OK, now that we've established that there is a queer sub-set of pitching records that consistently avoid zero-sum outcomes, what does that mean? Just what is it, anyway--what games are these, how many of them are there, and what do we do with it now that we've discovered it?
A "non-save situation" (which we'll sometimes abbreviate as "NSV") has several possible identities. Let's examine them.
First, they are relief appearances. You could say that all starter innings are "non-save situations," but that's silly--sillier, in fact, than the definition of a save (which contains a few moments of unintentional hilarity).
We are looking at what relievers do when they are not in a game moment where they are trying to protect a close lead. What are the components that are part of what is a compound class of situations?
For one, you have games where a reliever comes in with the score tied. (These will prove to be the ones that contribute most to the wins and losses which fall into the "non-save" situation, as we'll see shortly.) Then there are games when the reliever enters with his team behind. (A smaller number of decisions--which, as you can surmise, are tilted heavily toward wins--manifest themselves from these situations.) And there are games where the reliever enters with a lead too large to be considered a save situation (most +4 situations, and everything above that). And, finally, there are games where the reliever enters with a lead that would be a "save situation" except for the fact that the inning in which he enters is too early (prior to the seventh).
Now, yes, there's some artificiality in this construct...BUT, it's interesting how this grouping of components creates a steady winning result. To get a sense of how many wins and losses come from where, we're going to use one of our favorite "punching bag" franchises, the Kansas City Royals, to walk us through how it works.
It turns out that the Royals' relievers, unbeknownst to even their ardent admirers, turned in a near-historic performance in "non-save situations" in 2013. The two charts at right demonstrate this in different ways. The top chart shows the top ten ERAs turned in by teams in "non-save situations" over the past fifty-six years (from 1958 to 2013...we picked 1958 because it's the first year where the records are 100% complete). The 2013 Royals are ninth on that list.
You'll notice, though, that most of the teams on this list come from the 1960s and early 1970s; thus they reflect an age in which pitching was at its zenith (and when run scoring was at its nadir). To correct for this, we created the second chart, which uses a version of ERA+ that's customized to the aggregate major league performance (measured in ERA) for "non-save situations.
When we apply that, and run the numbers again, we see that the 2013 Royals are actually the fifth best in terms of ERA+ during "non-save" situations. That's an incredible performance, and in large part due to that level of effectiveness, the Royals had a 31-16 record in games that were decided from "non-save situations."
We'll look at that performance in greater detail below, but first let's focus on what else these charts can tell us. We included the next season performance for the teams on both lists. There are five common teams on each (1966 Dodgers, 1969 Orioles, 1972 Pirates, 1976 Yankees, and 2013 Royals.) Only the Royals failed to reach the post-season. They also had far more decisions (a total of 47) that came about as a result of these situations than the other; the '69 O's had 40, but the others had far fewer--with the '76 Yankees having only 20. That leads to a question about the historical rate of frequency for "non-save situation" decisions, but let's defer that for just a minute while we look at what the "next year" data tells us.
What we see from the "next year" data is what often happens in other partial measures of won-loss results (WPCT in one-run games, for example). We see a regression to the mean here (remembering that the "mean" for these "non-save situations" is .574, not .500. On both charts, the level of effectiveness (as measured by ERA) and in terms of wins and losses (WPCT) drops noticeably in the "year after" data. (Individual teams, of course, often "beat the odds" in multiple seasons--we see that in the case of several teams on each list--the '69-'70 Orioles and the '76-77 Yankees, just to name two from the "common teams" on both lists.
Won-loss performance in "non-save situations" might be a better yardstick of "team luck" than the Pythagorean Winning Percentage (PWP), though we've seen that it's more correlated with winning in general. Still, if we look at the Royals' .660 WPCT in these games and compare that to the historical average (574), we see that they're between four and five wins above expectations for the number of decisions they had. It looks like their relievers, pitching in the tighter spaces of games that were up for grabs, were able to snag extra wins for their team in 2013.
Now let's get to the components of this performance. (By the way, not easy to do with current breakouts: Forman et fils takes you only part-way. For pitchers who both start and relieve, you have to go into the game logs to break out the data, which is just what we did for Wade Davis and Luis Mendoza, who mostly started for KC in '13.)
The tie-game data shows us that these are often do-or-die situations. The average percentage of decisions accounted for by "non-save situations" is 20% over the 1958-2013 time frame, but the Royals (47 decisions) were closer to 30% overall, and 39 of those decisions came from 90 games in which relievers entered into tie games, a figure well over 40%. (Of course, you can actually win a game in which you give up the lead while you're on the mound, which is what happened to Louis Coleman: the Royals took back the lead after he'd surrendered it and he got a "cheap win" out of it.)
So the Royals went 23-16 in games where pitchers entered with the scored tied; they went 8-0 in games when they rallied from behind with a reliever on the mound. (Bruce Chen won three of those before he was put back in the starting rotation; Louis Coleman got two wins in "come-from-behind" appearances.)
So this little exercise with the Royals gives us some perspective on how decisions break out of "non-save" situations. It appears that upwards of 85% of them derive from relievers coming into tie games, and that the overall winning percentage for this component is probably right around .500. The rest of them derive from "come from behind" scenarios, and the overall winning percentage for that component is around .975 or so--there just aren't all that many games where a pitcher blows a five-run lead and is tagged with a loss.
Right now we don't know where that total of eight come-from-behind reliever wins that the Royals posted in 2013 sits in the relative order of things. Is it high, low, or somewhere in between? It's not easy to generate right now. What's clear is that the Royals' relievers pitched extremely well in all of the other component situations that go into the "NSV" bucket: their collective ERA outside of tie games was 2.02. That, too, isn't easy to anatomize with currently available data: do relievers generally pitch better in "low leverage"? The Royals' relievers certainly did in '13; we'll need to look at that data in much greater volume and detail to understand it.
Let's circle back to the question of the percentage of decisions that happen in NSV situations, and how that may be changing/fluctuating. Our final chart (for this installment, at least) shows how that percentage has changed. This running three-year chart actually undersells the trend--in 2013, this percentage hit an all-time high at 26%.
This is almost certainly due to the increasing use of the "squadron" approach to relief pitching. If the Tango Love Pie™ adherents have their way, and starters get pulled in an uber-Sparky Anderson scenario just after their second time 'round the batting order, we'd see this figure jump a good bit higher, probably somewhere close to 40%. That would tend to knock down the WPCT in NSV situations, because that seems to be the correlation (2013's highest-ever total of decisions produced only a .551 WPCT for the total number of decisions in NSV situations). That'd actually muddy the waters in terms of identifying the value in this data, but that's what happens when modelers go beyond predictive functions and decide to be prescriptive instead.
More data to come in our next several installments. 'Tis a brave new world for "neos"--"neo-traditionalists," that is.