Comments for Normal Deviate

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Paul Puglia

Paul Puglia — Sun, 26 May 2013 01:14:57 +0000

This page, written by a meteorologist, while a little outdated, has some musings on possible approaches to the calculation, and more interestingly, also lists some references on the distributions of violent tornados http://www.flame.org/~cdoswell/tor_probs/vtornado_prob.html

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Ken

Ken — Sat, 25 May 2013 23:18:25 +0000

Given that you know the tracks for a number of years, a more accurate estimate can be obtained by dividing the area into house size blocks and then seeing how often the tornadoes cross. Then modify this probability based on number of blocks occupied by houses.

In many ways I find this type of question a bit pointless. After all, it happened, so it has probability one of happening. Chances of it happening again are probably very small.

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Mike Anderson

Mike Anderson — Sat, 25 May 2013 22:37:53 +0000

Here’s an order-of-magnitude estimate using a very crude Poisson model, suitable for my freshman undergrads.

Let the estimated annual rate of destruction be the ratio of destroyed houses in Oklahoma to total houses in Oklahoma (a la Banana’s comment). Then find p = Pr(D > 1 | rate*t) where D ~ Poisson(rate*t). Remember, this is like the Birthday Problem: any house, just like any day of the year, will do for the improbable TwoFer. For example, t=15 years, and if rate = 0.001, then p ~= 1.1×10^(-4), about 1 in 9000; rate = 0.0001 gives p ~= 1.1×10^(-6), about 1 in 900,000.

So Sawyer would have been off by anywhere from 8 to 10 orders of magnitude. No wonder we think journalists are clueless gits.

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Fran

Fran — Sat, 25 May 2013 21:03:52 +0000

In fact, we can forget about the lower risk area, using all the tornado areas already gives a minimum value estimate so, even easier.

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Fran

Fran — Sat, 25 May 2013 20:52:52 +0000

Gee.. I mixed up higher and lower everywhere! It should be the houses destroyed twice in the lower risk area among the tornado areas… I really need to read twice before I shoot Enter.

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Fran

Fran — Sat, 25 May 2013 20:38:20 +0000

I would do a simulation for an accurate estimation on this one but, for the purpose to check how right Ms. Diane Sawyer is with her “one in a hundred trillion chance”, I would calculate a minimum value by considering the total number of houses in the highest tornado risk area in the US and the number of houses that were destroyed twice in that area.

The true odds will be higher than this calculation, but this one is simple to do and my guess is that it will be way, way bigger than Ms. Sawyer estimation and enough to prove her way, way wrong… which seems the whole point of the calculation in this case.

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Corey

Corey — Sat, 25 May 2013 20:29:08 +0000

‘She also did a long series on “Buy America” encouraging people to shun cheap goods from abroad. This is like telling people who live in Cleveland to shun buying any products and services not produced in Cleveland.’

I disagree. I think it makes more sense to view Sawyer’s series as an effort to deliver a positive shock to net exports. Valid criticism might be made on this basis, but not on the basis that promoting intra-country trade is very similar to promoting intra-city trade. After all, the U.S.A. is a fiscal union in which internal barriers to trade are low and worker mobility is high, permitting a lot of local specialization.

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Corey

Corey — Sat, 25 May 2013 20:19:38 +0000

Let S be some set of people, and let property U be unluckiness in the form of having experienced at least two home destructions due to F4 tornadoes. Here are two events that might be of interest: (1) There exists a person in S with property U; (2) a person selected uniformly at random from set S has property U.

Comment on Brad Efron, Tornadoes, and Diane Sawyer by Banana

Banana — Sat, 25 May 2013 17:57:24 +0000

For any location in the US, there are very detailed data showing the history of tornado frequency, strength, path, and time of year. So it should be pretty straightforward to compute the frequency of tornadoes of a given strength in a given location. See http://www.spc.noaa.gov/exper/envbrowser/ for a nice visualization of the data.

Comment on Aaronson, COLT, Bayesians and Frequentists by tdietterich

tdietterich — Thu, 23 May 2013 03:05:24 +0000

Can you explain why asymptotic consistency is needed? In the PAC model, I specify epsilon and delta, and I only need a confidence interval of width epsilon with confidence delta. As a machine learning practitioner (not particularly fluent in the theoretical analysis of PAC algorithms), it would seem that once I have the required sample size (which is polynomial in 1/epsilon and 1/delta), I don’t care how the estimator behaves for larger samples.

Is the issue that I can specify epsilon arbitrarily close to zero, and hence drive the required sample size arbitrarily high? This would never be done in practice. I think most machine learning folks would be happy to bound epsilon away from zero.