Run Differential and the Mets

A lot has been made about the Mets run differential so far this year, and how it suggests that the Mets should have lost a lot more games than they have.  They’re 32-29, but the Pythagorean formula says that with 262 runs scored and 281 runs allowed they should be 29-32.   They’re at a point where winning a 3-2 game may get them a win, but will hurt their Pythagorean record.

 

Math does not have a memory.  If the Mets play well enough to win a one run game, then they get a victory which suggests a .6774 winning percentage (110 wins) via Pythagorean.  However if they get blown out 14-5, they get one loss but Pythagorean suggests a .1319 winning percentage (21 wins).  Combined that’s a .2195 winning percentage (36 wins).  Take the Mets record over the last 10 games.  They are 4-6, but their 42:37 runs scored to runs allowed ratio suggests they should be 6-4.

 

This is all nonsense really, because I’m just cherry-picking games.  The logic to it all is that overall you need to score more runs than you allow to win baseball games.  There is no inherent talent (besides things like using mop-up relievers that make blowouts bigger blowouts) or ability that allows teams to scatter those runs in a way that groups all of the runs allowed into a couple of games and spread the runs scored more evenly so that they win a lot more games than they lose.  That’s simply luck.  A team that scores 3.5 runs a game but allows 4 runs a game should lose more often that it will win.  For the entire season.  Runs do group though through luck and randomness and late inning replacements in out of reach games, and a near even run differential rarely means a team loses or wins mostly close games.

 

Manny Acosta has given up 29 runs this season.  Last season he gave up only 21 runs all year (although he started in June) to a rate of .4468 runs given up per inning.   If we’d have had that Manny Acosta so far, he’d only have given up nine runs, rounded up.  Acosta is pitching to a 11.86 ERA, which has accounted for a ton of those runs allowed.  This is why run differential is not predictive of how the Mets will do going forward.  The Mets removed a key input, Manny Acosta, from the equation and he won’t be around to give up more runs.  If he does return, it’ll be because he’s pitching better and won’t be giving up runs at the same rate.

 

Let’s compare the run differential if we’d had the 2011 version of Manny Acosta.  Instead of 262 runs scored and 281 allows the Mets would have scored 262 runs and allowed 261.  Aha! A positive run differential.  As I’m sure you can guess, that comes out to a .502 winning percentage via Pythagorean and 81 wins.  The Mets are only three games above that, which isn’t an unreasonable deviation.

 

What run differential predictions forget is that teams and managers make changes.  Manny Acosta is only one example of that.  The Mets will make, and have made, other bullpen changes.  They’ll make offense changes to to try to score more runs.  Ike Davis will hopefully learn how to play baseball again, Daniel Murphy is almost a lock to score more runs than he has been.  The Mets current run differential has roughly zero correlation to what we can expect from this team going forward; it only tells us about what they’ve done so far.