Directed Acyclic Graphs for Causal Inference
A 2-Day Seminar Taught by Felix Elwert, Ph.D.
This seminar won the 2013 Causality in Statistics Education Award, given by the American Statistical Association.
To see a sample of the course materials, click here.
This seminar offers an applied introduction to directed acyclic graphs (DAGs) for causal inference. DAGs are a powerful new tool for understanding and resolving causal issues in empirical research. DAGs are useful for social and biomedical researchers, and for business and policy analysts who want to draw causal inferences from non-experimental data. A major attraction of DAGs is that they are “algebra-free,” relying instead on intuitive yet rigorous graphical rules.
The two primary uses of DAGs are (1) determining when causal effects can be identified from observed data, and (2) deriving the testable implications of a causal model. DAGs are also helpful for understanding the causal assumptions behind widely used estimation strategies, such as regression, matching, and instrumental variables analysis.
To provide a flavor for what we will cover in this class, consider the following DAG:
This is a stylized causal model relating an early health variable, income, smoking, cancer, and death. DAGs help us to see which inferences we can make and which inferences we cannot make. For example, if this model describes the true data generating process and if traumatic brain injury is unmeasured, then it would be possible to identify the causal effect of income on smoking but NOT the causal effect of income on death.
DAGs can also help us avoid common errors in interpretation by allowing us to derive associations between variables from a causal model. The graph below illustrates Berkson’s bias, a famous example of selection bias.
If both traffic accidents and the flu land people in the hospital, then a study of hospital records would show that accidents are negatively associated with having the flu. Does that mean that having the flu makes me a better driver? Or that getting in an accident is as good as a flu shot? Obviously not! But we will learn why researchers make these kind of errors regularly, how to avoid them, and what to do about them to make correct inferences.
The seminar will start by introducing the essential elements for causal reasoning with DAGs and then use DAGs to discuss a range of important challenges in observational data analysis. Topics include: conditions for the identification of causal effects; d-separation; the difference between confounding, over-control, and selection bias; identification by adjustment; backdoor identification; what variables to control for in observational research; what variables not to control for in observational research; structural assumptions in regression; and instrumental variables analysis.
This seminar will empower participants to recognize and understand problems and to spot fresh opportunities for causal inference in their own data. This is a hands–on course with at least one hour each day devoted to carefully structured and supervised exercises. Many of these exercises will use the freeware package DAGitty which allows users to draw and analyze causal graphs. See the computing section below for more details.
For background and preparation, we recommend reading:
Keele L, Stevenson RT, Elwert F (2019). “The causal interpretation of estimated associations in regression models.” Political Science Research and Methods. Click here.
To fully participate in the course, participants should bring a laptop computer. Demonstrations and exercises will make use of the freeware package DAGitty, which is available at daggity.net. We will use DAGitty in its web browser mode. However, it can also be downloaded for offline use and is available as a package for R.
DAGitty computes covariate adjustment sets for estimating causal effects, enumerates instrumental variables, derives testable implications (d-separation and vanishing tetrads), generates equivalent models, and includes a simple facility for data simulation.
Who should attend?
If you want to understand under what circumstances you can draw causal inferences from non-experimental data, this course is for you. Participants should have a good working knowledge of multiple regression. Some prior exposure to causal inference (counterfactuals, propensity scores, instrumental variables analysis) will be helpful but is not essential.
LOCAtions, Format, And Materials
The class will meet from 9 am to 5 pm each day with a 1-hour lunch break at Temple University Center City, 1515 Market Street, Philadelphia, PA 19103.
Participants receive a bound manual containing detailed lecture notes (with equations and graphics), examples of computer printout, and many other useful features. This book frees participants from the distracting task of note taking.
Registration and lodging
The fee of $995.00 includes all seminar materials.
If you cancel your registration at least two weeks before the course is scheduled to begin, you are entitled to a full refund (minus a processing fee of $50).
Lodging Reservation Instructions
A block of guest rooms has been reserved at the Club Quarters Hotel, 1628 Chestnut Street, Philadelphia, PA at a special rate of $177 per night. This location is about a 5-minute walk to the seminar location. In order to make reservations, call 203-905-2100 during business hours and identify yourself by using group code SH1017 or click here. For guaranteed rate and availability, you must reserve your room no later than Tuesday, September 17, 2019.
If you need to make reservations after the cut-off date, you may call Club Quarters directly and ask for the “Statistical Horizons” rate (do not use the code or mention a room block) and they will try to accommodate your request.
1. Counterfactual causality
2. Directed Acyclic Graphs (DAGs)
b. Graphical display of the data generating model
c. The importance of causal assumptions
3. Associational implications of a causal model
a. Association vs. causation in DAGs
b. Three sources of association and independence (d-separation)
c. The difference between confounding and selection bias
d. Deriving testable implications
4. Graphical Identification Criteria
b. Control for confounding via adjustment
c. Adjustment and ignorability
d. Back-door identification
e. Front-door identification
f. Some helpful sufficient rules
5. Selection Bias
a. Post-outcome selection examples
b. Intermediate-variable selection examples
c. Pre-treatment selection bias examples
d. Why selection and confounding are distinct causal concepts
6. Graphical insights for common methods
a. Identification in matching and regression
b. Hidden causal assumptions in regression analysis
7. DAGs for mediation analysis
a. “Controlled” and “natural” effects
b. Identification requirements
“I wholeheartedly recommend this course to all empirically minded doctoral students – as early as possible in the process, before statistics/econometrics courses shift their attention away from theory/causal mechanisms towards estimation/data.”
Thorsten Sellhorn, Ludwig-Maximilians-Universitaet
“The seminar offers immediately applicable, lucidly presented lessons on applying causal logic to research design and analysis.”
Michael Lorber, New York University
“I have spent years reading literatures on this topic and Dr. Elwert virtually provided a complete illumination of Causal Inference with Directed Graphs in just two days. Save yourself time, struggles, and confusion by taking this wonderful class!”
Joseph Staats, University of Minnesota Duluth
“Felix is an excellent instructor of this course and going through his training has helped me to understand the important basics of DAGs and graphical analysis as well as their importance in causal inference. I would highly recommend this course to anyone involved in causal inference in whatever area of specialty because, believe you me, this instructor is good!”
Edmore Chamappiwa, University of Manchester, UK
“My understanding of research design changed after this course. This is one of the most powerful courses in methods I have ever encountered.”
Anibal Perez-Linan, University of Pittsburgh
“This course helps reinforce epidemiologic methods through DAGs. It was extremely valuable, fun and worth participation. It gave me a great mental tune-up.”
Ximena Vergara, Electric Power Research Institute/UCLA