Tribute to Roy Halladay – Master of Short Games

Introduction

The baseball world was stunned last week with the news of Roy Halladay’s death at age 40.  Being a Phillies fan, I was a quick admirer of Halladay’s work ethic and his great pitching skills when he joined the Phillies, as illustrated by his two no-hitters in the 2010 season.  The first no-hitter was a perfect game against the Marlins and the second was in his first post-season start against the Reds.  In Ian Hunter’s recent tribute to Halladay, he describes six “Halladays” — complete games that finished in under two hours.

Last week, I described the duration of the 2017 World Series games, getting some insight how games can be so long.  Since Roy Halladay appeared to be a quick worker on the pitching mound, I was inspired by Hunter’s article to explore the durations of the games when Halladay was a starter.

Exploring Game Durations

Since Halladay started all but two games during the seasons 1999-2013, I looked first at the game durations (in minutes) for all games in these 15 seasons.  On average, we see that the median game duration has slowly increased from 1999 to 2013.  For a given season, the game durations are symmetric in the middle, but there are number of outliers at the high end (due to high scoring and extra-inning games).

An Aside: The Increase in Game Durations

Making this plot made me wonder how the median game duration has changed over, say, the last 50 seasons.  I have shown the pattern below.  There was a steady increase from the middle 70’s to 2000, then there was a drop until 2005, and there has been a steady increase until the current season.  The average game duration in the 2015 season is about 30 minutes longer than the average duration in the last 1970’s.

Games Started and Not Started by Halladay

Getting back to the main issue, were games started by Halladay unusually short?  To answer this, we draw boxplots of the game durations for games started and not-started by Halladay.  We see several interesting things:

• In Roy’s early seasons (1999 to 2001), his game duration was about average.
• Starting with 2002, his games were getting shorter.  The most interesting season was 2005 when the average in games he started was about 30-40 minutes shorter than games when he didn’t start.
• In later seasons (2006 and on), Roy was consistently shorter than average, but the advantage seemed to be smaller in his final seasons.

Quickest Starters

Did Halladay rank among the quickest pitchers in baseball?  To answer this, for each season, I ranked the top-ten pitchers (the ones that started at least 19 games) with respect to the median duration of the games that they started.  For the seasons 1999 through 2013, I constructed this top-ten list, and then I counted the number of occurrences on this top-10 list.  The below graph shows the seven pitchers who appeared at least three times on the top-10 list.  The king of short games during this period was Mark Buehrle (10 occurrences), followed by Greg Maddux (6 occurrences).  Halladay is one of the five pitchers who appeared in the top-10 list three times.

Maddux versus Halladay

Since I also was a great admirer of Greg Maddux who was noted as a quick worker, I thought it would be interesting to compare the game durations of the two pitchers.  Below I construct parallel boxplots of the durations during the seasons 1999 through 2008 when the two pitchers overlapped.   Generally, they seemed to be similar with respect to game duration, but again we note that Halladay was remarkably quick during the 2005 seasons.  (Things changed during 2006 and Maddux was the quicker worker on average.)

Takeaways and R Work

Okay, Roy Halladay appears to be a quick pitcher, although the game duration is not a precise way of measuring speed.  Since Halladay often pitched deep into a game, I imagine that game duration is a reasonable proxy for the more accurate measurement.  I think it is interesting that Halladay’s speed was not always fast — as he matured into one of the best pitchers in baseball, he tended to speed up.  This suggests that pitching speed might be related to confidence or experience.

All of the R code for this study can be found on my gist github site.  I have previously downloaded all of the Retrosheet game l data for all seasons and the game logs contain the starting pitchers and the game duration for all games.  The dplyr package is used for all of the data manipulations.