troparra said:
Well, you can use the assumption that the average of the Patriots' readings was 12.50. You cannot assume that each Patriots ball was at exactly 12.50 psi because this is literally impossible given the equipment alone.
As for Exponent, they are just as guilty as Wells for this mess. It's not a case of bad data ruining good scientific methodology. The Exponent section of the report is riddled with bias, fake data, misleading statements, etc.
Absolutely. The core problem is terrible data, research design, and dubious assumptions. But they did a huge amount of deceptive sleight-of-hand to shine that shit up, including not just examples of "ehhh, this is a semi-defensible decision among several options" but also just simple bad statistical practice.
One thing that bugs the hell out of me that never got that much attention (to my knowledge) is the decision to treat the data like they had 22 independent observations for the Patriots and 8 for the Colts. There were only 11 Patriots balls and 4 Colts balls. Just because they were measured twice doesn't multiply the number of balls. This is like having a sample size of 5 people, asking each respondent about their intended vote choice 20 different times, and treating it like a sample of 100.
Even if you look past all the other shenanigans in this section (like assuming the balls were measured at the same time so not factoring in that the Colts' balls would have more time for their psi to rise, or assuming all the Pats balls started at 12.5 and the Colts balls at 13), the simple choice to treat this as a 30 ball sample completely drove the "findings." If they do the reasonable thing, which is to say that they don't know which gauge was correct but they'll run the numbers separately for each gauge, then they find that its impossible to conclude that the differences in the Patriots and Colts balls were meaningful. Because of course its fucking impossible with 15 closely clustered observations, including 4 in one group.