And the Winning Number is…OPS!


by Cyril Morong



Click here to see my sabermetric blog called Cybermetrics



Well, okay, maybe it is not THE WINNING NUMBER, but it is a pretty good one. Here’s why.

Most of you probably know that OPS is on-base percentage (OBP) plus slugging percentage (SLG). OPS is highly correlated with winning. The correlation between team winning percentage OPS differential is .933. OPS differential is the difference between a team’s OPS and the OPS they allow their opponents. A perfect correlation is 1.00. The teams I looked at were from 2001-03 (data from ESPN).

The graph below illustrates the relationship.

As you can see, most teams are pretty close to the trend line. As your OPS differential rises, so does your winning percentage.

The correlation is often referred to as r. Another statistic is the r-squared, or square of the correlation. In this case it is .933*.933 or .87. This means that 87% of the difference in winning percentage across teams is explained by the OPS differential. How does this compare to other stats? The r-squared for batting average differential is .738, on-base percentage is .849 and slugging percentage is .786. So OPS tops them all in its ability to explain team winning percentage. This tells us that getting on base and hitting for power (and stopping your opponents from doing so) are very important in baseball.

One problem with OPS, however, is that it adds together two numbers that are highly correlated. In fact, the correlation between OBP and SLG here (only on the hitting side) is .806. So, I looked at an alternative measure. Branch Rickey developed it. It adds OBP plus .75*ISO (Isolated Power, which is slugging percentage minus batting average). The correlation between OBP and ISO was .643, not as high as it is for OBP & SLG. I then found the r-squared between winning percentage and the Rickey stat. It was .863. So again, a measure similar to OPS, which takes into account power hitting and getting on base, does a good job of explaining winning percentage.

Now it is possible that other factors, like stealing, affect a team’s OPS. This is not likely to be true. See “Other Factors” below.

The linear relationship between team winning percentage and OPS differential is


Pct = .496 + 1.3*OPSDIFF


Where OPSDIFF is each team’s OPS differential. This formula predicted about two-thirds of the teams to within 5 wins of their actual total. I then looked at which teams won more games than expected according to the formula. Five of the top ten on this list had less than .57 SBs per game, which was the major league average for the three years. I also looked at which teams won fewer games than expected according to the formula Four of the bottom ten had less than .57 SBs per game, almost the same. The top ten averaged .65 SBs per game while the bottom averaged .52. No big difference here. Perhaps some other reason explains why some teams won more than expected while others won less. It might be luck or it might be the bullpen.

            NOTE: THESE NEXT TWO PARAGRAPHS WERE ADDED ON JULY 10, 2005. I also looked at all teams from 1989-2002. The regression formula for that data set was


Pct = .49996 + 1.26*OPSDIFF


This is virtually the same as the regression for the 2001-03 period. Then I determined how many more (+) or fewer (-) games each team won for each season. Then all of those plusses and minuses were added up. The A’s won 45 more games than the formula predicted over the 14 year period. That is just a little over 3 a year. All but four of the 30 major league teams won within 2 games a year of what the formula predicted. Only one team was off by more than 4, the Red Sox, who won about 4.06 fewer games a year than predicted.

            If a team happens to win more games one year than expected, then next year they likely win fewer games. The Twins won 13.5 more games than expected in 1994, but won 1.9 FEWER the next (and just 4.6 more games more than predicted in 1993). So one could argue that a team wins more games than OPS would indicate because they play “small ball,” like stealing and moving the runners along through “productive outs.” But this is not likely to be true. If a team is supposedly good at moving runners along one year, it should do so the next. This is not happening. See also Productive Outs Are Not Productive.

            One last thing to remember here. I am basically saying that if you “out-OPS” your opponents, you will do very well. This could be done with better hitting, better pitching, better fielding or some combination of the three. But this analysis shows that comparing two players based on OPS is not a bad way to go.


Other Factors

Take base stealing for example. If a man is on, this distracts the pitcher. Maybe he throws more fastballs and the batter can be ready for this. Or a hole opens up that makes it easier to get a hit. So some of the 87% I attributed to OPS might, in fact, actually come from another source. So I looked at how both good and bad stealing teams hit with a runner on first base only.

I looked at the top 10 teams in SBs from 1982-92 and the bottom ten. Then I determined how much their AVG, SLG, OBP and OPS differed between having a runner on 1st or no runners on at all. I determined the runner on first data by finding the difference between the runners on base and the runners in scoring position data (from Retrosheet).

The top ten teams in SBs had the following increases when there was a runner on first compared to no runners on (the average across the teams)







That is, with a runner on first, these teams had a .025 higher batting average than they did when there were no runners on base. Slugging went up .030, OBP .008 and OPS .038.

The bottom ten teams in SBs had the following increases







The top ten teams averaged about 240 SBs and the bottom around 40. The one difference that is big is the AVG difference (.025-.019=.006).  But in general, the best stealing teams had little additional benefit over what the worst stealing teams.

The best stealing teams had in the 900 range of ABs with a runner on first. The bottom in the 1100 range. This makes sense because the best steal and they won’t be on first as often. Also, who is most likely to be left on first base on those teams? The few guys who don’t steal, like Jack Clark (5 of the teams were Cards). But those bottom teams must have rarely had a good base stealer on, a lot less often than the best. I think if the runners bother the pitcher, we should see a bigger effect here. After all, we are comparing the best stealing teams to the worst.

The change in OPS for both teams is just about the same. I am still skeptical that having a good stealer on first helps a lot. Maybe the change in AVG is simply a result of the hole opened up at first. There is little change in SLG. Maybe the fast guy bothering the pitcher and making it easier for the hitter is not happening.


Sources: Retrosheet and ESPN


Back to Cyril Morong's Sabermetric Research