Build a Better WAR Metric, Part 8
The worst time to strike out is with a runner on third base and less than 2 outs. In fact, it’s runner on 3B and exactly 1 out. The pitcher knows it, the batter knows it, the fielder knows it, the runner knows it. The fans know it. Everyone is extremely aware that getting the second out changes the entire dynamic of the situation, since now only a positive event can score that runner on third base.
We can even quantify this situation rather precisely. With a runner on third and 1 out, we expect around 0.94 runs to score to the end of the inning. But with 2 outs, it’s all the way down to 0.37 runs. A strikeout in that situation is worth an astounding -0.57 runs.
On the other hand a sacrifice fly clears the bases, gets an extra out (run expectancy down to 0.09 runs), but adds a run to the scoreboard, so the total run value is 1.09 runs, or a net gain of +.15 runs. The sac fly is one of the very few times that trading a base for an out is a net positive.
We have a potential swing of 0.72 runs between a “bad” out and a “good” out. With everyone in the ballpark well aware of the situation, and with outs so common place, it’s clear the pitcher is going to do his best to pitch in a way to increase strikeouts at the cost of perhaps an increase in low impact walks with first base open. And the batter is going to do his best to counteract that, and not try to strike out at the cost of perhaps a tiny bit of power. Everyone is changing their strategy.
But remove the specific context, and an out is an out.
So, how do you want to account for this specific play?
I champion context neutral stats so hard, but I can’t do it here. An at bat with a runner on third and less than two outs is a game within a game. Pitchers pitch dramatically differently, batters swing dramatically differently. Batters will sacrifice a ton of power for putting wood on the ball and getting that run in. Pitchers avoid the zone like crazy. This changes the context of what a plate appearance is, and thus should have different rules for being valued. Give me the context stat.
I wonder if the result would have been different if there hadn’t been suh a leadng question.
I think there should really be two types of WAR, one looking back and one going forward. When doing things like deciding awards, getting some context in there is probably a good thing, but remove what is not sustainable year-to-year out of the projection WAR. It still leads to these types of questions to see how much context should be included, however.
I love that readers were polling 65-95% on all the questions leaning toward less context than more.
And here we find the opposite of that, quite strongly.
This is the equivalent of saber-porn: you know it when you see it.
I voted for context neutral, but isn’t this a false/loaded argument? Wouldn’t any good WAR system treat a sacrifice fly different than a K? It’s not the context that is divergent and important to me, it is the actual result.
I voted for the context-sensitive option on gut feel, but shouldn’t it be easy to find some stats to verify this intuition? If we see stuff like higher K%, higher FB%, and lower GB%, this would be evidence that pitchers and batters really do change their approach.
The pitcher is trying to go for the K and the batter is trying to avoid the K. If those cancel out, you won’t find the evidence.
That said, yes, you would be able to find evidence if you look close enough.
I voted for the context-sensitive option but it does depend on the game situation. If it’s the 9th inning and you’re being by 2+ runs, they’re the same. If it’s the 8th inning and you’re behind by 4 runs, it doesn’t matter much.
I voted for context neutral because I don’t think the pitcher should be punished more for the fly out resulting in a sac fly. But I would support an argument that a pitcher receives more for any strikeout than any fly out (presumably the gain is split between the fielder and the pitcher). So in a way I think the sacrifice fly and strikeout should be treated differently but not because this specific fly out occurred with a runner on 3rd and 1 out.
This isn’t really the same type of context we have had in the past. In other cases, we were trying to take out the leverage of the differing circumstances and give credit for the individual player’s result. In this case, it is the same scenario, same leverage, etc… and so obviously, more credit is given for the SF over the K. Isn’t this already the case? In the aggregate, aren’t K’s generally more negative than a FB out?
If you remove the runner on third and less than 2 outs, there’s virtually no difference between the two.
Doesn’t this come down to intent-measuring?
This same question could be addressed to a shift–wouldn’t the batter adjust his approach to hit it the opposite way? In that situation should a dribble to the left be more heavily weighted than the average dribble to the left? Regardless, most hitters don’t do that, let alone even bunt.
So, if we can’t 100% rely on psychology/prediction then aren’t we inserting a bias in the cases where it fails?
For example, over the long-haul, who does the polled situation favor? An intent-consistent contact hitter or an intent-consistent power hitter?
Are they being rewarded (or punished) based upon their adaptability to situations presented to them or based upon their inherent skill-sets and stubborn approaches?
A dribble to left IS given more credit in the shift situation, it’s a double or triple rather than an out. Even “context neutral” stats include that context.
You’ll never be 100% context neutral, because some contexts change the very things we can measure.
The question to my mind is, does this context change outcomes in a measurable way. Clutch skill appears to exist, but it’s so small an effect that it’s not measurable on an individual batter level, so I have no trouble saying “ignore clutch, ignore leverage”.
But I believe that man on third, 1 out, the context is so important that it does change the outcomes. It’s possible that the very small clutch skill that does seem to exist is actually skill at these specific situations and that we’re having trouble seeing it because it’s buried in the noise from the far more common situations where behavior and intent don’t change.
If you imagine that a manager tells every batter what to do as he walks out of the duggout, 95% or so of the time the instruction would be “don’t get an out”. One out, man on third, the instruction will be “get that man in”.
The context is so significant that it’s actually changed the batter’s presumed intent for the at bat. His entire goal is supposed to be different.
If there is a change in batter behavior with a measurable effect due to the context then that should be taken into account. I think man on third, 1 out, is one of the most obvious cases of that situation.
@Doug Lampert Sorry. By weighted, I meant context weighted. So independent of actual results. So, two dribbles to the left that turn out to be singles. Does the one that we give agency to–the intentional dribble to the left during the shift–deserve more value?
