iussp2005 presentation1
TRANSCRIPT
Statistical Modelling and Causality
Federica Russo*, Michel Mouchart**, Michel Ghins*, Guillaume Wunsch***
* Institut Supérieur de Philosophie, Université Catholique de Louvain** Institut de Statistique, Université Catholique de Louvain*** Institut de Démographie, Université Catholique de Louvain
Structure of the paper Scientific knowledge Data Causality and statistical modelling
The statistical model Statistical inference and structural models Conditional models and exogeneity Beyond exogeneity Hypothetico-deductive methodology
The population and the individual
Scientific knowledge We are moderate realists:
Models grant cognitive access to some unobservable parts of reality
Modelling is constructing a simplifiedrepresentation of a complex reality
Data To acquire causal knowledge
we try to make sense of observations
However, collecting data isproblematic under several respects
Causality and statistical modelling The statistical model
A stochastic representation of the world Analyze data asas a realization
of a family of distributions
Causality and statistical modelling Statistical inference and structural models
Structural models make statistical inferenceoperational and meaningfulthrough a learning-by-observinglearning-by-observing process
StructuralStructural models: a representation of theworld that is stable under a large classof interventions
Causality and statistical modelling Conditional models and exogeneity
p (x | ) = p (z | ) p (y | z , )
The conditional part is structuralstructural and represents the data generating process
Z is an exogenousexogenous variable in a structural model,that is Z is a causalcausal variable
Causality and statistical modelling
Causality: exogeneity in a structural model
operational concept, internal to the model
Beyond exogeneity
Temporal and atemporal features grant causality
Hypothetico-deductive methodology Theorizing out-of-sample information
Choice of variablesFormulation of the causal hypothesis
Iteratively: Building the statistical model Testing the adequacy model-data
The population and the individual Methodological issue:
Detect causal variables, provide a sufficient list Describe the causal mechanism
Practical issue: Take decisions about individualsindividuals
based on knowledge about the populationpopulation
Epistemological issue: Causal knowledge about the population
guides causal attribution about individualsthrough Bayes’ theorem
To conclude… Causality: exogeneityexogeneity in a structural model
Thus defined, causality is internalinternal to the model
Structural models mediateepistemic accessepistemic access to causal relations