Baseball’s Most Surprising Seasons, Good And Bad

Which baseball players have had the most surprisingly bad and surprisingly good seasons in recent years? I wondered about this while researching an article on whether spring training performance foreshadows regular-season production.

I calculated the uncertainty in the Marcel forecasting system projection for batting wOBA — a measure of a hitter’s overall offensive production per plate appearance — based on the reliability of the forecast. This gives us a range of expected results and allows us to look at which players’ regular-season performances were the least likely going into the year. It’s a nice way of quantifying unexpectedly good and bad campaigns.

First, the most surprising strong seasons in the dataset I used in my article, which extends back to 2006 (minimum 200 plate appearances):

Before the 2013 season, we would have expected there to be a 50 percent chance that Hanley Ramirez’s wOBA would be above .346. If you’d asked us what the odds were that Ramirez’s wOBA would reach or beat .444, we would have said practically zero — 0.1 percent, to be exact.

The fact that Ramirez’s wOBA was .444 was an outcome in the 99.9th percentile of his preseason wOBA distribution.

Flipping things around, here are the most disappointing seasons of the past eight years:

Travis Hafner, if you’ll recall, had been one of the best hitters in baseball in the four years leading up to 2008, which was one of the big reasons why another statistical system for forecasting player performance, FiveThirtyEight founder Nate Silver’s PECOTA, called for the Cleveland Indians to win 91 games that year. Instead, the Indians went 81-81 as Hafner’s wOBA sunk to .270 — an outcome that seemed almost impossible (hence, the 0.0 percentile score).

Keep in mind that this method is based on the Marcel reliability score, which essentially measures how much of a given projection is made up of the league mean and how much belongs to the player’s statistical record. It employs a generic age adjustment, but it does not look at similar players at similar ages, as Silver did with PECOTA.

Hafner, Andruw Jones, Chone Figgins, Adam Dunn and others on the second list hit their early 30s and, rather than declining gradually, completely fell apart. Marcel has no way to determine whether players with certain tendencies or body types are more likely to completely crater, which would affect our confidence intervals. Cleaning up projections on the margins like that is one of the ways a system such as PECOTA is superior to Marcel, even though the returns diminish sharply with extra complexity.