OBP....and its importance

[NJOriolesFan]

Team's have won games while being no-hit so it's not about hits vs. walks; it's simply about outs.

Most sports are timed events. The game is over when the clock runs out. Even in bowling the game is over after you've completed 10 frames. Baseball's "time-keeping" system is simply outs. You want to not make outs plain and simply. Not making outs is the same as getting on base.

SG's original premise was getting on base is so important. Getting a walk and getting a hit are still both getting on-base.

If a team had a 1.000 OBP for an entire game (impossible) it would score the same number of runs as a team that got a 5.000 SLG percentage. It would be infinite in both cases because no outs are made. The scoring would just keep going.

Power is important and so is plate discipline/batting eye and so is making contact. The skills that "create" the stats are all important.

The only problem I have with OPS is that it counts batting average twice since batting average is both in OBP and in SLG. Quite frankly, and although I never correlated the data, I think the more important stat would be OBP+SLG-AVG, so you don't count average twice. In other words, you are adding walk rate, contact ability and power together and counting each skill as equal value.

Building a team, I would always take the player that makes outs at a lower frequency. ALWAYS, ALWAYS, ALWAYS....did I say ALWAYS. However, in certain situations you would rather have your power guy up. Just like I hate bunting and don't think it is a smart thing to do typically. However, there are SOME instances where it is a good thing.

You can't pick a specific isolated sample and say it is better to have the hit than the walk. They are the same in that neither produce an out and that is how you extend an inning.

[Frobby]

Building a team, I would always take the player that makes outs at a lower frequency. ALWAYS, ALWAYS, ALWAYS....did I say ALWAYS.

That is way too extreme. So you'd take .300/.375/.375 over .300/.340/.550? No way.

[PaulFolk]

If a team had a 1.000 OBP for an entire game (impossible) it would score the same number of runs as a team that got a 5.000 SLG percentage. It would be infinite in both cases because no outs are made. The scoring would just keep going.

Technically, it's not impossible. Every batter could get on base, but then they could keep getting picked off or caught stealing, or thrown out on the basepaths.

Granted, it would have to be the worst baserunning team of all time, and Tippy Martinez would have to be prominently involved, but there ya go.

[Lucky Jim]

That is way too extreme. So you'd take .300/.375/.375 over .300/.340/.550? No way.

I assumed he meant w/ equal OPS.

[KAZ97]

First, I'd like to say that I've appreciated reading this tread, and the discussion of causality vs correlation.

That said, it seems prudent to be precise in differentiating the talent or ability and the estimation or measurement of that talent. For example, we can talk about the talent of getting a hit, of getting on base, of hitting for power but that is not the same as BA, OBP and SLG.

To see the difference, ask what precisely is the talent estimated by batting average as defined by hits per at bat instead of hits per plate appearance. Personally I feel that as a measure OBP is far superior to BA and SLG because it accounts for all the player's plate appearances and does not (arbitrarily?) ignore walks, sac flys etc.

Also note that these correlations being reported are not consistently on the same unit. Each plate appearance can be represented by an OBP of either 1 or 0. Does that mean that unit of OBP is responsible for the runs scored during that play?

If I were to assess the correlation between OBP and runs scored (and given the current weather conditions I might have time to this weekend) I would base it on the smallest unit of measure possible. For me that would be innings. Each inning starts with zero runs scored and in each inning we can measure the probability of the batters reaching base (or not creating an out). I would then estimate the correlation between OBP and runs scored in each inning.

For another time: correlation DOES mean causation. A few more "winter warmers" and I could probably be baited into that discussion.

[backwardsk]

Team's have won games while being no-hit so it's not about hits vs. walks; it's simply about outs.

Most sports are timed events. The game is over when the clock runs out. Even in bowling the game is over after you've completed 10 frames. Baseball's "time-keeping" system is simply outs. You want to not make outs plain and simply. Not making outs is the same as getting on base.

SG's original premise was getting on base is so important. Getting a walk and getting a hit are still both getting on-base.

If a team had a 1.000 OBP for an entire game (impossible) it would score the same number of runs as a team that got a 5.000 SLG percentage. It would be infinite in both cases because no outs are made. The scoring would just keep going.

Power is important and so is plate discipline/batting eye and so is making contact. The skills that "create" the stats are all important.

The only problem I have with OPS is that it counts batting average twice since batting average is both in OBP and in SLG. Quite frankly, and although I never correlated the data, I think the more important stat would be OBP+SLG-AVG, so you don't count average twice. In other words, you are adding walk rate, contact ability and power together and counting each skill as equal value.

Building a team, I would always take the player that makes outs at a lower frequency. ALWAYS, ALWAYS, ALWAYS....did I say ALWAYS. However, in certain situations you would rather have your power guy up. Just like I hate bunting and don't think it is a smart thing to do typically. However, there are SOME instances where it is a good thing.

You can't pick a specific isolated sample and say it is better to have the hit than the walk. They are the same in that neither produce an out and that is how you extend an inning.

This is where I get lost in the conversation.