Anyway, my point really isn’t so much about that case, but more just any example where we can attempt to assign intent. I aimed to bring in more examples and question the standards. 90% of batters assume this intent in this situation? 95% do? 99.9%? If a batter came out and said he never attempts for sacrifice flies, and any that come about are pure circumstance, should he still merit value in these situations? I understand that this is in the weeds, but I assumed that was somewhat the point with this amount of polls (there I go assuming intent myself).
If your approach is taken: “If there is a change in batter behavior with a measurable effect due to the context then that should be taken into account. I think man on third, 1 out, is one of the most obvious cases of that situation.” Then I have a further question.
Would this context vary based on other states of the game? Do you think that a lack of leverage should eliminate the consideration. I doubt that 95% of managers would make that instruction in the 1st inning, whereas more may do it in a tied-game late. Likewise with great hitters versus an average hitter. Do Trout, Harper, Cabrera, Goldschmidt, etc. really change their approach? If so, are they still less likely to change than lesser hitters. Does Altuve even need to change his approach to avoid a strike out? Should Yelich be awarded more value for putting the ball in the air?
There are other considerations that are even more obvious in terms of the intent of the batter, pitcher and fielders shifting so dramatically.
It’s the bottom of the 9th, the game is tied, and there’s a runner on third and 1 out. All the outfielders are conceding the medium to deep flyball because they know the only chance at getting the runner is to get a flyout that is shallow-to-medium in depth.
And if you look at the stats in these do-or-die situations, you will see numbers all over the place, enough that I would exclude them from the stats, since they don’t match up well. It’s not akin to pulling your goalie with a minute to go in hockey, but it’s that kind of thing. Or in football, if you have to move 80 yards in under 90 seconds, the rate of interceptions must double or triple, simply because there’s no downside to these situations. In basketball, the fouling to end the game is another that is completely divorced from the normal routine of play.
So, the situation I presented is ALMOST like trying to come up with a “game within a game”. It’s probably as close as there is. So the question is designed to see if the readers think of it that way, or they still don’t think of context as being necessary.
If you acknowledge that there is a certifiable “game within a game” then do you go on to see that situation as discreet or a piece of the spectrum?
How granular can this go? If one type of PA can be singled out, can the others? Can all PAs be compared to their peer PAs by the 27 outs in an average game? The eight base states? Game scores? Shifts, such as the do-or-die shift above or the the normal pull-hitter version? Position players pitching in blow-outs? Combinations of all the above?
Expected intentions, approaches, and average outcomes can be outlined for all of these states. 25 outs, runner on third, tie game, do-or-die shift, closer pitching is almost completely discounted. 22 outs, runner on third, tie game, no shift, set-up man pitching would warrant more value to fly-balls. Same scenario but with 1 out in the game would warrant slightly less value to fly-balls relative to more outs. 26 outs, runner on second, tie game, no shift would award less value to a walk than if there were only 5 outs in the game, everything else equal. Et cetera.
If there is going to be a cut-off for the sake of material impact, then is it the best approach to base it off public perception? As Daniel Kahneman has found, even experts in related fields lack an intuitive understanding of statistics. Or is this just the best available resource? Can the influence of game states on approach and intention be accurately measured?
My opinion is that all batters and pitchers change their approach based on the plate count (3-0 approach is materially different from an 0-2 count). Even the umpire responds to that.
I don’t see how we can expect a batter/pitcher to change their approach with that, but NOT change their approach if first base is open or not. Since both sides know the walk has less impact with first base open, both sides are going to devalue the effect of a walk, and change their approach. They’d have to.
The fielders physically move with a runner on 1st and less than 2 outs, in order to go after the DP.
We can presume they change their approach with runner on third and less than 2 outs.
Other states are more nuanced, but each of the 24 base out states offers a different approach.
Morgan Ensberg for example reduced the 24 base-out states into 8 approaches (for him):
http://www.insidethebook.com/ee/index.php/site/comments/morgan_ensbergs_philosophy_of_hitting/
If you ask other hitters, they’d have their own philosophy. And certainly for the pitchers.
Certainly there are changes to approach for all those reasons you stated–even common situations like end-of-season, on a bad team, and near a career-high in X stat or entering FA, etc. But there comes a point where you have to put your foot down, otherwise the system of measurement will become as big as that which it is measuring. A reproduction.
I guess what I’m struggling with is not where to put that line but why.
In that link Ensberg put “drive the ball” foremost in every situation. I would be curious to see if that pattern extended to all 27 outs in a game. Regardless, his ranking would appear to run contrary to this poll result above. “Flyball swing” was his third approach. Why reward an outcome that is a tertiary intention? (Then again, he may be an outlier. Other players may have it their first priority.)
Has there been research on the 24 base-out states’ distribution of outcomes relative to all outcomes regardless of base-out state? If so, would this provide a better proxy for intent?
The issue you will have is both sides know what they want to do and what to avoid. It has a tendency to cancel out.
If first base open makes a walk lower impact, the batter is less motivated to take a pitch, while the pitcher is more motivated to throw it at the margins.
Basically, whatever one side wants, the other side is trying to avoid, and vice versa.
@Tangotiger Is that reflected in the data? Are there no distribution changes across the base-out states?
So, am I understanding it correctly that the attempt is to award more value to an outcome than it receives in a neutral state based on it coming in a situation where intuition dictates that the outcome is more difficult to attain due to opponent actions? And there is no way to validate that intuition beyond common sense or observation?
You could validate it, but not as easily.
Intuition should prevail nonetheless here.
@Tangotiger It still feels a little unsatisfying to me to say that a statistic measures X as Y in all cases except for one case Z.
But thanks for sticking it out with me, I appreciate the area as more grey now (which is good in my mind).
I’m not justifying it. I’m trying to explain how the readers are responding.