Instrumental Variables
The USAID-funded MEASURE Evaluation project is hosting a series of webinar discussions of the popular MEASURE Evaluation manual, How Do We Know If a Program Made a Difference? A Guide to Statistical Methods for Program Impact Evaluation (Lance P, Guilkey D, Hattori A, Angeles G, 2014). Each webinar in the series reviews key topics from a chapter through verbal discussion and graphical presentation. The webinar series will enable participants to understand the resources offered in the manual—a learning tool for use of methods to estimate program impact. The series also provides stand-alone training tools for the topics covered. MEASURE Evaluation’s goal with these webinars is to provide a highly interactive learning opportunity to participants (please ask questions!).
The fifth and final webinar in the series, “Instrumental Variables,” will take place May 11 at 10am EDT. This webinar will consider impact evaluation estimation when program participation is non-randomly determined by the observed and unobserved, frustrating straightforward estimation of impact by simple comparison of average outcomes between samples of participants and non-participants. The instrumental variables approach appeals to the notion that, despite the overall non-randomness of program participation, there might be variables, called instruments, that offer experimental channels of program participation within which participation is effectively randomly determined. For instance, in a randomized control trial with incomplete adherence to experimental assignment, that original experimental assignment is a kind of instrument. This analogy also highlights one of the complexities of this method: the instrumental variables method identifies a local average treatment effect—which is program impact among those who comply with the experimental assignment implied by their value for the instrument. We will conclude with a discussion of one particularly popular recent application of instrumental variables, the “fuzzy” regression discontinuity design.
The first webinar in the series, held March 31, 2016, was entitled “Fundamentals of Program Impact Evaluation.” It addressed the basic challenges of program impact estimation. The second webinar, held on June 29, 2016, and entitled “Randomization and Its Discontents,” considered program impact estimation based on randomization of program participation status (typically though a randomized controlled trial). The third webinar in the series, entitled “Selection on Observables” and held on September 13, 2016, considered impact evaluation methods that assumed we can observe all factors shaping both the participation decision and the outcome of interest. “Within Estimators,” the fourth webinar held on December 15, 2016, considered methods of impact estimation that assume that the outcome of interest and non-random program participation is driven by a fixed unobserved characteristic. All of these earlier webinars can all now be viewed online. This final remaining webinar will discuss instrumental variables methods.
