Cologne, Germany

Using Directed Acyclic Graphs for Causal & Statistical Inference

online course
when 3 August 2022 - 5 August 2022
language English
duration 1 week
fee EUR 200

This online short course uses causal graphs (or “directed acyclic graphs”, DAGs) as a remarkably simple, yet general and powerful framework to describe and discuss a large set of problems that empirical social scientists need to tackle. Is my question of interest descriptive or causal? How can I communicate my assumptions effectively to others, and can I test them? How can I tell correlation from causation? How do I choose control variables for my regression models? After discussing how DAGs can be used to answer these foundational questions, the course also covers basics of causal interaction and effect heterogeneity, causal mediation, nonresponse/selection bias (and adjustments for it) and, if time permits, instrumental variables and panel data analysis from a DAG perspective.

Course leader

Julian Schuessler is a post-doc at the Institute for Political Science, Aarhus University, Denmark.

Target group

Participants will find the course useful if they are:
- interested in causal questions and want to understand the assumptions associated with regression control, mediation analysis, and instrumental variables better:
- interested in non-causal questions, want to use data suffering from nonresponse, and want to understand how to use causal assumptions in this case.

Prerequisites:
Participants should be willing to learn and use formal reasoning and must have at least Bachelor-level knowledge of statistics. Basic knowledge of R is helpful.

Course aim

By the end of the course participants will:
- know how to use causal graphs to visualize causal assumptions, define quantities of interest, and determine testability of assumptions via d-separation;
- know how to graphically determine the identification of causal and descriptive quantities like average causal effects, causal interaction, effect heterogeneity, natural direct and indirect effects, and population distributions from data with nonresponse;
- know under what graphical assumptions instrumental variable and panel data analysis typically operate;
- will have some basic knowledge about how all of this relates to implementation in standard statistical software.

Fee info

EUR 200: Student/PhD student rate.
EUR 300: Academic/non-profit rate.
The rates include the tuition fee and course materials.