Proposed Baseball Statistic
Posted by
tgirsch
I’ve said before, including in recent threads here, that I think WHIP (walks and hits per inning pitched) is probably the best stat for evaluating a pitcher’s effectiveness — better than wins, and better than ERA. Better, even, than BAA. But even this statistic is flawed, because a pitcher who gives up just three walks in a 9 inning complete game will have the same WHIP as a pitcher who “only” gives up three home runs in a nine inning complete game.
So I’d like to suggest that what we should be tracking is TBIP - Total Bases per Inning Pitched. It’s not an acronym like “WHIP” is, so there’s no easy way to pronounce it, but I think it would be more meaningful. Both pitchers in my example would have a WHIP of 0.333, but the first pitcher (obviously the better performance) would have a TBIP of 0.333, while the latter pitcher would have a TBIP four times as high — 1.333.
Discuss. Otherwise, open thread.
would you include stolen bases, pass balls, wild pitches, sacrifices, etc. or just what happened with the player at the plate. And if you just count the player at the plate, how do you score errors or fielder’s choices??
Comment 8/6/2008
I would include any advancement that is the fault of the pitcher. So a wild pitch would count against his TBIP, as would a balk. But fielding errors would not. I’ll have to think about the fielder’s choice thing, but instinctively, I’d say not to include those (or sacrifices). Only situations where a runner advances, nobody’s out on the play, and the pitcher is at fault. I’d say that a stolen base is more the responsibility of the catcher than the pitcher.
Comment 8/6/2008
In essence, you’re proposing for something that is basically like OPS against (except OPS double-counts hits). Perhaps even better, for comparisons spanning eras would be OPS+ against. I remember seeing a figure in that format that basically showed the league collectively hitting at the level of an average NL pitcher against Pedro Martinez in 1999.
Offensive metrics have been damn near perfected (at least in comparison to fielding or pitching metrics). You could take any of the advanced rate metrics, like Eqa, or anything based on linear weights and flip it for a pitcher to represent it as (stat x)against, if you want.
Comment 8/7/2008
Those wouldn’t account for things like balks and wild pitches, but I suspect those would amount to little more than statistical noise, anyway.
Comment 8/7/2008
No, but the benefit of using linear weights to determine the offensive value of events against the pitcher would probably outweigh the smaller stuff it misses. The idea simple slugging, or stats based on total bases are linear, but the way of escalating events isn’t. A double isn’t worth twice as much as a single, etc.
Here’s a quick rendering of an example of a table of run values based on linear weights; the last column is supposed to be negative. This table is courtesy of stats guru, Tom Tango. The bottom row represents the average value of each event.
MOB 1b 2b 3b hr bb k out
— 0.29 0.49 0.68 1.00 0.29 -0.20 0.20
x– 0.49 0.97 1.36 1.74 0.43 -0.32 0.36
-x- 0.72 1.00 1.16 1.60 0.23 -0.39 0.34
–x 0.72 0.86 1.00 1.51 0.21 -0.48 0.29
xx- 0.93 1.54 1.94 2.38 0.56 -0.52 0.48
x-x 0.88 0.93 1.77 2.22 0.38 -0.61 0.46
-xx 1.17 1.46 1.62 2.07 0.23 -0.70 0.56
xxx 1.38 2.00 2.40 2.86 1.00 -0.82 0.68
ROB 0.73 1.14 1.49 1.92 0.42 -0.44 0.42
avg 0.49 0.79 1.06 1.42 0.35 -0.31 0.30
Comment 8/7/2008
Sorry, I guess should have figured it formatting would disappear once I posted.
Comment 8/7/2008
And, with that, you have officially lost me.
Comment 8/7/2008
The idea of linear weights to determine the average “run value” of an offensive event.
So, by running thousands of games through whatever program they use, you can find the average worth of different offensive events. Based on the same data, you can construct a run expectancy matrix to show how many runs a team can be expected to score from any combination of men on base and outs (”base-out states” - there’s 24 of them, like declensions in Latin, I think.) Anyway, an event’s linear weight is the amount by which the event changes to overall run projection from before the event to afterwards. Once you get passed the idea of the heavy math, it’s actually rather intuitive.
My point is that there are offensive metrics that use linear weights as their basis, and thus those metrics can be reversed and used as “against” stats for pitchers.
Again, a homerun is not the same value as four singles, as Total Bases or Slugging PCT treat it. Any forumula’s result is only going to be accurate as its inputs.
A lot of this stuff is sort of jumping to senior seminar sabermetrics (perhaps not even, there’s stuff out there beyond my comprehension). But, I think you would enjoy some of the baseball research along this line. I don’t fully subscribe to all of it, but it’s certainly worth seeing what’s out there. There’s really a lot of fascinating stuff going on beyond the Luddite MSM and their irrational jonesin’ for Wins and RBI.
Comment 8/7/2008
The most popular and useful statistics are ones that are (A) easy to understand and (B) meaningful. If, say, ERA fails to meet the latter criterion, the stat you’re talking about fails to meet the former criterion. The reason things like OPS and WHIP are so popular is that they’re pretty good metrics while also being easier to understand. The stats you’re suggesting might be the Greatest Stats Ever, but they’re wholly inaccessible to anyone with anything less than a Master’s degree in Statistics.
Comment 8/7/2008
The concept of “run value” seems intuitive enough. But the numbers themselves leave me scratching my head. Why on Earth would a HR with the bases loaded be worth only 2.86 runs? Seems that for this table to be taken seriously, “run value” for that HR column needs to be whole numbers equaling the actual number of runs scored!
Bases empty, 1.00, sure. One man on, oughta be 2.00; two on, 3.00; etc. A triple with the bases loaded, 3.00. Must be something I’m not seeing here, too?
Comment 8/28/2008