Monday, March 24, 2014

dubiously rational preferences, continued

So I've established that my preferences are definitely "behavioral", but possibly still technically rational. Maybe...

After a lot of agonizing over choices, I'm in fact pretty sure that my preferences are at least transitive, but only with a lot of intentional mental effort, which doesn't bode well for less thoroughly considered preferences. But hopefully lack of consideration is directly correlated with inconsequentiality, so that it's still rationalizeable.

The difficulty with transitivity, I'm pretty sure, has to do with comparing multiple attributes simultaneously. Small differences in one dimension are easy to ignore compared to single large differences, and that leads to cycles:

Hummus crackers or broccoli and cheese? That's easy, the latter is vastly more delicious, and they're both reasonably healthy and easy. Cheesy broccoli or ramen? That's easy, ramen you just stick in the microwave for a couple minutes and cheesy broccoli requires an actual pan on an actual stove. Hummus crackers or ramen? That's easy, hummus crackers are way healthier.

Cheesy broccoli > hummus crackers > ramen > cheesy broccoli.

But with additional mental effort, I can take multiple attributes into account simultaneously. Broccoli wins with 17 points. Trust me, this is much much harder with schools with a dozen attribute dimensions with much higher stakes...

But why is it so hard? If it's hard to combine multiple attributes it must be that they're not so intuitively fungible. But if that's true, why is it easy to discern small differences from big differences? Are we only gauging differences in a single dimension compared to the possible maximum difference in that dimension, rather than according to its actual influence on total overall utility? That seems about right, maybe.

Maybe having to make decisions that necessarily involve trade-offs along many dimensions is the only thing that forces us to assess overall utility, and all those dimensions really aren't fungible, and behavioral economics boils down to the problems with treating them like they are.


So after I wrote that, Eva (thanks for the link!) posted something extremely similar, and linked to this, which goes over some of the classic examples of nontransitivity, which I always found slightly unsatisfying since they rest on indifference. My example, if you were to translate it into that language, would be exactly the same as the ornament example, but I still don't like the requirement, necessary to generate a prediction, that certain items are indistinguishable on certain dimensions. I'm completely sure that hummus crackers are better than cheesy broccoli on both the health and convenience dimensions, but I ignore those small differences and focus on the one large difference instead. Salience (or something more fundamental that produces an illusion of salience) is the issue.


Sheng said...

You've seen Botond & Adam's paper on salience based decision weights? The intuition is pretty much exactly like in your example, that people over weigh attributes with large differences.

Interestingly, Matthew Rabin, Josh Schwartstein and Ben Bushong are working on their own paper on the same topic, but they make the exact opposite assumptions as Botond and Adam. Their intuition is something like diminishing sensitivity, so that the attributes with large contrast are under weighted. They end up with the exact opposite predictions as Botond & Adam, it's kind of funny.

Vera L. te Velde said...

hah that's funny! I didn't know about the new paper. Yeah I had something like the Szeidl Koszegi model in mind, or maybe Schleifer's salience stuff (been too long since I read either one to be sure which details are close). But vaguely hypothesizing that differences that are large relative to how large they could be are salient, rather than large differences according to some objective utility function of all attributes.