If you have two outs and a runner on second, why is a hit not better than a walk in that situation? Yes, both situations extend an inning, but even if the best hitter in the league comes up next, there's more than a 50% chance that he won't extend the inning. The hit knocks in the run (generally speaking), but the walk does not. It's more likely that the next guy will end the inning as opposed to extending it regardless of how good the next hitter is.

[now]

Thanks, NJOriolesFan - for bringing the discussion back to square one. This is where we arrive at the post-modern moment in baseball statistics when we realize that all the relative values we give to individual batting metrics, however logical, are subjective.

Sure, you can make a case that “A walk’s as good as a hit” and even prove it with logic and math, as you have. Outs, you say, are all-important, and so OBP (for or against) would be the best single measure of value. The limitation, however, is that it is only a single value, which doesn’t account for different weights of other important skills such as power, strikeouts, groundball rate, etc.

Another equally logical place to assign value is Runs (created or saved), because wins by definition ALWAYS depend on scoring more runs than an opponent. With this logic, the best way to accurately measure offensive value is to somehow account for how many runs a player produces. Yet for individual batters, the simple Runs scored measure is inadequate, as is the RBI measure, because these are team-related events. Homeruns are the only batting event that accounts for runs directly and completely credited to the batter (unless you add in considerations such as lineup protection, game situations such as bases loaded, etc.).

Similarly, on the pitching side, we like to point to strikeouts as the only event in the player’s control. In this case we cannot point directly to a “run saved” - but we can point to “creating an out,” which brings us back to Outs (created or saved) as the primary logic toward a win.

So, looking at individual performance alone, it would appear that, on the batting side, either homers or OBP would have the most value, depending on whether your objective is to “produce runs” or “save outs.” Or maybe you go with “low strikeout rate for batters,” since this event is also independent of team context. For pitchers the most value would attach to strikeouts or opponent-OBP, with both measures all about creating outs; or perhaps “low homerun rate,” if we aim to save runs directly.

When it comes to the crunch, however, do we sign homerun hitters or high-OBP guys? Do we draft the fireballer who might also be prone to the long ball, or do we stick with the groundball machine who gives up a lot of singles or misses a lot low in the zone? It gets messy fast.

If you follow the Outs logic, then OBP is clearly the primary word in player value. Yet it is not the end of the story, because player A and player B may produce outs at the same rate but offer different run-value with their respective power numbers. If you follow the Runs logic, you have homers (hit or allowed) and nothing else to go on, unless you bring the whole discussion up to a team level.

Enter Bill James and company in the sabermetric movement, who tinkered with complex calculations of on-field events in an attempt to gauge player value. Measures such as “Runs Created” and OPS were pieced together by logic and math and then tested against (correlated with) directly measurable team performance: team runs scored. Other sabermetric equations account for numbers of outs saved or created - but again, in the complex environment of many offensive and defensive variables carrying different weights relative to what we can measure directly: runs scored.

OPS is not a direct measure of individual performance, but an approximation arrived at indirectly from the team context. As Runs Created (RC) is found to be correlated with team runs, OPS is found to be correlated with RC. Therein lies the value of these equations. They seek to approximate - to correlate with - the measure of actual runs scored. This works out remarkably well in the team arena, and so by inference we try to apply them as measures of individual contributions to that team performance. What works neatly as a correlation on the team level doesn’t necessarily translate purely on an individual basis, as the earlier example of similar OPS producing different RC illustrates.

So what’s the point of this whole exercise? The quest is to look for concise and accurate measurement of player value. Where we can’t measure it directly with single events, we attempt to combine different measures into more complex equations, and to extrapolate from team results back to individual skill sets. Still we have no single ironclad measure that accounts directly and entirely for player value. So we pick and choose from the available tools - whether by logic, correlation, convenience, or our own preference.

[RShack]

First, I'd like to say that I've appreciated reading this tread, and the discussion of causality vs correlation.

That said, it seems prudent to be precise in differentiating the talent or ability and the estimation or measurement of that talent. For example, we can talk about the talent of getting a hit, of getting on base, of hitting for power but that is not the same as BA, OBP and SLG.

To see the difference, ask what precisely is the talent estimated by batting average as defined by hits per at bat instead of hits per plate appearance. Personally I feel that as a measure OBP is far superior to BA and SLG because it accounts for all the player's plate appearances and does not (arbitrarily?) ignore walks, sac flys etc.

Also note that these correlations being reported are not consistently on the same unit. Each plate appearance can be represented by an OBP of either 1 or 0. Does that mean that unit of OBP is responsible for the runs scored during that play?

If I were to assess the correlation between OBP and runs scored (and given the current weather conditions I might have time to this weekend) I would base it on the smallest unit of measure possible. For me that would be innings. Each inning starts with zero runs scored and in each inning we can measure the probability of the batters reaching base (or not creating an out). I would then estimate the correlation between OBP and runs scored in each inning.

For another time: correlation DOES mean causation. A few more "winter warmers" and I could probably be baited into that discussion.

Let's slow down just a tad here...

The correlations reported in the article that started all this confusion have nothing to do with talent or ability.

They also have nothing to do with how much OBP or SLG contribute to runs, either in isolation from each another or relative to each other.

