Causal inference in social science
Mikko Silliman University of Helsinki
January 2020 Department of Social and Public Policy
Course description:
What is the effect of Universal Basic Income on labor market and health outcomes? How does secondary school track affect labor market outcomes later on? How does social media affect political polarization?
These types of questions are central to work carried out by government administrators, researchers across the social sciences, business analysts, and journalists reporting on a wide array of social and economic issues. The objective of this course is twofold: 1) to provide students the toolkit with which to evaluate the credibility of answers to such questions, and 2) to gain familiarity with contemporary approaches by which students can use data to answer such questions themselves.
This course will cover the main empirical approaches by which data is used to answer causal questions – including the potential outcome framework, randomized control trials, instrument variable approaches, regression discontinuity designs, and difference-in-differences strategies. A primary focus of the course will be to develop the skills to assess the credibility of the assumptions underlying these approaches. While basic statistical concepts will be central to the course, these will be covered along with the above methods, and there are no pre-requisites for this course.
This course is intended for undergraduates, master’s students, and even doctoral students without prior exposure to empirical methods for causal inference. It may be particularly suitable for undergraduate and master’s students who plan to use statistics in their master’s thesis.
Time and place:
The course will meet 7-13.1.2020 from 9-11 and from 14-16, with a break in between for lunch and study. The first week the course will meet in Soc & Kom, room 210. The second week the course will meet in Aurora, room 225. I will hold drop in “office-hours” in Think Corner from 13-13:45 each day.
Slides
Causal questions and the potential outcome framework (PDF, POWERPOINT)
Experiments, randomized control trials (RCT's)
Regression continuity designs (RDD's) and the LATE framework
Difference-in-differences (DiD)
Instrument variable designs more broadly
Contact
silliman@g.harvard.edu