FP911 Login or Register Today
Follow FantasyPros911 Here
 |
 |
 |
 |
 |
Other Articles You Might Like
- Error
|
The sabermetric revolution is in full force. From VORP to WARP to MORP to BABIP to UZR, the stat-heads seem to be an acronym-loving bunch. One acronym that has gotten a lot of play is FIP. Our friend Derek Carty from THT wrote an article on Monday explaining why he doesn't use the metric to evaluate pitchers. That inspired me to do a complete evaluation of the various free ERA estimators available, which follows.
In fantasy baseball, and even real baseball for the most part, we are not concerned with what has already happened, but what will happen. If someone is offering me Jon Lester in a trade, I don't care that his ERA is 4.76. He would be coming to my team with a clean slate and all I am concerned with is how he will perform in the future. To help us accomplish the difficult task of figuring out what a young pitcher like Lester could contribute to our team, we could look at an ERA estimator. In general, what an ERA estimator sets out to do is strip out the effects of luck, and tell you what the pitcher's ERA should be based solely on his controllable skills. Have you ever wondered how to balance the effect of good luck in one category, but poor luck in another? Do they completely offset each other or is one category weighing on the pitcher's ERA more heavily? Let's look at Cliff Lee as an example. He has an inflated BABIP of .337, but his strand rate (LOB%) is an unsustainable 80.7%. So Lee has been victimized by poor defense, but has benefited from the ability of he or his bullpen to strand base runners at a higher than normal rate. An ERA estimator would mix all of Lee's skill components into a tasty concoction and produce a more accurate picture of exactly what Lee's ERA should look like, absent luck. Though some would argue that it is not fair to use for a backward-looking analysis, it is safe to assume that this is the best path to follow to project future performance. The only thing these ERA estimators do not take into account are a change in skills. It is not a projection tool, but rather looks at the process (the pitcher's current skills) and spits out an expected result (ERA) based on that process. Now that the purpose of an ERA estimator is out of the way, let's take a look at the various free ones that are available: FIP (Fielding Independent Pitching) Formula: (HR*13+(BB+HBP-IBB)*3-K*2)/IP, plus a league-specific factor (usually around 3.2) to round out the number to an equivalent ERA number Explanation: Easily the most famous of all the ERA estimators, this metric could be found on both The Hardball Times and FanGraphs. The numbers do differ on each though, most likely because each site is using a slightly different "league-specific factor". FIP strips out the effects of defense and assumes that every pitcher has the same ability to strand base runners. It also assumes that pitchers have varying degrees of ability to prevent home runs on fly balls, as you could see from its use of raw home runs allowed. FIP then looks at a couple more controllable skills that we are all familiar with, such as walks, strikeouts and hit by pitches. Intentional walks are removed from the equation as this is obviously just a strategic baseball move and not any fault of the pitcher. Pros: The formula is relatively simple and if you assume the league-specific factor is 3.2, you could calculate the number pretty easily yourself and come very close to what THT and FanGraphs show. The formula also correctly removes intentional walks from the equation and includes hit by pitches, while taking the most obvious controllable skills, the strikeout and walk, into account. Cons: The formula assumes that every pitcher has the same ability to strand base runners, which is false. The better pitchers will almost always have higher strand rates (with some exceptions; yes, that would be you Javier Vazquez) than weaker pitchers because they record outs more efficiently. Also, the better pitchers tend to have the highest strikeout rates and all else equal, strikeout pitchers will have higher strand rates than their soft-tossing counterparts. Last, FIP does not correct for an inordinately low or high home run per fly ball ratio (HR/FB). Numerous studies have shown that pitchers have little to no control over their HR/FB ratio and it will usually revert toward the league average of between 10%-11%. However, park effects do influence this number, and over a large enough sample (many seasons), some pitchers do show a slight ability to control it. Though, the effects are small enough that over just a couple of months of a season, your best bet is to assume a 10%-11% mark going forward. xFIP (Expected Fielding Independent Pitching) Formula: no exact formula is provided by THT; however it is similar to FIP above, but "normalizes" the home