Monday, August 02, 2010

PART II - Advanced Experimentation 410:
The Null Hypothesis Is Not A Dull Hypothesis  

In Part I, I explained why the 2010 Seattle Mariners' front office found themselves best served by a bold innovation.

In this part, I'll describe the innovation they chose to adopt as a working hypothesis, that run prevention could be amassed in such concentration that it could deliver escape velocity from the gravitational field of the Pythagorean Principle (that a team's runs-produced and runs-allowed strongly shape their win-loss record). 

If that Mariners hypothesis worked, it would...

  1. Free them from the expense of assembling a fully-featured offensive team, substituting lesser-valued players whose lead attribute, defense,  is cheaper to buy in the 2009-2010 market than those players with noteworthy offense.
  2. Ambush the scouting of other teams who would not at first know the workings of the hypothesis and once they knew it, would be challenging, energy-consuming and time consuming to respond to.

To put a rational bold experiment together (Beyond Baseball, too), it requires OMA...observation of past trends, measurement of past effects, and analysis of relationships between factors. Here's the OMA chain that led the Ms' front office to the experiment.

What to Implement to Maximize Run Prevention?
The 2009 and 2010 Seattle Mariners have a modern statistically-oriented analytical function in the front office. And the protocol among modern sabermetricians is that run-production (and by the perfect double-entry math of Baseball therefore, run-prevention too) is made up of three factors, one of which is heavily-affected by defense (the opponents' batting average of balls put into play), one heavily affected by pitching (the rate of issuing walks), and one moderately affected by pitching (home runs yielded).

Putting it Together #1 - Team Defense
The M's 2009 campaign featured the most hit-preventing team defense (as measured by relative Defensive Efficiency Rating) since 2001. The 2009 Ms allowed Batting Average of Balls in Play at 91.3% of the 2009 American League's composite average (.274 for the Ms, .300 for the AL), and the only team in the 21st Century to apply such asphyxiating team defense was the record-shattering 2001 Mariners that won 116 games while allowing Batting Average of Balls in Play at a rate of 88.2% of the 2001 American League's composite average.

As I stated in Part I, the 2009 Mariners exceed the number of wins the Pythagorean thumbnail estimation suggested they "should" have by eight, a very high and fairly unusual difference. The 2001 Mariners (a legend everyone in the Mariners' administration views with reverence) with their even-better relative team defense, won seven more games than the Pythagorean thumbnail suggested they "should". That is a lesson they couldn't overlook.

So, using a factor they could control, they were able to play around with team defense, at least at the edges, to try to exceed the rather exxxtreme accomplishment they'd notched in 2009. They moved a mixed bag of fielding talent, 2009 2bman Jose López, to 3rd base where his strong arm would have more positive value and his so-so range be less of a deficit. They gave the left field spot, for most teams, a place to locate one's weakest fielding big bat, to youngster Michael Saunders who'd looked a little promising in the field in 2009, even though his bat wasn't yet producing fully a major leaguer's output. They acquired the athletically-gifted Chone Figgins to play second base, and even though it hadn't been Figgins' primary position, he'd played at that spot occasionally over the first eight years of his career, and as a better athlete with an apparently-better baseball brain than his second-base predecessor, perhaps would add team defense (or perhaps wouldn't neutralize the hoped-for benefit of moving López to third). Finally, the team let the only possibly-legitimate middle-of-the-lineup slugger they had on the roster, the injured first-baseman Russell Branyan, leave, replacing him with a questionable bat with an acknowledged glove, Casey Kotchman. So the team had put into the two positions most usual in roster protocol to produce offense, LF and 1b, respectively, a very young, not-fully developed player and a guy who could hit some, sometimes. This was a risky pairing - if one or both produced offense near the top of their potential, this team's offense was going to be anemic, but if one or both didn't produce near the top of their potential, its was going to be sub-anemic.

But the Mariners front office put on the field a defense that should have been even better than the 2009 squad at hit prevention. What about homers and walks, the other two components that are part of the contemporary sabermetrics protocol for run prevention?

Putting it Together #2 - Homer Prevention
While the team can't control this factor day to day like Bill Veeck, the team's home park is the most extreme anti-hitting (ergo, pro-defense) park in the league over the last 9-1/2 seasons. Further, though, it is an intensely homer-negative park overall -- crushing right-handed pull power into dust while gently boosting left handers who loft the ball while pulling it.

So to put together this work of art and craft, the Mariners already had a great canvas for laying on the homer-prevention component of run prevention, the park they played 50% of their games in. In the other 50% of their games, on the road, their 2009 pitching staff was precisely league average in homer prevention.

Staff       Road HR
LAA           76
TEX           78
CHW           80
NYY           80
OAK           87
TOR           91
SEA           92
AL mean       92
MIN           92
BOS           95
DET           95
KCR           97
TBR          101
BAL          109
CLE          111

And keep in mind, three of the teams that allowed fewer home runs were in the same division as the Mariners, meaning not only did they get to face a marginally-powered Mariner offense quite often, but also got to face it in the Mariners' home park in a bunch of their road games (an affordance the Mariners never got to have, since the Ms played none of their away games in their homer-snuffing home park).

