This is a reading comprehension test.

First, the subjects had to read the questionnaire.  It appears they were not asked to orally defend their positions at this point.  This would have been confirmation that they actually read the statement correctly in round one.

Then, the statements were switched subtly to reverse meaning. 50% of people noticed the change and adjusted their response.  

Ok… Wouldn’t you want to know how many people read the question incorrectly in the first round? This would explain why their actual position was opposite their first round answer.    When asked for an oral defense of their position, they gave their real opinion.  This didnt match the first round of questioning because they read the survey incorrectly. There was no oral confirmation of their position in round one. No voodoo. No moral compass switch. Poor reading comprehension.