Last year, I did an analysis searching for indicators that can help predict which pitchers are most likely to miss extended time due to injuries or have a huge dropoff in performance. I followed that up with a post where I chose 20 Risky Pitchers for 2009 with the ambitious goal that 12 of the 20 would either fail to throw 2,000 pitches in the next season or see a FIP increase of 0.50 or higher (note: for the analysis, I’m switching to xFIP which is a new addition to FanGraphs and adjusts fly balls to the league average HR/FB rate).
The final (and humbling) results are below. 8 of the 20 pitchers or 40% of the pitchers dropped below 2,000 pitches or had an xFIP increase above 0.50. 40% sounds pretty good until you realize that about 40% of all pitchers coming off 2,700+ pitch seasons fall into one of these two categories the next year. Basically, my predictions were as successful as picking the names out of Kevin Mench’s ginormous hat.
|Pitches||xFIP Change (0.50+)||Dropoff|
Now I could try and make it look better by changing the criteria and saying my warnings of Nolasco saved teams of a 5.00+ ERA or a disappointing first half from Scott Baker but that wouldn’t be right. For every Nolasco (1.50+ ERA jump but a negative xFIP), there’s an AJ Burnett whose ERA was flat while his xFIP went over the 0.50 mark (note: you’d think this could be explained by the move to the new Yankee Stadium but his ERA was actually a run better at home vs away – 3.51 vs. 4.59). To add insult to injury, two of the selections (Vazquez and Greinke) turned out to be in the top 5 of our ‘best draft values‘ according to our Point Shares and their average ADP.
So f0r this post, a wiser and humbler Rudy Gamble will take another stab at the analysis. A future post will lay out 20 more predictions for 2010.
First off, there are a few aspects of the analysis that I think can be improved:
- Throw out any pitchers that met the 2,700 pitch marker but put up high numbers (5.00+ xFIP). This takes out a few players like Daniel Cabrera who aren’t going to be drafted anyway.
- Throw out any pitchers older than 37. The reasons why a Randy Johnson or John Smoltz missed time in 2009 is most likely different than the factors that would affect a 27-year old.
- Assume international pitchers like Dice-K and Kuroda had a similar pitch count the prior year (rather than credit them with a huge pitch increase).
- Increase the xFIP change from 0.50 to 0.75 to reduce the number of seasons that wouldn’t be viewed as a fantasy disappointment (e.g., CC Sabathia saw a 0.72 increase in FIP in 2009 but his 19-8/3.37/1.15 season was as good as any fantasy owner could rightfully expect).
With this revised ‘falloff’ definition, the amount of seasons that qualify move from ~40% to 27% (between 26-27% for 2005, 2008, and 2009 with an odd jump in 2006 to 36% and decline in 2007 to 16%). This represents 94 of 349 seasons between 2005-2009 with 72 fell below 2,000 pitches and another 22 had a +0.75 FIP increase.
The criteria we established last year after various tests were:
- 27+% of Sliders and Curveballs the year prior
- 700+ pitch increase the year prior (vs. the year before that) – inspired by the ‘Verducci Effect’
- Previous year was the first year above the 2,700+ MLB pitch threshold in 2008
Let’s revisit these assumptions based on some questions I had after the first analysis:
Revisiting pitch types
After my initial analysis, I exchanged a few messages with Disabled List Informer who ranked sliders as a greater injury risk than curveballs. I tested all pitch types again by comparing the averages of 2004-2008 seasons preceding dropoff vs. non-dropoff seasons in 2005-2009.
It appears sliders are negative indicators (11% more thrown in seasons preceding dropoff seasons) but there is no indication that curveballs are. Cut fastballs and split-finger fastballs are too small a percentage of pitches to take away any significance from this analysis. I did a subsequent analysis isolating pitchers who threw 10+% cut fastballs and split-finger fastballs. There were 42 seasons of 10+% cut fastballs which were succeeded by 12 falloff seasons. The 28% falloff rate almost exactly matches the average rate of 27% so we’ll rule out cut fastballs as a variable. Only 19 pitching seasons saw 10+% split finger fastballs (Roger Craig – the pitching coach not the 49er – sheds a tear) and 7 of those seasons were Dan Haren or Kelvim Escobar so I’m not going to make any conclusions on that given lack of sample.
