time doing this can turn up slight wording and style differences used for different genders and be a a lesson for future letters.