Article: Geeking Out: OPS Minus Batting Average

[ThePuck]

And it's even worse than that: simply scaling SLG down by 4X doesn't help because the typical value is not 4X the typical OPS. Combining the two values is a doomed exercise.

 

OPS is just a quick-and-dirty, fun, number.

 

Nothing wrong with quick and dirty...oh wait...

[ashbury]

A walk in a low OBP environment isn't as valuable as in a high OBP environment (although still valuable). Likewise a home run in a low OBP enviroment is more valuable on average, than one in a high OBP environment.

 

Is this correct? It seems opposite of my intuition: that if you have a team full of table setters, one good slugger will stand to make a huge difference, and if you have a team of low-average sluggers, getting a good OBP guy at the top of the lineup can work wonders.

 

I don't usually ask "source, please", but if you have a link to something that explains what you described, I'd like to read it.

[ashbury]

Nothing wrong with quick and dirty...oh wait...

 

This isn't a game thread...

[ThePuck]

This isn't a game thread...

 

I'm going to get a warning for that, aren't I? :-)

[Willihammer]

Is this correct? It seems opposite of my intuition: that if you have a team full of table setters, one good slugger will stand to make a huge difference, and if you have a team of low-average sluggers, getting a good OBP guy at the top of the lineup can work wonders.

 

I don't usually ask "source, please", but if you have a link to something that explains what you described, I'd like to read it.

 

Adjusting Linear Weights for Extreme Environments | FanGraphs Baseball

 

You're less dependant on extra base hits in a high OBP environment because you'll be able to sustain rallies longer.

[Kwak]

I want to thank Brock and Jay fortheir links on correlations. The discussion of OBP/SLG/OPS s covered by those two articles--OBP showed the highest corelation over the range selected.

 

There was some math in my education as well, (ome chemistry too!). The ranges selected (based on actual team stats) is important. All of the posted analyses are working overtime to make a linear function of the variable(s) in the analysis. They do appear to agree with linear in that narrow range--but the funcions must actually be non-linear over their entire range (OBP 0 to 1, and SLG 0 to 4). To me, it seems that there must be a factor of:

 

K/ (1-OBP) exp (X)

 

because if a team had a perfect OBP (1.0) they would score an infinite number of runs.

 

I was reminded of the equations used for process control (Proportional, Integral and Derivative) that involve some advanced math. The Proportional eq'n includes a factor o the natural log e and loks something like: (sorry Ihave this dumb keyboard, no sub or superscripts, Greek letters)

 

 

{1 - e exp((-K1/K2 + 1)/K3}

 

Then there are the integration and derivative of the key variables (not shown)

but they do obtain the desired results.

 

OK, enough math. My point is if one is trying to obtain one statistic to evaluate a player's offensive capability. I believe they are like the alcheists of old trying to turn base metal (BB statistics) into Gold (true value of a player). There are just too many variables to reduce everything to just one simple calculation/number.

[Oxtung]

Right. wOBA judges every batter by the same weighted averages but team scoring environments aren't all identical. A walk in a low OBP environment isn't as valuable as in a high OBP environment (although still valuable). Likewise a home run in a low OBP enviroment is more valuable on average, than one in a high OBP environment. The Blue Jays are built this way.

 

The Red Sox are excellent in every respect, so Pedroia's OBP contribution, still very valuable, might not be as valuable to his team as Desmond's HRs are to the Nats who are pretty terrible offensively in every respect.

 

So after reading that article as well as the 2 follow ons I now better understand what you were saying. I misunderstood your point the first time. I thought you were comparing HR's to walks when in reality you were comparing HR's to HR's and walks to walks.

 

I need more time to digest the information I read. Here are links to the two follow on articles.

 

Part 2:Linear Weights + BaseRuns = Good | FanGraphs Baseball

Part 3:Team-Specific Hitter Values by Markov | FanGraphs Baseball

[2wins87]

All of the posted analyses are working overtime to make a linear function of the variable(s) in the analysis. They do appear to agree with linear in that narrow range--but the funcions must actually be non-linear over their entire range (OBP 0 to 1, and SLG 0 to 4).

 

I was going to point this out also. I'd guess that if you have a team of 8 hitters who all have around .400 OBP and .400 SLG, you'd rather have your ninth guy have an OBP around .300 and SLG around .500 rather than another .400/.400 guy. Maybe the example isn't extreme enough to be true, but you get the idea.

