Dealing with misfits in random treatment assignment Stata Journal 17, no. 3 (2017): 652-67 In this article, I discuss the "misfits" problem, a practical issue that arises in random treatment assignment whenever observations cannot be neatly distributed among treatments. I also introduce the randtreat command, which performs random assignment of unequal treatment fractions and provides several methods to deal with misfits.
Student Loans & Choices: Equilibrium Effects in the Higher Education Market
, with Christopher A. Neilson
Students loan programs are one of the leading policies for reducing the barriers to entry to higher education. There is growing evidence that access to credit is an effective way of promoting higher education and reducing the inequality in access. However, little has been said about how these massive demand-side subsidies might be affecting the supply-side of the market. This paper uses Chilean data on higher education institutions to analyze the extent to which the introduction of a large, government-backed student loan program potentially contributed to changes in the market structure of higher education. The proposed mechanism is that loans alleviate credit constraints, but consumers who are more sensitive to out-of-pocket price might also be less prepared and, importantly, less informed about the choices' net present value (NPV). In this scenario, introducing government-backed student loans might improve access to college, but in aggregate will make demand less sensitive to price and NPV, shifting incentives for higher education institutions.
Predicting ER Wait Times: Efficiency Gains for Hospitals and Patients The ability to accurately and reliably predict waiting times at walk-in hospital facilities can increase both patient satisfaction and hospital efficiency via a better management of patient flow. This paper studies the implementation of machine learning (ML) models to predict waiting times in the Emergency Room (ER) of the largest public hospital in Chile. Detailed administrative data on date and time, patient flow, and current and past examinations was provided by [Saltala](http://landing.saltala.com/), who developed a smartphone application for remote queuing. Several ML algorithms were evaluated to find the most accurate and useful prediction, including neural network, support vector machine, elastic net and multivariate adaptive regression splines. Elastic net performs best among a total of 8 explored models for predicting wait times, and the most important predictors are identified.
Inverse probability weighting for subgroup analysis in RD settings
, with Andre Cazor, Maria Paula Gerardino, Stephan Litschig, Dina Pomeranz
In this article we introduce the rddsga command, which allows to conduct binary subgroup analysis in regression discontinuity designs. Observations in each subgroup are weighted by the inverse of their conditional probabilities to belong to that subgroup, given a set of covariates. Analyzing the differential treatment effect in the re-weighted sample helps to isolate the difference due to the subgroup characteristic of interest from other observable dimensions. This methodology is illustrated with a real example based on Meyersson (2014). We find that the threshold-crossing effect is only significant for one of the subgroups, and that the difference between both subgroups is underestimated when not correcting by inverse probability score weighting.
Schedule Optimization and Teacher Allocation Frictions Teacher allocation ---both within and across schools--- is a major source of inefficiency in school districts.
The school teachers market is particular because teachers cannot freely supply any number of labor hours, because they also have to fit each school's schedule.
These schedules are determined via a costly process which is inflexible and produces sub-optimal results.
In this paper I investigate how inflexible scheduling of labor hours is an important source of frictions in the school market, which leads to inefficient assignment of teachers across school districts.