The problem is this: a large class of studies compare how subjects' choices influence each other: suppose A, B and C are comparable, and the subject prefers A to B, they find that more often than not the subject also prefers C to B. The researchers can't understand why that would happen when B and C are comparable, and conclude that when the subject compares A to B and prefers A, the subject also decides that B is bad. (Monkeys do that too, so the researchers conclude that monkeys suffer cognitive dissonance.)
Chen points out that, if you do have an internal ranking of A, B, C -- not visible on the point scale the researchers give you (say, you give them all 3 points out of 5, but you have finer preferences among them) -- then if you prefer A to B, the probability of your preferring C to B is well over 1/2. To be precise, it is 2/3. This is because there are only three ranking scenarios where A is preferred to B: ABC, ACB, CAB; and in two of these three, C is preferred to B. (2/3 corresponds quite well to what was seen in the experiments on monkeys.)
Can this trivial detail really have escaped the psychology community for half a century?
A large number of cognitive-dissonance experiments fall in another category, which is a bit trickier to analyse, but Chen does that in his working paper above and comes to a similar conclusion.
A final thought: if an otherwise rational psychologist does not admit the error, can we take that as evidence of cognitive dissonance?