 

 

Also, regarding SB. I looked into the risk-reward of different base stealing situations using a run-expectancy matrix a while back. It'd be good for it's own post, I'll have to find the spreadsheet, but bottom line, a lot of situations you need to do better than an 80% success rate to generate positive value. While some of the lowest necessary success rates actually come from attempting to steal home. The problem is just that it is difficult to find a situation where you are reasonably sure you will be able to steal home.

[jay]

I was going to point this out also. I'd guess that if you have a team of 8 hitters who all have around .400 OBP and .400 SLG, you'd rather have your ninth guy have an OBP around .300 and SLG around .500 rather than another .400/.400 guy. Maybe the example isn't extreme enough to be true, but you get the idea.

 

This seemed to make sense to me at first, but even your example turned out to be enough to prove otherwise. The infamous Tom Tango has a simple little Markov tool to find expected runs: Run Expectancy, Run Frequency, Runs Created, Linear Weights Generator. This is also linked to in the FG article that Oxtung posted about Markov a little earlier.

 

If that 9th guy was also .400/.400, that would also be your team OBP/SLG. (I input 40 AB, 12H, 1 2B, 0 3B, 1 HR, 6.7BB, 7K.) That lineup provides expected runs of 6.527.

 

If the 9th guy was actually .300/.500, your team OBP/SLG would end up .389/.411. (Here I input 40 AB, 12H, 1.45 2B, 0 3B, 1 HR, 5.8BB, 7K.) This lineup provides expected runs of only 6.246.

 

So, we're talking almost a full .3 fewer runs per game. I think what we're seeing here is the value of OBP compared to SLG. Adjusted OPS puts the value of OBP at 1.8 to SLG. The 'outs are precious commodities' statement holds a lot of water.

[Oxtung]

This seemed to make sense to me at first, but even your example turned out to be enough to prove otherwise. The infamous Tom Tango has a simple little Markov tool to find expected runs: Run Expectancy, Run Frequency, Runs Created, Linear Weights Generator. This is also linked to in the FG article that Oxtung posted about Markov a little earlier.

 

If that 9th guy was also .400/.400, that would also be your team OBP/SLG. (I input 40 AB, 12H, 1 2B, 0 3B, 1 HR, 6.7BB, 7K.) That lineup provides expected runs of 6.527.

 

If the 9th guy was actually .300/.500, your team OBP/SLG would end up .389/.411. (Here I input 40 AB, 12H, 1.45 2B, 0 3B, 1 HR, 5.8BB, 7K.) This lineup provides expected runs of only 6.246.

 

So, we're talking almost a full .3 fewer runs per game. I think what we're seeing here is the value of OBP compared to SLG. Adjusted OPS puts the value of OBP at 1.8 to SLG. The 'outs are precious commodities' statement holds a lot of water.

 

It doesn't matter what your lineup is, if you have two players with identical OPS/wOBA then the player with the higher OBP is going to theoretically provide more runs than the SLG. In the third article I linked to earlier Steven Staube shows this to be true using real life lineups and players and the Markov model. OBP scales better with Runs/Game than SLG does. So as scoring increases OBP becomes more valuable, the math behind this is all available in those articles if you're interested in reading why. Theoretically there is a low run scoring environment where a HR becomes more valuable than a walk/single, which is intuitive because if you're guaranteed to only get 1 hit per inning the only thing that will score a run is a HR, however in reality even during the deadball era run enough runs were scored that OBP was still a better driver of runs than SLG.

 

Now, the real tricky part comes when you're comparing two players, who are almost assuredly not going to have the same OPS/wOBA how much higher does player 1's OBP have to be to make up for a deficit in OPS/wOBA? As I alluded to earlier, who was the more valuable player last year Mauer, .416OBP .446SLG .861OPS, or Willingham, .366OBP .524SLG .890OPS? Willingham's OPS is higher but Mauer has the higher OBP.

[jay]

Now, the real tricky part comes when you're comparing two players, who are almost assuredly not going to have the same OPS/wOBA how much higher does player 1's OBP have to be to make up for a deficit in OPS/wOBA? As I alluded to earlier, who was the more valuable player last year Mauer, .416OBP .446SLG .861OPS, or Willingham, .366OBP .524SLG .890OPS? Willingham's OPS is higher but Mauer has the higher OBP.

