Woo hoo!! :) Really excited that LRC has finally published its Mike Rossi article. I thought the article did a good job in taking a firm line and laying out the evidence, and I was happy to see the many references to the good, hard work that was done on this thread. Good work everyone!
There were two great items from the this thread that I was sad to see not make it into the article. First was RobE's statistical analysis indicating that, based on a database of 100 runners who ran a half marathon followed by a marathon within the next 90 days (like Mike Rossi), there is only a 1% chance that Mike Rossi would have run faster than 3:24 in the marathon. This was based on a regression analysis of marathon times vs half marathon times. This line of argument is complementary to GregTR's analysis of Mike's race times quoted in the article. It is a way of making the point that not only is Mike's 3:11 at Lehigh an outlier from his other times (as GregTR demonstrates), but also that outliers of this nature are very unusual (as RobE demonstrates with his regression analysis). Go to page 79 of this thread for more details. In my opinion this analysis is awesome against a skeptic who wants to argue that such a big improvement isn't that unusual. Well yes, in fact it is quantifiably quite unusual!
Second, I was happy to see my 'z-test' analysis referenced in the article (an analysis of the number of locations that the 100 finishers closest to Rossi were captured on the course).... however, a much stronger version of this argument has been developed. Namely the 'checkpoint' analysis which looks at what fraction of the top 200 finishers are seen at each of 5 photo checkpoints. Based on this analysis the chances of Mike not being photographed assuming he ran the whole course are about 1 in 586,000. See page 422 of this thread for more details. Not only is this statistically more damning than the analysis of # of locations photographed, it is also simpler and thus more bulletproof statistically (for example it doesn't have to rely on a Pearson's chi-squared test to establish the normality of the # of locations). The article does talk a bit about the checkpoints, mentioning that all of the top 200 non-Rossi finishers are seen at one of them, and linking to the great 9min video which lays out the statistics. But I would've liked to see the statistical argument in black and white for those who don't click on the link. 1 in 586,000 chance of no photos at these checkpoints assuming he ran the whole race - powerful, no?
Still, these are relatively minor points. It would be hard to read this article and not be persuaded that Mike Rossi didn't run the race. BroJos, I'm really glad you finally published it!!!