Talking about the urge to quantify things – even the stubbornly unquantifiable – leads me back to what I spoke of earlier (“Faces in the Clouds”, Oct. 20) about finding patterns even in random noise. I think these are two aspects of the same phenomenon.
We seem to have this information-processing machinery in our brains, constantly grinding away trying to integrate the flood of sensory input. Back in the visual cortex, for example, there are layers of neurons that specialize in things like horizontal contrast lines and sideways-moving objects. Further up in the processing, we’re especially tuned in to important things like human faces and facial expressions, to the point that people see them in rock formations and half-cooked tortillas. (If anyone thinks I made that last one up, I’d be happy to cite chapter and verse.) Other senses seem to be broken down in the same way, with local processing picking out specialized patterns in the raw sensory stream.
We’re looking for ordered data, because random noise doesn’t give our brains any traction, and they can’t stand it. Noise is the enemy of sensory processing – consider, say, blank-channel TV static. “What do you mean,” says the brain, “random flashes of light all over the visual spectrum? That’s not how the world works. Things stay pretty much the same color on that time scale, and stuff doesn’t just pop in and out that way without leaving a trail of motion. Something’s wrong. I’ll figure it out, just give me a minute. . .”
If we use our brains to think about non-sensory abstractions, we tend to map them to sensory data so we can get a handle on them. “Employee performance” is a tough concept to picture, but how about a ranking from 1 to 10? That’s something we can grasp (whether we should, in the first place, is a topic for another day.)
So we look for lines and curves on our graphs, and clumps of points on our scatterplots. The same systems that served us to warn about crouching sabretooth tigers now try to tip us off to epidemiology. And it wouldn’t surprise me a bit if we uncover higher-order structures (or neuronal patterns, at least) that work in a similar way. Higher cortical functions might have taken sensory processing as their model, and set themselves up to do unconscious curve-fitting and shape-filling in the world of logic and causality. Being able to infer cause-and-effect must have been quite a survival advantage, too.
Steve Postrel at SMU wrote me after my earlier post on this subject. He pointed that it’s true that the general public gets basic statistical patterns wrong pretty regularly, but scientists don’t do much better once things get past that. He’s got a point: one of his examples was the handling of global warming data. There’s so much information out there that you can argue just about any direction you want to on the subject. I am not going to get that debate right now (neither was he!) but whichever side of the argument you take is a statistical minefield. There are so many things that can influence the presentation of the data, and the conclusions drawn from it (starting and ending dates for sampling, location of same, error bars of the measurements – when you can even state them, hidden variables or assumptions in the models – it’s a mess.)
No doubt about it, whatever the human brain is optimized for, statistics and probability isn’t it. (Quantum mechanics sure isn’t it, either, come to think of it – and depending on your take, that has a generous dose of probability in it, too.) I suppose we shouldn’t be wondering why we don’t do it better, and be impressed that we can do it at all. . .