 

True on all accounts. If you're defining value as helping the team score runs, Mauer would unequivocally have been more valuable.

 

We see the same thing in the Callaspo/Trumbo example used in the FG article. They had identical wOBA, but Callaspo was more OBP and Trumbo was more SLG. It would take an unrealistically extreme environment for Trumbo to have been "worth" more.

[Oxtung]

True on all accounts. If you're defining value as helping the team score runs, Mauer would unequivocally have been more valuable.

 

We see the same thing in the Callaspo/Trumbo example used in the FG article. They had identical wOBA, but Callaspo was more OBP and Trumbo was more SLG. It would take an unrealistically extreme environment for Trumbo to have been "worth" more.

 

Did you run Willingham vs. Mauer through Markov? I'm not sure who was more valuable that's why I posted it.

[jay]

Did you run Willingham vs. Mauer through Markov? I'm not sure who was more valuable that's why I posted it.

 

Looking at the exact Twins run scoring environment from last year (2012 season stats), I just divided each category by 162 to find single game averages and input those figures to the Markov calculator. This results in 4.313 runs per game -- pretty darn close to the actual 4.32 (699 projected, 701 actual).

 

If we then replace Willingham's 615 PAs with Mauer's equivalent performance, the output is 4.342 runs per game. If we replace Mauer's 641 PAs with Willingham's equivalent performance, the output is 4.277 runs per game.

 

So, having 2 Joe Mauers in the lineup would have been about 10 runs more valuable than 2 Josh Willinghams (equivalent to about 1 additional win).

[ThePuck]

What do you guys think of the runs created stat? There's different versions, like RC27, RC/G and wRC.

[jay]

What do you guys think of the runs created stat? There's different versions, like RC27, RC/G and wRC.

 

I like them. The math is a lot more complicated, but runs win games and are a concept that folks can comprehend. I'm not a fan of the 100 scale rate stats. Counting stats can serve a great purpose for easier comparisons.

[ThePuck]

I like them. The math is a lot more complicated, but runs win games and are a concept that folks can comprehend. I'm not a fan of the 100 scale rate stats. Counting stats can serve a great purpose for easier comparisons.

 

wRC is a good one. It's a counting stat. I also like RC27. Both take into account playing time.

[jay]

Seth / John,

It seems like you both dabble in this saber/stats stuff a bit (maybe more so John). Nick adds great insights and touches on it. Parker does amazing analysis more often related to mechanics. Have there been any efforts to recruit someone a little more saber-heavy to the regular rotation? I certainly wouldn't be that guy, but I do really enjoy reading and trying to understand it. I think it might reach out and draw in another segment of the audience -- especially if it were heavily Twins focused.

[Oxtung]

Looking at the exact Twins run scoring environment from last year (2012 season stats), I just divided each category by 162 to find single game averages and input those figures to the Markov calculator. This results in 4.313 runs per game -- pretty darn close to the actual 4.32 (699 projected, 701 actual).

 

If we then replace Willingham's 615 PAs with Mauer's equivalent performance, the output is 4.342 runs per game. If we replace Mauer's 641 PAs with Willingham's equivalent performance, the output is 4.277 runs per game.

 

So, having 2 Joe Mauers in the lineup would have been about 10 runs more valuable than 2 Josh Willinghams (equivalent to about 1 additional win).

 

Awesome, thanks!

[Oxtung]

Seth / John,

It seems like you both dabble in this saber/stats stuff a bit (maybe more so John). Nick adds great insights and touches on it. Parker does amazing analysis more often related to mechanics. Have there been any efforts to recruit someone a little more saber-heavy to the regular rotation? I certainly wouldn't be that guy, but I do really enjoy reading and trying to understand it. I think it might reach out and draw in another segment of the audience -- especially if it were heavily Twins focused.

 

If this were ever to be considered I know of a candidate I think would be a good fit. If you were to have some kind of application process I think you'd find there are many worthy candidates.

[ashbury]

If this were ever to be considered I know of a candidate I think would be a good fit. If you were to have some kind of application process I think you'd find there are many worthy candidates.

 

See if you can recruit Nate Silver.

[Oxtung]

See if you can recruit Nate Silver.

 

Now wouldn't that be a bump in prestige! Unfortunately my friend is only a "lowly" statistics Professor at a top 15 school with a background in computer modelling. He probably doesn't quite have the name cache of Silver.