5/24/2023 0 Comments Patrick mahomes statsI make this point just to say that removing outliers from football statistics with a low number of data points is not sharp statistical analysis, even though it can be a useful tool in other datasets. This was the true and correct statistical analysis of Danny's experiment, but instead his production went the other direction. Patrick Mahomes - NFL Quarterback - News, Stats, Bio and more - The Athletic Roster evaluation model 15 Chiefs Patrick Mahomes QB, 15 YDS 5,250 TD 41 INT 12 RAT 105.2 Player Bio HT/WT 6. Projecting for a 16 game season would have yielded the following stats.ģ96 completions, 568 attempts, 5,223 yards, 69.7% completions, 9.2 yards per attempt, 48 TDs, 8.9 interceptions, 120.14 passer rating. Due to the very low number of data points at the time, the effect on the rest of the data from removing them is large. This is because the only outliers based on his stats at the time were a couple bottom outliers on completions and an outlier game for too many interceptions. Had he used the 1.5 rule, the correct conclusion would have been that Mahomes was going to regress upwards and get better. But you're never going to have a QB with 1,700 yards passing or 27 TDs in a single game.ĭanny's original post, while turning out to be somewhat true on the bottom line, didn't demonstrate an understanding of statistics beyond "I think this is too many so it's an outlier." That's why neighborhoods normally use a median value for home prices, which is unaffected by outliers. This is unlike home prices in a neighborhood, where for example a couple 10 million dollar homes could fundamentally alter the average of an otherwise normal neighborhood. This is mostly because actual outliers are extremely rare and, in football, they will never be so extreme as to affect the bulk of a large sample size. The result is that Mahomes is almost exactly the same player. Removing both the upper and lower outliers of completions, attempts, yards, TDs, and interceptions, and using that data to calculate new values in completion %, ypa, and passer rating yields the following. At the very bottom I have all the numbers and definitions used here. You can immediately see how this becomes problematic for a dataset with only 20 values. Generally speaking, you would expect to find an outlier in less than 1% of your data, though it can be more or less depending on the data. After all, removing outliers from a dataset is a common tool in statistical analysis. Then I got to wondering, 3 years later with a larger sample size, what the result would be of doing his same experiment, but correctly, which Danny did not do, because he didn't understand what an outlier was. The Chiefs have had a more difficult schedule on average after his post than before his post. To me, this seems to largely be a result of more difficult schedules though, rather than sharp statistical analysis. The post-danny passer rating is still currently the 3rd highest career passer rating ever.
0 Comments
Leave a Reply. |