THE BOOK--Playing The Percentages In Baseball

OPS is just a relatively easy way of quickly and crudely assessing someone’s performance without much effort. I think it’s particularly useful for minor league stats, where you don’t have much time to pour over numbers. “So and so in AA is 21 and has a .847 OPS...” He’s having a pretty good year. We get it.

However, in terms of real analysis, there’s just no good reason to go to OPS when so many better metrics are available.

I prefer BRA (OPS*SLG)

It irritates me that Joe Posnanski leans so heavily on OPS+, and I’m a huge Poz apologist. For quick assessments, sure OPS is useful, but it’s just not very helpful for making comparisons or any sort of halfway serious analysis. Presumably a writer has decided to go with OPS, because that writer wants to understand the game better than the BA/RBI/HR trinity allows. So why, in the face of very clear reasons to switch to something better and with such better things very readily available, would someone keep using OPS?

I have no problem with OPS as a quick and dirty, and easy to compute, measure of offensive production.  No problem at all.

At the same time, I agree that writers like Neyer and Poz lean too heavily on it in serious discussions about player talent and value.  Then again, Rob (Neyer) leans WAY to heavily on current season numbers in general, when discussing serious personnel issues, which really ticks me off (but I have given up criticizing).

#2, I think you will find that OPS is slightly better than OBA*SLG for batters, although it is probably very close.  And if it is very close, why would you want to multiply two awkward numbers and come up with a really awkward number?

The problem I see with OPS, and I guess this is the main point Tango makes, is that it weights OBP and SLG equally. I think we should always be aware that OBP is more important. In my regressions from a few years ago where I found the run value of OBP and SLG for each lineup slot, OBP was always higher, even for the middle of the order guys.

But I also think that sometimes OPS is useful graphically. It helps you tell a story simply since OPS boils down to just 1 variable. An example of something I did on this is in the following link. I found the relationship between winning pct and teams’ OPS differential

http://www.beyondtheboxscore.com/story/2005/9/12/01529/1532

I think I also tried some type of weighting for OBP and my correlations and standard errors did not change much. OPS was just as good

I also did a clutch hitting study where OPS was on the horizontal axis and a player’s PWA/PA was on the vertical axis. The idea was to visually show that how well a guy does in the clutch (like probability of winning added) is highly correlated with his overall stats. That study was at

http://www.geocities.com/cyrilmorong@sbcglobal.net/totalclutch1.htm

Also, I think OPS is not bad for looking at a guy’s splits like home and road or lefties vs. righties to get a quick idea of where he does better (or OPS in the clutch vs. non-clutch). Now maybe there are some cases where we really need to know a guy’s OBP and SLG both home & road. But I think if we see that player A has a .900 OPS vs. lefties and .800 vs. righties, he is better vs. lefties (given enough observations). But like I said, maybe there are some guys for whom this would not be true and I would like to hear about those cases and I will be looking at that data more carefully in the future for individual hitters.

I’m all in favor of seeing RC+.  Maybe Fangraphs can do a wRC+ actually…

In one of my regressions on lineups, the run value of OBP for the leadoff man was 2.49. It was .97 for SLG. So a leadoff man with a .400 OBP and .400 SLG has a run value of 1.384. A guy with a .350 OBP and a .450 SLG has a run value of 1.308. Tango, maybe your Markov chain models could tell us a story like that. It would make my example a little more believable.

But getting back to my example. The guy with the .350 SLG would have to have a .528 SLG to match the other guy (of course, he probably is not batting leadoff then). But on the other hand, if we gave the guy with a .450 SLG a higher OBP, it would have to get up to about .380 before he could equal the .400/.400 guy.

Tango

In your example, suppose that we have a guy with a .350 OBP and .450 SLG. That is an OPS of .800 but if we weight OBP by 1.8, it gets us 1.080, less than the 1.120 for the high OBP guy. The high SLG guy would need his SLG to rise to .49 to get his weighted OPS up to 1.120. If he had an SLG of .48, say, meaning an OPS of .830, he is still not as good as the guy with .400/.400 or .800 OPS. My guess is that we could find some actual players like this. One with a higher OPS but actually lower value (not taking team, lineup slot, league, year and park effects into account).

Cy

Suppose we discovered that tRA was superior to FIP. Would we decide to use tRA exclusively? I wouldn’t--not for my blog posts, or casual discussion with friends, or in any context where I think tRA might be unfamiliar to my audience. I can explain FIP to anyone that will listen in one minute; tRA--not so much. FIP is easy to caluculate and scales with ERA (and so provides a more familiar value that tRA which is scaled to RA.) FIP and tRA give very similar ordinal rankings of players, and large gaps in one mirror large gaps in the other. In short, they correlate very well. When I really wanted to get my hands dirty, I would look at both, but for 90% of player comparisons, it’s not really necessary. Either one would do, even the lesser of the two. I think pretty much all of that goes for OPS and wOBA or 1.8*OBP+SLG.

