What is PCA?

[Honus Wagner Rules]

Just curious, what is PCA?

[Francoeurstein]

You know the famous school my brother went to : Pacific Coast Academy.

[538280]

Just curious, what is PCA?

PCA is a total player metrtic, similar to Win Shares, TPR, and WARP, that measures a players total contribution to his team, including hitting, fielding, and baserunning. It also measures pitchers.

PCA is short for "Pythagorean Comparitive Analysis". It is being created by out very own Matt Souders (SABR Matt). If you have any questions as to specifics, ask him.

[SABR Matt]

There hasn't been up to this point enough information out there about the system, and I still refer to it in discussion about players...for that I apologize.

PCA is a work in progress...answers to specific questions are likely to get you a "we're working on this idea.." at present...The first results from the new SFBT (Souders-Fiato Baseball Translations) system may be coming very soon...we're starting with Strength of Schedule and Park analysis which should be interesting.

[Honus Wagner Rules]

There hasn't been up to this point enough information out there about the system, and I still refer to it in discussion about players...for that I apologize.

PCA is a work in progress...answers to specific questions are likely to get you a "we're working on this idea.." at present...The first results from the new SFBT (Souders-Fiato Baseball Translations) system may be coming very soon...we're starting with Strength of Schedule and Park analysis which should be interesting.

I was just curious since I joined BF I've seen PCA referred to many times but never actually defined. Can you give a description of PCA and what it is actually measuring.

Thanks,

HWR

[SABR Matt]

HWR...a short description of PCA is nearly impossible....but boiled down to simplest terms...

PCA determines how many wins a team was actually worth (not always exactly the same as how many they actually had...this is dependent on strength of schedule, luck, and the consistancy of the team among many other factors) and distributes those wins to the players through a series of calculations based on an extension of Pythagorean logic...I used the pythagorean to place the margin (minimal expected production for a major league player) and to compare the player's actual performance with that margin, and divide up credit to various skill units on a team and the various players within skill units on a team.

OK...that wasn't terribly simple...LOL

PCA is a combination of the ideas put forth by Pete Palmer, Bill James, and Baseball Prospectus on the best ways to rate the players along with my own adaptations...it results in Wins Created ratings for each player in each season.

[Honus Wagner Rules]

HWR...a short description of PCA is nearly impossible....but boiled down to simplest terms...

PCA determines how many wins a team was actually worth (not always exactly the same as how many they actually had...this is dependent on strength of schedule, luck, and the consistancy of the team among many other factors) and distributes those wins to the players through a series of calculations based on an extension of Pythagorean logic...I used the pythagorean to place the margin (minimal expected production for a major league player) and to compare the player's actual performance with that margin, and divide up credit to various skill units on a team and the various players within skill units on a team.

OK...that wasn't terribly simple...LOL

PCA is a combination of the ideas put forth by Pete Palmer, Bill James, and Baseball Prospectus on the best ways to rate the players along with my own adaptations...it results in Wins Created ratings for each player in each season.

Is this based on how a team would do against an "average" team i.e, one with "average" offense and "average" defense? A very interesting concept. I look forward to reading your research some time.

[SABR Matt]

It is based on what the team actually did do...adjusted so that they appear to have faced average competition...so indirectly yes.

I'm in the process of reworking some aspects of the methodology...as new information becomes available you will hear about it here.

[Honus Wagner Rules]

Hi Matt,

Are you at a point where you can give a more detailed description of PCA?

[SABR Matt]

HWR...I give a complete description of PCA already. Or rather...I make the description available to anyone who wants it. If you're a member of the Yahoo group baseball-databank, it's in the file list (it's called The PCA Manifest) or I can e-mail it to you.

[Sultan_1895-1948]

Gee, HWR sure got very curious about PCA all of a sudden :hyper:

Matt, how far down the road can we expect some type of LQ adjustment? I realize that you have bigger (and more) fish to fry as of right now, but down the road, do you see yourself tackling that issue? And if you do, am I correct in assuming that not all players will be affected equally, since clearly the upper elites would still be just that. Thanks.

[SABR Matt]

The grave (IMHO) mistake of most league quality adjustments is that they treat every part of the quality distribution as though it can be manipulated in identical ways to make sense of the quality spectrum.

In answer to your question...league quality remains on the agenda for the overhaul which I expect to begin in earnest early this summer. There are three steps to total normalization (the process by which any baseball player can have his performance altered so that it can be seen to have occured on a level playing field with the rest of baseball).

1) We need to assess the strength of schedule each player faced. I'm in the process of working with a colleague of mine who is cooperating with baseball-reference.com in the hopes of acquiring and digitizing the boxscore database (which goes all the way back to 1876 though the level of detail is not particularly great). We will be able to use that data (we hope) to see how many PA or batters faced each player had against each team. Since the FSIA has already been used to calculate how intrinsically skilled each team was relative to their league, we can use that information to rate each player's strength of schedule faced.

2) We need to normalize PCA wins created (which themselves have already been adjusted using SoS calculations) rates through curve fitting so that all leagues and all players see their performance placed on one curve that defines the league-relative talent distribution identically for all time.

3) We need to see how players move from year to year along the all-time talent curve to judge how the quality of the league changes from season to season. If players from 1941 who also played in 1942 see their percentile ranking improve overall in 1942, we can conclude that 1942 is a weaker league and put a figure on how much weaker it is. This method has been done by the Hardball Times, but they only used production metrics normalized for league average (not totally normalized as I described above), and reached what IMHO is an overly smoothed looking nearly linear league quality curve.

I think if you do all of the outlined adjusting before you get to the last step, you're going to get something more useful...but I could be wrong. I will find out when I reach that point in the analysis.

[SABR Matt]

One important problem I will need to correct for is the fact that good players stick and bad players don't. I may need to group the data and see if there are different performance change curves for different subsets of the MLB population.

[Honus Wagner Rules]

Gee, HWR sure got very curious about PCA all of a sudden :hyper:

Matt, how far down the road can we expect some type of LQ adjustment? I realize that you have bigger (and more) fish to fry as of right now, but down the road, do you see yourself tackling that issue? And if you do, am I correct in assuming that not all players will be affected equally, since clearly the upper elites would still be just that. Thanks.

I stumbled across this old thread I started back in 2005. I just wanted to learn more about PCA.

[Honus Wagner Rules]

HWR...I give a complete description of PCA already. Or rather...I make the description available to anyone who wants it. If you're a member of the Yahoo group baseball-databank, it's in the file list (it's called The PCA Manifest) or I can e-mail it to you.

Can anyone join baseball-databank? I've decided to rededicate myself to sabermetrics and get back into the flow of things.