RESEARCH: Rebooting LIMA

There is a lesson in the success of this summer's blockbuster move, "The Dark Knight Rises", that can be extrapolated to our games: periodically, you need to reboot even highly successful franchises. For BaseballHQ.com, our hallmark franchise is the LIMA Plan, which happens to be overdue for the reboot treatment.


A quick refresher

Back in 1998, Ron Shandler first introduced the LIMA Plan with the seminal essay "The secret of my crappy pitching staff." From there, what was eventually labeled as LIMA became a dominant roto strategy for the early part of the 2000s, before Ron eventually pushed the bar higher with a number of plans that could be described as evolutionary descendants of LIMA. (For details, see this 2009 article "Is LIMA still relevant?")

As Ron notes in that latter piece, BaseballHQ.com has moved on from the LIMA Plan. But the legacy of LIMA, and specifically the core skills filters it introduced, is still engrained in our thought processes. Most readers will recognize these quickly, if not instinctively:

Ctl  <= 3.0
Dom  >= 6.0 (eventually lowered to 5.6)
Cmd  >= 2.0
hr/9 <= 1.0


The simplicity of those thresholds, all set at easy-to-remember whole number values, made those cutoffs "sticky" in our heads. But sometimes "sticky" things are a double-edged sword: as the MLB environment has changed over the years, those thresholds may now be obsolete. Continued reliance on them may even do us more harm than good.

Therefore, our goal this week is to "reboot" the LIMA skill filters, as they are the most widely-applicable component of the overall LIMA Plan, to make them relevant for another 15 years.


How have the filters held up over the years?

To evaluate how the LIMA filters have performed over the past 15 years, we compiled the aggregate stats lines of the pitchers who worked at least 100 IP in a season and met all of the Ctl, Dom, and Cmd filters. (We left the hr/9 component out of this exercise, for a couple of reasons that we'll discuss a little later.)

The results are rather consistent:

Year  Qualifiers   Cohort IP    Cohort ERA
====  ==========   =========    ==========
1998      37         7641          3.67
1999      22         4309          3.75
2000      21         4388          4.02
2001      36         6719          3.70
2002      35         6358          3.64
2003      34         6385          3.51
2004      36         6459          3.90
2005      33         6727          3.48
2006      42         7496          3.95
2007      39         7259          3.87
2008      38         7221          3.59
2009      38         7525          3.62
2010      45         8494          3.56
2011      62        11829          3.50
2012**    71         9368          3.80

** used 70 IP (as of August 21) as the cutoff for 2012

So, over the sample size of 15 years, the performance of full LIMA qualifying pitchers has fluctuated within a narrow range. Other than a couple of outliers on either side of the bell curve, we can observe a normal range of 3.60-3.85 ERA from the qualifying cohort in any given year.

And as one might expect, as overall offense has been somewhat muted in recent years, the number of qualifying pitchers per year has been rising, with 2012 tracking toward a second consecutive all-time high-water mark.

So, all is well, right? Not so fast...

Context is everything

We have established that the performance of our cohort is fairly stable over a 15-year study period. However, the performance of the league as a whole is anything but stable. Here is the above table again, with league-wide data added for context:

Year  Qualifiers   Cohort IP    Cohort ERA       MLB ERA     Cohort gap      % gap
====  ==========   =========    ==========       =======     ==========      =====
1998      37         7641          3.67           4.43        -0.760          -21%
1999      22         4309          3.75           4.71        -0.965          -26%
2000      21         4388          4.02           4.77        -0.755          -19%
2001      36         6719          3.70           4.42        -0.723          -20%
2002      35         6358          3.64           4.28        -0.638          -18%
2003      34         6385          3.51           4.40        -0.888          -25%
2004      36         6459          3.90           4.46        -0.565          -15%
2005      33         6727          3.48           4.29        -0.808          -23%
2006      42         7496          3.95           4.53        -0.578          -15%
2007      39         7259          3.87           4.47        -0.601          -16%
2008      38         7221          3.59           4.32        -0.731          -20%
2009      38         7525          3.62           4.32        -0.700          -19%
2010      45         8494          3.56           4.08        -0.517          -15%
2011      62        11829          3.50           3.94        -0.445          -13%
2012**    71         9368          3.80           4.04        -0.238           -6%

** used 70 IP (as of August 21) as the cutoff for 2012

We have added three additional columns to the previous table here, but the right-most one alone tells the tale. That percentage is measuring the inherent advantage held by the pool of full LIMA qualifiers, as compared to the MLB pitching population at large. In the early/prime years of LIMA, that edge was consistently in the 20% range. Since 2010, that has dipped to 15% and below, even cratering well into the single-digits this year.

