Question
Choice and Choices: What is the best way to order a list of options for a decision if you are optimizing for the satisfaction of the decider?
Answer
If your confidence in your prediction (either predicted #1 preference, or a complete predicted ordering) is low I would suggest a pseudo-randomized list, where the best and worst options are inserted somewhere in the middle of the list. The rest can be in any order.
This is for exactly the reason you mention: anchoring bias (called primacy and recency effects for the beginning and end biases, or more generally, a serial position effect: http://en.wikipedia.org/wiki/Ser...)
If on the other hand, your confidence is high, present the ordered information.
Why? The satisfaction of a bounded-rational decision-maker will be a function of decision cost and decision quality. Decision cost is basically processing time in this case.
Decision quality is of course, how close to "best" the choice is. Taking too long to pick #1 is just as bad as being rushed to process a random list and picking #n by accident.
The one asset you have to design the decision is the ordering prediction. The higher the confidence in that prediction, the more you can lower decision cost. You cannot actually increase decision quality, since the person can still make the wrong choice in terms of future regret due to a sudden impulse or something. So canceling out the one known bias: serial bias, you present as much of the prediction information as your confidence allows.
So the general answer is: trade-off randomness versus use of prediction, modulo cancellation of known systematic biases.
The empirical unknown here is the threshold for showing #1 on top (i.e. taking advantage of the primacy bias to speed up the selection of a high-confidence correct answer). You have to run actual experiments to measure that kind of thing.
There is also a epistemological problem here. The assumption that the decision-maker has something like a static preference order is problematic. What does that mean? Should you ask the person to rationally take their time when they are fresh and rested, and come up with an ordering and use that as ground truth? Should you measure regret rates later? How much later? Tomorrow? Next week? Deathbed? Is their inhibition-free ordering when drunk the real best ordering?
Behavioral economists solve this problem by assuming that there is a ground-truth mental model within which a "rational" solution exists once you strip away the known "predictable irrationalities." I disagree philosophically with this view. There is no such ground truth state. Stripping away known predictable irrationalities still leaves you with the problem of justifying what's left behind.
My own solution is to assume that all optimality is relative to a mental model or narrative. For a goody-two-shoes type, the preference order he comes up with when sober and rested is the right one that will move his life according to script. For an artistic type, the preference order he comes up with when drunk might be the right one. For a Nietzsche type, the preference order that leads to the most future self-destructive misery might be the right one.
For a reflective-hedonist type, future regret minimization might be the way to test for correctness. For an in-the-moment type, immediate gratification following the decision might be the right one (to claim that deferred gratification is the "right" way is to pretend that a given social script is "truth"). For a cerebral thinker, some NPV type function of regret across the rest of life (an integral cost function) might be the right one.
There is a lot to a list of options.
This is for exactly the reason you mention: anchoring bias (called primacy and recency effects for the beginning and end biases, or more generally, a serial position effect: http://en.wikipedia.org/wiki/Ser...)
If on the other hand, your confidence is high, present the ordered information.
Why? The satisfaction of a bounded-rational decision-maker will be a function of decision cost and decision quality. Decision cost is basically processing time in this case.
Decision quality is of course, how close to "best" the choice is. Taking too long to pick #1 is just as bad as being rushed to process a random list and picking #n by accident.
The one asset you have to design the decision is the ordering prediction. The higher the confidence in that prediction, the more you can lower decision cost. You cannot actually increase decision quality, since the person can still make the wrong choice in terms of future regret due to a sudden impulse or something. So canceling out the one known bias: serial bias, you present as much of the prediction information as your confidence allows.
So the general answer is: trade-off randomness versus use of prediction, modulo cancellation of known systematic biases.
The empirical unknown here is the threshold for showing #1 on top (i.e. taking advantage of the primacy bias to speed up the selection of a high-confidence correct answer). You have to run actual experiments to measure that kind of thing.
There is also a epistemological problem here. The assumption that the decision-maker has something like a static preference order is problematic. What does that mean? Should you ask the person to rationally take their time when they are fresh and rested, and come up with an ordering and use that as ground truth? Should you measure regret rates later? How much later? Tomorrow? Next week? Deathbed? Is their inhibition-free ordering when drunk the real best ordering?
Behavioral economists solve this problem by assuming that there is a ground-truth mental model within which a "rational" solution exists once you strip away the known "predictable irrationalities." I disagree philosophically with this view. There is no such ground truth state. Stripping away known predictable irrationalities still leaves you with the problem of justifying what's left behind.
My own solution is to assume that all optimality is relative to a mental model or narrative. For a goody-two-shoes type, the preference order he comes up with when sober and rested is the right one that will move his life according to script. For an artistic type, the preference order he comes up with when drunk might be the right one. For a Nietzsche type, the preference order that leads to the most future self-destructive misery might be the right one.
For a reflective-hedonist type, future regret minimization might be the way to test for correctness. For an in-the-moment type, immediate gratification following the decision might be the right one (to claim that deferred gratification is the "right" way is to pretend that a given social script is "truth"). For a cerebral thinker, some NPV type function of regret across the rest of life (an integral cost function) might be the right one.
There is a lot to a list of options.