Statistical Modeling, Causal Inference, and Social Science: History of Monte Carlo

History of Monte Carlo

Monte Carlo is the ubiquitous little beast of burden in Bayesian statistics. Val points to the article by Nick Metropolis "The Beginning of the Monte Carlo Method." Los Alamos Science, No. 15, p. 125, 1987 about his years at Los Alamos (1943-1999) with Stan Ulam, Dick Feynman, Enrico Fermi and others. Some excerpts:

TCO

I'm just starting to check this site out.

a. What is the theme?

b. What is y'all's political leaning?

c. What do you think of MAKING IT COUNT (Leiberson)?

d. What do you think of the statistics argument wrt MBH98 hockey stick and such on the Climate Audit blog? http://www.climateaudit.org/

Martin Ringo

Andrew,

While I knew of Metropolis, I had never read anything directly written by him, and I found the Metropolis article interesting. When I started doing Monte Carlo experiments in the late 60s, my old man gave me, in an oral form, much the material in the article. As it turned out he -- my dad -- was a year ahead of Metropolis in the U of C physics program and may have been the only nuclear physicist in the program who did not work on the Manhattan project. The Navy would release him from the Naval Research Labs (Sonar) for a wild idea like nuclear fisson. Anyway after hearing much of the original and later uses in nuclear modeling, it was kind of nice to read about it from one of the early practioners.

Regarding the substance of the review, Metropolis views Monte Carlos like a physicist where it is often used as an exploratory tool. After nearly four decades of designing Monte Carlo experiments, I have a much simpler view of the technique, at least for the great majority of the work. Monte Carlos simply give the research the ability to integrate distributions that may be so intractable that even trying to write them down is difficult. This observation lead, in the 60s, to brief period in my field, econometrics, in which the tool was considered, shall we say, somewhat lowbrow. Fortunately, the field woke up and realized that integration by generation of (pseudo ) random numbers by arithmetical means, while sinful, is quite useful.

sounds sort of like doing numerical approximation for an integral that is difficult to evaluate analytically.

Gloria

I found a downloadable book on design and analysis of Monte Carlo experiments.
http://ideas.repec.org/p/dgr/kubcen/200417.html
http://www.exphs.org