? Posh and Becks, cont. | Main | Don't F*** With Hitch ?

## July 18, 2007

### If all else fails: The Mule

J. Richard Gott III has forecast, amongst other things, the longevity of Broadway plays, newspapers, dogs and the tenure in office of hundreds of political leaders around the world. Also, that humanity will be around for at least another 5100 years. Shades of psychohistory... John Tierney echoes Gott calling it a Copernican Principle (via Instapundit).

Suppose you want to forecast the political longevity of the leader of a foreign country, and you know nothing about her country except that she has just finished her 39th week in power. What are the odds that she’ll leave office in her 40th week? According to the Copernican Principle, there’s nothing special about this week, so there’s only a 1-in-40 chance, or 2.5 percent, that she’s now in the final week of her tenure. It’s equally unlikely that she’s still at the very beginning of her tenure. If she were just completing the first 2.5 percent of her time in power, that would mean her remaining time would be 39 times as long as the period she’s already served — 1,521 more weeks (a little more than 29 years).

So you can now confidently forecast that she will stay in power at least one more week but not as long as 1,521 weeks. The odds of your being wrong are 2.5 percent on the short end and 2.5 percent on the long end — a total of just 5 percent, which means that your forecast has an expected accuracy of 95 percent, the scientific standard for statistical significance.

Which is very interesting though I am not certain Tierney's math is quite working. Still. The snag: Gott thinks our long term survival depends on a Mars colony up and running in the next 46 years. I say we make it.

Posted by Ghost of a flea at July 18, 2007 07:03 AM

## Comments

Hmm, yeah. I thought the "magical" number for statistical significance was about .66. Or, sixty-six percent. Or, about the size of the area you would expect to find within the first deviation of a normally distributed curve. If the correlation fits in 95 out of one-hundred cases, that is. Measures of confidence are not the same measures of significance. We also want to test to a .99 confidence interval, if we're able to test for heteroskadisticity, no?

Posted by: OregonGuy at July 18, 2007 04:05 PM