The change in the CBA which now will end the experiment to make the outcome of the All Star Game determine home field advantage in the World Series has met with general acclaim; this is mostly because people didn’t like the idea in the first place, not because it’s being replaced with something better. It’s being replaced with better regular season record, which is simply not a very good idea at all.

Let’s start with the fact that it doesn’t matter very much. My previous statistical exploration in Braves Journal was devoted (unsuccessfully) to explaining why baseball’s home field advantage is so small, much smaller than any other sport. That series focused on the regular season, but it’s not much different in the World Series. The AL won 11 of 14 “This Time It Counts” All-Star Games, but only 6 of the resulting 14 World Series. That’s barely even evidence; in fact they won just over half of their advantages (6 out of 11) while the NL won all three of theirs, but that is entirely consistent with the advantage being worthless.

For Braves fans, 1991 still sticks in the craw as the World Series Determined by Who Played At Home and The Hulk Hrbek And Those Stupid Baggies In The Outfield, but for every 1991, there’s a 1996: home teams went 1-5. And, while not a very good measure, home teams in Game 7 in the ASG-determined years are 1-2.

That said, on the assumption that fans of a team want the 4-3 advantage, there is a case to be made that the better deserving team gets the “advantage.” There might even be some extra revenue in it.   People seem to forget that the old system was for the home field advantage to alternate. The NL got the odd years and the AL got the even years. Nothing a team did determined whether or not they got the advantage. That said, ever since the playoffs began in 1969 (remembering Braves debacles seems to be my specialty in this essay) the better regular season record got home field advantage, so it was somewhat natural to think of carrying that system into the World Series.

But in an unbalanced schedule, it really makes very little sense, beyond the fact that it’s easy to calculate. After all, a coin flip is pretty easy to calculate too. Teams with better records are very often worse teams. This is particularly true when the records are close. I don’t think this surprises anybody, but we put it aside in determining, for example, who gets in the playoffs except when we want to argue about some really good team that didn’t get in.

And to be honest, I think the quest to get the best team is a little silly anyway, so I’m OK with using a not-particularly-good index of goodness to measure it. But for those who want a better measure, they abound, and we don’t use them for three reasons: (1) they’re more complicated; (2) they are more out of a team’s own control than their own win-loss record; and (3) people don’t care enough.

A simple robust measure is a Bradley-Terry ranking. A variant of this (ELO ranking) is used to rank chess players and it is pretty standard in comparing college hockey teams, where it goes by the name KRACH, which stands for Ken’s Rankings of American College Hockey, after Ken Butler who first used it in this way. His original explanation of how it works is clearer than I’m going to be here, so people who want the details can go there, but I’ll give a little flavor here, for the MLB version. In Ken’s spirit, I’m naming it JOBA, for Jonathan’s Overall Baseball Assessment. I chose this name because it will take the world by storm, be attacked by a swarm of small insects (metaphorical critics) be ridiculously overused and then fall into obscurity.

JOBA is a vector of values, one per team, which summarizes their chances of winning every head-to-head match between the two teams. If The Braves have JOBA-value B and the Mets have JOBA-value M, then the chances that the Braves will beat the Mets in a head-to-head match is B/(B+M). That’s it. We then pick the 30 JOBA-values to best explain how teams did head-to-head against each other. In fact, what we do is pick JOBA values that get the aggregate win-loss numbers for each team exactly right given the schedule they played. The only data you need is the 30 x 30 matrix of head-to-head wins. And the programming to get the ratings is not that complicated.

So while all JOBA does is recover the exact win-loss record for each team, it does it in a way that accounts for scheduling differences. 93 wins is a much better record in a good division (the AL East, last year) than 95 wins is in a division that has some bad teams in it (looking at you, Nationals.) We know this, and we talk about it, but JOBA makes a unique simple adjustment for it that in some ways explains where the won-loss record comes from by eliminating strength of schedule considerations.

JOBA ratings aren’t actually unique. Note in the example above that if we multiply B and M by the same constant we get the same prediction. So while the relative ratings are unique, we can change how we express them up to a constant. I have chosen to make the Atlanta Braves have a permanent JOBA rating of 1000. Every other team is determined relative to the Braves. This decision doesn’t affect the relative rankings in any way. When ESPN takes over this idea, they’ll do something like have the Red Sox and the Yankees add to 100… doesn’t matter. Any single constraint that fixes the level of any one team will do, as will any scaling constraint that sets the total range.

So, based on the 2016 regular season, here are the JOBA rankings:

Team JOBA Wins
  CHC   2297    103
   TEX   2156     95
   BOS   2093     93
   CLE   2078     94
   BAL   1940     89
   TOR   1916     89
   WSN   1836     95
   SEA   1759     86
   DET   1731     86
   NYY   1723     84
   HOU   1681     84
   LAD   1646     91
   KCR   1536     81
   NYM   1526     87
   STL   1525     86
   SFG   1510     87
   CHW   1445     78
   LAA   1362     74
   PIT   1288     78
   MIA   1283     79
   OAK   1205     69
   TBR   1195     68
   COL   1157     75
   MIL   1121     73
   PHI   1061     71
   CIN   1024     68
   ARI   1010     69
   ATL   1000     68
   SDP    987     68
   MIN    923     59

 

First off, while these ratings don’t mirror wins, they go in pretty much the same direction, as you’d expect. (The correlation coefficient is almost 95 percent, for those who care.) And where they diverge (as with Boston versus Washington) they go in exactly the way you’d expect: Boston was a much better team than Washington last year despite having two less wins.

The other thing that leaps out at you in these rankings is how badly the NL sucks right now, top to bottom. Tampa Bay is a pretty bad team, but they are actually slightly better than Colorado, who won 7 more games. Note by the way that the interleague games are telling you everything you can possibly know about the relative strength of the two leagues, but that instead of just looking at the interleague record, they look at whether strong NL divisions played weak AL divisions, and vice versa.

There are another couple of advantages to using JOBA. First, we would no longer care how the interleague schedule works out. Plus, teams would get full credit for how they did against who they played. Second, makeup games at the end of the season and ties leading to one-game playoffs could be played if you wanted to, but you wouldn’t really need to: the 162 game JOBA and the 161 game JOBA are not going to be very different, and you could break regular season ties with JOBA. Even better, you could use JOBA to pick wild card teams. (The only difference last year is that Cardinals would have gotten in instead of the Giants, so maybe records are better: the Cardinals should never get in.)

So there you have it: JOBA. I’d use it for standings, wild cards, and every playoff matchup, but that’s definitely just me. W-L isn’t that bad. But if you think home field advantage in the WS matters, if you aren’t going to use JOBA, go back to alternation. It’s fairer.