In the social and behavioural sciences we have many questions that concern both causal relations and the dynamics of processes: We are often interested in the way that internal and external forces give rise to the momentary state of a system (e.g., a person, dyad family, profession), how well we can forecast the future states of a system, and how we could intervene on the system to change its state.
To answer these research questions, we need both a conceptual framework that allows us to think clearly about causal and predictive relations, and statistical techniques that allow us extract information from different types of data.
In this course students will learn the important difference between causal and predictive or descriptive questions, and get acquainted with diverse techniques to study causality and forecasting in the context of both crosssectional and longitudinal data.
We begin with an introduction to the core areas within the interventionist framework for causal inference, and discuss: a) potential outcomes and how to account for confounding in our analyses; b) graphical approaches to causal modelling based on directed acyclical graphs (DAGs) and how these can be used to guide design and analysis choices; and c) how causal models themselves can be learned from data.
Students obtain handson experience with different techniques, including regression, stratification, matching, inverse probability weighting, dseperation, identifying adjustment sets, and conditional independence testing.
Subsequently, we see how these techniques can be used if we have repeated measures of the outcome, and when we have repeated measures of the cause (i.e., treatment) and the outcome. In the final part of the course, we focus on time series data (i.e., many repeated measures from a single case), and consider different forecasting and causal inference techniques for such data.
In the social and behavioural sciences we have many questions that concern both causal relations and the dynamics of processes: We are often interested in the way that internal and external forces give rise to the momentary state of a system (e.g., a person, dyad family, profession), how well we can forecast the future states of a system, and how we could intervene on the system to change its state.
To answer these research questions, we need both a conceptual framework that allows us to think clearly about causal and predictive relations, and statistical techniques that allow us extract information from different types of data.
In this course students will learn the important difference between causal and predictive or descriptive questions, and get acquainted with diverse techniques to study causality and forecasting in the context of both crosssectional and longitudinal data.
We begin with an introduction to the core areas within the interventionist framework for causal inference, and discuss: a) potential outcomes and how to account for confounding in our analyses; b) graphical approaches to causal modelling based on directed acyclical graphs (DAGs) and how these can be used to guide design and analysis choices; and c) how causal models themselves can be learned from data. Students obtain handson experience with different techniques, including regression, stratification, matching, inverse probability weighting, dseperation, identifying adjustment sets, and conditional independence testing.
Subsequently, we see how these techniques can be used if we have repeated measures of the outcome, and when we have repeated measures of the cause (i.e., treatment) and the outcome. In the final part of the course, we focus on time series data (i.e., many repeated measures from a single case), and consider different forecasting and causal inference techniques for such data.
Research goals can be divided into focusing on description, prediction or causation. Each of these goal poses different challenges, and it is therefore of critical importance to recognize what goal is being pursued; only then can data scientists make informed decisions about design and analysis.
A particularly challenging task is how to properly use nonexperimental (also referred to as observational or correlational) data for causal inference, both in the context of crosssectional and longitudinal research. The latter offers unique opportunities to study process as they unfold over time, but also pose additional challenges about how to decide what variables to control for and how to define a causal effect.
The broader context is provided by also introducing the exploratory technique of causal discovery, and forecasting techniques for time series data. This will help students to recognize the difference between the diverse research goals, and to make informed decisions about how to approach them
Course form
Weekly lectures and lab sessions.
The lectures will focus on the technical background of studying causality and forecasting with crosssectional and longitudinal data.
Homework exercises will allow students to gain handson experience with the techniques.
In preparation of f the lab session, students perform the techniques that were introduced in the lecture (aims ac).
In addition, the exercises stimulate students to critically evaluating the diverse approaches and their appropriateness in different contexts (aim d).
Literature
Several introductory articles and blogs that discuss the kind of research questions and analytical techniques that are fundamental in studying processes in the social and behavioural sciences
