Clarification on Correlation

Here's something that I think is true. Hume notes that we can never observe causation in the world, rather we can only observe "constant conjunction." What does this mean? Suppose that your picture of human observation is that we can download an image of the world apart from the concepts that we apply to it.

Let me use an example to illustrate this: I think that apples are tasty. When I see an apple, I think "Damn, that apple is tasty." Does this mean that I'm observing tastiness in the world? Quite plausibly, no, it doesn't. Instead, what I'm doing is observing the apple and once my brain knows that the apple is tasty it applies "tastiness" to it. Contrast this with my actual observation of the apple's presence with my hand. With my hand I sense it's firmness, it's smoothness and this tells me that the apple is extended in space and of a certain texture. Now, does this work like tastiness? That is, do I first observe the apple and then apply to the apple the properties of "extended in space" or "firm"? No, this doesn't seem quite right. Rather, what I'm directly observing is that it fills my hands or pushes back when I squeeze.

(Of course, you might disagree. You might be Kant, for instance, and think that we only observe the world through the a priori categories. Or you might disagree in many other ways.)

The point in the above example is something close to what Hume wants (this is contingent on my memory of Hume from my freshman year course is true). That is, there are some things that we observe directly in the world--primary qualities or something--and there are things that we apply to our observations in the world.

Causation belongs to the second category according to Hume. His argument, if I recall, relies on his specific empiricist theory on the origins of knowledge (that it must have its source through the senses or through something like the analytic manipulation of concepts, and the source of all concepts is from experience, and the way that concepts are gained from experience is that there is a picture of what you experience in your thoughts [I think this is something close to what he says...I need a refresher.]) But we can make the point on our own way too. Take a billiard ball hitting another billiard ball. A very reasonable thing to say is that the first billiard ball caused the second billiard ball to move. Without a doubt, this is very reasonable. But how do we know this? From experience? What experience? The experience that other billiard balls cause other billiard balls to move? That certainly won't do, because it's clearly circular. So instead we want to say something such as, we know that Ball1 caused Ball2 to move because whenever we see Ball1 collide with Ball2, Ball2 moves. This gets more clear on the point--what we really observe is just a collision event; causation is more like tastiness, something that our faculty of reason applies to the observation. Just as my mind applies "tasty" to applies, my mind applies "caused" to a collision event. And even if Ball1 collides in this way with Ball2 many times and each time Ball2 moves, all we have is more of the same collision observation. For Hume, it's unclear where causation gets in.

All of this is to say that I realized that I usually tell this story wrong, both to myself and to others. I usually say that the difference is between causation and correlation, even though I know that Hume talks not about correlation, but rather "constant conjunction." I realized today that there's an important difference here. I think that Hume would probably say that correlation, as we use the term today, is just like causation in being outside of observation.

Correlation refers to a trend, or a tendency for two things to go along together. When I took a statistical inference seminar what the teacher reminded us is that there's alternate explanations for a correlation between two events, other than one causing the other. For instance, one further event might cause both of these events. Or something like that.

Another alternative is that it could be some sort of coincidence that these two events occur with each other all the time. In such an event, the proper thing to say is that there is actually no real correlation between the two events. Rather, there has been some sort of mistake--maybe a sampling error, leading us to think that these two events go together, when in fact they do not. In such a case the proper thing might be to say that there was just a coincidence when we observed these two events occuring together.

My point is that this is what Hume needs. His point is that all we see are coincidences, snapshots of the world as it is at a particular moment, not extended through time and not extended through space. All we have in experience are a series of unrelated flash-frames of the world, and all you get are coincidences, and you can't know causation through experience. But I think that also means that you can't get correlation either.

(Another way of getting to this point much more quickly: Hume also famously argues that we have no empirical grounds for believing in induction. How can you establish a correlation without relying on induction?)