A few times I've received this question, usually from people whose work intersects with economics, yet without those people needing to produce econometric studies themselves. "How can I better understand the empirical papers I am reading?" I have a few suggestions:
1. Attend some (empirical) economics seminars first, to get a sense of what you need to learn and how discourse proceeds and what sort of points end up being contested. Subsequent class learning will be more focused and productive.
2. Often a good hands-on undergraduate class is more useful for these purposes than a graduate class. The latter might have too much econometric theory and theorem-proving.
3. The quality of an econometrics class (for these purposes, putting aside frontier work) is not well correlated with the quality of the college or university it is being taught at. The quality of the class is instructor-specific.
4. The quality of econometrics in the profession has continued to rise. That is good news, but for the purposes of this discussion there is a downside, namely that mistakes are much less transparent. For an increasing number of papers, it is hard to judge the final quality of the work without spending a lot of time on it. Whether or not a paper can be replicated is a more important question, given that more researchers are operating with frontier-level techniques.
Do you have other suggestions?
Addendum: Here is Mark Thoma's course, entirely on-line and free.
Amen on #2. My graduate econometrics course was taught from the dreadful Greene book, which I'm pretty sure you could read cover to cover without actually learning how to run a regression.
"Whether or not a paper can be replicated is a more important question..": do authors make the data available to permit this? Or is Econometrics like Climate "Science"?
I recommend James Hamilton's "Time Series Analysis". Lot's of proofs, but every step is included.
Tyler could delete the word "econometrics" from point #3 and make a very deep and useful statement about "how best to learn" (a subject which he's covered before!)
I agree with many of the bloggers above. Kennedy for basic intuition and also Woolridge for Cross section and Panel data. Finally and most importantly by doing it.
After having read Green and Kennedy I now read Gelman s book. The latter one makes me see some things much clearer but maybe that is just because I read the other two beforehand.
You could download the two handouts I made available on this topic:
http://marcfbellemare.com/wordpress/2011/01/contr…
Agreed w/ anon. Playing around with the data helps make things click.
Agreed with the above. Forget about econometrics courses – they are taught by econometricians and thus useless. Instead, take a good applied empirical micro class – labor, development, public – and you will learn so much more. Larry Katz's Labor course at Harvard is legendary.
For reading, try Angrist and Pischke's "Mostly Harmless Econometrics".
Dougherty's intro book is one of the best textbooks I have ever encountered, perhaps _the_ best.
My solution is three parts.
Realize econometrics uses statistical methods.
Realize econometrics is supposed to test economic theory.
Learn statistics and economic theory.
Some healthy skepticism is due with respect to claim #3. Yes, the quality of a class certainly depends, all else equal, on the instructor. But it is misleading to claim that the quality of university/college is uncorrelated with the quality of the course. Far more than casual empiricism shows that: (1) the quality of students is increasing in the quality of the university/college; and (b) professors/instructors condition the rigor/challenge of their course on the quality of the students they have. Most professors/instructors who deny (b) are either liars or are experiencing a severe hallucination. Accordingly, the higher is the quality of the university/college, the higher will be the intellectual challenge of the econometrics course. If you doubt the educational efficacy of undergoing such an "intellectual challenge," chances are good you frequently delude yourself into thinking you "understand" things you are, in fact, quite clueless about.
That said, Wooldridge's undergraduate text and Angrist and Pischke's "Mostly Harmless Econometrics" book are both excellent sources.
Time series is so fraught with scientific hazards, it's best to just read the abstract and conclusion and rely on the skill of the authors to do the proper modelling and diagnostics. They don't provide the data for you to check their work anyway, so you have to accept their methodology at face value unless you're an econometrician. If there are modelling or estimation problems, someone is going to write a follow up.
Not all expert econometricians, though, get it right. Case-Quigley-Shiller is plagued by endogeneity. Card and Krueger was based on awful data collection techniques (survey instead of payroll) and the worst sin in econometrics: improper selection of the dependent variable (they used employment level instead of labor hours employed). C&K, addressed these shortcomings in a subsequent paper, but it doesn't excuse shoddy work in their original paper.
Gujarati is my favorite undergrad econometrics book. I also like Wooldridge. Enders is my favorite book on Applied Time Series.
The Chartered Financial Analyst (CFA) Level II, Volume 1 book or Schweser notes for that book have a brief but good overview of time series analysis in a single chapter. You should be able to find a used 2009 edition fairly cheap now that the exam is past. It's a good primer for someone with very little background who doesn't want to get wrapped up in too many deeply theoretical discussions.
Unluess you want to do work of your own, you don't really need to buy the Angrist & Pischke book – just read their 2010 article in JEP, which is available for free for anyone with an internet connection, to get a sense of how the causality problem can be tackled. To echo others, Introductory Econometrics by Wooldridge is pretty good as far as textbooks are concerned. Greene is the devil.