Fastballs and changeups appear to be mild positive indicators. When I isolated pitchers who threw a below average % of Fastballs and Changeups, the results were promising for indicating potential falloff candidates. But after taking out those with a high % of sliders, the remaining seasons came in about average. So we’ll be taking curveballs out of the equation and focusing on sliders thrown as an indicator.
Below are the dropoff rates of those throwing sliders as 15, 20, and 25+ of their pitches indexed against the overall rate (27%). We can see that the higher the slider rate, the higher the percentage of dropoffs. That said, even at 25+%, the dropoff rate is only 35% (which is 32% more likely than a random pitcher). The fact that CC Sabathia threw 25% sliders in 2009 doesn’t really give me any pause in drafting him.
|Season Prior (2700+ pitches + xFIP < 5.00)||Seasons||Dropoff Seasons||% of Dropoff||Index|
|Slider > 15%||158||47||29.7%||110|
|Slider > 20%||79||25||31.6%||117|
|Slider > 25%||31||11||35.5%||132|
Perhaps a stronger argument for not using this one statistic alone is isolating pitchers who threw 15% sliders but they didn’t have a 700+ pitch spike in the previous year nor was it their first year > 2,500+ pitches (loosening this up vs the previous 2,700+ pitches). Of the 106 seasons that fit that description, 28 had falloffs or 26.4%. So basically an experienced slider pitcher is no more likely to have a dropoff than the average pitcher. In retrospect, this line of thinking would’ve taken Javier Vazquez, Ted Lilly, and Scott Baker off last year’s dropoff list.
I was curious to see if my initial heralding of a ‘700+ pitch increase’ and ‘first year above 2,700+ MLB pitches’ were just hiding an age-related skew – e.g., younger pitchers are more likely to drop off than players in their prime years.
Below is a distribution of all pitching seasons by age indexed against the 27% rate seen across all 21-37 year old pitchers. As you can see, there is no rhyme or reason here. I’m not reading into that dip at 27 given that 26 and 28 overindex. So scrap player age as a consideration.
|Age||Total Seasons||Dropoff Seasons||% Dropoff Seasons||Index|
700+ Pitch Spike
This criterion was inspired by the ‘Verducci Effect’ which theorizes that pitchers with a 40+ IP increase year over year is more at risk for injuries the next year. His theory seemed to have a level of success over the years although last year’s predictions – based on my dropoff criteria – were subpar. The only big dropoff on the list was from the worst (or 2nd worst) pitcher on the list (Eveland) and John Danks and Jonathan Niese are marginal cases.
|‘Verducci Effect’ Choice||2009 Pitches||xFIP Change||Dropoff|
|Jonathan Niese||1906 (estimate MLB + minors)||NA||YES|
(Quick Update: SI.com just posted Tom Verducci’s ‘Verducci Effect‘ 10 for ’10 today. I really like his work on SI and MLB. But my first allegiance is with Fantasy Baseballers so I need to point out that his success metric of ‘year without injury and with a lower ERA’ is a rather low bar. How low? Of the 349 pitcher seasons of 21-37 years olds following years of 2700+ pitches and < 5.00 FIP in 2004-2008 (translation: generally healthy years with a modicum of success), a full 60% of them saw a decrease in their FIP the next year. I’m assuming ERA follows the same path. Of the remaining 137 pitcher seasons, another 58 saw < 3,000 pitches thrown (a liberal proxy for no injuries as a healthy seasons is about 32 starts/3200+ pitches). Net result: 23% (79 of 349) of all pitchers might ‘succeed’ based on his criteria. So his 4-for-34 (12%) stat – which sounds amazing – is a little bit like taking credit for predicting a Jersey Shore character might do something embarrassing in an episode.)
