Statistical modelling and causality
Federica Russo, Université Catholique de Louvain
Michel Mouchart, Université Catholique de Louvain
Michel Ghins, Université Catholique de Louvain
Guillaume Wunsch, Université Catholique de Louvain
The authors consider causality from an epistemological perspective; they do not address the nature of causation but rather how we come about to accepting causal relations by way of statistical models. In their view of moderate scientific realism, models are not the locus of final truth but they can give a partial and approximate knowledge of external realities and their interrelations. The selection of relevant data is a crucial step in scientific knowledge-building; it depends upon the current state of knowledge. A statistical model is a stochastic representation of the world. Its randomness delineates the frontiers of one’s explanation. Conditional statistical models characterized by parameters that are stable over a large class of different environments, i.e. structural models, can lead to causal statements. However, the concept of causality is viewed as internal or relative to the structural model itself. Statistical modelling per se can never prove that causes are true.