Tuesday, May 3, 2011

"Scientism" accusations need meat on them - math helps us understand the social and natural world

One of the things that frustrates me to no end about accusations of "scientism" and "physics envy" is that they are always hopelessly vague and wishy washy. What precisely do you have concerns about? It's rarely made explicit because in my opinion the critique is rarely carefully considered by those making it. Constrained optimization is occassionally cited specifically for censure. It's a convenient case because Samuelson explicitly mentioned thermodynamics in his early work on constrained optimization, which painted a bulls-eye on him for people who were more interested in painting him as a physics-envier than they were in good economic science.

Outside of that, few specific critiques of why a technique is appropriate or not are offered.

I thought of that this morning when I was looking at this new NBER working paper titled "Quantile Regression with Censoring and Endogeneity". This is precisely the sort of thing a lot of the "scientism/physics envy" people would talk about: fancy math for fancy math's sake according to them. That's really unfortunate. Here's a summary of what this paper does:

"In this paper, we develop a new censored quantile instrumental variable (CQIV) estimator and describe its properties and computation. The CQIV estimator combines Powell (1986) censored quantile regression (CQR) to deal semiparametrically with censoring, with a control variable approach to incorporate endogenous regressors. The CQIV estimator is obtained in two stages that are nonadditive in the unobservables. The first stage estimates a nonadditive model with infinite dimensional parameters for the control variable, such as a quantile or distribution regression model. The second stage estimates a nonadditive censored quantile regression model for the response variable of interest, including the estimated control variable to deal with endogeneity. For computation, we extend the algorithm for CQR developed by Chernozhukov and Hong (2002) to incorporate the estimation of the control variable. We give generic regularity conditions for asymptotic normality of the CQIV estimator and for the validity of resampling methods to approximate its asymptotic distribution. We verify these conditions for quantile and distribution regression estimation of the control variable. We illustrate the computation and applicability of the CQIV estimator with numerical examples and an empirical application on estimation of Engel curves for alcohol."

I wish they came out with this five years ago, because this actually sounds like a very useful technique. We could have made good use of this in an evaluation of child welfare reform that I helped perform for the Department of Health and Human Services a couple years back (the evaluation itself is still working its way through the DHHS bureaucracy, but background is available online and we've presented preliminary results). We used quantile regressions for the analysis, but one thing we didn't have was an IV version of a quantile regression. This would have helped tremendously because we were investigating a series of highly endogenous child welfare outcomes - but we did have a quasi-random treatment which would have worked with the CQIV presented here. Some of our variables were also censored because of institutional time constraints within the child welfare system. A quantile regression approach that deals with both censoring and endogeneity in such a clean way is great stuff.

The trouble is, if you open up that working paper you see a lot of math. I would have had to spend a fair amount of time sitting down and reading it very closely to understand the gist of it. Then I'd probably have to take the time of walking down the hall to Doug Wissoker's office - a senior econometrician here, whose door I knock on often - and humbly profess my remaining ignorance and ask him to help me. And then he'd have to sit and look at for a while to figure out what they were getting that. And then we'd also share what we eventually did with it in meticulous detail to our technical working group and then they'd make sure we were doing something reasonable. It would have been quite a process - but it would have considerably improved our understanding of how abused and neglected children fare in the child welfare system in this country. And that is a very good thing.

These are the sorts of papers, though, that often just get dismissed as "scientism" with no details on exactly what is wrong with investing the time to learn and use it. That bothers me a great deal.

17 comments:

  1. How would you describe usage of math for post mortem information from the past and usage of math for predicting the future?

    Your evaluation of child welfare reform relates to discussing past data.

    But what would you say about economics that uses mathematical models for made-up hypothetical situations or mathematical prediction of future outcomes?

    There lies the problem. One might say that nobody can predict the future anyway, so blaming math is pointless. And such an objection would be right. Nobody can predict the future, using any method.

    So surely you would object to the idea of mathematical prediction of human behaviour?

