Not understanding OPS+

[brett]

Can someone please explain OPS+ to me. I though it was like ERA+ in that it was simply a ratio of a players OPS to the ballpark adjusted league OPS but this clearly is not the case. For example Ruth has an OPS of 1.164 and the league adjusted is .753 for a ratio of 1.55, or 155 with 100 taken as average, but his OPS+ is listed as 207.

[SABR Matt]

OPS+ = Rel OBP + Rel SLG - 100.

In otherwords, it's the ratio of OBP to league OBP plus the ratio of SLG to league SLG. If you're 15% better than league OBP and 25% better than league average SLG, you get a 140 OPS+

[Tango Tiger]

It is of course completely stupid. Every other "+" stat is metric divided by league average of that metric.

Furthermore, by taking the relative values first, you are changing the OPS ranking! For example, let's say you have these two .750 OPS guys, in a league where the average is OBP = .340 and SLG = .410 (i.e., .750 OPS):

player1, OBP = .250, SLG = .500
player2, OBP = .370, SLG = .380

What is their "OPS+" using the stupid way? The first guy has an OPS+ of 95, and the second guy has an OPS+ of 102, even though both their OPS is .750!

Now, as it turns out, the stupid way is actually better, since the effect is to make it: 2.4 * (OBP*1.2+SLG-.41)

Forgetting about those constants, we see that OBP will get a slightly greater weight. It should be much higher, at 1.8, but sobeit.

In any case, if you want to use the construction of "OPS+", we shouldn't be calling it OPS+. Call it something else.

[Sliding Billy]

What bothers me about OPS in general is interpreting it.

I can understand SLOB--(H+BB/PA*TB/AB--as a rough model of the interaction between getting runners on base and advancing them. I can understand the different results of an increase in one H, BB, or TB on the overall RC/AB.

But what exactly is (H+BB)/PA + TB/AB supposed to represent?

I realise that both account for around 92% of the variation in runs scored, but one of them makes sense to me; the other is a mystery.

[SABR Matt]

Precisely Sliding Billy. I have always believed (and I know Tango does too) that you should create a stat that says what you mean it to say...that is a direct representation of a real mechanism...not something that happily resembles what you're trying to model by pure luck.

[John Beamer]

Billy

The reason why OPS works is because it assigns near-correct weights to different offensive events. Yes, it is cumbersome and a little awkward, and has the dual denominator issue, but in general it works. This shows that the precision we all strive for in baseball isn't necessarily *that* important. OPS is only marginally worse than other run models, such as BaseRuns. It is the most simple, good metric we have.

This is why it works:

OPS = [(1B+2B+3B+HR+HBP+BB) + (1B + 2*2B + 3*3B +4*4B)*1.07]/PA

Assuming that 7% of all plate appearance are BB or HBP or some such.

Now if you work out the relative values of the events we get, for instance:

2b/1b ~= 3/2

LWTS tells us that 2b/1b = .75/.47 ~= 3/2

Again OPS for 3b/1b = 4/2

LWTS tells us that 3b/2b = 1.03/0.47 ~= 4/2

So actually OPS, although a but clumsy actually works quite well.

The reason why people like 1.8*OBP + SLG is because it gives makes the relative value of the events much more accurate that crude OPS

[Sliding Billy]

Billy

The reason why OPS works is because it assigns near-correct weights to different offensive events. Yes, it is cumbersome and a little awkward, and has the dual denominator issue, but in general it works. This shows that the precision we all strive for in baseball isn't necessarily *that* important. OPS is only marginally worse than other run models, such as BaseRuns. It is the most simple, good metric we have.
.........

The reason why people like 1.8*OBP + SLG is because it gives makes the relative value of the events much more accurate that crude OPS
Thanks for the explanation, John.

Precision wasn't really the issue: Obviously OPS is close enough for a one-liner in a HOF argument. Interesting that it happens to jibe with LWTs that well at present. Maybe John Henry Waugh is a sabremetrician.