← volver

Ciclo seminarios Management Science

23Ago

13:30 horas, sala 31, República 701, 3er piso (entrada por Domeyko 2338, Santiago)

Invitado: Juan Pablo Vielma, Profesor Asociado de MIT Sloan School of Management

Título de la conferencia: “Using Mixed Integer Programming to Build a Representative Matched Sample with Treatment Doses”

Abstract

Matching procedures provide a non-parametric adjustment method which has many advantages, such as comparing observations that are alike while maintaining the units of analysis intact. However, as many other estimation strategies, traditional matching often provides estimates of local treatment effects for a particular sample, which might not be useful from a policy perspective. In order to address the representativeness issue and inform policy, we propose the use of template matching to estimate causal effects for a target population of interest. Additionally, this method allows us to expand matching from a binary treatment to matching with doses, balancing the distribution of covariates by construction. Applying standard graph-matching-based formulations to model the associated combinatorial problem results on an extremely large mixed integer programming problem (MIP) that is intractable both in theory and practice. Fortunately, by applying known results for the perfectly matchable subgraph polytope we are able to significantly reduce the size of the formulation without any loss of its strength. In particular, we show that the resulting formulation is integral or ideal when only two covariates are balanced and is tractable in practice for the general case.

We illustrate the representative matching procedure by estimating the effect that the different intensities of the 2010 Chilean earthquake had on educational outcomes. Using data from the Chilean Education Quality Measurement System (SIMCE) for 2008 as our baseline for matching, we find a negative effect of the earthquake on school attendance for a representative sample of senior high school students in Chile, where individuals who suffered the most intense earthquake missed on average 14% more of regular classes than students who did not suffer the earthquake or experienced it at a very low intensity. However, we find no significant effect of the earthquake on the probability of registering for the University Selection Tests (PSU) or taking the PSU, and further, no significant impact of this natural disaster on PSU scores.

This is joint work with Magdalena Bennett and Jose Zubizarreta.

Organizan: Ingeniería Industrial U. Chile e Instituto Sistemas Complejos de Ingeniería (ISCI)

Contacto: Pamela Tapia / ptapia@dii.uchile.cl

Confirmar asistencia AQUÍ