If OPS were as bad as, say, K/BB, I would be the first one to abbandon OPS. But OPS does remarkably well for an extremely simple metric. Unless I need the precision of the more accurate metrics, I’m probably not feeling the pressure to use them.

Do you still think I should stand down, or are we just talking at cross-purposes?

Addendum to #9: K/BB for pitchers is what I had in mind. It’s not a great metric for hitters, but at least there I feel like it isolates a skill or combination of skills that we might want to assess. In the case of pitchers, any metric that makes Joel Pinero look like James Shields is bunk.

I said this:

“How much bend are you willing to accept?  Well, about as much as your seriousness on the matter. “

So, if you want something quick and dirty (i.e., a “1” on the 0-10 seriousness scale), then fine.

When you say this:
“Do you still think I should stand down, or are we just talking at cross-purposes? “

You need to stand down when I will object to a study of yours that shows its bias.  A list of HOF candidates by OPS or OPS+ is a stand-down position.  A trade discussion using OPS+ is a stand-down position.  A “look at his home/road OPS+ splits” is likely ok.

You need to stand down when I will object to a study of yours that shows its bias.

Yep.

A list of HOF candidates by OPS or OPS+ is a stand-down position. A trade discussion using OPS+ is a stand-down position.

Maybe.  When Albert Pujols retires with a career OPS+ second among RHB to Rogers Hornsby, we can go through the exercise of wOBA, but OPS+ told all the tale we needed to tell. wOBA might surprise us by showing that Albert was less good than we thought, or even better than we though (based on OPS+), but it’s not going to show us anything radical, like that he was actually only as good as Ryan Howard. My point is that it’s not just a matter of seriousness, it’s a matter of OPS being sufficiently accurate that, for some comparisons, there’s no need to dig deeper. Questions like “Is Andy LaRoche a good hitter?” or “Are there many middle infielders who hit better than Chase Utley?” are answered readily by OPS. Other questions--is Willy Taveras an improvement on Cory Patterson?--should appeal to a more precise metric, and one that accounts for the particular skill set of the two players.

Indeed, and I think you will agree, the practice of ignoring a player’s defensive skills far more frequently introduces misevaluations of trades than the use of OPS.

OPS+ is almost necessary as its the only widely available park-adjusted rate stat. 

Batting runs are park-adjusted on fangraphs (but only the runs at the bottom of the page in the “value” section), but their wOBA isn’t adjusted (as far as I can tell).

OPS+ is listed on the front page of every BBRef page.

EqA is park-adjusted.

***

“My point is that it’s not just a matter of seriousness, it’s a matter of OPS being sufficiently accurate that, for some comparisons, there’s no need to dig deeper.”

Well, yes, if you want to compare Pujols to Casey Kotchman, we don’t need to go beyond OPS+.  Or even runs produced for that matter.

I am talking about those close calls, when it matters how you are bending your metric.  Those guys like Sammy Sosa or Juan Uribe or Pedro Feliz or Kenny Lofton or Willie Wilson.  Too often, the Church of OPS+ is used in those cases. 

For the non-close calls, even Runs Produced per out is fine.

I like By The Number because they do pretty in depth work, but when I read the comment, I couldn’t believe it.

StatsCorner publishes a wOBA* that is park adjusted as well. It’s probably the most useful single hitter stat anyone publishes. StatsCorner, by the way, is probably the most under-publicized baseball cite for great stats.

Since you use it to refute OPS, you evidently think 1.8*OBP+SLG is a pretty good stat. How do you reconcile this with your view that BA+OBP+SLG is NOT a good stat, since the two are practically equivalent?

For instance, major-league hitters batted a combined .264/.333/.416 in 2008.

(1.8 * .333) + .416 = 1.015
.264 + .333 + .416 = 1.013

Only two-tenths of one percent difference.  How come you like one stat but not the other?

"Since you use it to refute OPS, you evidently think 1.8*OBP+SLG is a pretty good stat. How do you reconcile this with your view that BA+OBP+SLG is NOT a good stat, since the two are practically equivalent?”

They’re “practically the same stat” to the degree that BA = .8*OBP, which for many players is accurate but for many players not. BA+OBP+SLG is a poor measure of the abilities of Adam Dunn or Christian Guzman. The mathematical complexity of the two functions is the same but the results are more accurate for one than the other.

