Re: Interpretation of reaction time results. Reaction time is one of the most ambiguous measures (and among the most frequently used) in psychology (that's not just my opinion, but also that of R. D. Luce in regard ot reaction time results generally). It is not clear, for example, whether what was shown was indicative of prejudice, or knowledge of prejudice, self-awareness of being tested for prejudice (which is very obvious--in fact, far too obvious...this artifact alone brings the interpretation of the results into doubt). It is necessary to obtain other measures in conjunction with reaction time in order to get greater confidence that what is being measured truly is what is intended to be measured. (This is in addition to having a large number of different conditions, which help to rule out alternative explanations...but alas this becomes increasingly and often prohibitively expensive). In addition, other tasks should be used to show that the effect is not specific to the type of task in question. (Other ways to ensure generalizability of the results: test different age groups, test different populations (i.e., test people in different countries).

Re: Height of CEO's. The irrational belief that taller people make better leaders may indeed be a factor here. But there are a bunch or intercorrelated factors here, eg., height is correlated with socio-economic status (SES); SES is correlated with education; one's SES is associated with one's parents' SES; height is correlated with parents' height; SES is correlated with nutrition; nutrition is correlated with height; education is correlated with nutrition; education is correlated with IQ...and there's probably more I'm leaving out. I'm not sure the research has been done to sort out the roles of all of these factors determining CEO's heights. Confidence and dominant style of social interaction might be related to height insofar as playground, high school, and college socialization may establish this pattern. Yet another factor to be considered is women's preference for taller males, which has been shown in 37 out of 37 cultures studied. (Many opportunities for preferential treatment could be affected by this bias).

Some of our biases are truly staggering; by my above cautions regarding methodology I don't mean to downplay the biases or deny that they could exist, of course. For example, people will give a higher grade to an essay attributed to an author judged to be good-looking vs one judged to be not good-looking. (The same essay is assigned to different groups, with a picture of the author on the front. E.g., an essay would be assigned with a good-looking author's picture to one group, and the same essay would be assigned with a not-good-looking author's picture. The groups then were given a set of criteria and were asked to grade the essays accordingly. They were not asked to consider anything else such as the author's appearance; appearance wasn't mentioned. Other judges had previously judged the pictures independently of the essays to obtain the "attractiveness" ratings). If memory serves me correctly, there was actually a 10% difference in the average marks, with the advantage going to the good-looking authors.

The point as I understand it is, if people were just repeating previously learned patterns, then why would the results in general skew so dramatically toward a particular set of associations? Since the category order is randomized, if that was the error people were making, the associations should be randomly distributed. But they're not. There is a consistent association, across many test-takers, between certain items and categories.

I don't think your conclusions are valid at all, it's just establishing a memorized rote which a person follows. I'd be more impressed if the word order were actually randomized and isn't influenced by the memory of the previous test.

In short, people who are not racist may hesitate to answer not because they are unconsciously prejudiced, but because unconsciously they want to take care that their answers won't be perceived as racist.

I wonder if part of the problem might be the learned habit of reading left to right and a tendency to think of these words in a certain order ("male and female" being less phonetically awkward than "female and male").

These are both interesting suggestions to explain these results - I don't actually know whether the word order is reversed from test to test, though it certainly should be, to exclude that as a possible source of error. Even if not, though, one might argue that viewing "male and female" as more natural than "female and male" may itself be a type of that subtle subconscious bias this study was designed to measure.

As BlackWizardMagus pointed out, it's an open question whether this phenomenon actually affects how people treat each other consciously. But there's no question that subtle factors can affect our decisions in ways we're not explicitly aware of. Gladwell points out in the same chapter of Blink that 58% of CEOs are six feet tall or more, while only 15% of men in general are that tall. Even more striking, only 4% of men are six foot two or taller, but among CEOs, it's almost one-third. Clearly there is a subconscious bias here (an irrational belief that tall people make better leaders) that is affecting conscious decision-making in a strikingly obvious way. We don't yet know if implicit racial or gender bias affects decisions in a similar fashion, but it would be foolish to dismiss the possibility.