There is a long tradition in social science research, going back at least to Converse (1964), holding that most people's political views are relatively incoherent, poorly thought-through, and unstable. They're just making up responses to survey questions on the spot, in a way that can involve a lot of randomness.

This study demonstrates that plus confabulation, in a way that is particularly compelling because of the short time scale involved and the experimental manipulation of what opinion the person was defending.