Before I go over the final "score" from the 2007 season, I'll remind you all what this here PythagenMatt thingy is.  Quick review: PythagenPat is a modified version of the original Bill James discovery of the Pythagorean equation RS^2 / (RS^2 + RA^2) = W%. Patriot (as he calls himself) realized that the exponent was dependent on the run scoring environment for a given team over the course of the season and so wrote the 2s and Xs where X = ((RS+RA)/Games)^0.285.  That is still the best seasonal pythag equation you will find in terms of lowest RMSE and highest correlation with real W%.

PythagenMatt, as I've dubbed it (following in the footsteps of Davenport's PythagenPort and Patriot's PythagenPat), is PythagenPat…but calculated one game at a time.  What this does is forces the run scoring for each individual game to only count for at most one whole win.  When the Orioles got beat 30-3 this year, it completely threw off the seasonal PythagenPat numbers for both them and the Rangers and this happens all the time with teams during blowout games.  It also allows me to correctly reward teams for winning a lot of close games and get a better handle on who's hot and who's not over a given recent stretch.

Let's start with the big summary chart.  It's similar to the chart I updated daily during the regular season until the Mariners' collapse took the wind out of my sabermetric sails. :)

Team G# W Matt Pat M-P Closer ME PE
ANA 162 94 92.4 90.0 2.4 MATT -1.6 -4.0
ARI 162 90 84.4 78.9 5.5 MATT -5.6 -11.1
ATL 162 84 84.5 88.7 -4.1 MATT 0.5 4.7
BAL 162 69 75.7 70.3 5.4 PAT 6.7 1.3
BOS 162 96 99.2 101.8 -2.6 MATT 3.2 5.8
CHA 162 72 69.0 66.5 2.6 MATT -3.0 -5.5
CHN 162 85 88.4 87.5 1.0 PAT 3.4 2.5
CIN 162 72 74.4 74.3 0.0 TIE 2.4 2.3
CLE 162 96 93.0 91.8 1.2 MATT -3.0 -4.2
COL 163 90 89.1 90.8 -1.7 TIE -0.9 0.8
DET 162 88 87.5 89.4 -1.9 MATT -0.5 1.4
FLO 162 71 69.5 71.6 -2.1 PAT -1.5 0.6
HOU 162 73 72.4 72.0 0.4 TIE -0.6 -1.0
KCA 162 69 70.1 73.6 -3.5 MATT 1.1 4.6
LAN 162 82 78.7 81.8 -3.1 PAT -3.3 -0.2
MIL 162 83 84.1 83.5 0.6 PAT 1.1 0.5
MIN 162 79 78.2 80.3 -2.0 MATT -0.8 1.3
NYA 162 94 98.2 98.3 0.0 TIE 4.2 4.3
NYN 162 88 90.4 86.4 4.1 TIE 2.4 -1.6
OAK 162 76 75.7 79.3 -3.5 MATT -0.3 3.3
PHI 162 89 90.4 87.6 2.8 TIE 1.4 -1.4
PIT 162 68 67.6 69.0 -1.4 MATT -0.4 1.0
SDN 163 89 92.3 89.0 3.3 PAT 3.3 0.0
SEA 162 88 84.9 79.2 5.7 MATT -3.1 -8.8
SFN 162 71 74.2 77.1 -2.8 MATT 3.2 6.1
SLN 162 78 74.1 70.7 3.3 MATT -3.9 -7.3
TBA 162 66 63.9 66.2 -2.3 PAT -2.1 0.2
TEX 162 75 72.5 78.3 -5.9 MATT -2.5 3.3
TOR 162 83 86.9 86.6 0.3 TIE 3.9 3.6
WAS 162 73 68.1 69.6 -1.5 PAT -4.9 -3.4
                 
SYS W RMSE            
MATT 18.5 2.972            
PAT 11.5 4.169            

 

Notice, I was incorrect in my memory about how the Mariners actually finished in terms of PythagenMatt.  I did continue to track a few teams of particular interest (all of the AL West, Arizona because of the Pythagorean implications of that team, the Cubs, and a couple of others that had my attention all year) so I had some end-of-seaso data even before I went and acquired the gamelog data for 2007 and did the full report.  But I was remembering 86.8 PythagenMatt wins for the Mariners and it was actually 84.8.  Sorry about that. :\

I referred to Root Mean Square Error (RMSE) when I talked about the strong-form of the statistical argument in favor of using PythagenMatt over seasonal Pythag earlier.  Looking at data from 1900-2006, the PythagenPat error on a 162 game season is very close to the error represented above (about 4.2 Wins).  Using the same data set, the RMSE on PythagenMatt is 3.4 (I was wrong about that too, sorry Doc…I thought it was 3.7…LOL), so we did a little better than average in 2007, but the point remains clear.  Not only does PythagenMatt do about 4% better correlating with real W% (going from 0.9416 to 0.9805!)…it reduces error by 25%.  That's LARGE, folks.  The common wisdom before PythagenMatt was that we had maxed out our ability to estimate W% using RS data and that we would never do better than being off by 4 or 5 wins on a full season.  This is the equivalent (in terms of statistical achievements) of the jump from OPS to Linear Weight Runs Created in terms of explaining team run scoring.

Now then…how about some pretty pictures?

If you want to see that in a larger size, click on the thumbnail.  Here is a graph of the Seattle Mariners' season viewed through the eyes of PythagenMatt and PythagenPat.  You can see that Matt had the drop on Pat from the very beginning…that the Mariners consistently outperformed Pat in good times and bad times…streaks and slumps.  There was NEVER a point when this wasn't true.  There is clearly something intrinsic about this team that makes it win games in a way that PythagenPat wouldn't see (and I've already discussed the two main culprits so I won't rehash here. 

Here are the 2007 Diamondbacks viewed similarly:

 

Once again, in the final analysis, PythagenMatt did significantly better in predicting the performance of the team, but this time it took half the season for the difference to become apparent.  Although to be fair, it also took half the season for the D'Backs to get really hot and start winning a bunch of close games since it took half the season for their bullpen to fully gel.

How about a couple of "underachievers" in PythagenPat wisdom?  Let's start with the Oakland As, shall we?

 

The teams with large biases all look like this…PythagenMatt takes the correct side of the over/under with PythagenPat on projecting the rest of the season in almost every instance.  This is why I was never worried about the As even when their Pythag was over .600…they were a bully team early on and then they collapsed completely.

If you folks want to see other teams, I can produce those graphics…just ask and you shall receive.  Also…if you want to see other data displayed, feel free to chime in.