Weili Ding: Estimating Context Independent
time: 2015/12/6 views 710

Estimating Context Independent Treatment Effects in Education Experiments

Empirical evaluation researches identify the economic and social effects resulted from policies or programs. However, its methodology depends on some pretty critical assumptions which might not be satisfied in reality. This seminar, held in Room 710, Fanhai Building, on 24th, Nov. 2015, had the honor to invite Professor Weili Ding of NYU Shanghai & Queen’s University to present her latest co-authored work on the improvement of evaluation research methodology and its application in education experiments. Professor Ming Lu hosted the seminar with the attendance of Professor Zhao Chen and some of the PhD students.


Professor Ding’s research was motivated by Class Size Reduction Initiatives throughout North America over the past decade. “This program brought incredible attention into Tennessee’s Student Teacher Achievement Ratio project, a.k.a. STAR, in policy debates and decisions. Nevertheless, there is substantial heterogeneity in the estimated treatment effects across schools, which remains unexplained.” Professor Ding said.

To explain this heterogeneity, Professor Ding presented a “puzzle” which suggested that class size treatment effects in Project STAR were driven by the excellent small class performance of students attending schools where a smaller fraction of students had received treatment. “This variation in how the treatment is offered across schools provides a clear case where the stable unit treatment value assumption, a.k.a. SUTVA, is violated.”

Professor Ding introduced new empirical strategies that identify and estimate causal effects when SUTVA is not maintained. “Our strategy is to exploit the random variation in the fraction of individuals offered treatment across locations (schools in Project STAR) to explain treatment heterogeneity. This variation in the experiment was not induced by the experimenters. And the treatment effect explained by this variation is not the intended effect by research design.” Professor Ding said.

Using the dataset from Project STAR, Professor Ding concluded that treatment effect heterogeneity in Project STAR was driven by the performance of students in treatment classes within schools with low treatment intensity. The context specific effects accounted for 60-85% of the total treatment effects in Project STAR. In the meantime, Professor Ding improved empirical methodology to identify causal parameters especially when either context effects or social interactions were present which implied that SUTVA might not be satisfied. “By relaxing SUTVA we may also have benefits of increasing the external validity of a study by randomizing not just treatment but also the context variable.” Professor Ding said.






By SHI Shuo

©2008 www.cces.fudan.edu.cn All rights reserved.