THE BOOK--Playing The Percentages In Baseball

Very helpful, Tango. No one that I am aware of has commented on the D’Back hitters “win efficiency”.  Incidentally, it appears to relate to “game clutch” performance rather than “situation clutch”.  The club actually hit less well with runners in scoring position that one would expect, but hit particularly well in tie and 1 run games. 

Our pre-conception of a team that hits well in “game clutch” situations is a veteran one, filled with players like Gary Sheffield, not one filled with younger players like Chris Young (who was a monster in the clutch in 2007).

I should note that I got all the data from Fangraphs:
http://www.fangraphs.com/winss.aspx?team=Diamondbacks&season=2007

In terms of “situational” performance (meaning how they did with men on base/outs, regardless of score), their hitters were -47 runs, or almost -5 wins, which is along the lines of their overall numbers.  So, it’s definitely in high leverage situations where the Dbacks hitters performed well.

***

As for relievers, you can look at a guy like this: Jailen Peguero.
http://www.fangraphs.com/statsd.aspx?playerid=2913&position=P&season=

In the 6 games that he entered that were NOT blowouts (4.1 IP, 18 PA, excluding IBB), he allowed 6 baserunners, which is an average performance.  His total WPA in those games was +0.1 wins.

But, in the blowout games (10.1 IP, 52 PA), he allowed a whopping 23 baserunners.  His total WPA however was 0.0.  That is, while he pitched horribly, it didn’t affect the outcome of the game at all.

Obviously, all teams have such pitchers.  The key is to add it up all for all pitchers and compare it to all the teams.  And, from that standpoint, the timeliness of the relievers’ performance led to 4 (we estimate) more tangible wins.

Very interesting.  Everybody seems to be “crediting” the efficiency of the DB starters and relievers for Arizona’s success and treating it as a team skill.  But I bet those same people would NOT consider offensive clutchiness a skill.

The data points towards Arizona being much worse next year, but they can also expect their young team to show improvement.

Interesting, I just finished (well, almost finished) an article for the THT Annual where I look at, among other things, each team’s expected w/l this year, based on their underlying offensive, defensive, and pitching performance.  I did not break down pitching into starters and relievers. I used park and context-neutral lwts for batters, UZR for defense, and park and context neutral component lwts (adjusted for team defense) for pitchers.  I also looked at what their offense, defense, and pitching was supposed to look like based on actual playing time of each player and their pre-season Superlwts and pitching projections (per me).  Here are the numbers I get:

ARI

Projected offense -18
Actual offense -41
Projected pitching 36
Actual pitching 45
Projected defense 3
Acutal defense -1

So they were “supposed” to be (again, based on whoi actually played, not who was supposed to play before the season started) a 2 win below average team in offense.  They were 2 wins worse than that in underlying (context-neutral offensive lwts) performance.

They were supposed to be +3.5 wins in pitching and they were actually a win better than that (based on park and context-neutral component lwts of all their pitchers, giving higher weight to the closer and set-up man, and lesser weight to the long relievers).

In defense, they were basically expected to be average and were average, more or less.

Based on their underlyng performance (-41 in offense, +45 in pitching and -1 in defense), they were expected to win 80 games.  They scored and allowed (at least relatively speaking) about what they were expected based on the underlying component performance, as their Pythag win total was 79 (one less than that expected based on that underlying component performance).

Of course they won 90 games, 10 or 11 games more than “expected.” You can debate or point out “why” all you want.  My money is on the fact that most of that was sheer, plain, ordinary luck.  Assuming around the same talent next year, if anyone wants to take the over next year on, say, 86 or 87 games, I’ll gladly bet as much as they want!

BTW, it does not interest me at all when a team far exceeds their pythag wp.  Not at all.  Going into just one season, we KNOW for a fact, that if there was no such thing as leveraging peformance or clutch hitting or pitching (as a “skill") that 1 in 10 teams would be off from their pythag win total by MORE THAN GAMES, BY CHANCE ALONE!  (One SD for team wins in a season, due to chance, is around 7 games.) Given that, why should we get excited when a team exceeds or underperforms their pythag record by 10 games?  That should happen more than one time every single year, by chance alone.

Oops, that should be “BY MORE THAN 10 GAMES” above…

MGL:  the SD of 7 is variation of W-L from true talent, which includes over/underperformance in scoring or run prevention.  Presumably, departures from pythag record will be smaller, since you’ve now controlled for actual offensive and defensive performance.  And in fact, +/- 10 games is unusual.

That said, I agree with your general perspective.

Yup, guy, you are right of course.  I am not sure that the SD between pythag and actual is a whole ot less than that though (70%?).  How would you figure that?

