Printer-friendly

Contact Author

April 9, 2009

Checking the Numbers

Cliff and the Gang

by Eric Seidman

Taken as a group, ground-ball pitchers are fascinating, in that they can succeed at the major league level with what might be seen as relatively average stuff. The group doesn't miss many bats, but they do tend to have the benefit of both command and control, hovering around the strike zone in order to pitch to contact, and preventing balls from being hit in the air. Ground-ball rates themselves are stable and immune to large fluctuations; take any five-year span and run an intraclass correlation—testing the year-to-year stability on the individual pitcher level—and it will likely show a strong relationship somewhere in the 0.55-0.70 range. When a large fluctuation does surface, due diligence would require asking if a change in approach or skill has been observed, instead of a knee-jerk reaction that might dismiss the statistical shift as a luck-based indicator that is bound to regress.

All of which brings us to Cliff Lee, the reigning AL Cy Young Award winner, who experienced a large jump in his ground-ball rate last season, almost doubling his career GB:FB ratio to that point. There were plenty of cynical analysts on the web who fell prey to confirmation biases last year, performing exploratory statistical surgery in an attempt to diffuse the mainstream obsession with Lee's success. When some of the telltale signs of luck normalized, such as his BABIP and strand rate, and Lee sustained his minuscule walk rate and newly increased ground-ball rate, it became less reasonable to disagree with the consensus that Lee was just flat-out dominant. When the season came to an end and Lee became the second straight Cleveland hurler to take home the Cy Young, questions arose once more, this time pertaining to whether or not Lee could build upon, or sustain, his incredible 2008 season.

When Marc Normandin, Kevin Goldstein, and I penned our Player Profile on Lee last year, we combined the quantified performance aspects with scouting information, concluding that Lee had altered his approach and was also pitching with more confidence than ever before. One of the major causes for his progress in both of these areas was his improved mechanics, as the team spent time getting the lanky lefty to repeat his release point consistently on each pitch. Nobody really expects Lee to repeat his lofty 2008 performance, but his ability to sustain even a fraction of his production from last season will hinge on whether or not the spiked GB:FB rate is random, or the result of a documented change in approach. While Lee himself will need to pitch another couple of seasons before this question can be completely answered, we can investigate how others with similarly large spikes fared in the seasons following their higher ground-ball rates; the information can at least provide some cursory insight into what might be expected.

Coding for balls in play is a bit problematic in the sense that different systems score plays in different fashions. I use a Retrosheet database for studies like this, which potentially carries different data than those from Baseball Info Solutions. For that very reason, the specific ground-ball rates will not be mentioned here and a bit more leeway will be given to the spikes and year-to-year differences. Once the ball-in-play rates were calculated, I queried the database for situations in which a pitcher experienced a ground-ball rate increase of at least 8% between two seasons while facing 500+ batters in the initial year as well as the year of the spike. To clarify, increases in percentages can be confusing depending on whether or not subtraction or division is used to make the calculations. The subtraction method is much more common so when referring to an 8% increase, an example would be a pitcher seeing his rate rise from 45% to 53%, even though this rate-change would actually constitute an 18% increase. Of the 86 pitchers since 1954 that met this criteria, the next step involved comparing their ground-ball rates over the next several seasons to the year of the spike; essentially, if a pitcher jumped from 45% to 57%, were the rates in the following seasons close to 57%?

Comparing the future years to the year of the spike is more accurate than looping the subtraction to occur each year, since our examples could then consistently drop off by a slim margin to the point that the ground-ball rate four years later would actually be lower than the year before the spike. Keeping with the theme of a minimum of 500 batters faced, I decided to set the low-end rate discrepancy to -0.03, so that the pitcher could fall no lower than 54 percent in any of the next few seasons. This made it easier to separate the flukes from those who had legitimately developed some new skill.