This note shows how big data can helpcombine experimental with non-experimental approaches inimpact evaluations when take-up is low. In this study,author have access to a large administrative data set (of660 MB), which follows the monthly financial indicators ofeach client for up to 18 months prior to the interventionand 6 months after it. Moreover, from the experimentalapproach their also had a large pool of clients randomlyassigned to the control group. This data enables us toobtain credible estimates by combining the experiment withtwo non-experimental approaches. Their first use propensityscore matching to find, among the clients in the controlgroup, a subset of clients that best mimics thepre-intervention financial trajectories of clients in thetreatment group that received treatment. The effects of theworkshops on the treated clients are summarized. Under ourpreferred specification, the author finds that participatingin the workshop increases by 11 percentage points thelikelihood of paying more than the minimum payment, andreduces by 3.4 percentage points the likelihood of delayingpayment. Monthly credit card spending increases by 63.7percent, and the likelihood of owning a deposit account withour partner bank also increases by 2.7 percentage points.The two financial education interventions help clients reachthe minimum payment and pay their bills on time more often,without reducing their credit card spending. Bothinterventions increase the likelihood that clients areprofitable for the bank.
PDF
download
878987797
028-85220240
