Traditionally, economists have put the facts in a subordinate role and theory in the driver’s seat. Plausible-sounding theories are believed to be true unless proven false, while empirical facts are often dismissed if they don’t make sense in the context of leading theories. This isn’t a problem with math -- it was just as true back when economics theories were written out in long literary volumes. Econ developed as a form of philosophy and then added math later, becoming basically a form of mathematical philosophy.His proposed solution is to turn the thinking over to machines instead of people. I wonder how well that will work? Economic activity is based on decisions made by people for fairly complex reasons that machines don't experience. On the other hand, they'll be considering the data at a level of abstraction that will mask the actual causes of the decisions, and treat the decisions themselves as the data. Does losing the causality hurt anything? We'll find out.
In other words, econ is now a rogue branch of applied math. Developed without access to good data, it evolved different scientific values and conventions. But this is changing fast, as information technology and the computer revolution have furnished economists with mountains of data.
My concern, of course, is that if it does work well the people literally drop out of the equation. We'll make policy as if the decisions were what counted, and not the people who make those decisions -- and especially not what they hoped to achieve by making the decisions. But the purpose of economic activity is that very thing we'll be dropping out of our system for thinking about the activity. The real question by which you should judge economic activity is how well it helps the many, many people involved achieve what they'd hoped to achieve by undertaking an economic action. If you get a job, does it help you flourish, or does it lock you into a company store? Are you blocked from making economic decisions about selling your milk or pumpkins because of oppressive government regulation, and your inactivity doesn't show up in the data because it never rises to the level of a decision that can be counted?
Just the things I think we ought to care about are the things that will drop out of consideration.
17 comments:
"We'll make policy as if the decisions were what counted, and not the people who make those decisions -- and especially not what they hoped to achieve by making the decisions."
Not sure why that would be. Surely we'll make policy as if we cared about the results that matter to people? Of course we'll have to try to collect data in a way that helps us think through not only what happened but what might have been prevented from happening. But the mere act of caring about results over wishes won't prevent us from considering things like whether power imbalances or dumb regulations trapped people in places where neither they nor we want them to be.
There are ways to collect data on those problems, too. Will passing Law X leader to more or fewer people stuck in a company store? Will Law Y lead to more or fewer people discouraged from selling milk or pumpkins? But the answers will depend on data, not on arguments over whether one is a good or bad person for wishing for this or that result.
What I mean by that is that the kind of data that you can collect easily is purchasing or sales decisions, the election to take a job or open a business.
You can guess less effectively about some things that don't show up as positive signals: how many people stopped looking for work? We don't really know, but we have a sense in terms of how far off the positive numbers are from what we expected for various reasons. Unfortunately, that kind of data -- which would be hugely helpful for the kind of thing you bring up re: selling milk -- come with a lot of noise, because they really are based on assumptions and guesswork rather than positive countable signals. That's a problem in several ways.
All of this, though, is a question of "What did they do?" or "What didn't they do that we expected to be done?" What traditional economics -- Adam Smith, for example -- is concerned about is why they did it, and what will maximize outcomes for all those various whys. Now we're counting actions instead of people, and we aren't interested in why they do what they do: we just want to count what they do.
I don't know anyone who advocates limiting ourselves to counting what people do. That's not the only kind of data that we can collect or calculate, whether or not we turn our calculations over to machines.
I took the article to be making the same kind of criticism of economics that many of us make of climate science: it's a bad idea to make up the theory according to what seems plausible, then discount the facts if they don't fit. Sometimes the advantage of using an automatic analysis at certain stages is that you reduce the bias of wanting to make the data fit the theory because the theory is plausible (or familiar, or convenient, or compatible with other philosophical positions, and so on). If the data don't fit your theory, you're still free to conclude that they're not the most important data, but that's better than convincing yourself that the data are different from what they are.
Then your job is to figure what what information you think is more important to your ultimate goals for the economy, and go look at it instead. Maybe it's numerical, maybe not. But it's hard completely to exclude the numerical from economic analysis, unless you conclude that it doesn't matter how much wealth people have. Of course it's a defensible position that it doesn't matter how much wealth people have, but it's incompatible with most of what we call economic inquiry.
I guess I was thinking less about climate science than about your recent post on 'stands to reason medicine.' In a way, there's a valid insight here: what 'stands to reason' can turn out to be wrong. Certainly lots of economic theories that have sounded valid to very smart people have proven not to be at all accurate. So there's something to be gotten out of it.
On the other hand, I think there's a danger in moving from the qualitative to the quantitative. We don't really get involved in economic activity to make more widgets. We get involved in the business of making widgets only because there's something else we value, and making widgets for people is a way of supporting that thing (whatever it is). So it's that thing (whatever it is, and it may well be different for different people) that was really the important thing, the goal of the enterprise. And I don't see how we capture it by looking at the data on economic activity, which is really about what and not why.
I don't mean to dismiss the whole idea, but to point to a danger that seems to be lurking there for me.
Agreed. We rarely get involved in any quantitative analysis for the sake of the numbers themselves.
The way I might put it is that economics won't help you decide what kind of world you want. It's just supposed to help you understand how things work out if you arrange your economic affairs this way or that. A doctor uses medical science to decide which cures are most effective, not to decide whether health is a good thing, or even which parts of health are most important to each patient: strength? resilience? ability to live in challenging disease conditions? longevity? freedom from unpleasant medical intervention? No one wants to consult a doctor who doesn't know the difference, or read an economist who can't understand why people care about the results of economic issues on their real lives.
