Let's see if I can take your college example and fit it to what Freakonics is investigating.

Before you roll the dice, you are asked how confident you are that if the dice roll 6, you will in fact enroll and pay the first semester's tuition at school X and still be attending classes there two months from now. You can choose from:

(a) Very likely

(b) Somewhat likely

(c) Somewhat unlikely

(d) Very unlikely

Then you're asked to give a probability estimate that you will not show up, pay up, and stick it out for two months.

Let's say you're highly motivated to do school and all three school choices are equally wonderful to you. But you don't have the tuition money and all three schools have turned you down for a scholarship. You are determined to work your way through school, but you know that the odds are against you being able to work full time and go to school full time at the same time.

So you estimate the odds against paying the first chunk of tuition and carrying a full load of classes and performing well enough to keep your job at 75%. You know it's going to be pretty damned hard.

All the same, you are confident that you are more likely than not to succeed anyway. You pick "somewhat likely" as your confidence of success.

These two estimates are logically incongruent. What interested me about the Freakonomics study is that the software challenged me on the mismatch. It popped up a dialog that said, in effect, you said you're more likely to succeed than not, but you estimated a 75% chance of failure. It sounds like you've already decided to quit. Are you sure you want to roll the dice?

You can end it there or go on with the roll.

Now doesn't that challenge change the whole feel of the decision for you? It sure does for me.