Jim Manzi: Correlation, Causation, Understanding, and Predicting
Jim Manzi, curious as always (especially about how to evaluate government policies), tries to plumb the problem of causality. Here’s where he begins:
Consider two questions:
- – Does A cause B?
- – If I take action A, will it cause outcome B?
I don’t care about the first, or more precisely, I might care about it, but only as scaffolding that might ultimately help me to answer the second.
I’ll risk going Manichaean here in an effort at (simplistic) clarity: I think this encapsulates a key difference between scientists and engineers.
Scientists want to understand how something works — not just “does A cause B?” but “what is the mechanism whereby A causes B?” Successful prediction is valuable (mainly? only?) because it validates or invalidates that understanding. Their main goal is to build coherent theory and understanding. Prediction is a happy by-product and necessary corrective.
Engineers want to understand how something works so that they can predict things — and create things that capitalize on that predictive power. They’re perfectly happy if they can predict successfully without understanding why the prediction works. Coherent theory is generally necessary to achieve this — they need to understand how things work — but it’s not their ultimate goal. Theory is a necessary evil.
An assertion of causality requires both. You need to show that B follows A reliably, but to be confident of causation, you need to explain — really tell a story — about how that causality works.
Both of these approaches are necessary and proper, of course, and they’re complementary. But I’d suggest that theory — the goal of science and most scientists — is what really matters in the long run. It’s easy enough to predict that the sun will rise in the east every day. Successfully predicting that yields many happy benefits. But understanding why — heliocentrism, earth rotation, gravity, momentum, etc. — now that is really profound. That coherent theory provides engineers with their necessary evil, so they can create and capitalize on further successful predictions.
With his “I don’t care about the first,” Jim puts himself squarely in the “engineers” camp that I’ve (again simplistically) described. Don’t get me wrong: I’m quite certain that Jim Manzi is quite curious and hungry for understanding (even while he’s skeptical about our ability to understand how complex human systems work). But by putting himself in that camp, he is aligning himself with, and providing aid and comfort to, a group that actively distrusts and dislikes egghead elitist types with all their fancy theories about evolution and climate and yes, economics. He aligns himself with a group that doesn’t really care about understanding, that just wants to know “which button should I push?”
It’s a group that tends to come up with answers like “Just cut taxes.”
Cross-posted at Asymptosis.
Or come up with ‘just raise taxes.’ Apparently there’s little to no correlation between lowering taxes and raising economic activity. OK. I don’t need to know why exactly, so much as I need to know raising taxes helps pay the bills, and I’m sort of risk averse about not paying bills. So I raise taxes and understand that doing so does not hinder economic activity. My guess is that diddling with tax rates that are reasonable to begin with, is really a waste of time. Leave them alone and keep them progressive and make them high enough to pay the bills. Meanwhile the other, as yet apparently identified as causal towards economic activity, will do their thing. Someday we may actually know exactly what these dynamics are, but in the meantime I’m paying the bills.
It is the manager who simply wants to know which button to push.
Economic “scientists” seek simplified models and limited experiments that allow their theories to work. Engineers study science so that they will know whether they can make it work reliably. Since economic science requires ignoring much of the real world, there aren’t any economic engineers.
That does not keep there from being managers.
We could also raise taxes and not pay the bills. A very likely scenario.
This is why I have been so cautious in my recent posts about drawing any causation conclusions, though I firmly believe that they would have been valid.
I just don’t want the content of my post to get lost in a correlation-causation side bar.
But what you can conclusively demonstrate is that in the absence of correlation you absolutely cannot have causation.
The data tells us conclusively that tax cuts over the past several decades did not help the economy, and suggests that they have been counterproductive.
Ditto CG tax cuts and investment. The correlation is higher CG taxes and higher investment.
What one needs to do is construct a narrative that is both internally coherent and consistent with the data. But that still is not conclusive.
For real world data where you cannot run confirming experiments, I just don’t know how you get there.