Weekly links June 29: cash does more good things, interpreting IHS, technical assistance to banks increased credit, and more...

• Dan Kopf at Quartz has a nice summary piece on “the transformative power of giving young women cash” – covering work in Malawi (by Berk and co-authors), in South Africa, in Bangladesh, and a forthcoming WBRO paper on cash transfers and intimate partner violence that overviews 14 studies “eleven of these studies found a reduction in domestic violence, two found no change, and only one found an increase”.
• Scott Guggenheim offers his response to/critique of the recent 3ie evaluation of CDD programs on the From Poverty to Power blog.
• In a new Finance & PSD impact note, Miriam Bruhn, Rekha Reddy and Claudia Ruiz summarize work they have done in Mexico that uses matched diff-in-diff to evaluate the impact of offering technical assistance to rural banks -  “technical assistance allowed rural credit unions to increase their operating efficiency and reduce their non-performing loans ratio. Part of these gains translated into higher returns for the credit unions, but they were also passed on to the final borrowers in the form of more credit at lower lending interest rates.” We are now up to 50 of these 2-page summaries of finance and private sector impact evaluations.

• On VoxDev, Feng, Lagakos and Rauch on how unemployment increases with development.
• Dave Evans shares slide decks on How to influence research with policy and How to get the most out of education impact evaluations.
• Marc Bellemare has a teaser post on elasticities and the inverse hyperbolic sine transformation – he advertises a new working paper he has on this, but doesn’t explain in the post what his results are.  He has results for several cases, but the most common case of interest for treatment effect estimation is likely to be a regression of the form:                Arcsinh(y) = a + b*Treatment + e

              The question is then how to interpret the estimated coefficient b. The nice result for us is that “standard logarithmic adjustments for semi-elasticities in arcsinh-linear equations with dummy variables can be used with little error for dependent variables with untransformed means roughly greater than 10.”. That is, for y large enough (he says y>10 as a guideline), then the just like in a log-linear specification, the percent change in y due to treatment changing from 0 to 1 is approximately exp(b) – 1.
 
Note that Development Impact will be on break next week for the 4^th of July. See you back on July 9.