How do you know “BA+OBP+SLG is a poor measure of the abilities of Adam Dunn or Christian Guzman”?  Linear weights may tell you this, but how do you know linear weights apply at the level of individual players, since their values cannot be verifed below the level of team scoring?  After all, with the exception of solo home runs, all scoring plays are co-operative efforts.

"but how do you know linear weights apply at the level of individual players, since their values cannot be verifed below the level of team scoring?”

I’m very confused by this. The run value of a walk vs. an out or a double vs. an out is calculated by figuring out how many runs those events usually produce. What more could you want to know about a player’s performance than what, typically, his performance would produce? “Verified below the level of team scoring"--what other court would we look to besides team scoring? The goal of the game is to score runs--wouldn’t looking at any other effect it had just be irrelevant?

“all scoring plays are co-operative efforts.”

Yes! Which is why every linear weight can be decomposed into to its ability to (1) move runners on base over, (2) put runners in a position to score, and (3) end an inning, thereby removing runners from the bases.

Maybe this will make my point clear:  I contend that, for example, Babe Ruth’s walks were more valuable than Barry Bonds’, because Ruth had Lou Gehrig hitting behind him instead of a collection of sub-replacement level hitters.  I think Bonds walked 232 times in 2004 not because of his “on-base skills”, but because opposing managers and pitchers knew that his walks, in that lineup, were worth less, in linear-weight terms, than walks are worth on average, and they weighed that value against the chance of him hitting a home run if they tried too hard not to walk him.  I think it is possible to generalize from this that many power hitters who walk a great deal have inflated linear-weight values because linear weights look at what a walk is worth for the average player on the average team rather than what it is worth for a weak-hitting team’s best power hitter.

Bluzer, yes that is true for every player.  A player’s lwts or wOBA is his theoretical value when batting in an average lineup slot for an average team.  That is in fact the definition of lwts or wOBA as it applies to a an individual player.  You are not pointing out anything that is not already included in the definition of those metrics.

Now, why would “many” power hitters (as opposed to “a few") be overrated by virtue of their walks?  Do power hitters generally play for weak hitting teams?  Obviously some power hitters, like any other kind of hitter, play for weak hitting teams, but the average power hitter plays for an average team, such that on the average the value of their walks and their other offensive components are the same as everyone else’s.

In fact given that power hitters, or any good hitter for that matter, tend to be surrounded in the lineup by other good hitters, you could make a good argument that even on a weak hitting team, a good (power) hitter’s stats are not going to have less value than their average lwt value suggests.  Keep in mind that the more removed in the lineup you get from a player, the less those other players influence the value a particular player’s offensive events.  For example, the primary value of the walk is the on base component, obviously.  If a player is followed by 2 or 3 very good hitters and then by 2 or 3 poor hitters after that, the poor hitters will have little effect on the value of that hitter’s walks.

Now, if you want to argue that power hitters tend to get walked with base open and in high leverage situations, such that the value of their walks is less than average, I’ll buy that.  We have discussed that before. I think that we also found that the difference in value of Barry Bonds’ walks and Rey Ordonez is not as much as you might think - maybe a couple of hundreds of a run at most.

The power hitters who are overrated by virtue of their walks are the ones who walk a lot because the hitters following them are unlikely to capitalize on the walk.  The fact that they walk is still valuable, it just isn’t as valuable as it would have been with more potent hitters coming up next.  If power hitters with high walk rates had more potent hitters behind them, they would not walk as often, so in what sense is their walk rate attributable to any skill on their part?

I did a study called “The Value of OBP and SLG by Lineup Position for High-Scoring and Low-Scoring Teams” It is at

http://www.beyondtheboxscore.com/story/2006/9/26/93215/5836

If you go to this link, you will see that the marginal run value for the cleanup hitter’s OBP is actually higher on the low scoring team.

My study on walks to Bonds is at

http://www.geocities.com/cyrilmorong@sbcglobal.net/Walkvalue2.htm

It looks like his walks had pretty close to the normal value

Why doesn’t someone just do a study on which team metric—1.8*OBP+SLG or BA+OBP+SLG—better correlates to runs scored?  I would be very interested in seeing that.

Cyril, your work looks interesting and I will give it a closer look later.

Bluzer,

That sounds like a good idea and I might try it if I get a chance. I also wondered if we might check the correlation for players between 1.8*OBP+SLG and BA+OBP+SLG. If the correlation is not low and the former is more higly correlated with team scoring, then we need to stick with it. Feel free to send me any comments on my work.