    ReplyDelete
  2. Hi Daniel, If you are interested in learning more about Mises' epistemology here is something I wrote on it. http://mises.org/daily/5158/Mises-on-Mind-and-Method

    (Let me say, in case that would you to read it, that your fellow econ blogger Jonathan Catalan thought highly of it.)

    Mises laid out his epistemology in 4 different treatises written throughout his career. In the article above, I try to distill it down to 23 pages.

    ReplyDelete
  3. I thought your comments on Keynes preface to the german edition of the General Theory showed more knowledge of the literature than I think typical of young economists. Not so much here, though.

    Knowledgeable people don't call constrained optimization scientism just because of some remarks by Samuelson. One idea is the structure of neoclassical economics was created by coping classical mechanics. It thus incorporated certain conservation laws. Unlike in physics, economists do not organize an empirical research program around those conservation laws.

    The economists to read here are, for example, Philip Mirowski and Nicholas Georgescu-Roegen. It is true that Mirowski amusingly dissects Samuelson.

    Are you aware that, over the last century, many mathematicians and physicists have looked at neoclassical economists and found it lacking?

    ReplyDelete
  4. It's always more fun to be just a Keynesian on here than it is to be a neoclassicist, Robert. When I'm just a Keynesian, I get disagreement from only one side. When I'm a neoclassicist, I get disagreement from both :)

    I would say this - you are holding constrained optimization in economics to the wrong standard. In physics, because of conservation laws and the precision of mechanics, classical mechanics is rightly treated as a "law of physics".

    No one makes such a claim for the math of utility or profit maximization in neoclassical economics. The real "physics envy" is in the assumption that because physicists make such strong claims, economists are also bound to make such strong claims - that if economists use any precise math it ought to be comparable to a "law". When you hear physcists critique economics, they are often making this same mistake you are.

    Economics is not physics. Constrained optimization in economics is not a law - it's an assumed relation that we use to build a model that we test against observation. Models aren't laws. You should not compare economists' use of constrained optimization to physics - you should compare it to things like the Lotka-Volterra equation in biology. No biologists suggests that that is a "law" of biology. It's a model of population dynamics that gets tweaked and tested and helps us understand the underlying properties of population dynamics without making any strong assumptions the way physicists do (and are able to do) with classical mechanics.

    Economists model things to test out theories. We think people try to increase how happy they are and that firms try to increase how profitable they are, but that both are constrained by budgets, time limits, etc. If we want to model anything (NOT establish a law) a little optimization seems appropriate. That's all there is. If you can accept that consumers try to increase their happiness but are subject to important constraints, then you should have no problem at all with constrained optimization in economics.

    ReplyDelete
  5. "Outside of that, few specific critiques of why a technique is appropriate or not are offered."

    Specific critiques:

    1)You can't control enough variables because the system is too complex RIGHT NOW to make sense of it.

    2)You can't control enough variables because the system is about to change in a big way.

    3)Approximately the most useful thing we know about economics is that economists exhibit confirmation bias all of the time. They're human and highly politicized.

    Here's a fine example of the kind of thinking that leads to ever more programs making it "easier" to buy a house. Notice any problems of the types 1 through 3?

    From 2003:

    http://www.newtowncdc.org/pdf/social_consequences_study.pdf

    1. Children of homeowners are likely to perform higher on academic achievement tests and are more likely to finish high school. Furthermore, children of homeowners have fewer behavioral
    problems in school and are less likely to become pregnant as teenagers. These outcomes survive many controls for parental education, martial status, and other statistical comparisons, as well as neighborhood characteristics.

    2. Political activity, like voting, as well as participation in civic organizations is higher among homeowners than renters after controlling for personal characteristics and socioeconomic status.

    3. Homeowners, again once controls are in place, are more satisfied with their lives and are happier.

    4. Some of the most recent research suggests that a high level of homeownership in neighborhoods enhances property values.

    In closing, I wish you would get a real job in real science. Anyone who can make sense of that math could be doing R&D in the energy sector, helping to make us all rich by making energy radically cheaper, obviating carbon taxes and the guns they come with or the endless political waste and nonsense associated with the arguments. Helping, maybe, to end wars over resources. Instead, you want to eat my tax money, dream up new programs, and slice the pie.

