Does OPS+ matter?

[Josquin]

I was curious how well the OPS+ statistic reflects a player's offensive value in terms of run production. So I compared my team's OPS+ to my own formula for measuring runs produced. Below are the results.

The first column is the player's run production (per 550 AB) according to my formula. The second column is the player's OPS+.

In general, a higher OPS+ does indeed correlate with more runs produced, but it doesn't tell the whole story. For example, there are two players with almost identical OPS+ who are nevertheless 15 runs apart in terms of production.

Bottom line: OPS+ by itself doesn't tell the whole story.

The highlighted number (726) at the bottom is the predicted number of runs scored by my team based on the run production formula. My team's actual number of runs scored was 722, so the two agree very well.

[Orcin]

Since you haven't given me your formula, I will have to be content with the tools in the game.

I prefer wRC+ over OPS+, but they are very close.

P.S. Congrats on clinching a playoff spot already.

[Josquin]

Originally Posted by Orcin

Since you haven't given me your formula, I will have to be content with the tools in the game.

I prefer wRC+ over OPS+, but they are very close.

P.S. Congrats on clinching a playoff spot already.

You are having one heck of a season yourself. That George Brett is having a monster year.

[CrazyWR]

wouldn't the problem here be that OBP is worth more than Slugging percentage?

[X3NEIZE]

IMO OPS+ is the fastest indicator of how well that player is performing offensively... it's my go-to KPI day-to-day.

My rule of thumb is, #2, #3, #4 and #5 hitters need to be >=140 OPS+.

[Josquin]

Originally Posted by CrazyWR

wouldn't the problem here be that OBP is worth more than Slugging percentage?

Something like that... I've noticed that the players most short-changed by OPS+ are the ones that draw a lot of walks.

[chazzycat]

Originally Posted by Josquin

Something like that... I've noticed that the players most short-changed by OPS+ are the ones that draw a lot of walks.

yeah...OPS is a bad stat for this precise reason.

It's just two other numbers (OBP and SLG) added together. That assumes OBP and SLG are worth the same amount - but they are not. OBP is worth more by a significant amount.

OPS+ is slightly better (normalized for park factors & league average) but since it's based on a flawed statistic (OPS) it's really not that great.

[Josquin]

Originally Posted by chazzycat

yeah...OPS is a bad stat for this precise reason.

It's just two other numbers (OBP and SLG) added together. That assumes OBP and SLG are worth the same amount - but they are not. OBP is worth more by a significant amount.

OPS+ is slightly better (normalized for park factors & league average) but since it's based on a flawed statistic (OPS) it's really not that great.

Looking at wRC+ seems to be more of the same. Usually tracks fairly closely with OPS+ and undervalues walks.

[chazzycat]

Hmmm...I wonder how the game calculates wRC+ then. According to FanGraphs glossary, wRC+ uses wOBA instead of OBP/SLG so it should be much better.

[Josquin]

Originally Posted by chazzycat

Hmmm...I wonder how the game calculates wRC+ then. According to FanGraphs glossary, wRC+ uses wOBA instead of OBP/SLG so it should be much better.

Looking at the way wOBA is calculated, it should indeed be much better. The run values are a bit off from what I'm getting for the Perfect League environment, and it doesn't take into account baserunning or the way outs are made (strike outs vs. balls in play), but the overall approach should be more accurate than OPS.

[SpacePope]

Is the goal of your formula to predict run production?

In both Google Docs and Excel there's a spreadsheet function, =CORREL, that will give you the Pearson's correlation coefficient. I think what you'd want to do to see if your formula is working better than OPS+ (or wRC+) at predicting run production is to compare the correlation you get between your formula and runs produced against the correlation you get between OPS+ (or wRC+) and runs produced. (Your formula has a .95 Pearson's correlation, where 1.0 and -1.0 are perfect correlation and between .1 and -.1 are what you might see with random numbers. EDIT: you can actually see much stronger correlations over just 14 data points of random data)

If your goal is just to predict run production, however, can you really do better than just dividing runs produced by plate appearances?

Depending on what you're using it for, it might also matter how you count runs produced. If you use RBI + R - HR, you double count non-home runs when totaling for your whole team.

[Josquin]

Originally Posted by SpacePope

Is the goal of your formula to predict run production?

Yes.

If your goal is just to predict run production, however, can you really do better than just dividing runs produced by plate appearances?

The trick is figuring out what "runs produced" means, based the player's statistics for the season.

Depending on what you're using it for, it might also matter how you count runs produced. If you use RBI + R - HR, you double count non-home runs when totaling for your whole team.

RBI and R are not too useful as ingredients, since they depend too much on the quality of the team around the player, i.e. they don't isolate the player's individual value enough. There are much better ways of going about this by assigning run values to the outcomes that the player produces (which is what wOBA does, but it leaves out certain important aspects of a player's overall offensive value).

[chazzycat]

In any rate, it's always been a fools errand to try and boil run production down to just 1 metric in my opinion. Hitting for average, hitting for power, drawing walks, & running the bases are just different skills which are largely independent of each other.

