Hello Outlier

Word trickled out this morning that the Houston Astros had re-signed LaTroy Hawkins to a one year deal for $3.75 million. That seamed a bit cheap to me, given what pitchers are getting in free agency these days, so I took a closer look at Hawkins’ numbers to see if I was missing something. What I found was one of the strangest outliers I’ve ever seen.

Check out Hawkins’ batted ball graph:


Look at that ridiculous spike in GB% during 2007. His whole career, Hawkins has been an average groundball guy, but in 2007, he became the relief pitching version of Brandon Webb, just without all the strikeouts. It wasn’t a small sample size thing, either – Hawkins faced 225 batters in 2007 while pitching a full year in relief for the Rockies. Out of nowhere, with seemingly no explanation, his GB% jumped from 44% to 63% in a single year.

As you’ll notice from the graph, the trend didn’t continue in 2008, as his GB% fell back to 46.1%. He offset the lack of groundballs by returning his strikeout year to 2005 levels, posting a 6.97 K/9. His 3.28 FIP in 2008 was the best he’s posted since 2003.

So what happened? GB rates have extremely high year to year correlations, and they normalize very quickly. You just don’t see huge swings in groundball rates from year to year like this. The odds of it being random variation are extremely slim. It is much more likely that Hawkins changed his approach in some way during the ’07 season, decided he didn’t like the results, and scrapped it for 2008. However, I haven’t been able to find any evience that this is true.

So, Rockies/Astros fans, what have you got? What happened to LaTroy Hawkins in 2007, and why didn’t it carry over to 2008?

Print This Post

Dave is a co-founder of USSMariner.com and contributes to the Wall Street Journal.

13 Responses to “Hello Outlier”

You can follow any responses to this entry through the RSS 2.0 feed.
  1. Jon says:

    He threw a lot of changeups that year…?

    Vote -1 Vote +1

  2. Tom Au says:

    In my experience, one year is a “small sample size” for baseball players. It takes about three years to make a valid sample. So the fact that this, or another pitcher, couldn’t “keep it up” this year suggests that last year’s “trend” isn’t valid.

    Yes, there is often high correlation between one year and the next, because three of them suggests a trend. But this happened to be an exception to the rule, when this year reversed the previous year’s “trend.”

    Vote -1 Vote +1

  3. Dave Cameron says:

    Sorry Tom, but the idea that a three year sample is required for every single statistic just doesn’t hold. Basic statistical principles – the lower the variance in a population, the smaller sample size you need.

    GB% has one of the lowest variances of any individual performance statistic. If this was common, you’d be able to point to a whole bunch of guys who had one year flukes where their GB% rose or fell 20%.

    If you can find another example, feel free to point them out.

    Vote -1 Vote +1

  4. David Foy says:

    Has anyone looked at his Pfx? Was there a greater amount of a certain pitch? Was there a change in location of his pitches? An increase in velocity?

    Vote -1 Vote +1

  5. Per Tom’s comment, from what I’ve read, more specifically, it is a relief pitcher’s “season” that suffers from small sample randomness and requires three seasons to smooth out their tendencies.

    But to Dave’s comment, just because variance is low does not mean that a relief pitcher had enough “events” to meet the sample size requirement. Have you done the calcuations (or seen it somewhere)? If so, what is the calcuation for sample size for this situation?

    And just in general, the right sample size varies greatly in baseball, I’ve found. For example, Tom Tango a while back calculated that it would take 7 seasons of BABIP for a starting pitcher to (sorry don’t remember his exact words or technical stat jargon) “prove” that his BABIP is lower than the generally accepted .290 or .300 BABIP that “every” pitcher regresses to the mean towards.

    Vote -1 Vote +1

  6. I did some rough analysis. He had roughly an extra 31 extra ground balls in 2007. Looking over his games, I went through every game with 3 or more ground balls, and found that a lot of them – 10 – were at home, leading to at least 13 extra ground balls (assuming 2 was “average” or the most he could have expected).

    So I dug deeper and calculated the GB% for home vs. road. At home, he was at 70.5% GB, on the road 56.2% GB. That worked out to roughly 9 extra ground balls on the road (out of 89) and 22 extra ground balls at home (out of 88). So it was his home that really drove his high GB% in 2007, though his road was also on the high side for his career.

    Also, looking at the FB%, it looks like it was FB that took the brunt of the switch to GB in 2007. On the road, his LD% was virtually the same as his career, but FB% fell from 34.6% to 23.6%. At home, however, both fell by approximately half. FB% fell from 34.6% to 18.2% and LD% from from 19.6% to 11.4%. Fell, meaning differed from his overall career numbers.

    Something about those humidor balls in Colorado really deadens the ball, killing FB and LD and inducing more GB.