Putting it Together #3 - Walk Prevention/Overall Pitching Quality
Tweaking the 2009 team as a base would require care. The 2009 team had great-looking pitching overall (ignoring the home-away splits). At 4.27 runs per game surrendered, the team was #1 in the American League

  R/G    Tm     W    L W-L%  ERA   HR ERA+  WHIP H/9 BB/9 SO/9
 4.27    SEA   85   77 .525 3.87  172  112 1.30  8.4  3.3  6.5
 4.52    CHW   79   83 .488 4.14  169  112 1.35  9.0  3.2  7.0
 4.54    BOS   95   67 .586 4.35  167  108 1.41  9.4  3.3  7.7
 4.57    DET   86   77 .528 4.29  182  106 1.41  9.0  3.7  6.9
 4.57    TEX   87   75 .537 4.38  171  106 1.37  9.0  3.3  6.4
 4.65    NYY  103   59 .636 4.26  181  101 1.35  8.6  3.6  7.8
 4.65    TBR   84   78 .519 4.33  183  104 1.36  9.0  3.2  7.1
 4.69    MIN   87   76 .534 4.50  185   98 1.38  9.6  2.9  6.5
 4.70    LAA   97   65 .599 4.45  180  102 1.41  9.4  3.3  6.6
 4.70    OAK   75   87 .463 4.26  156  103 1.39  9.2  3.3  7.0
 4.75    LgAv  82   80 .505 4.45  178  100 1.40  9.2  3.4  6.9
 4.76    TOR   75   87 .463 4.47  181   98 1.42  9.4  3.4  7.3
 5.20    KCR   65   97 .401 4.83  166   92 1.46  9.4  3.8  7.3
 5.34    CLE   65   97 .401 5.06  183   83 1.51  9.9  3.8  6.2
 5.41    BAL   64   98 .395 5.15  218   88 1.53 10.3  3.4  5.9
Provided by Baseball-Reference.com: View Original Table
Generated 7/23/2010.

That 4.27 Runs/Game (the first column) is not only the best in the league, but by a stunning margin of .25 R/G over the second best team. Why is a quarter of a run stunning? Well, the difference between the 2nd best team and the 5th is a mere .05 R/G, and to get to the team that's 0.25 R/G less effective than the 2nd best team you have to get past the 12th best. It's such a significant difference, the difference between #1 and #2 just about spans the rest of the entire league.

The 2009 M's pitching is, by this measure, an extreme outlier. And it was, as I pointed out in Part I, also an extreme outlier in its W-L record compared to its Pythagorean projection. An inquisitive and observant manager in any field would wonder if there might be a cause (incredible out-of-scope run-prevention) and effect (outperforming pre-season expectations or outperforming Pythagorean projection). What IF way-above normal defense bent the space-time continuum and changed the gravitational field the Pythagorean affects. She might not presume it existed, but she would at least be interested in if it was true.

Because if it was linked by some cause-and-effect, this not-broadly-known insight would be actionable knowledge, that is, a competitive edge. In the open market, pitching is seen as a very valuable commodity, so it's generally expensive, but defense has been both difficult to measure and not thought of as a critical factor in the high run-production era since the owners juiced the ball after the 1993 season. I'm fairly sure no team has tried to make defense its lead positive attribute since the end of practice of Deadball strategies in the early 1920s. And just as the insights described in Moneyball six years earlier had seemed a way to create more wins/dollar and something that others would then have to chase, and many couldn't chase even if they wanted to and knew how to (The Texas Rangers, for example, which have an offense-stimulating stadium), competitors could be slow to adapt to this possible Mariner revolution yielding a benefit that could last a while.

The coup de grâce for the 2010 rotation was Zdurienck's acquisition of a starter who may have been potentially the best starter in the American League, but certainly the best for this team. Because the acquisition, Cliff Lee, rarely walks anyone (in 2009, 1.7 walks per 9 innings pitched compared to the league average of 3.4), and his homers-allowed per 9 innings pitched was 0.7 compared to the league's 1.1). Roll in the Ms defense's ability to snuff hits as a complement, and it looked to Zdurienck as though Lee might notch his best season ever.

And compared to 2009, the season against which Zdurienck was looking to improve, Lee (potentially the best) would be replacing Carlos Silva (who in 2008 and 2009 had been 5-18 with an ERA of 6.81).

ASIDE: Z-Man had the opportunity to try to sign Cliff Lee, due to become a free agent at the end of 2010, to a long-term deal before the perfect Lee-glorifying scheme they'd put together had borne fruit, and many managers in all fields would have thought it best to get in front of that, getting Lee to commit before the probably-superb season played out. Z-Man didn't, and that was brilliant, a lesson I'll talk about in Part III and one you should follow.

So make a perfect alignment for run-prevention

  • Asphyxiating ballpark, unchanged;
  • Exxxxtreme defensive ability on the field, made more extreme;
  • Good pitching, improved

Thus, the hypothesis to be tested, perhaps offensive ability would become less material.

Because the Mariners weren't using any of the standard operating procedures for building a winning team, there was a big chance the hypothesis would be null, that the experiment would not pan out.

And while the bold experiment didn't work for reasons I'll describe in Part III, Z-Man's brilliance made it possible to come away with positives even if all the Ms could prove was the null hypothesis, as I'll tell you about in Part III.

This page is powered by Blogger. Isn't yours?

free website counter