run component Explanation: This is FIP's less well-known cousin that is only available on The Hardball Times. It does everything that FIP does, except also adjusts the pitcher's raw home run total allowed to reflect the league average number of home runs allowed per fly ball. Pros: Though not as simple to calculate as FIP, it is still simple enough to figure out on your own with a little extra work and data. You just need to look at how many fly balls the pitcher has allowed and that information is readily available. Besides correctly looking at the various controllable skills that FIP takes into account, xFIP rightly adjusts the pitcher's raw home run total to reflect the research done that has proven that pitchers have little to no control over their HR/FB ratios. Cons: This suffers from the same weakness as FIP in that it assumes every pitcher has the same ability to strand base runners. It is also a very simplified formula like FIP, not fully taking batted ball data into account or park effects which could have a material effect on HR/FB ratios. tRA (True Runs Against) Formula: no exact formula is provided by StatCorner, but a full explanation could be found here Explanation: This is one of the newest ERA estimators we have to work with and it was developed by Graham MacAree and could be found at StatCorner. The basic idea of the formula is to assign run and out values to all events under a pitcher's control to derive an expected number of runs allowed and outs generated in a defense and park neutral environment. It is actually on a Runs/9 scale, as opposed to Earned Runs/9, so an adjustment needs to be made to find the equivalent ERA estimate. The formula includes the following components: strikeouts, walks issued, hit by pitches, line drives allowed, ground balls induced, outfield fly balls allowed, infield fly balls allowed, and home runs allowed. Each of these events are assigned specific expected run and out values, thrown into the equation, adjusted for park and base running effects, and cooked up to produce the delicious tRA result. Pros: This metric includes the most components of the ERA estimators we look at in this article. The more controllable skills included, the more accurate the result, assuming the math behind the equation is sound. Cons: If you took a peek at the link above that explains the metric in further detail, you realize this is nearly impossible to calculate on your own. It might be more difficult to embrace if you do not fully understand how it is calculated. In addition, because it is on a R/9 scale and not an ER/9 scale, you will have to multiply it by about 90% to get the equivalent ERA estimate. Next, like the knock against the FIP metric, tRA does not adjust the home runs allowed component to reflect a league average HR/FB ratio. Also, like the two variants of FIP above, this equation does not seem to account for a pitcher's differing ability to strand runners. Last, the formula makes an adjustment for park. If we were in a Major League front office, we might care about a park neutral metric. However, we actually want to know how park effects will influence a pitcher's ERA assuming he remains with the same team. If Jake Peavy was traded to the White Sox, it would have helped to know his park neutral ERA. Since he wasn't traded and remained a Padre, he will continue to benefit from pitching half his games in the pitcher haven known as PETCO Park and a good ERA estimator for fantasy purposes should take this into account. tRA* (Regressed True Runs Against) Formula: no exact formula is provided by StatCorner, but it is a regression version of the above tRA equation Explanation: The regressed version of tRA takes each of the pitcher's component stats and regresses them back to league average; the magnitude of the regression is based on the total number of batters the pitcher has faced. tRA* is not necessarily an ERA estimator that uses a pitcher's current skills like the other estimators described, but it is a system that attempts to estimate a pitcher's true talent level looking solely at current season statistics. Many pitching stats will fluctuate wildly from year to year, especially in-season, so in order to correct for this, every event that the basic tRA takes into account is regressed toward the mean here, with less regression being applied the larger the sample size. Pros: If we were interested in estimating a pitcher's true talent level, this looks like a fantastic stat. This is obviously most helpful for Major League front offices, but it also could come into play if you are trying to evaluate a pitcher's true talent or feel like having a debate with a friend (hopefully he won't tune you out when you start throwing tRAs and BABIPs at him). Cons: From a fantasy perspective, we might not be as interested in regressing a pitcher's component skills toward league average. It would