Wednesday, April 11, 2012

Assault of Thoughts - 4/11/2012

"Words ought to be a little wild, for they are the assault of thoughts on the unthinking" - JMK

- Noahpinion on the new Wieland paper, and when it makes sense to use structural or non-structural models ("ad hoc" always seemed a little harsh to me... they're not ad hoc, they're just not structural! It's not like you just write up a non-structural model on a whim or something!). He distinguishes between two uses of models: forecasting and policy analysis. For forecasting, "all of the above" strategies are fine. For policy analysis, you want structural models. I think this is basically right. However, there's a third thing that we want to get out of economic models that is actually the most important reason for me: explaining the economy. Not predicting its future path. Not advising policymakers. But simply understanding how it works. And for that, I think both non-structural and structural models are useful. Since explanation is preferably causal rather than just descriptive, we don't want to ignore the microfoundations.

- The IMF says that we should look to the 1930s for housing policy. As Keynes might say, appeasing the rentier will not get us out of the slump. I'm interested in Andrew Bossie's thoughts on this.

- Speaking of Keynes, Jared Bernstein discusses George Bush the Keynesian. Now I kind of miss the guy. Maybe we should have had Gore 2000-2004 to keep us out of Iraq as we prosecuted the war on terror, Bush 2004-2012 to ignore the deficit when it actually made sense to ignore the deficit, and Obama 2012-2020, to swoop in and fix health care and entitlements.

- More Keynes: Jonathan on Chapter 3 of the General Theory. Haven't gotten a chance to read closely yet, but you should not make the same mistake.

- I am trying to pull together a power analysis for our Sloan Foundation proposal, at their request. I know generally what these entail, but it's not something that economists generally use - usually it's used by people with smaller scale experiments in fields like psychology. Although just out of curiosity I've picked the brains of a friend (who just finished a psych PhD) and my sister-in-law (who is in the midst of a psych PhD) about it, and neither of them are big fans of the approach. But you do what the funders ask you to do (so long as it isn't blatantly wrong, of course). Anyway - because economists don't generally use these methods, I wanted to share what I found out about power analysis in stata in case any of you are interested. The UCLA site has two commands of interest: powerreg and sampsi. You can import both by typing "findit powerreg" and "findit sampsi". Powerreg does a power analysis for regressions and sampsi does it for difference of means tests. I didn't even know you could do power analysis for regressions, and it's a little screwy - you ultimately get out the improvement in R-squared that you can detect with a given sample size and power level... but who thinks about regressions in this way? I would have thought you'd be most interested in the change in coefficient that you'd be able to detect with a given sample size and power level. So I think I may stick with the more traditional difference in means to satisfy Sloan's curiosity. But stata can do both if you ever have a need for it.


  1. One could make that argument that the DSGE is based on Brownian motion, which is flawed.

    The IMF's policy recommendations seem to be heading in the right direction, but then again, there's the question of moral hazard. At least it isn't as much of a question of moral hazard, when compared to Steve Keen's proposal of a debt jubilee.

    Even though George W. Bush had N. Gregory Mankiw as a policy adviser, that doesn't do much to redeem him in my eyes, despite my support for Keynesian economics.

    Chapter 3 is only an introductory treatment for what Keynes does later in Chapter 20. Keynes himself provides a footnote that refers to Chapter 20 on Page 31 (I have the 1964 Harcourt, Brace and World edition). It is in Chapter 20 that you can find Keynes's technical formulations for a model.

    As for power analysis...I hope you aren't making the mistake of assuming there's a normal distribution without having a goodness-of-fit test. While I'm not that far into statistics, I do know that the misuse of the Gaussian distribution is common in the social sciences.

    1. I think the normal distribution criticism is mostly off-base. Yes, actual variables of interest may not be normally distributed. But social scientists are usually concerned with the distribution of given estimators - say, a regression coefficient. And while the variables of interest may not be normally distributed, the error around an estimator of interest is. Now, of course you have to think about whether the model specification that that estimator is a part of is a good one, but the normality assumption is perfectly appropriate as it pertains to the estimators, unless you can explain to me why I should think otherwise.


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