Philosophy, Psychiatry, & Psychology

Erwin

Erwin states that I claim the following principle is a method for inferring causes: “if X is causally relevant to the occurrence of Y, then the incidence of Ys in the class of Xs and Ys will be different compared with the incidence of Ys in the class of non-Xs and Ys” (Erwin 2012, 217). This is not so. The method I attribute to Grünbaum is given in what follows the principle in the quotation given on p. 199, namely, “To validate a claim of causal relevance, we must first divide the reference class C into two subclasses, the Xs and the non-Xs. And then we must show that the incidence of Ys among the Xs is different from what it is among the non-Xs.” And this clearly is epistemological, not ontological, because it speaks of what we need to do to validate a claim. Furthermore, Grünbaum repeatedly employs this method in examples, such as those regarding visits to the library and derogatory remarks (Grünbaum 1984, 73; 1993, 164).

Grünbaum’s statement of the method is itself grounds to attribute to him the view that to validate a causal claim, statistical knowledge, namely the relative incidence of Ys among Xs and non-Xs, is necessary.¹ And it is striking that Grünbaum’s statement of the method is made without qualification. Erwin notes the only two examples, I believe, where Grünbaum provides any indication that statistical data might not be needed. One is his response to Erwin’s example of the autistic boy whose head-banging ceases after shock treatment is administered as a punishment. Grünbaum comments that the inference, that the shock punishments caused the head-banging to cease, is justified because we know the ‘natural history’ of the condition. But such knowledge is knowledge of what behavior, symptoms, and so on, are likely to occur at each stage of the condition, for example, knowledge of the probability of symptoms such as head-banging going into spontaneous remission. The second is Grünbaum’s remark that statistical knowledge regarding the particular case can be replaced by background theoretical knowledge in astronomical inferences regarding planetary motion. But such theoretical knowledge may yet be statistical, and in the case of Newton’s laws of motion, statistical evidence was relevant (indeed, it was precisely this kind of evidence that led to their [End Page 221] qualification and the confirmation of Einstein’s theory of relativity). Nor does Erwin’s analogy between watermelons and baseballs provide a counterexample: Although we may rely on the analogy to infer that a watermelon will break a window, this is insufficient without the additional background knowledge that baseballs very often break windows, which is statistical knowledge. I suspect that Grünbaum would similarly only allow analogy to have heuristic, not probative, value.²

However, the central question is whether the kind of background knowledge that enables causal inferences based on thematic affinity need not be statistical. Grünbaum notes that there are such “special further conditions under which thematic affinity does indeed warrant causal inferences” (1993, 129). He never specifies the conditions but presents six examples.³ In all six, the background knowledge described involves statistical information, in most cases, knowledge of the probability of the effect occurring by means other than the thematically similar cause. Grünbaum even comments of two cases that the warrant is derived from Mill’s joint method of agreement and difference (1993, 132). The centrality of statistical knowledge is reaffirmed in Grünbaum’s (1993, 219) remark that “To guard against inductive fallacies of causal inference, methods of controlled inquiry or their equivalents from relevant...