Based on my new dropoff criteria, 38 of 112 (33.9%) seasons following a 700+ pitch spike saw a dropoff. This is a 126 index which is better than the 15% slider threshold. BUT if we isolate this criteria from the others (< 15% sliders and not the first year with 2,500+ pitches), it results in 6 of 22 seasons or 27%. So this factor alone isn’t a good predictor. Note that none of the 20 risky pitchers I picked last year fit only the 700+ pitch spike criteria.
(Note: It is possible this could be improved by factoring in minor league pitches as well. It’s a big pain to cobble together the stats for all the minor league divisions though and only innings pitched are available.)
First Season Above 2,500+ Pitches (in MLB)
This is the rarest of the three criteria with 78 seasons (or 22% of all seasons) and 34.6% of the instances (27) followed with a dropoff season. This 129 index slightly edges out the other two criteria. Isolating this criteria is near impossible as it almost always occurs with a 700+ pitch spike – only 2 cases have occurred in the past 4 years. But as the next section will show, it does seem to do a good job of isolating the riskier players who satisfy one of the other two criteria.
Testing 2+ of the Criteria
As noted above, while it appears that each criteria by itself is a positive indicator, isolating it from the other criteria saps it of any power.
The below chart shows the various combinations of the three criteria. A combination of 2+ of the criteria nets a 34% dropoff rate (index 128) and all three criteria nets 40.6% (index 151).
|Season Prior (2700+ pitches + xFIP < 5.00)||Shortcut description||Seasons||Dropoff Seasons||% of Dropoff||Index|
|Slider > 15%||A||158||47||29.7%||110|
|Pitch Diff > 700||B||112||38||33.9%||126|
|First Year > 2500||C||78||27||34.6%||129|
|None of three||-(ABC)||126||27||21.4%||80|
|1+ of three||A OR B OR C||223||67||30.0%||112|
|2+ of three||2 OR MORE||93||32||34.4%||128|
|All three||A & B & C||32||13||40.6%||151|
|*2 or more is the sum of ‘All three’ plus:|
|A AND C NOT B||3||0||0.0%||0|
|A AND B NOT C||17||6||35.3%||131|
|B AND C NOT A||41||13||31.7%||118|
If I had just limited my risky pitcher selections to only those that fit 2 criteria, Vazquez, Lilly, and Baker could’ve been replaced by Edinson Volquez (first MLB year with 2,500+ pitches and 700+ pitch spike), Manny Parra (same two criteria) and either Backe or Redding (whom I noted as risks) and netted a respectable 11-for-20 (with Greinke and Lester being painful selections).
Here is a breakdown of the 19 pitchers that qualified as a falloff with the # of criteria they met. 8 of the 19 (42%) fit 2+ criteria while 26 of the 73 (35.6%) overall fit the criteria (index: 118):
|2009 Dropoff Pitchers|
|Slider 15+%||Pitch Spike 700+||First Year Above 2,500+ MLB Pitches||Criteria Met|
Final point for this section: Of the 32 seasons that qualify for the trifecta, the ‘dropoff’ rate was 11-for-24 in 2005-2008 (46%) but only 2-for-8 in 2009 (25%). Andy Sonnanstine (whom was on my list) and Brandon Backe (who was noted but not put on because he wasn’t likely to be drafted) were the two players in 2009 that fit the bill. Todd Wellemeyer (2,117 pitches/0.72 xFIP increase), Armando Galarraga (2,453 pitches, 0.53 xFIP increase), and John Lannan (0.41 xFIP increase) did see regression with Ricky Nolasco, Johnny Cueto, and Gavin Floyd were able to maintain or progress. Just goes to show that even the highest indexing predictor can be ineffective for a season given its small sample.
That’s it for now. In the next post, I’ll list out the 20 riskiest pitchers to not draft based on these criteria.