Do Head-to-Head Regular Season Records Matter in the Playoffs?

Since I’m an obnoxiously determined Devil’s advocate, one of my favorite uses of data is tackling conventional wisdom. For example, one such bit of wisdom that always bugs me is when pundits insist that the best teams are the ones that win close games. In fact, the opposite is true. The most predictive run differential comes in blowouts — the good teams are the ones that are more likely to humiliate their opponents, not squeeze out a close one. This time of year, you start to see a lot of analysis asserting that X team is definitely blessed or doomed come playoff time because of some randomly chosen factor Y. We could do a column a day on these and still have dozens of unwritten pieces by the time the actual playoffs roll around, but let’s focus on a few specific ones, concentrating on who good teams beat rather than how many games they win.

First off, do regular season head-to-head records matter in the playoffs? Since the start of divisional play in 1969, teams that face each other in the playoffs have frequently met in the regular season. Interleague play added eventual World Series matchups to the regular season, and starting in 2023, every playoff matchup will have already occurred during the regular season. Given the sample size of playoff series, if we construct a simple model of series winning percentage that only consists of a team’s regular season winning percentage and its winning percentage in head-to-head matchups, the model horribly inaccurate, with an r-squared of 0.0886 and a mean absolute error of 275 points of winning percentage.

But including head-to-head winning percentage doesn’t really even have a marginal influence on the coin flip; without the head-to-head matchups, the model’s MAE increases to 276 points of winning percentage. Now, a head-to-head record may imply something about a team’s overall strength that isn’t captured in its overall record, but rather than pick up a small sample implication, we can use strength of schedule directly, which does help the model a tiny bit (playoff series are always going to be very uncertain unless we move to best-of-75 series or something wacky).

Just to illustrate this with actual teams rather than a bunch of swanky numbers, let’s look at the 25 most-lopsided regular season matchups of teams that went on to face each other in the playoffs. A team’s expected W-L is derived here solely from the regular season records of the team and its opponents. I ranked the teams by wins-under-even so that the biggest mismatches weren’t littered with a bunch of teams that went 3-0 in their one regular season series:

Three of the four teams that were dominated the most by their opponents in the regular season emerged victorious. The 1983 Phillies, 1988 Dodgers, and 2008 Red Sox combined to go 3-29 against the Dodgers, Mets, and Angels, respectively, yet won their playoff series. All told, these 25 teams were expected to go 52-62 based on regular season records and went 50-64. Once you make the model a little more complex and take into account that the 25 most-dominated teams had a few more road games in the playoffs than they did at home and a slightly worse strength-of-schedule, the records are within a win over their 114 games. I can’t conclusively say that there won’t ever be individual cases where head-to-head record matters, but on the macro level, there are probably a lot of things that are more important in the vast majority of cases.

Another question that comes up this time of year — and what it even means never seems to actually be agreed upon past the moment someone makes the statement — is whether it matters which kind of team playoff teams beat up on in the regular season. Is beating up the best teams or the worst teams predictive?

I started off by constructing a simple model of playoff winning percentage based on regular season records and records against the top quintile of teams (.572 winning percentage or better). I got a mean absolute error of 188 points of winning percentage with this model; it gets slightly worse when you only use the regular season record, with an MAE of 212 points. But there’s a big ‘but’ there. Again, teams that beat the best opponents more often, as a group, had a slightly stronger strength of schedule, causing their seasonal records to underrate their overall strength. As mentioned above, it was capturing strength of schedule indirectly rather than directly. Just cutting it out and using regular season record with an adjustment for strength of schedule improves the model and more than eliminates the value of knowing how teams fared against the best quintile of teams.

As before, here are the top 25 playoff teams by how well they fared against the best teams in their league:

Before we go, let’s look at the flip side, the playoff teams that dominated the weaker teams in the regular season, those historically in the bottom quintile of teams. The cutoff point here is a .429 winning percentage:

In the end, knowing which teams dominated the weak teams did nothing to improve the simple model, adding nothing to our understanding of why some teams win in the playoffs and others don’t (there was no change in r-squared, RMSE, MAE, or, well, anything).

So, we have two bits of information commonly used when talking about playoff matchups: head-to-head record and whether teams are beating up on the best or the worst teams. If you know a team’s overall record — and you absolutely are going to — and a team’s overall strength of schedule in achieving that record, it doesn’t make you better able to predict the coin flip. Even if these data have some hidden meaning on a league-wide level, there are far more important things to care about this time of year, such as whether a team has an ace with a sore elbow or an exploitable platoon split.