Hi, all. I got an email asking for my thoughts, on this thread. OK, listen. First, I am a Pats fan. This is my back.
View attachment 9636 Second, I am a statistician. Let me give some background. I saw a link to the AEI report with the excerpt : "our replication of the report’s analysis finds that it relies on an unorthodox statistical procedure at odds with the methodology the report describes". Since statistical analysis is my job, I decided to check it out, for the hell of it. A major piece of the AEI report (pages 2-3) is showing that they could not replicate Wells' results. When I ran the appropriate analysis, my results matched Wells exactly. So, I got pissed off. If a major component of Tom’s defense is this AEI rebuttal, I want it to be done correctly. That’s the main point. AEI used the wrong statistical method for the dataset, for their replication, and their other analyses. So, I emailed AEI, asking for an explanation. No reply. I forwarded it to Volin, because I saw he had interviewed them, and thought he might put me in touch with them. No reply. Then Volin emailed, wanting to get my story, and that’s when he wrote the article.
Overall, my opinion is that Wells’ conclusions are total B.S. There is no useable data. Here are my thoughts on the analysis I’ve been sending:
The primary result that Wells points out is the main effect of Team. So, the mean pressure drop for Patriots balls (-1.2023) is significantly greater than the mean pressure drop for Colts balls (-0.4688), with a p-value of 0.0042. However, on page 40 of Appendix 1 in the Wells report, they have the following description of the effect of the Ideal Gas Law on Patriots balls:
"For example, using the most likely pressure and temperature values for the Patriots game balls
on the day of the AFC Championship Game (i.e., a starting pressure of 12.5 psig, a starting
temperature of between 67 and 71°F and a final temperature of 48°F prior to the balls being
taken back into the Officials Locker Room), these equations predict that the Patriots balls should
have measured between 11.52 and 11.32 psig at the end of the first half, just before they were
brought back into the Officials Locker Room"
Which is the mean pressure drop of Patriots balls shown by the data. The mean halftime pressure of Pats balls for Gauge A is 11.3; for Gauge B, it's 11.5. So, the Patriots balls had the expected pressure drop, and this was significantly different from that of the Colts balls, most probably due to the fact that the Colts balls were measured when they were back at higher temperature.
The other issue with the analysis results is the Team*Gauge interaction. Looking at the simple effects of Team at each level of gauge: the difference in mean pressure drop between Colts and Patriots balls for gauge A is 1.0159 (p = 0.0005), while the difference in mean pressure drop between Colts and Patriots balls for gauge B is 0.4511 (p = 0.06). The difference in these simple effects is the reason for the significant Team*gauge interaction (p=0.0005). However, there is no reason that the Colts vs Pats differences should differ by gauge. Except that the data for which gauge was used when is incorrect.
So, the gauge variable data is unknown, the data was taken at different temperatures, and the exact pre data is unknown. Among other issues I'm sure I'm missing. So, as I stated, with this flawed, unreliable dataset, I don't think any definite conclusions can be made. I was just verifying results using the correct methodology. Overall, the Wells report was correct analysis of bad data, and the AEI report was incorrect analysis of bad data.
Now, although Wells’ conclusions are B.S. (and obviously this whole thing is complete B.S. and never should have been made an issue), AEI f—ked their statistical analyses.
I know, I probably shouldn’t have brought the issue up, and I’m starting to regret it, but my only original point was that AEI should not have made an issue of Wells analysis method, when theirs was actually wrong. And that if they’re going to back the Pats, get it right. Nothing about conclusions.
Mike DeSarno
