Experimental Design Details
After the couple is introduced to and practices the task, each spouse is asked in turn to make, privately, a certain number of draws with replacement from the urn. They are then asked, still in private, to guess how many red balls there are in the urn. Next, we bring the two spouses together and ask them to discuss what they saw and make a joint guess. Finally, we separate the spouses and elicit once again their private prediction (which may differ from the joint guess). To incentivise accurate guessing, we randomly select at the end of the experiment one of the guesses (individual or joint) that have been made, and pay couples according to the deviation of this guess from the true number of red balls. Couples stand to earn up to 210 rupees (about $3) from guessing accurately -- a significant amount for this population.
The key treatment condition is the number of draws from the urn each spouse has. We choose this number randomly and independently for each spouse, and make it clear to them that we choose it in this way. This creates random variation in which spouse is relatively better-informed.
In other rounds, we will extend the basic design to identify the mechanisms behind inefficiencies and asymmetries in aggregation. In particular, we consider the following mechanisms:
Do individuals do just as poorly if they privately receive all information themselves, suggesting that the underlying issue is not specific to social learning? We test this by having individuals play rounds which mimic the basic design, but in which they personally draw balls from the urn twice, instead of relying on a discussion with their spouse for the second set of information.
Are failures in efficient information aggregation explained entirely by a failure to share information in the first place? Or is information shared by others under-/over-weighted (conditional on being shared)? We test this in two ways: first, by recording the conversation during the free-form discussion section of the main experiment, and by contrasting performance in the free-discussion rounds with rounds in which we directly share information between spouses.
Are failures to share information or appropriately weight information provided by one’s spouse explained by (potentially inaccurate) beliefs about relative competence? We test this by directly eliciting each individual’s beliefs about the other’s competence using incentivized survey questions.
Alternatively, do individuals simply generically place greater weight while learning on their own experiences compared to the experiences of others? We see this as a very plausible residual explanation: that people may generally find their own experiences to be far more salient than others’ experiences, and may thus over-weight their own experiences. We test this by estimating how people update their private guess when told the information of their spouse (or paired individual).
In the case of disagreement between spouses, is the husband’s information weighted more heavily in the joint decision? Bayesian individuals should converge on the same answer after sharing information, but we will test this by allowing each spouse to make a further private guess after they have discussed and made a joint guess about the number of balls. If the two spouses disagree, we can infer the bargaining weights implicitly placed on each of their guesses in determining the joint household guess. We can then test whether who acts as the (more) dominant decision maker in the household is related to (actual or perceived) relative competence.
Are the failures of information aggregation we study specific to learning within the household? Or might such failures occur more generally between any two individuals of different gender in the Indian context, or even between individuals of the same gender? We study this by repeating the basic design of the experiment with same vs. mixed-gender teams comprised of strangers.