After talking about domestic violence measurement  and the need for some kind of model when you think about things like domestic violence with Toan last week , this week I look at a new paper  from Jonas Hort and Espen Villanger which both asks the question carefully and definitely makes me think hard about what the right model might be.
They have a great experiment. The flower industry in Ethiopia is growing rapidly and “because supervisors were already allocating job offers relatively arbitrarily when approached by the researchers, explicit randomization was a modest procedural change.” (OK, these are unskilled jobs, but if only more of research design was a modest procedural change). Now for some reason, these flower farms only employ women. So this gives them a chance to see what happens when women get a job.
And this job turns out to be a good one relative to the options that exist in these parts of Ethiopia. For the women who got the job, this represented a 154% increase on average (and a 28% increase in overall household income).
Now domestic violence, by a number of measures, is very high in Ethiopia. In this population of flower-job applicants, the baseline value was on average 0.96 violent incidents (e.g. pushes, slaps, punches or sexual assault) a month. For the women who get the jobs, this gets significantly worse: they are 13% more likely to experience physical abuse, 34% more likely to experience emotional abuse, and the number of violence incidents experience per month goes up by 32%. So these jobs clearly generate more domestic violence in the household.
Now, there are models out there in which an increase in a woman’s earnings is a good thing. It can increase her bargaining power, leading to her preferences playing a more significant role in household decisions. But the results here aren’t consistent with these. Nor does the increase in violence seem to be consistent with the husband using violence to get more of her money. A couple of additional results make it clear that things aren’t quite simple. First, there is no correlation at baseline between domestic violence and women saying they had money that they alone could decide how to spend. Second, the fraction of women who do control some of their own money does go up as a result of employment – from 37 to 47 percent. And these women are (again) not more likely to be abused. Third, husbands don’t seem to be turning to violence as a way to increase bargaining power: when economic independence (measured as a woman who says that she could divorce her spouse and support her family on her own) is interacted with employment, the interaction isn’t significant.
What is significant, though, is the interaction of employment with wife’s income as a share of household income at baseline. Citing arguments from sociologists that domestic violence is about men reinstating control over their wives, the authors argue that these results are consistent with a story that the improvement in the relative economic standing of women generates an emotional cost for their husbands, and the violence is the way men act on these costs.
In the end, these results are startling and disturbing enough to indicate a need for replication in other contexts (and they do echo results for some types of policy interventions). They also add significantly to the literature that argues when policy (or the movement of the economy) are going to result in a significant shift in gender roles/within household economic power, we need to think about policy to deal with the “frictions” that will result.