ABSTRACT
There is a way to fix this. Again referring to the tiny examples of Section 7, 1 could always give expressions for b0.i and bj, and so if I could observe every element of F then of course I could observe these, too. Suppose I were to imagine that, to the contrary, F contained factors that I could not measure. There are plenty of practical examples when this is the case, and in point of fact the case where all factors are observable is the highly unusual one. But once I allow unobservable factors into F, there doesn't seem to be any reason not to allow enough in to give a complete causal account of y. In other words, if some factors are unobservable, then saying C[y|F] = y is actually an assertion that y is always caused by something that I might conceivably know about, and there doesn't seem to be any logical reason why I shouldn't do this.
