I have suggested that a good way of categorizing different approaches to the nature of causation is in terms of the four alternatives of direct realism, Humean reductionism, non-Humean reductionism, and theoretical realism.  Let us now consider each of these in turn.

        Direct realism involves four main theses.  First, the relation of causation is directly observable.  Secondly, that relation is not reducible to non-causal properties and/or relations.  Thirdly, the relation of causation is also not reducible to non-causal properties and/or relations together with causal laws -- since such a reduction would entail that one could not  be directly acquainted with the relation of causation.  Fourthly, the concept of the relation of causation is analytically basic.

        A number of philosophers have claimed that the relation of causation is observable, including David Armstrong (1968, p. 97, and 1999, pp. 211-16), Elizabeth Anscombe (1971, pp. 92-3), and Evan Fales (1990, pp. 11-25).  Thus Anscombe argues that one acquires observational knowledge of causal states of affairs when, for example, one sees a stone break a window, or a knife cut through butter, while Fales, who offers the most detailed argument in support of the view that causation is observable, appeals especially to the impression of pressure upon one's body, and to one's introspective awareness of willing, together with the accompanying perception of the event whose occurrence one willed.

        Suppose that it is granted that in such cases one does, in some straightforward sense, observe that one event causes another.  Does this provide one with any reason for thinking that direct realism is true?  For it to do so, one would have to be able to move from the claim that the relation of causation is thus observable to the conclusion that it is not necessary to offer any analysis of the concept of causation, that it can be taken as analytically basic.  But observational knowledge, in this broad, everyday sense, would not seem to provide adequate grounds for concluding that the relevant concepts are analytically basic.  One can, for example, quite properly speak of physicists as seeing electrons when they look into cloud chambers, even though the concept of an electron is certainly not analytically basic.  Similarly, the fact, for example, that sodium chloride is observable, and that one can tell by simply looking and tasting that a substance is sodium chloride does not mean that the expression 'sodium chloride' does not stand in need of analysis.

        But might not it be argued in response, first, that, one can observe that two events are causally related in precisely the same sense in which one can observe that something is red; secondly, that the concept of being red is analytically basic, in virtue of the observability of redness; and therefore, thirdly, that the concept of causation must, for parallel reasons, also be analytically basic?

        This response is open, however, to the following reply.  If a concept is analytically basic, then one can acquire the concept in question only by being in perceptual or introspective contact with an instance of the property or relation in question that is picked out by the concept.  One could, however, acquire the concept of being red in a world where there were no red things: it would suffice if things sometimes looked red, or if one had hallucinations of seeing red things, or experienced red after-images.  The concept of red must, therefore, be definable, and so cannot be analytically basic.

        What is required if a concept is to be analytically basic?  The answer that is suggested by the case of the concept of redness is that for a concept to be analytically basic, the property or relation in virtue of which the concept applies to a given thing must be such that that property or relation is immediately given in experience, where a property or relation is immediately given in experience only if, for any two qualitatively indistinguishable experiences, the property must either be given in both or given in neither.

        Is the relation of causation immediately given in experience?  The answer is that it is not.  For given any experience E whatever -- be it a perception of external events, an awareness of pressure upon one's body, or an introspective awareness of some mental occurrence, such as an act of willing, or a process of thinking -- it is logically possible that appropriate, direct stimulation of the brain might produce an experience, E*, which was qualitatively indistinguishable from E, but which did not involve any causally related elements.  So, for example, it might seem to one that one was engaging in a process of deductive reasoning, when, in fact, there was not really any direct connection at all between the thoughts themselves -- all of them being caused instead by something outside of oneself.  Causal relations cannot, therefore, be immediately given in experience in the sense that is required if the concept of causation is to be unanalyzable.

        Let us now turn to objections to direct realism.  The first has, in effect, just been set out.  For if, for any experience in which one is in perceptual or introspective contact with the relation of causation, there could be a qualitatively indistinguishable, hallucinatory experience in which one was not in contact with the relation of causation, it would be possible to acquire the concept of causation without ever being in contact with an instance of that relation.  But such experiences are logically possible.  So the concept of causation must be analyzable, rather than being analytically basic.

        Secondly, it seems plausible that there is a basic relation of causation that is necessarily irreflexive and asymmetric, even if this is not true of the ancestral of that relation.  If either reductionism or indirect realism is correct, one may very well be able to explain the necessary truths in question, since the fact that causal concepts are, on either of those views, analyzable means that those necessary truths may turn out to be analytic.  Direct realism, by contrast, in holding that the concept of causation is analytically basic, is barred from offering such an explanation of the asymmetry and irreflexivity of the basic relation of causation.

        Thirdly, direct realism encounters epistemological problems.  Thus, features such as the direction of increase in entropy, or the direction of the transmission of order in non-entropic, irreversible processes, or the direction of open forks, often provide evidence concerning how events are causally connected.  Similarly, causal beliefs are often established on the basis of statistical information -- using methods that, especially within the social sciences, are very sophisticated.  Given an appropriate analysis of the relation of causation, one can show why such features are epistemologically relevant, and why the statistical methods in question can serve to establish causal hypotheses, whereas if causation is a basic irreducible relation, it is not at all clear how either of these things can be the case.