Nothing. That is not what the correlations reported in the article are about.

The only thing the correlations reported for OBP and SLG are about is the degree to which team-OBP and team-SLG can be used to track team-runs. That's all they are about, nothing else.

• The fact that team-OBP and team-SLG have virtually the same correlation means that they are equally good/not-good at tracking team-runs.

  .

• If the correlation was 1.0, then they track team-runs perfectly, albeit by some constant factor. This would mean that you could take team-OBP, multiply it by some Mystery-Factor-X, and get the number of team-runs. You could do the same thing by taking team-SLG and multiplying it by a Mystery-Factor-Y.

  .

• Because the correlation is less than 1.0, that tells you that they're not perfect for tracking team-runs. So, how not-perfect are they? Both team-OBP and team-SLG have a correlation with team-runs of ~0.91. One informal rule-of-thumb says you square that, get ~0.83, and give them a grade based on that, like a "low-B," as a predictor. When you combine them into OPS, you get a better predictor: take the OPS correlation of 0.955, square it, and get ~0.91, or a grade of "barely-a-low-A" as a predictor. While "low-B" and "low-A" sound good, both still contain lots of room for error and uncertainty. For example, if you had to have surgery, and your chance of survival was 0.83, which is a bit less than 5 out of 6, would you feel good about that? No, you would not: you'd want it way up in the very-very-high-90's.

  .

• How much you can use them as predictors says nothing about what Mystery Factors you need to use to quantify how much they each contribute to team-runs. How much they contribute relative to each other is what the 1.8 factor is about that doesn't show up anywhere in the correlations. The correlations reported in that article do not know or care about factors like that. That's not their job. Their job is to see past any and all Mystery Factors and tell us how real and trustworthy the relationship is between 2 variables regardless of whatever the appropriate Mystery Factors might be.

Everybody needs to just forget the idea that the correlations reported in that article have anything to do with talent, ability, or anything else about individual hitters. They are about summary team data and nothing else.

Everybody also needs to forget the idea that they have anything to do with how much OBP contributes to runs compared to SLG. They don't. You can look at those correlations This-Way and That-Way and the Other-Way, but they still don't, no matter how you look at them.

[Frobby]

There is a logical paradox in this thread that I want to point out.

If you had a choice between a team full of guys that alternates outs and homers, and a team full of guys who walk 100% of the time, you'd take the team full of walkers. Why? Because they'd never make an out, and score an infinite number of runs.

But now assume you have a normal baseball team, with players in the normal range, and you are adding a single player. Player #1 alternates between outs and homers. Player #2 walks every time up. Here, you'd want Player # 1. Why? Because you know his alternating homers are going to result in at least one run being scored, and maybe more, while a lot of the other player's walks aren't going to lead to runs. According to the linear weights described earlier in this thread, Player #1 would be producing about .55 runs per plate appearance, while Player #2 would be producing .35 runs per plate appearance.

So, the hypothetical about the team of players that always walk is kind of misleading in the real world. There's a tradeoff in real situations that has to be recognized.

[now]

There is a logical paradox in this thread that I want to point out.

If you had a choice between a team full of guys that alternates outs and homers, and a team full of guys who walk 100% of the time, you'd take the team full of walkers. Why? Because they'd never make an out, and score an infinite number of runs.

But now assume you have a normal baseball team, with players in the normal range, and you are adding a single player. Player #1 alternates between outs and homers. Player #2 walks every time up. Here, you'd want Player # 1. Why? Because you know his alternating homers are going to result in at least one run being scored, and maybe more, while a lot of the other player's walks aren't going to lead to runs. According to the linear weights described earlier in this thread, Player #1 would be producing about .55 runs per plate appearance, while Player #2 would be producing .35 runs per plate appearance.

So, the hypothetical about the team of players that always walk is kind of misleading in the real world. There's a tradeoff in real situations that has to be recognized.

This is a good point. The player who walks a lot is doing a great job individually for each at-bat, but he's not necessarily producing runs at the same rate as someone with power who is driving in runners and getting into scoring position efficiently. Furthermore, that high-BB guy, though he is gaining one base, is also largely passing on to the next hitter the responsibility to "save an out." When that probability in real life is almost always less than 50%, most of the time "saving an out" really means "delaying an out." Again the team context must be considered, and here runs take on more significance relative to outs.

[KAZ97]

How much you can use them as predictors says nothing about what Mystery Factors you need to use to quantify how much they each contribute to team-runs. How much they contribute relative to each other is what the 1.8 factor is about that doesn't show up anywhere in the correlations. The correlations reported in that article do not know or care about factors like that. That's not their job. Their job is to see past any and all Mystery Factors and tell us how real and trustworthy the relationship is between 2 variables regardless of whatever the appropriate Mystery Factors might be.

Right. I was speaking more generally about your "mystery factor" and not a correlation coefficient, sorry I wasn't precise. Maybe if I have time, this weekend I'll estimate the "mystery factor" for the Orioles. That "mystery factor" is dependent upon run environment and it would be interesting comparing how many extra (or fewer) runs the Orioles get out of some increment OBP as compared to other teams.