That 2012 number is a little misleading, since this year's cohort ERA is on the high end of the normal range. But overall, there is still a clear trend line to be drawn, from the last peak in 2005 down to today. Especially when you consider that fantasy leaguers aren't rostering 100% of the pool of MLB pitchers, the "LIMA edge" has suffered some significant erosion due to the recent reduction in run-scoring across MLB.

So what do we do about that?


Resetting our filters

If we want to look at how our filters need to be adjusted, we need to look at the league context for those metrics. Without going through the full 15-year history, here is an abbreviated look at where they were in the early days, and where they are now:

Year     Ctl   Dom   Cmd   hr/9
====     ===   ===   ===   ====
1998     3.4   6.6   1.9    1.0
1999     3.7   6.5   1.7    1.2
2000     3.8   6.5   1.7    1.2

2010     3.3   7.1   2.2    1.0
2011     3.1   7.1   2.3    0.9
2012     3.1   7.5   2.4    1.0

Relative to the current league context, a traditional LIMA-qualifying pitcher could now simply be exhibiting barely-average Control, and subpar Dominance and Command!

We took a trial-and-error approach to finding a new set of filters that would restore LIMA's advantage over the league-wide player pool, and cut the pool of qualifiers back down to a similar level of selectivity as the LIMA distinction used to carry. If possible, we also wanted to retain some of the convenience of round-number levels.

Not all of these goals were attainable, but we think we found levels that meet several of those objectives:

  • 2.8 Ctl
  • 7.0 Dom
  • 2.5 Cmd

In the following table, we compare traditional LIMA's first five years to these new levels :
 

Year  Qualifiers   Cohort IP    Cohort ERA       MLB ERA     Cohort gap      % gap
====  ==========   =========    ==========       =======     ==========      =====
traditional LIMA: 3.0 Ctl, 6.0 Dom, 2.0 Cmd
1998      37         7641          3.67           4.43        -0.760          -21%
1999      22         4309          3.75           4.71        -0.965          -26%
2000      21         4388          4.02           4.77        -0.755          -19%
2001      36         6719          3.70           4.42        -0.723          -20%
2002      35         6358          3.64           4.28        -0.638          -18%

"rebooted" LIMA: 2.8 Ctl, 7.0 Dom, 2.5 Cmd
2010      25         5065          3.40           4.08        -0.679          -20%
2011      33         6662          3.37           3.94        -0.569          -17%
2012      45         6022          3.61           4.04        -0.434          -12%

These new filters largely meet our goals:

  • The population selectivity is right in line with original LIMA;
  • The % gap is also back in line, although 2012 still lags a bit (again, 2012 is a high-side outlier for cohort ERA in both the original formula and the new formula. We don't want to correct any further and shrink the population too far).
  • 2.5 Cmd becomes the new "magic" level. Setting the new Dom threshold at 7.0 is about convenience as much as anything; you can wiggle the Ctl/Dom filters a bit as long as you stay locked on the 2.5 Cmd. For instance, if you want to really emphasize strikeouts, you can go to 3.0 Ctl/7.5 Dom as your baseline.

A word about hr/9

There are two reasons that we have ignored LIMA's original hr/9 filter in this analysis:

First, HR rates haven't moved as much as Ctl, Dom, and Cmd have in the past 15 years. After a brief spike to 1.2 hr/9 league-wide in LIMA's early years, HR rates have stayed at a consistent 1.0 hr/9 or 0.9 hr/9 in recent years. There was no sea change here; if you want to retain a hr/9 < 1.0 filter, feel free.

Second, the best argument for dropping the hr/9 requirement is that we have learned a lot over the years about where HR come from. Since HR are basically a function of fly balls allowed, we could just as easily set our filter based on FB%, for example FB% < 40%. You could even tighten or loosen that requirement a little bit based on a pitcher's home ballpark, and whether it favors hitters or pitchers. Basically, in original LIMA, the hr/9 component was a concession to the fact that we didn't understand much about how much pitchers influence HR allows. We don't have that problem anymore.

Conclusion

To regain the advantage that LIMA offered in the late 1990s, reboot your LIMA-trained eye to look for these levels:

Ctl <= 2.8
Dom >= 7.0
Cmd >= 2.5
FB% < approx 40% (optional)

Happy 15th birthday, LIMA! Here's to your next 15....


Click here to subscribe

  For more information about the terms used in this article, see our Glossary Primer.