Earthquake predictions: not likely!

So there's some website, called Quake Prediction, claiming that there is large chance that a large (6.5-7.0) earthquake will hit Los Angeles, and then another in San Francisco in the next few days.

RISK IN LOS ANGELES;
Oct 05: 05%
Oct 04: 14%
Oct 03: 42%
Oct 02: 72%
Oct 01: 97%
Sep 30: 98%
Sep 29: 90%

RISK IN SAN FRANCISCO;
Oct 05: 74%
Oct 04: 66%
Oct 03: 51%
Oct 02: 27%
Oct 01: 10%
Sep 30: 05%
Sep 29: 05%

The predictions above were taken from the website on September 29, 3:30 PM PDT (with links removed).  I note the time, because who knows how often the website changes.

I'm slightly confused by those numbers, because there seems to be some very bad math going on.  If there is a 90%+ chance of earthquakes on Sep 29, 30, and Oct 1, does that mean that there is at least a 70% chance of three earthquakes?  Someone doesn't understand probability... but let's ignore that for a moment.

There are many reasons to be very suspicious of this.  The first is that a lot of research has gone into forecasting earthquakes, but no one can achieve anything anywhere near this accuracy.  We don't expect there to suddenly be a breakthrough where we can start predicting earthquakes within a few days.  And if there is a breakthrough, we don't expect it to come from some standalone person with a poorly constructed website.  In fact, this is more what we would expect if there were some crank who just thought he could predict earthquakes.

Moreover, how does he know his predictions are that accurate?  The best and easiest way to know is by looking at past predictions and seeing how they aligned with reality.  Has he done this?  If he has, then why isn't it published?  Why isn't it all over the news?

In fact, I did find this guy in the news.  They called him the Quake Quack.  Note the date of the article: April 12.  It's talking about a prediction he made for a large (5.0-6.5) Southern California earthquake on April 12-14.  I don't remember that one ever happening, do you?

In fact, if you look back on QuakePrediction's twitter archives, you find that it's just littered with predictions that never happened.  He predicted a 5.0-6.5 earthquake on September 25, a 4.0-5.5 on September 14th, a 4.5-5.5 in September 6-7, and so on.  You can check these against the list of significant earthquakes on the US Geological Survey, and nothing matches.

It's a marvel how someone can be so confident when a cursory glance reveals that the predictions have nowhere near the claimed success rate.  Seriously, that kind of self-confidence is a Life Skill, but one that has been hopelessly misdirected.  A more cynical suggestion is that he's just throwing around lots of predictions just so he can brag about it when he eventually gets one right.

I was thinking of doing a more systematic analysis of the twitter archives, comparing his successful prediction rate to that of chance*, but then I remembered I have better things to do.  I was going to suggest that someone else do it, but then it occurred to me that if I have better things to do, you probably do too.

*Experimental note: Measuring the chance success rate is very simple.  Just take each prediction, and pretend he really made the prediction for a day N weeks later/earlier.  Try for many values of N and see if his success rate is significantly higher for N=0.

As a final note, this stuff is not harmless.  Here's a story from What's the Harm? on how a quake rumor disrupted a town.