Stats Modeling in PS 2006

[BryanK]

Before grad school applications and my banking job took away all my free time, I was pretty active here making mods and doing some analysis of Puresim's stats engine. Well, after an up hill climb I actually got into school. To celebrate, I've decided to quit my job and take the summer off ... seriously. I'll probably work as a counsellor at a children's camp or something, but there's going to be lots of free time for stuff like Puresim

That being the case, and with the recent talk of features in PS 2006, I was curious about if any changes to the stats modeling were in the works. Based on the changes made last year, real player associations are very accurate. I don't think I ever posted the results from my 1.21 test association, but the general issue I've been seeing in fictional associations is that stats, particuarlly home runs, ramp up after an initial period. I think league leaders in OBP were also a little low, but I don't remember...

Essentially what I'm asking is if there's any appetite to investing in the stats engine. I'll gladly help out with the analysis and number crunching if there is... but I was just curious is it's something other people had noticed or if Shaun was looking to make any changes here.

[puresimmer]

Absolutely.

I am looking into modeling extensively for PS 2006.

[BryanK]

Sweet! Shaun = Super Awesome

I still have a few details to iron out regarding my move to Conneticut, student loans, and the like, but I can probably have some spiffy analysis ready around the end of the month or the begining of May. I just got some statistical software that makes the analysis a lot easier than it was in excel, so I can look at a bunch more stuff. And looking back, I have no idea why I calibrated my earlier analysis on a per game basis instead of a per Plate Appearance or per IP basis... yeah... I'll fix that.

[BryanK]

Quick update...

I decided to calibrate the analysis on a per AB basis for hitting and a per BF (batter faced) basis for pitching. It just seemed most logical to track the number of times something happened relative to the number of times it *could* have happened, and I think per AB and per BF do the best job.

Based on the work I've already done, I think PS is getting the right level (mean) for the different statistics, but that the way it arrives at that level (i.e. the percentile distributions) is off. So, starting in 1901 I am going to calculate the mean and percentile distributions for a number of metrics, all of which are normalized on a per AB or per BF basis. Note that to get adequate sample with the historical data, I needed to group by decade: 1900, 1910, 1920, etc

I'm going to run all these numbers on both real-life data and numbers taken from a fictional PS association using 1.21. Then we can compare the two and see how the game engine stacks up.

This sound good?

[FranklinNoble]

Just thinking about all that math makes my head hurt... [:D][:D][:D]

[BryanK]

That's what statistical software is for
Still, it took about 5 good hours today to get the real-life data packaged and processed. My PC is simming the 1961 season right now, so hopefully I will have the PS data together by the weekend.

[puresimmer]

Wow, awesome. I look forward to seeing the data. Hopefully I can make some tweaks for PS 2006's engine based on this. There are so many variables in the engine, that it is a tenuous balance when tweaking it: Aging model/career arc, stats engine, AI, ball park dimensions, etc..