Similarly, a materials scientist tells you how to make a better steel, not whether your life will be happier with or without a new airplane. By the same token, if a philosopher concludes that a new airplane will reduce our quality of life, you don't want him in charge of the numerical analysis of the strength of the steel, unless you happen to know he's also a good and honest materials scientist.
The right tool for the job, and a respect for truth. They work wonders.
A couple of thoughts. One is that machine learning is just modeling with a different jargon. At bottom, the machine learns what the programmer told it to learn with the code he wrote--and all of the biases and prejudices, subtle and overt, the coder brings to his coding. In addition, what the machine "learns" via that program is entirely dependent on what a researcher applies as inputs. As we see in an extreme example, climate models can be made to produce whatever outcome--whatever outputs--the inputter wants for his results.
The second thing is this: On the other hand, they'll be considering the data at a level of abstraction that will mask the actual causes of the decisions, and treat the decisions themselves as the data. Does losing the causality hurt anything?
Economics, at least as much as any other human endeavor is an iterative thing. The decisions made are data as well as causes. A decision causes an outcome which informs a subsequent decision which....
What needs to be not lost is the fact that economic decisions are both causes and effects--data.
[E]conomics won't help you decide what kind of world you want. It's just supposed to help you understand how things work out if you arrange your economic affairs this way or that.
Indeed, and that's a too common misunderstanding of what economics is. Give 100 economists a set of inputs and ask them what will happen, and you'll get only one or two different answers. That's economics. Give 100 economists a set of outcomes (perhaps from those same inputs) and ask them what ought be done about what happens, and you'll get 90-95 answers. That's political economics.
Eric Hines
Economics, at least as much as any other human endeavor is an iterative thing. The decisions made are data as well as causes. A decision causes an outcome which informs a subsequent decision which....
It's the distinction between efficient and final causes, though. The economic causes you're talking about are things that move other things. The final cause is why anything was moved. It was the whole point: did we achieve it or not?
There are no final causes in economics. Everything that happens is interim, as it must be, since there is no single chain of events, linear or otherwise. There are a vast multiplicity of chains in progress, in various stages, every one open-ended, and each of them influenced by all of the others.
Even for an individual actor, the chain of his decisions, outcomes, further decisions, further outcomes does not end until his death. His sequence, though, lives on in the actors surviving him.
An economist explaining price movements on a supply/demand curve in a high school course might describe an increase in demand inflating price and stopping there, but that's only for simple illustration. The response to that is a decision to change the supply state, followed by other decisions by a range of actors, etc, etc, etc. We haven't even gotten to innovation, yet. Or to the flash storms of recessions or booms. Nothing ends. No final causes or outcomes.
Eric Hines
There are no final causes in economics.
That's not true of classical economics: Smith could talk about getting what you need from the butcher because of his self interest; Marx could talk about the good of the human being on which economic activity was premised; Schumpeter about the ways in which we could circumvent oppressive structures such as Marx envisioned and still achieve our ends. The problem is that it might become true of this sort of economics. That would be a weakness.
It's absolutely true of classical economics. Smith, for instance, could only talk about getting what you need from the butcher because of his self interest because of the iterative interaction of all of the self interests involved. He never said anything about a single exchange and then all was done.
The invisible hand was precisely that iterative, ceaseless cycle.
Eric Hines
So your opinion is that final causes don't exist just because there's too many of them? Everyone has a final cause for being involved in the market. To consider none of them, and get mired in the efficient causes, is to lose sight of what is really most important to all the economic actors. It's not one thing, yes, it's all their various final causes. But ultimately, those things are the important things.
No, I've been quite clear. I've said nothing about their number only their unending sequence and interactions.
It's those sequences and all the interactions with all the actors that keeps an individual, or any collection of them--related to each other or not--in the market. Or not.
I don't know where you're getting this "consider none," either; you're the only one talking about that.
Individual cause/outcome pairs are always considered, as are abbreviated sequences of interactions, for modeling the larger behavior. Just like any science or social pseudo-science does.
Eric Hines
"We rarely get involved in any quantitative analysis for the sake of the numbers themselves."
Do you know any mathematicians? But, yes, the occurrence of mathematicians among the population is rare.
It is the compilation of individual decisions, and to me the best analogy is that of nature itself. Each creature and specie of flora has its self interests, and works towards those ends, but it's impacted by the actions of other creatures and the availability of food sources, etc., and in turn their presence and actions affect others, repeat ad infinitum.
...you're the only one talking about that.
Yes, that's just what concerns me. The discipline will do just what you say it ought to do: deny final causes exist, as physics has tried to do (as might make sense at that level of organization), and no one will talk about them any longer.
I've been quite clear. I've said nothing about their number only their unending sequence and interactions.
What I believe you said was, "There are no final causes in economics." That sounds like a number to me: to whit, zero.
"We rarely get involved . . ."--Yes, I had mathematicians in mind, so I didn't say "never." :-)
"There are no final causes in economics." That sounds like a number to me: to whit, zero.
Quibble, I claim. My remark about "nothing about number" was in response to your mischaracterization that I'd said there were too many final causes. If I'm claiming there are none, I can't be beefing about there being too many of something that doesn't exist.
...deny final causes exist, as physics has tried to do (as might make sense at that level of organization), and no one will talk about them any longer.
What final causes do you think exist in physics? If there aren't any (as physics suggests), there aren't any about which to talk.
Eric Hines
I'm for better data and better analysis, but economics is one of the few fields that has not fallen entirely to the social justice warriors, and I highly suspect that this is where this push is really going.
Post a Comment