As we begin to untangle the wreckage of the Astros' 2011 season, the statistical pieces are strewn about like a big puzzle. Although the Astros were not, by any means, a good team in 2011, the Astros' biggest loser team of all time had a record which reflected significant under-performance, You can call it bad luck if you want, or you can call it a statistical outlier.
You probably know that the Astros' Pythagorean Record indicates that the Astros' should have been a 62 win team, instead of the actual 56 wins, based on the team's runs scored and allowed differential. Teams' over and under performance of their Pythag projection is shown here. The luck column displays the deviation of actual wins from the Pythag result. The Astros had the third biggest margin of under-performance (-6), behind the Padres and Royals at -7 and -8. The Brewers, Tigers, Giants, and D-Backs were "lucky" to the tune of +8 above the Pythag result.
The Pythagorean results show deviations in results at a relatively high level--runs. A more detailed view of performance can be seen at the level of hits, extra bases, and walks. OPS gained credence as a measure of offense because it correlates well with runs. Similarly OPS-against reflects pitchers' ability to minimize hits, extra bases, and walks. The Astros' runs scored and allowed significantly underperformed the offense's OPS and the pitchers' OPS-against. If the Astros had converted OPS into runs, and OPS-against into runs allowed, at the same rate as the league average, the Astros would have scored 7 more runs on offense and allowed 17 fewer runs, for a net gain of 2 - 3 wins. If the Astros' actual runs scored and runs allowed are normalized for the team's OPS and OPS-against, the Pythagorean formula would project the Astros as a 65 win team.
A number of factors can affect a team's ability to under or over perform its OPS and OPS-against. But one of the most important factors is the timing of the events which comprise OPS---whether the hitters and pitchers performed better in clutch situations. I use Fangraphs' clutch statistic, which is based upon players' win probability added (WPA) for high leverage situations compared to normal situations. Based on this statistic, the Astros were the 9th worst clutch team on offense and the pitchers were 4th worst in clutch situations. Undoubtedly, the Astros' poor clutch performances made the Astros less efficient in converting OPS into runs and preventing opponents' OPS from turning into runs. The Astros' -2.6 clutch hitting score implies that the Astros lost 2.6 wins due to poor hitting in the clutch.
Relationship Between Pythag and Clutch
After the 2010 season, I wrote about the Astros' clutch team performance. In 2010, the Astros over performed their Pythag projection by 8 games (68 win Pythag vs. 78 actual wins)---the highest over performance in the majors. The Astros also were the best clutch hitting team in 2010. The Astros also had the highest actual win margin above their Pythag for the three year period 2008 - 2010, accompanied by the highest clutch hitting statistic for the same period. The top five clutch hitting teams for the same three year period also over performed their Pythag projection for the same period.
Between 2010 and 2011, the Astros shifted from extreme over performance for both Pythag and clutch to extreme under performance for both Pythag and clutch. Is there a cause and effect relationship?
To test this question for the 2011 season, I performed correlations between clutch hitting and clutch pitching vs. major league teams' under or over margin for the Pythagorean projection in 2011. Both forms of clutch, as defined by Fangraphs, are positively correlated with the difference between Pythag and the teams' actual record. However, batting clutch is more highly correlated with the Pythag deviations than pitching clutch (correlation coefficients of 0.65 for batting and 0.31 for pitching). Combining both batting and pitching clutch stats as independent variables in a linear regression estimate of Pythag differentials produces an R-square of 0.508, meaning that the fangraph clutch statistic "explains" over half of the deviation between Pythagorean projections and actual win records in 2011. Since this analysis pertains only to the 2011 season, these results may not apply to all years. However, the results suggest that clutch hitting, as defined by Fangraphs, has a strong relationship with teams' over or under performance of Pythag.
Because bullpen performance frequently is cited as a factor in teams' deviations from Pythagorean projections, I also tested the correlation of bullpen performance. I calculated a measure based on shutdowns minus meltdowns, as those two stats are defined by Fangraphs; the Astros' bullpen had the worst result of any team for shutdowns net of meltdowns. A positive correlation exists between this measure of bullpen performance and Pythag deviations, but the correlation coefficient (0.24) is substantially smaller than for the clutch stats.
I'm not sure how teams' clutch hitting affects their deviations from the Pythagorean projections. Presumably the effect is indirect. Perhaps clutch hitting affects teams' win frequency in close games.
There is dispute over whether clutch hitting is a repeatable skill on the part of hitters or pitchers. Many in the sabermetric community would put clutch results in the "luck" category. Some believe that clutch skill may exist, but that the effect is small. Others think that there is insufficient evidence. If the clutch statistic is a luck measure, even if only partly so, then we might expect some beneficial regression for the Astros' W/L record in the future.
In my article after the 2010 season, I said:
If the Astros' clutch hitting has been critical to beating the Pythag win-loss record...and if the clutch results are basically luck...then this could be a bad predictor for the Astros in 2011, since it might indicate a regression in win-loss record.
That statement may have turned out to be correct. I also pointed out that two of the most significant contributors to the Astros' clutch hitting, Lance Berkman and Geoff Blum, would not return to the team in 2011. Furthermore, Michael Bourn and Hunter Pence, who are also among the better clutch hitting contributors were traded during the 2011 season. On top of that, Carlos Lee, who normally is one of the more consistent clutch performers in baseball, did not score well on the clutch stat in 2011.
The top Astros' contributors, based on the Fangraphs clutch stat, in 2011 were:
Best Clutch Hitters
Note: all of the Clutch values, above, are positive numbers.