BryanK
Posts: 140
Joined: 8/21/2005 Status: offline
|
I went back and validated the HR and BB findings from yesterday with a second associaiton and dove into the pitcher usage stuff I mentioned yesterday. Findings: 1. Based on what I saw in a second test association I ran last night, the best fix for the HR and BB level issue might be more involved than just scaling levels in the XML. It appears that for the first few years of an association, regardless of the year it starts begins, Puresim produces historically accurate HR numbers. Then it quickly regresses to an unrealistic level that is dead on with my earlier test. I observed similar results for the BB levels. 2. Pitcher utilization is perhaps the weakest part of Puresim's stat engine. A: From an IP perspective, there isn't enough distinction between durable starters and average ones B: All starters log pretty much the same number of starts C: Relievers who make any appearnaces make too many apperances The suggestions that come to mind for item 2 are generally related to scaling down pitcher durability or adjusting the distribution thereof. I go into more detail in the slides below. A quick thought on 2-B, though, would be to add a slot for spot starters who could take a pitcher's turn in the rotation if said pitcher is too tired to go. That might be a bit tricky to impliment properly, though. Granted, the game is using the pitchers it does as often as it does because they're better than the other arms available. If durability gets scaled down, the game will probably compensate by using bad pitchers more often, which could possibly lead to higher offensive stats. So... this might be a bit of a hornets nest. But based on the data, though, I think it's clear that pitcher usage is the area that offers the most room for improvement. For the new slides click HERE For the updated summary sheets, click HERE Also, does anyone know where to buy a second hand license for JMP v5 software... if in fact those licenses are transferable? I only have it on my work computer, and so I'll lose access to it when I leave the company at the end of the month. Without that package, it'll be nigh impossible to break out the distributions that made this analysis so cool, and the sticker price for the program is like $1600.
< Message edited by BryanK -- 4/18/2006 8:37:41 AM >
|