Dr. Tommy John

Of course that's true. I tried to do something similar at one point last year and I found a decent r, but also found that most of the correlation was coming from runs scored in the first three innings (when the reading would be inherently most accurate.) Again though, I was really spotty with doing it and kept no records, so I never got to actually look at whether incorporating or omitting specific data sets would improve my prediction or worsen it in the aggregate. For all I know, the days I used my hacky wind data were the days the wind was unpredictable in a statistically significant way. I'd have no way of regressing that. I would win after analyzing something new that I thought of and then keep doing it, but who knows how many wins came from random variance and how many came from an improved process (or at the expense of one) Thinking about it, it could even be advantageous for this format if the data does skew that way. I'll be really curious to hear about your findings when you've run the numbers on this. Any chance I could see the r's for various stats vs. windspeed once you've made them?

A friend of mine who knows I developed a much dumber, entirely manual version of your system the last couple years pointed me to this thread, so I signed up to thank the lot of you (and especially your robot overlord) for the enlightening 60 pages. I'll be sure to use your form filler after any contest to help pad your data points. I'm blown away that you managed to automate the entire process. My stats knowledge isn't so far behind so I understand and agree with your overall ideas on projection, but I have next to no programming knowledge so my research was limited to what I cared to scrape data for each day I chose to play. This is like being shown the internet after visiting your first library. e: On that note, I would be wary of putting too much weight on wind data. Unless you've got access to different data than I have ever seen, all you have is the estimated speed and direction at the time of the first pitch, which is introducing a scary amount of noise into any calculation unelss you regress it basically to infinity.