By trying to combine everything together, by its nature, the overall picture gets muddied. Consider for example, the importance of accounting for BABIP randomness. A good (lucky) BABIP can inflate ALL of these combined metrics. Even the "better" metric wOBA still assigns a flat run value per hit - not considering whether it was luck in the first place that the hit fell in. Simply judging whether a batting average is "lucky" or not requires a good deal of analysis on its own (batted ball data, StatCast etc.). So you generally want to analyze contact skill separately, in a vacuum, to prevent that randomness from bleeding out and polluting all your other data.

It's the same logic that dictates why you shouldn't use the OVR rating to pick your players. You should be looking at an array of different statistics that describe individual skills, never just a single one.

[pappyzan]

Estimated Runs Produced maybe? http://www.baseballthinkfactory.org/...s/jameserp.htm

TLDR: (2 x (TB + BB + HP) + H + SB - (.605 x (AB + CS + GIDP - H))) x .16 = Runs

[Josquin]

Originally Posted by chazzycat

In any rate, it's always been a fools errand to try and boil run production down to just 1 metric in my opinion. Hitting for average, hitting for power, drawing walks, & running the bases are just different skills which are largely independent of each other.

By trying to combine everything together, by its nature, the overall picture gets muddied. Consider for example, the importance of accounting for BABIP randomness. A good (lucky) BABIP can inflate ALL of these combined metrics. Even the "better" metric wOBA still assigns a flat run value per hit - not considering whether it was luck in the first place that the hit fell in. Simply judging whether a batting average is "lucky" or not requires a good deal of analysis on its own (batted ball data, StatCast etc.). So you generally want to analyze contact skill separately, in a vacuum, to prevent that randomness from bleeding out and polluting all your other data.

That may be true in real baseball, but PT is just a random number generator driven by the player ratings as inputs. Taking a long-term view, the player ratings will predict all aspects of the player's average run production. There will be random variation from season to season, but over the course of many seasons, clear patterns will emerge. And since the players don't get injured, don't age, and don't lose their skills, we have the luxury of ignoring the short-term randomness in favor of long-term averages.

[SpacePope]

Originally Posted by Josquin

RBI and R are not too useful as ingredients, since they depend too much on the quality of the team around the player, i.e. they don't isolate the player's individual value enough. There are much better ways of going about this by assigning run values to the outcomes that the player produces (which is what wOBA does, but it leaves out certain important aspects of a player's overall offensive value).

I agree (although I'd probably go with wRC+). I was confused cause you were using run produced to judge whether your formula was working. There is a strong correlation, but that doesn't mean the formula is doing what you want (predicting overall hitting performance) and could just mean your drawing on data points that necessarily map run production closely (and that you might be then over and under valuing the same things that those stats do in rating overall hitting performance).

[Josquin]

Originally Posted by pappyzan

Estimated Runs Produced maybe? http://www.baseballthinkfactory.org/...s/jameserp.htm

TLDR: (2 x (TB + BB + HP) + H + SB - (.605 x (AB + CS + GIDP - H))) x .16 = Runs

Definitely on the right track.

[chazzycat]

Originally Posted by Josquin

That may be true in real baseball, but PT is just a random number generator driven by the player ratings as inputs. Taking a long-term view, the player ratings will predict all aspects of the player's average run production. There will be random variation from season to season, but over the course of many seasons, clear patterns will emerge. And since the players don't get injured, don't age, and don't lose their skills, we have the luxury of ignoring the short-term randomness in favor of long-term averages.

That's true. But as long as the game lets me modify my park factors, hence creating an incentive to prioritize certain skills over others, I'm going to use individual statistics that analyze those skills separately from each other.

[Josquin]

Originally Posted by SpacePope

I agree (although I'd probably go with wRC+). I was confused cause you were using run produced to judge whether your formula was working. There is a strong correlation, but that doesn't mean the formula is doing what you want (predicting overall hitting performance) and could just mean your drawing on data points that necessarily map run production closely (and that you might be then over and under valuing the same things that those stats do in rating overall hitting performance).

Here's a very simplified example. Suppose the only way to score runs is via solo HR. Then if I tell you how many HR a team hit, you'd be able to immediately tell me how many runs that team scored, right? And then it would also be simple to calculate each player's contribution to the team's run total: just count the number of HR they each hit.

If we could extrapolate this reasoning to all the different outcomes that a player's plate appearance can produce, and somehow deal with the messy details, then we'd have a pretty good way of measuring each player's contribution to the team's run total.

[Orcin]

Originally Posted by Josquin

Here's a very simplified example. Suppose the only way to score runs is via solo HR. Then if I tell you how many HR a team hit, you'd be able to immediately tell me how many runs that team scored, right? And then it would also be simple to calculate each player's contribution to the team's run total: just count the number of HR they each hit.

If we could extrapolate this reasoning to all the different outcomes that a player's plate appearance can produce, and somehow deal with the messy details, then we'd have a pretty good way of measuring each player's contribution to the team's run total.

Isn't this exactly what Runs Created does?