My wife, on occasion, has indicated that I have not paid attention to what she has said. Whether or not I might not be alone in this behavior is the subject of a fascinating new paper by John Conlon, Malavika Mani, Gautam Rao, Matthew Ridley and Frank Schilbach.
Let’s start with the setup. They recruit a bunch of couples in Chennai. They also recruit a bunch of strangers of different sexes. And they recruit them to play a prediction game. Participants are faced with an urn which contains some combination of red and white balls for a total of 20 (with red ranging between 4 and 16).
Initially, there are two rounds (in random order). In the individual round, individuals privately draw (with replacement) 1,5, or 9 balls (randomized). They make a guess on the number of red balls. Then they draw again and guess again. This is the base case.
In the discussion round, individuals make a set of private draws and guesses. Then the couple holds a face-to-face discussion and makes a joint guess. And then they each make a private guess. This is what we’ll compare to the guess from the individual round.
Conlon and co. want to deal with issues of information frictions so following these two rounds they throw some complications in the form of additional rounds into the mix (again the order of this set is randomized). First up is “draw-sharing” where the experimenter replaces the discussion with the experimenter telling the participant what their spouse drew, followed by a private guess and then the rest of the discussion round. Next is “guess-sharing” which is the same except that the experimenter tells the participants what their spouse guessed and how many draws this was based on. Finally, they repeat the experiments but with strangers instead of their spouse.
There are real incentives here: the closer you are to the right number (from one of the couples’ guesses chosen at random) the higher the payout. The maximum payout is worth about $3, versus a daily wage of about $5 and you lose about 40 cents for every ball you are off.
OK, so what happened? First of all, people do update! As Conlon and co. describe it: “participants’ performance lies roughly halfway between the random guesser and the risk-neutral Bayesian.” And men and women (in the individual game) perform the same. Interestingly, when asked to predict performance, both men and women overestimate their own performance, and women go further by overestimating the quality of their spouse (whereas men get it right by saying their wives are as good as they are).
Now what happens when they have a discussion? Women learn from their husbands at the same rate as if they had drawn the balls themselves. Men, on the other hand, discount their wives’ information by about 60 percent. This pattern repeats even if the experimenter is the one who tells him the information (the draw-sharing round). So, it’s not the wife talking per se – its where the information comes from. And this difference between husbands and wives in behavior doesn’t go down when controlling for a pretty diverse set of covariates. All of this costs them: the men earn significantly less due to the fact that they’re not listening to their wives’ information (particularly when she has rich information).
So, is this about male reactions to women or just to wives? When they play with strangers both men and women discount their (stranger) partner’s information. In the discussion round, men look a lot like when they are playing with their wives. Women also discount partner information; although these results are somewhat weaker, they aren’t significantly different from men. And there isn’t a significant difference if they are playing in a mixed-sex pair or a single-sex pair for either men or women.
OK, so within households this seems to be a men treating-wives-like-strangers thing (or a women trusting their husbands thing). Until now, we’ve been focusing on private decisions. What happens when they have to put their heads together and make a joint decision? Men finally get it. Here they put equal weight on men and women’s information and, what’s more, they (and their wives) weight more heavily towards the information of the spouse who is better at the task (both as objectively and subjectively measured).
Conlon and co. do a fair amount to rule out alternate explanations (e.g. risk aversion, trust in the experimenter). They also run a version of the experiment (with the strangers) where folks actually observe their partners drawing the ball out. While this might put a small dent in the discount for partners information (significant at 10 percent), folks still underweight this information.
All in all, these results suggest that joint decisions, in the context of discussions, can get folks to better outcomes. I’ll have more to say about this in a future blog – we’re working on this in the context of farm planning. And we know that joint decision making is often correlated with better outcomes for kids and lower levels of intimate partner violence (e.g. this recent paper). Conlon and co. do a really neat job of showing us some insight into the mechanisms behind these kind of outcomes.
Oh, and the title of this blog? Credit goes to my wife.