Van den Haak and de Jong (2005) compared the interaction between the test moderator and test user in two different settings: using thinking-aloud method alone and paired user testing that they call constructive interaction test. They analysed parts of the users’ conversations in both settings, and made (quite bold) conclusions about them. Although they had no setting in which the user(s) had been alone in the test, they concluded that all the abnormal behaviour and apologetic attitude was due to the moderator presence without even considering the possibility of the mere test setting having an effect on the users’ behaviour.
Nevertheless, their results also indicated that the test users working as a pair paid less attention to the test moderator, and behaved more securely than the users in a single thinking-aloud setting. This is something that I have observed as well.