    ReplyDelete
  6. mobsrule, I'm not going to respond to this except to say that the fact that poor papers are written is not an indictment of the idea that economics is an empirical and theoretical science that can profitably use mathematical and statistical techniques.

    Good science critiques, refutes, and responds to such papers.

    Poor philosophy, poor logic, and poor prose is written too and I can provide you links to that. Is that an indictment of a more deductive epistemological approach to economics? Of course not.

    ReplyDelete
  7. Did you expunge my most recent comment on the Mises quote or is that just Blogger?

    ReplyDelete
  8. The following is false:

    "In physics, because of conservation laws and the precision of mechanics, classical mechanics is rightly treated as a 'law of physics'. No one makes such a claim for the math of utility or profit maximization in neoclassical economics."

    Philip Mirowski, for example, has demonstrated the falsity of the above.

    ReplyDelete
  9. Saying "no one" was ill-advised on my part, but "the large majority" should suffice.

    The fundamentals of your critique are still fallacious for the reasons I provided. I've browsed Machine Dreams before and recall finding it unconvincing for the same reasons you are unconvincing here: you're trumping up what economists claim and then holding them to a completely irrelevant standard, and then you're finding them wanting.

    ReplyDelete
  10. "...but it would have considerably improved our understanding of how abused and neglected children fare in the child welfare system in this country."

    I didn't need to do any of that tell you what their situation is like; it sucks.

    ReplyDelete
  11. What sucks Gary?

    The removal? The OTC hearing? The CMO process? The counsel appointment process? Visitation? Pursuance of relatives caregivers (does relative care make it more or less sucky?)? The continuances? Or is it the adjudication process that sucks? Service referal? The disposition? The placement process? Does reunification suck more or does it suck less Gary? What about the TPR hearing? Does that suck more or less than the alternative?

    And our fundamental question - does the CMC make each of these things suck more or less for these children?

    And that's just in the courts. The kids haven't even entered care yet.

    If one wants to actually help these children you have to know a hell of a lot more than "it sucks", and you have to know you have the best empirical evidence possible because there are vested interests who don't want change in the system and who are waiting to pounce on your analysis.

    I came on this project after coming out of an economics degree. Learning all these details about the court system was definitely not what I had in mind for my career at that point. The project turned out to be professionally satisfying because we did have a lot of interesting empirical and econometric questions to work out. And it did important work because my PIs recognized that you can't just say "it sucks" and leave it at that.


    You might have been entirely facetious with that comment - I recognize that. I just want to provide a detailed illustration of why precision and detail matter and why "first principles" and vague sympathies don't get you all that far.

    ReplyDelete
  12. Daniel,

    I am more than personally aware of the process.

    ReplyDelete
  13. You were involved in it, Gary?

    ReplyDelete
  14. Daniel,

    I think of the process as being industrial in nature; children are just widgets in the system, to be stamped out and sent on their way.

    Anyway, what a lot of the anecdotal scandals regarding the process have shown is (a) there isn't a lot of daylight regarding these programs and (b) there is very little incentive for the population at large to care outside of a major league scandal which quickly fades from the scene.

    When it comes to children libertarianism (as does any ideology) always has a tough nut to crack. I am very skeptical, however, whatever the good intentions, that the current state system of child welfare is going to have much headway.

    ReplyDelete
  15. Prateek,

    Given the criteria (then and now) I should have been and there was some movement as I recall towards that when I was a child (that memory could be incorrect); but looking back on it, I'm glad I wasn't. I have, in my travels, met a number of people who were; I can't say that I envy them even though ostensibly I should.

    That's really about as personal as I want to get on the matter.

    ReplyDelete
  16. Well... Amanda Kowalski has a code to perform CQIV on her website... and there is an algorithm. Exposition could have been done better... suggestions are welcome. I could tell you that that the proofs were freakishly hard to do. It took me about 4 years to figure that one out.

    ReplyDelete

All anonymous comments will be deleted. Consistent pseudonyms are fine.