Information détaillée concernant le cours

[ Retour ]

Causality and Machine Learning with Applications in Psychology


6, 13, 20 & 27 October 2022, 14:00-18:00


Dr Matthew Vowels, UNIL


What is the current problem? In spite of a general recognition that causality is a fundamental part of the scientific endeavour to understand the world, causal approaches to analysis are still taboo in much of psychology and social science. Furthermore, the validity of most of the popular techniques being used rests on assumptions which we already know not to hold in practice (linearity, parametric form, etc.). The result of these two shortcomings is poor replicability, meaningless interpretations of results, and theories which cannot be formalized or tested.

Solution - This course offers PhD students conceptual and methodological state-of-the-art approaches from the intersection of causal inference, causal discovery, and machine learning to overcome the current limitations. We focus on Directed Acyclic Graphs (DAGs) for causality in general, constraint-based approaches for causal discovery, ensemble methods for machine learning, and combinations of these for testing our theories and hypotheses.

Benefits - This will help students to (a) develop, enrich, and formalize their theories so that they can test them unambiguously, (b) identify key control variables in order to render interpretable and causal results, and (c) use modern approaches to predictive machine learning. Together, this brings psychology in line with modern approaches to analysis without sacrificing interpretability.

(Overview of this topic: )

Lecture 1 – Causality & Theory

  • Causal relationships are the foundation of psychological research and theories
  • What is theory?
  • What is causality?
  • What are the limitations in current approaches to theory development and causality in psychology and social science? What is the cost of not improving?
  • Proposed alternatives, and justifications/benefits?
  • Overview of required tools
    • Directed Acyclic Graphs (DAGs)
    • Causal inference
    • Machine Learning
  • Discussion of examples with group and applicability of the approaches


Lecture 2 – How to formulate psychological theories mathematically (by graphically representing them)?

  • DAGs and graphical models
  • How we use DAGs to specify theory
  • Example structures and help developing group's projects


Lecture 3 – Causal Inference and Control Variables

  • Identifying the target effect – backdoor paths and confounding
  • Conditional independency rules of DAGs
  • Backdoor paths for DAGs – good and bad control variables
  • Specifying a hypothesis in terms of our DAG/theory
  • Improve your own research questions using DAGs


Lecture 4 – Estimation

  • Use the specified hypothesis expressed in terms of the DAG/theory
  • Estimate this quantity
  • Why machine learning can be useful (causal approach versus predictions)
  • Basics of machine learning and plug-in estimation
  • Example causal effect estimation
  • How we interpret the results
  • How can you implement these steps into your research

UNIL, Géopolis building, room 2218



Délai d'inscription 06.10.2022
short-url short URL

short-url URL onepage