by Anna de Bruyckere
Dr. Tobias Henschen, research fellow at the Zukunftskolleg at the University of Konstanz, gave a talk in the Birley Room in Hatfield College that proved a worthy end to the 2014 CHESS/GPI seminar series in Philosophy, Economics, and Public Policy. Henschen is a philosopher and economist speaking to us about causal epistemology and non-cognitive values regarding modelling for policy purposes.Henschen’s point of departure is his observation that structural modelling constitutes the main methodology of those social scientific practices that produce results with the purpose of informing public policy making. Structural models are systems of linear equations representing the co-variation of variables. These variables can be endogenous or exogenous. In other words, for some variables, the values are explained by reference to the values of the other variables within the model. For others, this is not so; their determinants fall outside the model. Congenial to an interventionist account of causation such as James Woodward’s, Henschen takes a structural model “to represent a stable and autonomous mechanism if the value of Y changes as a result to an intervention on the value of X.” He labels such an account as objectivist, as “it assumes that stable and autonomous mechanisms exist, that we can identify them, and that we can manipulate them for policy purposes.” He notes the widespread confidence in the possibility of obtaining evidence for interventionist causal claims regarding the relations between variables of structural models. Such evidence is to be obtained methodically such as through randomized control trials (RCTs), analyses of natural experiments, and all sorts of statistical independence tests. Henschen argues that the confidence placed in our usual practices of gathering empirical evidence for claims about causal relations is not warranted. The main problem consists in the fact that evidence will always remain inconclusive, in the sense that we can never be certain on its basis that the value of Y changes as a result to a change in the value of X. It remains, on principle, always possible to question the assertion of the existence of a causal relation: any evidentiary propositions that the proponent will bring into the debate can itself be debated. This is so because whatever kind of evidence the proposition is based on, its establishment will come with its own assumptions that either opponents can show to be violated, or the proponent cannot show to have been met. More specifically, all methods of producing evidence rely on strong background assumptions which economists rarely have proper grounds for. In particular, these assumptions cannot be defended on statistical grounds. Pointing to the overwhelming centrality of number crunching in evidentiary economic practices, Henschen holds that other kinds of evidentiary reasoning, such as based on historical or institutional or otherwise qualitative data, cannot carry the same consensus-forming force. Hence, this principled lack of conclusive statistical support for evidentiary propositions and their methodical assumptions is highly detrimental on Henschen’s view—as statistics are the only means to bring about consensus among economists. For that reason, Henschen holds that objectivist accounts of causation (such as the interventionist account) are unsatisfactory: regardless of their metaphysical value, they come with an epistemology that is beyond our epistemic capacities. Consequently, Henschen proposes a subjectivist variant of an interventionist account of causation to save the legitimacy of model-based policy: “a definition of causality in terms of the subjective belief that the value of a dependent variable would change as a result to a change in the value of an independent variable.” In the absence of conclusive objective evidence for such beliefs, Henschen further argues that these beliefs are at least partly determined by non-epistemic values, where the meaning of ‘value’ seems determined as exactly this absence of conclusive objective evidence. He draws on Hempel’s account of the influence of values on scientific activity, scientific knowledge, and scientific method, respectively. On Henschen’s account, empirical evidence does still play a (large) role in the development of subjective beliefs about causal relations between variables of the model (and, supposedly, the real world target the model is meant to elucidate). RCTs, natural experiments, statistical independence tests, and the like are joined by value judgments, group interests, and ideological commitment in order to determine social scientific policy makers’ beliefs that a policy intervention on variable X of magnitude a would produce a change in the value of dependent variable Y of magnitude b.