Quantifying Carroll

Projecting playing time accurately is one of the keys to predicting fantasy baseball value and forecasting injury risk is a large part of that.  Rick Wilton recently wrote, and Tango blogged about the need for an actuarial table for injuries.

With these thoughts in mind and armed with Josh Hermsmeyer’s Injury Database I decided to evaluate the success of Will Carroll’s Team Health Reports.  Each year, Carroll puts out Team Health Reports prior to the season in which he assigns players green, yellow and red lights based on their level of injury risk.  Will’s THR’s are based on an actuarial table with 12 factors including position, age, team, body mass, injury history, recovery time, role or position change and conditioning.

So, should fantasy players stop at Carroll’s red lights?

2009 Results:

Position

Players

Avg DL days

Avg DL stints

Hitters

115

14.9

0.41

84

12.7

0.39

71

25.1

0.55

Starters

48

21.0

0.40

51

38.6

0.61

51

41.1

0.65

Relievers

27

11.0

0.33

17

24.2

0.41

16

35.9

0.56

Yes.  For hitters, it looks like green and yellow light players are similarly risky but a
red light player spends almost twice as much time on the disabled list.  For starting pitchers, only green lights look safe and by safe I mean “still projected to spend more time on the DL than the average hitter.”  Notice that not only are starters in each group far riskier than hitters in that same group but starters are also more likely to be red-lighted (34% of pitchers to 26% of hitters). 

So, can Carroll’s expertise help us project playing time?

Marcel used community projected playing time (an average of fan projections).  I regressed  actual 2009 plate appearances against the communities projected playing time and a “Red Light?” variable that had a value of 1 or 0. 

Real PA = 206 + 0.613*Community Forecast – 52.1*Red

According to the regression (with a sample of 252 hitters from 2009), you subtract 52 plate appearances (with a standard error of 22 PA) when you see a red light.  A larger data set would allow us to narrow down the size of the red light effect but with a p-value of 0.02, I’m fairly confident that there is one. Is this just because fans are hopelessly optimistic whereas an expert in projecting playing time is essentially taking all of this actuarial data into account?

In our forecast evaluations, Fantistics proved to be the expert on playing time.  Do red lights add information to their playing time projections?

Real PA = 132 + 0.652*Fantistics – 36.6*Red

From Fantistics, we subtract 37 PA for a red light (with a standard error of 21 PA and p = 0.09).  It looks like there might be room to combine Fantistics eerie ability to see into the minds of MLB managers with Carroll’s actuarial table and project playing time even more accurately.  (Note: Peter Rosenbloom has already been playing around with the 2010 THR’s in hopes of using them to improve the upcoming Steamer Projections.)

What about the playing time projections in Baseball Prospectus’s depth charts from last year? They’re already taking Carroll’s actuarial data into account, perhaps.

Real PA = 188 + 0.630*BP – 46.6*Red

From Baseball Prospectus’s depth charts we subtract 47 PA’s for a red light (with a standard error of 22 PA and p = 0.03)

Finally, I wanted to see whether red hitters underperformed their projections.  Although, I wish I had a larger data set here, it looks like they did, or more precisely, yellow and green players outperformed their projections.

Injury Rating

Hitters

Projected OPS*

Actual OPS*

Green/Yellow

187

768

790

Red

65

770

771

*weighted by 2009 PA

 

A red rating foretold an OPS loss of 22 points relative to green/yellow (with a standard error of 12 OPS points, p = .06).  Breaking this down by the number of 2009 DL stints instead of Carroll’s injury rating we see:

DL stints

Hitters

Projected*

Actual*

0

151

759

779

1

84

767

759

2

15

753

691

3

2

740

682

Clearly, we shouldn’t pay too much attention to the third and fourth rows of this table given the sample size but it seems worth noting that players who had one DL stint ended up 8 points below their projections, while players who were (serious) injury-free outperformed their projections by 20 points.  I’m assuming that the effect of DL time on performance has been studied before but I can’t find anything like that.  I could argue that players who are performing badly are more likely to be placed on the DL but that doesn’t explain the red light effect.  Does this effect hold for previous years? This might be worthy of further investigation.

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>