I had an interesting discussion with my mom today about a new working paper on arranged marriage. The basic finding in the paper is that, among a large sample of families in Mumbai, India, men who find their partners via arranged marriage are much less likely to marry college-educated and/or working women. The paper makes a valiant attempt to show that this is being driven by arranged marriage specific parental preferences rather than omitted variable bias.
My mom, an extremely intelligent person who, like all mothers, "knows best", had the following comments:
1) I am not sure whether this finding would hold up in other parts of India.
2) I don't buy the finding because I know so and so, and they married someone highly educated through an arranged marriage process.
My mom makes a valid statistical argument with the first point: underlying parameters may indeed differ across populations. There is no way to know whether these findings replicate in Chennai without getting data and giving it a go. Go mom.
Its the second point that interests me more. Regression coefficients basically give you some average effect, conditional on whatever else you control for. Put less loosely, the regression model in this paper tells you the expected change in spousal education for a given change in type of marriage holding constant family background, income, etc. While any handful of individuals might behave contrary to the prediction, these folks are averaged out the the large number of people who follow rank-and-file.
What's interesting is how my mom completely rejected the average effect based on her experience with a few relatives of ours. I think this is an extremely common phenomenon and speaks towards to power of personal experience in shaping expectations and behavior. No doubt, behavioral economics probably has a lot to say about this kind of thinking (which I am prone to on almost a daily basis!).
The whole episode reminded me of a comment made by Yale School of Management behavioral economist Keith Chen in our departmental seminar. Paraphrasing a bit, his point was that people are much more likely to rely on their next door neighbor's experience in making a decision to buy, say, a car rather than Consumer Reports. This is irrational since the former represents a single draw from a distribution of reviews, whereas the latter gives the expected value of this underlying distribution.
3 comments:
Hi Atheendar,
I'd call this an example of the "law of small numbers" bias. The idea of this bias is that people think that small samples will represent the underlying population much more than they actually do on average.
You point out an interesting application of the law of small numbers bias; people will seek out too little information because they falsely over-estimate how well their small amount of information represents the population they're interested in.
Thanks Santosh!
I figured this kind of bias would have been explored many times, but didn't realize it had a name. Is there any work on the consequences of this bias as far as mistakes in decision making and welfare? Also, are there ways to intervene with policy? (Though I can't imagine much beyond marketing the existing information better...)
This type of thing is frequently discussed in Overcoming Bias. If you guys didn't know about it already, I think it's worth a read.
Post a Comment