domingo, junio 21, 2009

Extensions to the marshmallow experiment

First of all, if you don't know what the marshmallow experiment is you can find it out on this wikipedia entry or through this hilarious video from Joachim de Posada. For the few who found the topic interesting, I'd recommend you this great article from The New Yorker too (HT to TBP).

That being said, here are some proposed extensions to the marshmallow experiment that I'd love to see (or do) some day. Any new ideas are welcome in the comments.

  1. Try different candies!
    I just noted there are two strong underlying assumptions here: i) all kids like marshmallows; ii) all kids like them in the same way. I know is almost impossible to find kids who doesn't like candies, but I'm pretty sure is less improbable to find kids who does not like marshmallows as much as average kids. What if some of the kids "delayed" just because they didn't like that specific candy as much as others do? Couldn't that be a source of bias?

    A solution may be to show different candies at first, rank kids' preferences (or even measure them in some way), and then propose the deal. In other words, control for preferences' intensity.

  2. Try more candy!
    This works as a solution to the problem described in 1., but also as a way to see if it's important to care about it. It would be interesting to see what would have happened if all the kids were offered 5 additional marshmallows instead of just one. Or maybe what if some kids had been randomly offered between 1-5 additional ones? Would results hold?

    By doing this, we could find if kids' ability to delay gratification depends on how they value the source of gratification. By common sense we would expect that the more they value marshmallows, the less they'll delay eating them; that having less to do with self-control abilities.

  3. Define the level of uncertainty
    I'm not sure if kids were heavily assured they'll have the second marshmallow in the deal. What if some kids thought there was a chance of getting nothing at all if they didn't ate it as soon as they can? Then different levels of risk aversion would have done the rest. ("there might be some trick here... I'd rather play sure!! ...glup!).

  4. Explain, then repeat the experiment
    Try doing the experiment one time, then explaining the kids what they just did and put them on trial again. This time set a longer awaiting time (let's say 25 minutes or even half hour). Would kids learn self-control? Can they?

  5. Not if, but when will they eat it
    It would be fascinating to see some survival analysis techniques applied to answer this question. Among those who ate the marshmallow before the adult comes back, what determines that some ate at minute 1 and some others at minute 14? Would it be the same factors that explain why they finally ate it (or at last the same factors but in the same way?)?

Well, that's pretty much it. The new york article blew my mind with more ideas, but I think I'll take a couple of days reflecting about them before putting them here.

sábado, febrero 14, 2009

Perfil de Roland Fryer

"He is a scientist, he explains, devoted to squeezing truths from the data, wherever that may lead"

Toward a Unified Theory of Black America, NYT 03/2005