What if I thought everyone around me was more regressive than they actually were? And then what if you told me I was wrong?
That’s the experiment in a recentby Leonardo Bursztyn, Alessandra Gonzalez and David Yanagizawa-Drott. They are working in Saudi Arabia, where female labor force participation is 18 percent. They recruit 500 married Saudi young men in groups of 30 from different neighborhoods – so guys come with a set of people where they will already know some of the folks in the room.
Once participants show up, Bursztyn and co. have them take a survey on their phones. The survey gets demographics and then asks whether the participant agrees or disagrees with a number of statements about the labor market. Topics include things like unemployment insurance and the minimum wage. But the one Bursztyn and co. are really interested in is: “in my opinion, women should be allowed to work outside of the home.” Then participants are asked to guess what fraction of the other participants agree with the various labor market statements – including female labor force participation (with a $20 Amazon gift card for the closest guess).
Result #1: 87 percent of participants think it is ok for women to work outside of the home. When asked about the views of others though, around 70 percent of folks underestimate their peers – guessing that 63% agree with this statement on average. So: individuals think their neighbors are more restrictive than they actually are.
Next, Bursztyn and co. assign half of the respondents to get information that lets them see the average of what others actually said. The other half get nothing. Everyone is then provided with information about a Saudi startup that helps connect Saudi women looking for a job with potential employers. Participants are then asked to choose between a $5 Amazon gift card or the ability to sign their wife up for this service for free.
Result #2: 23 percent of the folks who get no information about social norms register their wives for the job match service. Those who get information opt for this at a significantly higher 32 percent – and this effect is stronger among those who underestimated their neighbors.
OK, but does this translate into actual labor market outcomes? To answer this, Bursztyn and co. follow up with participants by phone three to five months later. Now the respondents get a more detailed (and more obvious) survey about their wife’s job seeking and employment outcomes. They also check in on the participants’ views on what others in the neighborhood think about women working outside of the home to see if any of the updated information stuck.
Result 3: Indeed, the information stuck. Those who got the information about other participants report significantly higher pro-female labor force participation views of a randomly selected 30 people in their neighborhood than those who did not get the information.
Result 4: What about her? Wives of men who got the updated information on social views are significantly more likely to have applied for a job outside of the home: 16.2 percent of them do, relative to 5.8 percent in the control group. And these women have more interviews: 5.8 percent of them do, relative to 1.1 percent in the no-information/control group. Employment outcomes also seem to move – but here Bursztyn and co. are underpowered and the result is not statistically significant. But there is a bonus result: Bursztyn and co. find that men who got the information are significantly more likely to indicate that they would be willing to sign their wives up for driving lessons – so there may in fact be spillovers to other dimensions.
All in all, this is a neat paper. Bursztyn and co. are quite careful about how they set things up and I learned a couple of things about how to set up this kind of experiment.
First, they are really careful about anonymity, in order to make respondents feel as comfortable as possible giving truthful answers. Indeed, they identify respondents by the last 3 digits of their phone number and session number, and then use this to match them through to the follow up survey. They also make sure there are no westerners present when respondents are answering the questions in order to avoid potentially biasing the answers.
Second, they pay attention to measurement. This includes using list experiments (see Berk’s It also includes incentivizing folks to think hard about what they really think that their neighbors think.here for some discussion of these).
Finally, they devote a significant amount of discussion to external validity – using not just one, but two other surveys in Saudi Arabia to show us that the norms about women working that they observe aren’t crazily out of line.
Coming back to the results, it would be interesting to explore which are the norms which we systematically overestimate as more restrictive than they actually are. And what kind of interventions can start moving the needle. And can we keep the needle moving beyond the community average?