"The future ain't what it used to be." --Yogi Berra
If Bo Porter wants to look for a place to improve the Astros, an easy place to start is winning extra inning games. The Astros had a 1-11 record (8% W/L) in extra inning games last year. Maybe the manager has some influence on winning close games and extra innings affairs. But I suspect that the Astros could gain 3 or 4 wins in extra innings next year without Bo doing anything special. That's called regression or reversion to the mean--something we say a lot in the talking sabermetric series.
Last month, I wrote "The Astros and the Plexiglass Principle," which discussed the general tendency of teams at the extremes of W/L record to regress in the direction of a .500 record. This article follows up on that discussion with a more specific look at 1 run games and extra inning games. Let's parse, dissect, and reconstruct the 2012 season in search of potential improvement in 2013. (Why? Well, that's what we do.) If you have been reading this column in the past, you know that a team's Pythagorean Record is a formula predicting the W/L record based on a league average distribution of runs scored and runs allowed across the season. W/L records in 1 run games and/or extra inning games can skew a team's ability to meet it Pythagorean expectation (Note that the Astros W/L record in 2012 underperformed the Pythag Record by 4 games.)
Background: One Run Games
Thanks to the outlier of outliers--the 2012 Baltimore Orioles--with the highest win percent in 1 run games (29-9 or 89%) in MLB history, we have seen some recent debate about the repeatability of 1 run margin W/L records. Blake Murphy's Beyond the Boxscore article, "One Run Game Performance is Unsustainable," puts to rest the idea that an extreme winning percent in 1 run games is a repeatable skill. For teams comprising more than 2.300 MLB seasons, he examined the ability to predict 1 run game records based on the previous season's record for 1 run games. He found:
- Team record in 1 run games has no explanatory power for the next year record in 1 run games (R-squared less than 2%). Year to year records in 1 run games are basically random.
- Teams with a presumed elite ability to win 1 run games (i.e., very high win percents in those games) regress toward .500 records to the same extent as other teams (R-squared of 1.2%).
- A team's overall Pythagorean Record has only a minor effect on explaining the regression in 1 run game records (R-squared of 9%). With explanatory power that small, perhaps we should expect regression to .500, rather than the team Pythag winning percentage.
- Bullpen FIP has some explanatory power for 1 run game records, but it is small (R-squared of 11%).
Extra Inning Games
Since most extra inning games are won with 1 run, extra inning games are practically a sub set of 1 run games. Not surprisingly, team winning percentages in 1 run games was reasonably correlated with extra inning win percent in 2012 (I found a 27% R-squared).
The Astros' winning percent (39%) in 1 run games in 2012 was relatively close to the Astros' overall Pythag Record. However, based on runs scored and runs allowed in 1 run games, the Astros under performed; the Pythag formula predicts a 47% winning percent. Given that 1 run games, by definition, have a small margin between runs scored and allowed, we shouldn't be surprised that the Pythag predicts close to a .500 record in those games. In fact, this should give us some insight as to why we should expect teams' 1 run game winning percents to regress to .500.
However, the Astros had an extreme record in extra inning games (8% winning percentage), which is much lower than what we would expect from the Astros' runs scored and allowed in extra inning games (40% winning percentage, according to Pythag formula). The Orioles had the best win percent in extra inning games (16-2 or 89%), and the Astros had the worst. By all appearances, the Orioles were very lucky in extra inning games and the Astros were very unlucky. The Orioles shocked the projection systems by winning the AL wild card...and the Astros won the first pick in this year's draft.
In order to conjecture about the possible causes of 1 run and extra inning records, I examined whether player clutch performance, as measured by fangraphs' clutch stat, can explain the differences in teams' win percents for both types of games in 2012. The clutch statistic is based on the win probability family of stats, and reflects the extent that a player's performance in a season was better or worse in high leverage situations than his normal performance. (This stat is not measuring players' or teams'overall talent levels.) I summed each team's batting and bullpen clutch stat to arrive at an overall clutch measure applicable to late and close games, as well as examining batting and bullpen clutch stats separately. I used R-squared as a test of explanatory power.
1 Run Games
- The combination of batting and bullpen clutch explained most of the variation in 1 run game win percent (R square of 64%).
- Examining batting and pitching clutch separately, both components influence the 1 run game record (.32 and .42 R squared), but bullpen clutch performance is the bigger influence.
Extra Inning Games
- The sum of batting and bullpen clutch explains a significant amount of the difference in extra inning game win percent (30% R squared), but not to the same extent as 1 run game records.
- Batting clutch has virtually no impact on extra inning records (3% R squared), but bullpen clutch explains a lot of the variation in extra inning win percent (45% R squared). Compared to 1 run games generally, extra inning games are more influenced by bullpen clutch performance.
I expected to see a pattern of extra inning losses falling on end of the bench pitchers, which might reflect poor bullpen depth and possibly ineffective management. But that wasn't the case. Only 3 of the 11 extra inning losses were attributed to middle relievers. In large part, the failures can be attributed to late inning relievers--the part of the bullpen that the Astros viewed as most reliable. And only one of the losses involved a real end of the bench reliever (Storey). The reliever (Lopez) who achieved the Astros' single win in extra innings didn't cover himself with glory, blowing the save before winning the game.
Interestingly, none of the losing pitchers remains on the Astros' roster. (Should that give any comfort to Bo Porter? Who knows?) Here are the losing pitchers:
Brett Myers: 3 losses in 0.2 IP
David Carpenter: 2 losses in 2.2 IP
Brandon Lyon: 2 losses in 1 IP
F. Rodriguez: 2 losses in 1 IP
Wilton Lopez: 1 loss in 2 IP
Storey: 1 loss in 2 IP
One surprise is that the closer, Brett Myer lost the most extra inning games. And the three losses only required 2/3 of an inning. In these extra inning games, Lyon came into a tie game and immediately couldn't get outs. If there is a pattern, it is that Mills held Myers back in regulation tie games and saved him for extra innings. But he didn't wait until the Astros had a lead in extra innings to use Myers. Does this support the theory that you shouldn't bring a closer into a non-save situation? That's hard to say. Maybe it affected this particular closer, but wouldn't affect other closers. Myers said he was puzzled by his usage after one of the extra inning losses; so maybe it got in his head. Or maybe that is just rationalization on Myers' part.
In general, it doesn't appear that the problems arose because a reliever had to pitch too many innings in extras. Only Storey and Lopez had 2 inning stints, and that wasn't unusual for their typical usage. In most cases, the losing extra inning pitchers ran into immediate trouble and lasted less than an inning.
Really, it seems like the Astros' extra inning record leans more toward bad luck than managerial usage. Perhaps a "closer by committee" approach might help in the future, if it prevents relievers from letting up in non-save situations.
It will be interesting to see how Bo handles the bullpen, but in any event, I think the Astros are likely to revert to more normal W-L results in extra innings.