Household dynamics and female labor force participation: discussion edition

• Email
• Share
• Tweet
• Share
• Comment

In a context of lower labor force participation, does it matter who in a couple gets information about a job opportunity for the woman? What if the spouse doesn't know about this information? What would encouraging a discussion do?

The setting is India (Uttar Pradesh). Female labor force participation in the focus communities is quite low. Think about your answers to these questions. Maybe jot them down if you want.

The answers to these questions is the subject of a neat recent paper by Matt Lowe and Madeline McKelway. In addition to the usual economic work, they surveyed a bunch of intrahousehold-research experts about their priors against which Lowe and McKelway benchmark their results. (I'll pause here in case now you'd like to write down your answers)

Let's start with the set up: A large carpet producer is opening up a bunch of new workshops. They have job opportunities (which include four months of paid training) for women. Lowe and McKelway visit the communities and, first of all, randomize whether the ticket for enrolling for one of these slots is given to the man or the woman (So: initial private information, but it's worth noting at the time of actually enrolling both of them have to show up).

Lowe and McKelway then cross-randomize the information folks have: in some cases the spouse who didn't get the ticket is given no information (noinfo), in some the other spouse is told (info) and in others the couple is both given the information about the ticket and they are encouraged to have a short discussion (discuss). And 10 percent of households don't get a ticket, so folks in the noinfo group can plausibly deny getting a ticket.

Let's start with the info effect. Relative to the spouse having no information, providing information to the spouse who did not get the ticket has no significant impact on enrollment for the job. Indeed, the point estimate is negative and pretty much the same whether the husband or the wife gets the ticket. And the effects are again pretty much the same where the person who got the ticket was less supportive of women working than the non-ticket spouse.

Overall, information is likely to be passed around - in the noinfo condition, more than 70 percent of the spouses who did not get the ticket know about it. The info intervention bumps this up by about 10 percentage points but, as we just saw, this doesn't translate into enrollment.

Discussion, on the other, significantly lowers enrollment (yes, I got that prediction wrong but, hey, so did almost all of the experts). Relative to noinfo, those who discuss are 9 percentage points less likely to enroll, and this is 6 percentage points less than those in the info group. Given that enrollment in the noninfo group is 18 percentage points, these are large effects.

So bottom line: it doesn't seem to matter if the husband or wife gets the ticket (in the noninfo group there is no significant difference). Information doesn't matter. And discussion makes it less likely.

In addition to proving the predictions of a bunch of experts wrong, this also goes against some standard theory. So what could be going on?

Maybe folks are actually arguing when they have a discussion. Lowe and McKelway rule this out both from enumerators' (who observed these discussions) reports and the fact that those participants aren't more likely to report disagreement in a follow-up survey some days later. They also rule out a bunch of other possible explanations.

Instead, Lowe and McKelway offer some suggestive evidence for an alternate view: "(i) intra-household decision rights are not clearly delineated, such that the ticket-holder has de facto power to make the decision in noinfo, and (ii) interventions that make decision-making more joint, like our info and discuss treatments, give both spouses veto power." To support this, they start with the fact that a lot of couples ex ante disagree over whether a woman should be a weaver (58 percent in fact). Then, when they look at effects of info or discuss, they are zero for those who agree, but reduce enrollment by 11 percentage points (significant at 10 percent) for those who disagree. Lowe and McKelway do acknowledge that this doesn't totally nail things down - it's still not clear why discuss delivers lower enrollment than info. Nonetheless, this is an interesting idea to work with and gives us a more nuanced and perhaps fluid view of within household power than our standard models.

Before wrapping up, let's get back to the experts. Lowe and McKelway sent out a survey to a bunch of folks who had done research and published on intrahousehold decision making. First of all, response rates weren't great: 25 percent completed the survey and 19 percent offered a set of complete predictions on what the results would be. But that's still 70 people. And those predictions were wrong. Lowe and McKelway go on to analyze inaccuracy by rank and Google scholar citations. Care to make one last prediction?

Yup, full professors and those with more Google scholar citations were less accurate.