be better to look at the pitcher's historical track record to make the decision as to whether he is likely to continue showing the skills he has thus far during this season. Besides that, the formula suffers from the same weaknesses as the more basic tRA equation above. ERC (Component ERA) Formula: (((H + BB + HBP)×PTB)/(BFP×IP))×9 − 0.56, where PTB = 0.89×(1.255×(H − HR) + 4×HR) + 0.56×(BB + HBP − IBB) Explanation: This is the famous Bill James formula that was added to ESPN.com's stats page in 2004. Unlike any of the previous formulas, ERC uses a pitcher's actual hits allowed and does not include strikeouts. However, it does include walks issued, home runs allowed, hit by pitches, and removes intentional walks. Pros: The formula could be entered into Excel, so you could calculate ERC yourself with relative ease. Cons: It makes the fatal mistake of including actual hits allowed that completely goes against Voros McCracken's famous discovery that pitchers have little to no control over their batting average on balls in play allowed (later this was changed to little control). The formula uses home runs allowed and ignores the research on HR/FB ratio discussed previously. It also makes the common assumption that pitchers all have the same ability to strand runners. Last, there appear to be some random numbers in the formula, such as 0.56, 0.89 and 1.255, of which it is difficult to tell exactly what these constants actually represent. So there you have it, a complete rundown of the free ERA estimators available online. I did leave one out, DIPS, which is also available on ESPN.com. It is very similar to FIP, which is just a simplified version of the metric, and suffers from the same weaknesses. Verdict: No formula is perfect, but without a doubt, my favorite free ERA estimator is xFIP. It is the only one that adjusts the home runs allowed component to reflect the research that pitchers have little control over their HR/FB ratio. However, FIP is by far the most popular, and to be honest, I have no idea why. I will typically ignore FIP completely unless the pitcher's HR/FB ratio is within the range of the league average rate. The StatCorner tRA stats are excellent, but I don't think they have as much application for fantasy leaguers with the park adjustments and regression and it is also tripped up by the lack of a home runs allowed adjustment. With no slight to the Godfather of sabermetrics Bill James, ERC is by far the worst ERA estimator, in my opinion. An equation that still uses actual hits allowed is outdated and simply cannot stand up to more current formulas that remove the impact of a pitcher's defense. Trackback(0)
Comments (5)
  written by Derek Carty, June 22, 2009
Good stuff, Mike. I hadn't read the comments yet, though, so in regard to HQ's xERA explicitly controlling for LOB%, it calculates an expected LOB% using K, BB, and GB. A stat like LIPS or xFIP should do a pretty good job of implicitly accounting for this since it uses K, BB, and GB.
...
written by John Arias, June 22, 2009
What a great article. I really appreciate the research and the summary, it is a must read for all baseball fans.
...
written by Brian Joura, June 18, 2009
This is a great article and one that every fantasy player should read.
If I may suggest a potential follow-up piece I would find it useful to see players that the various systems disagree on - whether due to HR rate or strand rate or whatever else. If FIP shows Player X as a sell candidate but xFIP shows him as a hold, I would like to see how the various systems "grade out" in this result. Over an extended period of time the various rates will gravitate toward the norm for most pitchers, but I don't think you can say with certainty that they will in any particular season.
...
written by Mike Podhorzer , June 18, 2009
Thanks for the comment Jonathan. The reason I brought up strand rate is because Baseball HQ's xERA metric explicitly calculates an expected strand rate component using the pitcher's underlying skills. It is the only formula I know of that acknowledges the varying abilities in stranding runners. I guess it is possible that the formulas discussed in the article do indirectly assume a different strand rate based on the other skills, but I really don't know.
written by Jonathan H, June 18, 2009
Mike -
This is a great summary. I am not sure I understand your comment in the FIP part about FIP (and xFIP, I guess) not controlling for differences in ability to strand runners. By including K's and some sort of HR likelihood, these measures are partly controlling for "strand ability" to the extent that some pitchers can strand more runners via more K's and fewer HRs. K and HR don't capture strand ability perfectly, I'm sure, but I wonder if they do a pretty good job (despite Vasquez).
Write comment
|
TrackBack URI for this entry
Subscribe to this comment's feed
More articles :