In this article (http://www.hardballtimes.com/main/article/pondering-pythagoras/), DG puts out a number, 2.67, but I am not exactly sure what that is.

"How would you figure that?”

Not a clue.  But Sal might—he’s written a lot about the distributions of RS and RA, and why the intersection of the two yield the pythag relationship to W-L.

The reason I was guessing that the random w/l standard error around the pythag w/l might not be that much less than the actual w/l record is that the pythag w/l record is not that much more of a predictor of future w/l record than actual record.

I can surely use my sim to figure out an approximate answer.

OK, I ran just one matchup (team X versus team Y, each with a certain fixed wp at home and another won on the road) through my sim 81,000 times (500 “seasons").  I kept track of the average deviation from each team’s true wp and each team’s pythag wp.  Here is what I got:

The average difference between a team’s actual record and their true talent was 7.0.

The average difference between a team’s actual and pythag record was 5.4.

The square root of the average squared difference was 8.7 and 6.4, respectively.

I am not sure why the actual standard error (8.7) is so much higher than one estimated from the binomial (6.36).  One reason is that I play half the games at home and half away, which is two different true talent levels.  That increases the variance a little.  Is it because I am using a true sim (I am not just using wp for team A and wp for team B and a log5 matchup formula)?  IOW, does a team actually have a standard error of wins around 9 per season, even if we assume a realtively static true talent wp?

In any case, it looks like the standard error of a team’s actual win total versus its pythag estimated win total is around 74% of its standard error in actual win total versus average true talent for the whole season.  I think I got lucky in #7 (I guessed 70%).

If you sqrt the sum of their random squared differences, i.e, sqrt(6.36^2+6.36^2)=9.00, which is close to your 8.77.

Can you detail more of your process, with some illustrations?

I just simmed 2 teams who were close in talent (I don’t think that matters) and for every 162 games, I kept track of each team’s average wins and pythag wins, based on runs scored and runs allowed.  Each tame played half their games at home and half on the road (you have to do that otherwise the pythag record will never equal the actual record on the average because of all the 1 run wins by the home team).  From there I did the above calcs.  I was just wondering why the SD deviation in team wins per 162 games is so much higher than that estimated from the binomial assuming a p of around .5 (6.37).

OK, I realized one thing I did wrong.  I ran 162 games for each team at home, then 162 on the road, so and so forth, alternating 162 game blocks between home and road.  That created more variance among blocks than the binomial would suggest if p were constant (in this case, p was fluctuating from block to block).

Here are the new numbers:

Standard error of games won per season = 6.49
Standard error of difference between pythag and actual record is = 3.94

Now that is more like it.  That expected from the binomial is 6.33.  The difference between the 6.49 and 6.33 is the fluctuating p (home and away) from game to game within the 162 game block.  In reality, that p will fluctuate even more, with their pitcher and qaulity of the opposing team and pitcher, so the real standard error will be closer to 7 I would assume, which is the number I usually use anyway.

The other standard error though, the difference between actual and pythag came up pretty small.  So, for example, the D-backs are around 2.5 standard deviations to the good.  You expect that around .5% of the time.  For 30 teams though, you still expect that to happen around every 7th season or so (at least one team to exceed their pythag record by 10 games or more).  So it is not that rare.

So if exceeding or undershooting pythag record were a significant skill, we would epxect to see the distribution look different from what we expect by chance.  Anyone want to look at the last X years and see what the distribution actually looks like.  If the SD of the bell curve is not signficantly greater than 4, then we can say that it is almost entirely due to chance when a team under or overperforms their pythag record, no matter by how much.

BTW, I have not heard anyone talk about this, but if a team wins more (or fewer) games at home than expected, their record will automatically exceed (or undershoot) their pythag record.  This is because of the extra 1 run wins you get at home by not having to complete the inning when you go ahead in the bottom of the 9th or later.  Did AZ win an inordinate number of home games this year, given their overall record?  I have no idea, but I’ll guess that they did.  Plus since the average team wins more games at home due to the HFA, the average team will actually slightly outperform its pythag record.

Did AZ win an inordinate number of home games this year, given their overall record?

Just did a little calculation:

Home (+9.3)
Actual 51-31 v.s Pythag 41.7-39.3

Away (+3.0)
Actual 40-41 v.s Pythag 37.0-44.0

So would a better Pythag use R/out instead of R/G?  Anyone have the capability of running that to see if it makes any significant difference??

Sorry, in #15, it’s:

Home (+8.3)
Actual 50-31 v.s Pythag 41.7-39.3

Yes, runs per out would be better, but even then you are missing out on a few expected runs.

Looks like they only won 2 more games than expected at home, given that they are a .555 team overall (home expected wp is around overall times 1.08).