Cy

By coincidence I was just working on calculating linear weights by lineup position.  Bluzer you are right and you are wrong.  Walks do vary a fair amount, but it is not because of the quality of players following the player walking vary from team to team, it is because those following players vary from an average player due to lineup position.  LW is calculated by adding the runs scored on the play plus the runs expected from the base out state following the play and subtracting the runs expected before the play, and then dividing that number by the number of plays.  A weak hitting team will affect both the runs expected after the play and the runs expected before the play almost the exact same amount.  Since you are subtracting one from the other the effect cancels out.  Same with a strong hitting team.

But the effects of lineup do not cancel out.  They vary because the BaseOut situations in which a batter gets walked vary considerably from lineup position to lineup position as MGL mentions above, and they also vary by the amount each lineup position’s average skill varies from the average league player.  The results for a NIBB are shown below.

Lineup Pos------Lineup_LW
1-----------------.360
2-----------------.372
3-----------------.294
4-----------------.279
5-----------------.283
6-----------------.285
7-----------------.296
8-----------------.302
9-----------------.406

Also, OPS makes no distinction between NIBB and IBB and the rate of IBB for slugging middle of the order batters will be much higher than for beginning or end of the order batters.  This means that the general calculation of ModOPS as 1.8*OBP + SLG should have a lower multiplier for middle of the order batters and higher for top of the order hitters. Just how much depends on how batting order effects all batting events.  The NIBB is the largest factor but it is offset somewhat by the higher value of the home run for middle of the order hitters.

"the general calculation of ModOPS as 1.8*OBP + SLG should have a lower multiplier for middle of the order batters and higher for top of the order hitters”

Peter, traditional managers tend to put high-BA, non-power hitters at the top of the order, so BA+OBP+SLG would in effect provide a higher multiplier for those hitters, while providing a lower multiplier for low-BA power hitters like Dunn.

Yes Bluzer, I was agreeing with you that BA+OBP+SLG may actually be a decent substitute for calculating separate formulas for each lineup position.  But as I said above one must look at the totality of all the effects of lineup position on LWs, not just walks.  And correlation on the team level will have absolutely no meaning of whether it should be used for individual players.  You would need to calculate lineup adjusted linear weights for individual players and then see how well BA+OBP+SLG correlates with them.

Peter, those are some huge differences between lineup slots! For #9 hitters, what did you do about pitchers batting?

In general, I’m uncomfortable with this kind of study because it assumes a conventional lineup order. I guess it can answer the question, “in a conventional lineup, what is the best slot for player X?” But if player X is actually on your team, and you’re considering moving his lineup slot, I don’t see how such a study would help. Better to use a sim.

If you are including IBB and treating it the same as a NIBB in ANY metric, then all bets are off of course.  If you know what I mean.

We’ve had this discussion before.  An IBB is probably best treated as a non-entity for purposes of any rate stat (like lwts per PA, or wOBA, or OPS) and as equal to the average value of a particular player’s PA for purposes of a counting stat (like VORP or WARP) or lwts.  Of course, if you do the latter, the former is obvious…

dcj - I didn’t believe those numbers either so I went back and checked and found my error.  Corrected numbers are below.  I hate it when I put something in print too fast and it isn’t right.  I hope this is correct.

Lineup_Position--------Lineup_LW_NIBB
1--------------------------.359
2--------------------------.360
3--------------------------.311
4--------------------------.305
5--------------------------.297
6--------------------------.296
7--------------------------.304
8--------------------------.296
9--------------------------.336

Those numbers seem a little weird, Peter. From 2-6, the numbers are going down, which seems to make sense. Then it goes back up at 7, so I’m thinking maybe that’s because the top of the order is up soon (with better hitters). But then the value goes down again at the 8th spot, so I’m a little perplexed.

What if you did this for AL games only? I think the numbers would follow a much better progression than they do now. It’s just the values from 6 through 8 that don’t make any sense to me.

Dan - I probably shouldn’t have shown them to the third decimal since no LW is going to be that accurate.  If you round to the nearest hundredth it looks pretty reasonable.

Peter, in Table 51 (I think) I have the LWTS by batting order.  Feel free to post those walk numbers and compare it to yours.

Tango - How did you get the run values for the events in the 24 baseout states that you present in Table 50?

I start with the RE chart.  And I simply weight the “final” state based on the frequency.

So, for a single with runner on 1B and 1 out, I figure out how often each of the final states occurs and apply the RE value to each of those states.  That gives me the average final RE value from which I subtract the initial RE value. 

Pretty standard…