I think a good number of people are familiar, even without knowing what it is specifically about, with the classic example:
Linda is a shop assistant in a department store
Linda is a shop assistant in a department store AND she is active in the feminist movement'.
Where, as a rule, people respond to the second option as being more likely than the first. This fallacy is quite persistent, and is discussed in more detail in Kahneman & Tversky's paper 'Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment'.
Another cognitive scientist (whom I greatly appreciate), named Gerd Gigerenzer, tries to show that the heuristics and biases approach to analysing human reasoning tends to be flawed but, unfortunately, he is wrong.
How it is wrong is easier explained than understood. Before you continue reading, I recommend you read, in order:
Frequentist Probability
Classical Probability
Bayesian Probability (subjective)
Propensity Probability
Dutch Book
Now, if you have read:
Your thesis tends to be wrong because it is based solely on the frequentist approach:Probabilities cannot be assigned to single events or situations of subjective uncertainty, so that people cannot be wrong in test scenarios.
The fundamental problem I think is the way the question is presented, try it yourself:
I tell you that 'Linda is a shop assistant in a department store
Linda is a shop assistant in a department store AND she is active in the feminist movement', and the answer should (mostly) be the one above. But if I said:
"400 people fit Linda's profile. How many of them are:
shop assistants in a department store
shop assistants in a department store and active in the feminist movement', what would you answer?
You would definitely give the first answer as more likely than the second. And you would be right.
In the first example, the probability is presented subjectively, in an isolated case. In the second, the question is posed as frequentist.
In the first case, 85% of people answer by adding characteristics (it can be feminist movement or cheerleading club, it doesn't matter as long as it is a plausible possibility), but in the case where the pool widens the error drops to 22%.
The simple rephrasing into two simple views substantially changes the accuracy of a judgement, curbing the possibility of error.