6 Comments

I'm not exactly a fan of OPS+, but if you interpret a 118 OPS+ as "18% more productive than a league average hitter" rather than "18% higher OPS than a league average hitter", it is a correct statement (at least to the extent that OPS+ is a predictor of team runs scored). The negative values are an unavoidable consequence of a linear run estimator applied at extremes of offensive ineptitude, although I certainly understand why one might find them obnoxious.

I wrote a little bit about the interpretation of OPS+ in response to Bill James' comments in his 2023 Handbook (RIP) last year: https://walksaber.substack.com/p/rehashing-runs-created-and-ops

Expand full comment

But is that even true? I could agree that OPS/(league average OPS) works to use with a true percentage. The idea of adding one average to another and subtracting one is just weird and mathematically illogical. It might give answers that are ballpark-ish “correct” but that doesn’t make the process correct.

Expand full comment

I concur that it is weird and that the process is not correct. But if forced to use OPS, some kind of OPS/LgOPS, or OPS+, I'd personally take the latter because:

1. I think it actually makes more sense to add OBA/LgOBA and SLG/LgSLG than it does to add OBA and SLG in the first place.

2. If you accept the constraint (although I certainly don't think anyone should) of a linear equation using OBA and SLG relative to the league average (I'll call then aOBA and a SLG) to estimate team runs (per game or per out, which I'll call aR), and that they must be weighted equally:

aR = m*(aOBA + aSLG) + b

you will find the best fit from coefficients very close to m = 1 and b = -1. If you do the same with OPS:

aR = m*aOPS + b

you will find the best fit coefficients very close to m = 2 and b = -1. OPS+ has a 1:1 relationship with runs; OPS/LgOPS has a 2:1 relationship

3. OPS+ inherently weights OBA more heavily than OPS because in live ball baseball LgSLG > LgOBA. The ratio of LgSLG/LgOBA is usually something around 1.2 so it functions as weighting OBA around 1.2x SLG, although if you really wanted a best fit with runs it would be something like 1.8. But it's closer than OPS' 1x.

All that being said, I would be perfectly pleased to send all forms of OPS to the landfill (I don't think they are recyclable).

Expand full comment

Thanks for the detailed explanation. I too much prefer wOBA to OPS and wRC+ to OPS+. Playing around with Retrosheet data, it seems like the relationship was that wRC+ was close to the square of wOBA (perhaps like the Pythagorean Wins, the exponent might be something other than exactly two). Using that assumption, the m=2 and b=-1 is a close linear fit to a square function. For example, 0.8^2 is 0.64 and 2*0.8-1 is 0.6. Similarly 1.3^2 is 1.69, and 2*1.3-1 is 1.6. So basically the weird calculation is a linear approximation of a square function that is close for values near 1. It obviously fails for values close to zero or greater getting close to two or higher.

Expand full comment

Using mathematical terminology, 2x-1 is the first two terms of the Taylor expansion of x^2.

Expand full comment

While OPS is bad enough, I cringe most at the definition of OPS+ which is OBP/(leave average OBP) plus SLG/(league average SLG) minus one. How did that originate? People erroneously talk about percentages (e.g., an OPS+of 118 they say is 18% above average). This obviously breaks down for very low OPS+ which can theoretically bottom out at -100. There are pitchers with a negative career OPS+ like Sandy Koufax with a -39 OPS+. What percentage worse is he than average? 139% worse?

Expand full comment