A philosophical analysis of causality in econometrics
This thesis makes explicit, develops and critically discusses a concept of causality that is assumed in structural models in econometrics. The thesis begins with a development of Herbert Simon's (1953) treatment of causal order for linear deterministic, simultaneous systems of equations to provide a fully explicit mechanistic interpretation for these systems. Doing this allows important properties of the assumed causal reading to be discussed including: the invariance of mechanisms to intervention and the role of independence in interventions. This work is then extended to basic structural models actually used in econometrics, linear models with errors-in-the-equations. This part of the thesis provides a discussion of how error terms are to be interpreted and sets out a way to introduce probabilistic concepts into the mechanistic interpretation set out earlier. The resulting analysis is then critically compared with similar work by economists, Stephen LeRoy (1995) and Kevin Hoover (2001a) who both develop Simon's work on causal order in different ways. In the latter part of the thesis, the mechanistic interpretation set out at the beginning is used to interpret identification conditions. Typically, these are presented in econometrics as mathematical conditions for determining whether unknown parameters in equations can be measured from observation. In the thesis it is shown that the identification conditions imposed on sets of equations when interpreted mechanistically require a sparseness of causal structure that ensures that experiments are hypothetically possible of the causal structure. It also analyses the role of identifiability conditions in causal inference. The final part of the thesis shows that the mechanistic interpretation developed in the thesis succeeds, unlike Simon's own methods for analysing spurious correlation, in avoiding important criticisms by Nancy Cartwright (1989) whose own approach to inferring causal structure from observations is also critically analysed.