Thursday, December 19, 2013


Modeling continues to be the "rage" in the post-neo age of sabermetrics, though it is being supplanted by a hybrid approach that is desperate to granularize that process down to the lowest possible level. Such approaches are really trial and error (mostly the latter...), but it's part of a process that is exceptionally eager to leave behind anything resembling traditional statistics.

But what if certain breakouts of traditional data, sub-sets of data that comprise their own niche, can lead us to perspectives and insights that do as much or more than the increasingly hyperbolic uber-modeling efforts? (You could say that DIPS--Defense-Independent Pitching Stats, if you've been down in a bomb shelter for the whole of the 21st century--is modeling based on traditional stats. But what we want to find out is whether there are combinations or selected subsets of traditional stats that can, without recourse to extreme levels of modeling abstraction, tell us things that we don't already know.)

It's just possible that one of these might exist in the form of a stat that's been hidden in plain sight as part of the data set at Forman et fils (and, to defiantly digress for a moment...we wonder if with the departure of Neil Paine for the greenback pastures of the shiny new, this means that we need to lop off the "s" in "fils" again...but we'll let that pass--for now, at any rate).

What's that stat? It's called the "non-save" situation. It captures most, but not all, of the wins and losses that are credited to a team's relief pitchers. We'll define just what that means in greater detail in part two, and we'll go much further in subsequent installments.

For now, however, consider the 50+-year history of wins and losses as they manifest themselves in the "non-save situation." This is no zero-sum scenario, as the chart demonstrates. Since 1958 (which is when the data starts to become complete), pitchers who record decisions in "non-save situations"--let's start using NSV as our abbreviation du jour--have a winning percentage of .574.

Yes, that's for all teams over that 56-season timespan. The average team, over all that time, wins games at a clip that would produce a 93-69 record over a 162-game season. (Of course, no one gets close to that many decisions--and the number of decisions fluctuates from team to team in each and every season.)

The chart also shows the NSV winning percentage of teams that make it to the post-season. If this were a random thing, that WPCT should be equal to or lower than the overall NSV winning percentage.

Of course--otherwise we wouldn't be writing this--it's not. The average figure (which, like the overall NSV WPCT, fluctuates from year to year) turns out to be .637.

There's something odd happening here, and, as our getting-really-long-in-the-tooth pals from Buffalo Springfield would say, what it is ain't exactly clear. We're going to spend some time over the next week trying to clear it up for you, can't say you haven't been warned.