    Given this evidence of ball deadening, and given the rumor that Colorado sometimes slips in non-humidor balls, hence those old wild games of old in humidor days, it would be interesting to separate the low-scoring games from the high scoring games in Colorado and seeing how different the GB%, FB%, and LD% are. Obviously, to have a high scoring game, there necessarily has to be a lot of FB and LD, but perhaps there is a stat expert reading who can calculate the odds of a high percentage of FB% suddenly in one game and seeing if that proportion holds across a full Colorado season, or even multiple seasons.

    Also, just noticed his BABIP was way low in 2007, a result, obviously of all the GB.

    Vote -1 Vote +1

  7. Roman says:

    Coors field affect? focused more on staying low in the zone?

    Vote -1 Vote +1

  8. DropDeadFred says:

    I think the change from AL East to NL Central also helped Hawkins in the middle of 2008.

    His numbers as a Yankee in 2008: 41 IP, 23 K, 16 B, 42 H, 78 ERA+, 1.44 WHIP, 7 out of 12 inherited runners scored. Opponents hit .275/.339/.386 against him.

    His numbers as an Astro in 2008: 21 IP, 23 K, 5 BB, 11 H, 992 ERA+, 0.76 WHIP, 0 out of 5 inhertied runners scored. Opponents hit .151/.203/.205 against him.

    So his K rate and really all his other numbers didn’t pick up until he moved to the NL.

    Vote -1 Vote +1

  9. Derek Carty says:

    StatCorner has Coors’s GB Park Factor at 99.6, very close to league average. FBs are 95.3 and LDs are 102.7, so I’m not so sure park effects explain it, and it doesn’t look like Coors deadens balls in the air since FB are on the low side and LDs on the high side.

    A quick look at PITCHf/x doesn’t explain things either, though. He didn’t throw a sinker or a change-up with a lot of sink, and he located pitches down in the zone at about a league average rate (you can check out a chart here: http://www.hardballtimes.com/images/uploads/LaTroy_Hawkins_2007_Vertical_Location.jpg). I really expected to see a whole bunch down in the zone, but that’s just not the case.

    Very strange stuff.

    Vote -1 Vote +1

  10. Watcher says:

    I think its clear that Hawkins, who spent the majority of his career until that time in the AL, went to the NL and was faced with the prospect of staring down Mike “Juggernaut” Jacobs, he completely changed his style out of fear.

    Vote -1 Vote +1

  11. C Johnson says:

    I’m an Astros’ fan and I saw most of HAwkins’ pitching with Houston.

    Normally, I would guess that a spike like that would be caused by throwing more 2 seam fastballs. Hawkins can throw both 4 seam and 2 seam fastballs. It would be natural to expect that Hawkins might fear the flyball in Coors Field and shift to more sinking fastballs. His average pitch speeds don’t give any indication of that, though; maybe a Rockies fan could tell us if my guess is correct.

    As for his time with the Astros, he was getting Ks with a rising 4 seam fastball and a good slider. After the hiatus subsequent to his release from the Yankees, two of the Astros top baseball people worked out Hawkins and reported that his velocity had returned, relative to his period with the Yankees. The Fox network guns were reporting velocity of 95 – 97 mph as an Astros player. The swings and misses seemed consistent with that. Also, Hawkins has a huge differential in his righty/lefty performance. In 08, his WHIP was 0.86 against RHB and 1.74 against LHB. Cooper did a good job of protecting him from pitching against too many LHB, particularly in leveraged situations.

    Vote -1 Vote +1

  12. Jim says:

    I think there may be something to the fact that his pitch selection looks very different in 2007. Maybe the extra change-ups caused more GB, especially with over-eager batters in CO. Interesting stuff.

    Vote -1 Vote +1

  13. Wesley says:

    Hawkins dramatic and unexplained spike in ground ball % was a subject of interest in Colorado as the 2007 season played out.

    I never heard anything in the local media from Hawkins himself or any other players or members of the organization that shed any light on the anomalous results.

    The only two factors that I can think of are these:

    1) he had an arm injury early in the year that robbed him of his ability to throw his prefered breaking ball properly (this was the explaination for his horrible performance at the start of the season) and even when he returned he seemed to still be trying to find new ways of getting people out rather than reverting to the pitch selection and style that he had used the rest of his career.

    2) Apodaca does place a heavy emphasis on pitchers keeping the ball down in the zone and getting ground balls and there is some record of success in this aspect with various pitchers over the last several years.

    On a final note, having watched the vast majority of Rockies games on TV in 2007 I feel confident in saying that Hawkins results did not feel ‘flukey’ or random at all. Totally unexplained – yes. Flukey? No.

    Vote -1 Vote +1

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>