Here is a write-up on a pretty highly referenced study:

Monin, B., & Miller, D. T. (January 01, 2001). Moral credentials and the expression of prejudice. Journal of Personality and Social Psychology, 81, 1, 33-43. (http://www.ask-force.org/web/Discourse/Monin-Moral-Credentials-Prejudice-2001.pdf)

Abstract: Three experiments supported the hypothesis that people are more willing to express attitudes that could be viewed as prejudiced when their past behavior has established their credentials as non-prejudiced persons.

One manifestation of people's anxiety about appearing prejudiced (to themselves or to others) is their hesitancy to act on non-egalitarian attitudes unless those attitudes can plausibly be attributed to something other than prejudice. (p33)

For example, the more a man has shown that he is not a sexist, the less he will fear that his current behavior might be attributed to sexism and the more comfortable he will be expressing a pro-male attitude. "Moral credentials" of this type are most available to individuals who customarily behave in a non-prejudiced fashion. However, even prejudiced individuals sometimes engage in ostensibly non-prejudiced behavior, if only out of deference to the anti-prejudice norms described earlier, and, thus, they too can establish and be liberated by moral credentials...people who have previously expressed antiprejudiced attitudes are more likely to express their true attitudes in contexts in which there exists the potential for accusations of prejudice.” (p34)

Study 1:, participants given the opportunity to disagree with blatantly sexist statements were later more willing to favor a man for a stereotypically male job.

Details: Base rate group did not get a survey. Other participants got a survey that asked for opinions on: “SOME women”, or “MOST women”. For example: “Some/Most women are better off at home taking care of the children. “ Participants who got the “Most” survey were therefore able to prove their moral credentials by disagreeing with overtly sexist remarks. Participants were then handed a job description for a stereotypically male job (dealt with construction, foremen, confidence, etc) and asked if they felt the job was more suited for a specific gender.

Result : There was no difference among females in the various groups who all felt the job was slightly more suited for a male, but males who had just gotten to prove their moral credentials by answering “no” to the overtly sexist remarks about “most” women were extremely strong in their preference for a male in that position. “Presumably, the opportunity to disagree with sexist items left the former participants feeling they had demonstrated that they were not sexist, thereby liberating them to respond more honestly on the dependent measure.”

Study 2: participants who first had the opportunity to select a member of a stereotyped group (a woman or an African American) for a category-neutral job were more likely to reject a member of that group for a job stereotypically suited for majority members.

Details : People were asked to choose best candidate of 5 applicants. The fourth candidate was designed to be the obvious best choice (however 17% still chose someone else), and in various conditions was either a white female, a black male, or a white male. The other candidates were all white males. Participants who had an opportunity to show their credentials, via recommending a female or black person, were less likely to find a minority member suitable in a following question. This time, there was no difference between male and female responses to the sexism portion.

It suggests that a decision that favors one minority member (even if it is totally deserved) is sufficient to liberate people to act on an attitude (often based mainly on prejudice) that is detrimental for other minority members.”

Study 3: participants who had established credentials as nonprejudiced persons revealed a greater willingness to express a politically incorrect opinion even when the audience was unaware of their credentials. The general conditions under which people feel licensed to act on illicit motives are discussed.

Details: Same as in Study 2, except this time, for one of the groups, the credentialing task (choosing the best candidate) was administered by one experimenter “then a second experimenter knocked on the door, asking whether there was time left to do his questionnaire. After the first experimenter checked her watch, she agreed to let the second experimenter hand out his task, which was the police force scenario. Participants in this condition were thus led to believe that the person reading their answer to the police force scenario would not be aware of their answer to the consulting firm task.” The self-credentialing effect was almost as strong as the credentialing effect with an audience (the experimenter).

it is not critical that others know of one's credentials for them to have a licensing effect. In other words, moral credentials do not serve solely to make one appear less prejudiced to others; they also serve, at least partially, to reaffirm one's self-image as a nonprejudiced person.” (p39)