Tel: +49 221 47694-579
Tel: +49 221 47694-579
Tel: +49 221 47694-162
Tel: +49 221 47694-162
Week 2: Causal Inference in Observational Studies
Dr. Krisztián Pósch, Thiago Oliveira
Date: 08.03 - 12.03.2021 ics-file
Location: Online via Zoom / Course language: english
There is a growing expectation from governmental, and non-governmental organisations for social scientists to produce research which are capable of assessing the impact of certain changes to policies or the emerging effects of interventions. Yet, it is often difficult to analyse such changes when traditional experiments cannot be carried out either on the grounds of feasibility or due to ethical constraints. This course will provide participants with a toolkit to analyse so-called 'quasi-experimental' designs and draw causal conclusions using observational datasets.
We have put together a course focused on addressing practical considerations including when certain designs can be credibly employed and how the emerging results can be interpreted. In particular, the five most commonly used families of methods will be discussed: matching, difference-in-differences, instrumental variables, causal mediation, and regression discontinuity designs. We are to demonstrate each of these approaches by discussing existing applications from across the fields of the social sciences. We will also provide a tutorial for the 'R' statistical software and share with the participants the code for the methods covered by the course. We are also going to provide ample opportunity for participants to discuss their research plans, ask for advice regarding their own data, and recommend cutting-edge methods to address their research questions.
Participants will find the course useful if:
- They are social scientists (economists, political scientists, sociologists, criminologists, psychologists, etc.) who are planning to engage with non-experimental research designs that will allow them to derive causal estimates. Researchers who plan to use or evaluate robust methods which are capable of going beyond mere associations and establish causal relationships, will greatly benefit by taking the course.
By the end of the course participants will:
- Be capable of designing their own studies to derive causal estimates in observational settings.
- Acquire an in-depth understanding of and the pragmatic (programming) skills to carry out five family of methods: matching, difference-in-differences, instrumental variables, causal mediation, and regression discontinuity design.
- Become familiar with interdisciplinary applications of the methods covered by the course, including the fields of economics, political science, criminology, education, sociology, epidemiology, and so on.
- Be able to engage the contemporary literature of causal inference and identify state-of-the-art methods which might be most relevant to their specific research question.
- Have a good understanding of the potential outcome framework, causal diagrams, and the counterfactual way of thinking.
- Become adept users of the statistical programme, R.
Organisational Structure of the Course:
Participants will be expected to attend both the synchronous lectures and seminars. Each day, a two-hour pre-recorded lecture will outline the basic concepts and features of the method(s) discussed. These lectures will be sliced up to shorter segments making each set of concepts and methods easier to grasp. Each short segment will be accompanied by a forum where every participant can ask follow-up questions. Each day at 11 am, there will be another hour of synchronous lecture, where applied examples will be discussed to show how each of these approaches have been adopted by researchers in various disciplines across multiple settings. These synchronous lectures will also provide a chance to ask questions of the lecturers regarding the pre-recorded materials.
In the afternoon, the first two hours of the seminar will be dedicated to a worked example where each participant will have a chance to implement the method covered in the morning using the statistical programme R. Each day will end with another hour of computer session, where participants can ask questions from the course convenors about the newly covered materials or their own research designs, datasets, and analytical plans.
Participants will be assigned to smaller groups and encouraged to post their ideas and questions to an online forum where those will be commented on by the course convenors and the other interested participants. We hope that in addition to the course materials, these forums will provide another knowledge base to which the participants can refer to in the future.
- Knowledge of basic statistical concepts, including the principles of null-hypothesis significance testing (NHST), multilinear regression, moderated effects (interactions), and binary logistic regression.
- Basic understanding of designing quantitative studies.
- Background in statistical software is not assumed.
Software and hardware requirements:
The course will be taught using R and RStudio, both of which are freely available for anyone to download. The primary environment will be PC, but modified scripts will be provided for Mac users to assure cross-platform compatibility. Where possible, we will provide the equivalent STATA .do file to run the analysis but all participants will be encouraged to use R.