Does Education Really Cause Domestic Violence? Replication and Reappraisal of "For Better or For Worse? Education and the Prevalence of Domestic Violence in Turkey"

Using the 2008 Turkish National Survey of Domestic Violence against Women, Erten and Keskin (2018, henceforth EK), published in AEJ–Applied Economics, find that women's education increases the psychological violence and financial control behavior that they face from their partners. The authors also claim that the incidence of financial control behavior rises because women become more likely to be employed—supporting the instrumental violence hypothesis. They present this evidence only for women who live in what they call "rural areas during childhood". EK's findings are an artifact of the way the authors create two key variables: the variable that classifies women into rural vs. urban childhood location and the variable measuring financial control behavior. EK misclassify the variable on childhood rural status. We find that once this variable is defined properly, the evidence for all their findings vanishes. EK make use of two of the three variables related to financial control behavior in the dataset. We show that using all three variables—or any other combination of two of the three variables—generates no evidence of a policy effect on financial control behavior. Even after ignoring these problems, the evidence EK provide in their paper is highly specification sensitive and the standard checks of the continuity assumption in RDD fail for their key outcomes. Moreover, the results obtained from the analysis of urban areas—not provided in their paper—are inconsistent with the instrumental violence hypothesis. In addition, EK claim—using RDD graphs with high-order polynomials but no estimation results—that the policy has no effect on men's schooling, contrary to the findings of the previous literature. However, we show a clear and substantial policy effect on men's schooling, resulting in the failure of their exclusion restriction assumption.