Via Pharyngula.

A recent set of studies by Gervais, Shariff, and Norenzayan tested whether public dislike of atheists was based more on distrust or revulsion.  The first study simply asked directly, and found that American adults report a strong distrust of atheists.  However, they worried that explicit answers might be more about signaling, so they did a second study with a more unusual methodology. Questions they asked included (paraphrased) "Is a person who steals money out of a lost wallet more likely to be 1) a teacher or 2) a teacher and an atheist?" and "Is a person who goes all day without noticing he has phlegm on his tie more likely to be 1) a teacher or 2) a teacher and an atheist?"  Students, especially religious ones, often answered 2) to the first and 1) to the second, suggesting that it's more about mistrust.

I suspect this isn't actually a more effective way of eliciting stereotypes than asking directly.  I think signaling concerns will be just as active in the second study, and there will be a skewing of results in that anyone familiar with the conjunction fallacy (or even the importance of base rates) will answer 1) even if they distrust or are disgusted by atheists.  The result will inevitably underestimate dislike of atheists and evince a spurious or exaggerated correlation between such dislike and statistical innumeracy.  I think a better way to look for implicit, rather than explicit, stereotyping would be to create an Implicit Association Test.  That said, I think the study is still meaningful and I'm intrigued by the methodology.

Thanks to lukeprog, here's the full paper.  Relevant excerpt:

Study 1 demonstrated explicit distrust of atheists, but it is possible that, instead of being representative of personal feelings, participants’ explicit responses may have instead reflected cultural norms determining which groups are fair game for criticism and which should be insulated. The varied permissibility of such criticism is itself an interesting indicator of prejudice, but it does not specifically map on to the questions of distrust at the heart of this project. As a result, in Study 2, we adapted a classic conjunction fallacy paradigm (e.g., Tversky & Kahnemann, 1983) to create an indirect measure of distrust for various groups of people.