To any person interested in learning econometrics and statistical methods in general, please remember GIGO.
From Wikipedia entry on Garbage in, Garbage Out (copied today 02/20/11)
Garbage In, Garbage Out (abbreviated to GIGO, coined as a pun on the phrase First-In, First-Out) is a phrase in the field of computer science or information and communication technology. It is used primarily to call attention to the fact that computers will unquestioningly process the most nonsensical of input data (garbage in) and produce nonsensical output (garbage out). It was most popular in the early days of computing, but applies even more today, when powerful computers can spew out mountains of erroneous information in a short time. The actual term "Garbage in, garbage out", coined as a teaching mantra by George Fuechsel, an IBM 305 RAMAC technician/instructor in New York, was soon contracted to the acronym "GIGO".[citation needed] Early programmers were required to test virtually each program step and cautioned not to expect that the resulting program would "do the right thing" when given imperfect input. The underlying principle was noted by the inventor of the first programmable computing device design:
On two occasions I have been asked,—"Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?" … I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.
—Charles Babbage, Passages from the Life of a Philosopher[1]
It is also commonly used to describe failures in human decision making due to faulty, incomplete, or imprecise data.
The term can also be used as an explanation for the poor quality of a digitized audio or video file. Although digitizing can be the first step in cleaning up a signal, it does not, by itself, improve the quality. Defects in the original analog signal will be faithfully recorded, but may be identified and removed by a subsequent step. (See Digital signal processing.)
Garbage In, Gospel Out is a more recent expansion of the acronym. It is a sardonic comment on the tendency to put excessive trust in "computerized" data, and on the propensity for individuals to blindly accept what the computer says. Because the data goes through the computer, people tend to believe it. Decision-makers increasingly face computer-generated information and analyses that could be collected and analyzed in no other way. Precisely for that reason, going behind that output is out of the question, even if one has good cause to be suspicious. In short, the computer analysis becomes the gospel.[2]
End of Wikipedia entry.
FOOTNOTE: When working for my Ph.D. in 1967-70, I took advanced courses in statistics and econometrics (I worked every theorem in E. Malinvaud's Statistical Methods for Econometrics and I studied most of the work done in the Cowles Commission; see http://cowles.econ.yale.edu/archive/reprints/50th…. In my Ph.D. thesis on Argentina's money and bank credit, I learnt how unreliable data could be, and later in China I learnt how terrible data often was and how little economists were concerned about the data they were using (I still don't agree with Professor Gregory Chow's view about the quality of Chinese data). I strongly believe that in the past two years, almost 100% of the research work in the U.S. and other countries about the Great Recession has relied on terrible data about national accounts, public finances, monetary and financial statements, balance of payments, price indexes, and in particular labor. MR posts and comments on the Great Recession have often illustrated how much the old GIGO has become Garbage In, Gospel Out.
Rahul, there are ways to deal with the problem but unfortunately in most cases they are too costly to apply. You may benefit from reading these two papers http://www.radpro.com/GIGO-RPM.pdf http://www.imf.org/external/pubs/ft/wp/2008/wp082…
If you want to learn econometrics, you hopefully have a reason-a data set that you're interested in.
Learning econometrics is much easier in a class than by yourself.
In Washington, DC USDA used to offer applied econometrics.
You can also go to ICPSR in Ann Arbor, MI. They do one month intensive courses. The instructors are picked based on their ability to teach not research.
What is the history behind this thing called "econometrics"? Basically it's statistics with an emphasis on time series analysis. Every quantitiative discipline (physics, engineering, epidemiology, even quantitative sociology and psych) has its particular bag of common statistical tricks, but none of them elevate that bag of tricks into a seperate discipline. A physicist who has a particularly thorny data analysis question talks to a statistician, not a "physimetrician".
What bartman said. Do some modeling in excel so that you actually have to set up the matrix manipulations to impose constraints, etc. Take the best math stats course you can.
First off, there is a big difference between the econometrics of macro and finance (mainly time series stuff) and microeconomics.
For time series (which, admittedly, is not my thing), I refer to "Applied Econometric Time Series" by Walter Enders and "Analysis of Financial Time Series" by Ruey Tsay (the latter is a pretty easy read with software based examples).
For Microeconometrics, I agree whole-heartedly with the advice that someone take a course in labor, development, etc. That is the best way to learn how to do applied work. There isn't really a good textbook in my opinion.
I couldn't disagree more with number 2. Some of THE MOST important things one can learn about econometrics is just how wrong a simple regression can be due to issues like measurement error, endogeneity, selection bias, censored or truncated data, unobserved heterogeneity, etc. I can pretty much guarantee that undergraduate courses don't cover the majority of these issues. In short, what you learn as an undergrad only works when data is perfect. In the real world, data always sucks and you have to correct for that as best you can. If you don't your results could be really messed up.
Comments on this entry are closed.