So… if travel doesn’t seem to do much, what about teams? Once we go to teams, we can no longer just look at home and road winning percentages, since those will depend greatly on how good the team was. Instead, we now switch to home-road splits: winning percentage at home minus winning percentage on the road. Franchise-by-franchise since 1966 (now including 2012) here are the teams ordered by mean splits.
There are several interesting things about this table. First, except for teams with very short histories (the Seattle Pilots — SE1, the Washington Senators — WS2, the Miami Marlins, and the current Washington Nationals — WAS) every team has at least one year of negative splits where they played better on the road than at home. I should note here that Retrosheet changes the team name when the team changes cities. This seeems slightly controversial to me in the case of Miami, but it’s only one year. Also, every team has at least one year of monster positive splits.
To make a direct comparison with what we’re calling home field advantage, you have divide the splits in half, since a team with exactly a 54 percent probability at home and a 46 percent probability on the road will have a split of 8 percent. So there aren’t many teams with high splits, really just Tampa Bay, Houston and Colorado of teams with any reasonable number of years under their belts.
These three teams are the poster children for the “stadium effect.” Oddly enough, Houston is the poster child in two very different stadiums, the Astrodome and Enron/Minute Maid Park. But from the standpoint of what we’re measuring that shouldn’t matter, because we’re measuring the ability to tailor your team to your park, and it shouldn’t matter in how many parks we do it so long as the parks don’t change too often.
Even within these findings there’s a lot of variance. No one will be surprised to see Colorado as the highest average split team. Houston is next highest, combining the experience in an extreme pitcher’s park with a pretty friendly hitter’s park. (I haven’t yet looked at these data by park, but I could.) Even Colorado had a year when they played better on the road — but it has an asterisk, since it’s a partial year, 1994.
I invite people to look over this list and form hypotheses about what kinds of teams (or parks) generate high home/road splits beyond Colorado. Given the extreme variance of home-road splits, I suspect no simple hypothesis will survive a statistical test for robustness.
Several people have commented that good teams (or bad teams) may be prone to different behavior in home-road splits. Not so. First, let’s look across all teams and all seasons and plot season winning percentage against home road splits. This is what you get:
There’s no trend to eyeball and the statistical tests back it up: there is a very slight negative relationship, statistically insignificant, between how good your team is and what your home-road splits are. A ten percentage point increase in your winning percentage (a gigantic move) is associated with about a one percent decrease in your home-road splits.
OK. That’s in aggregate. What about team by team? Get out your magnifying glasses (or open the image in a new tab) and take a look:
If you squint, a few teams might seem to slope upward a little bit and a few more teams seem to slope downward, but this definitely doesn’t seem like anything obvious even at the team level. If you run a rigorous statistical test, only Houston (down), Minnesota (up) have statistically significant relationships between winning percentages and home-road splits.
Finally, let me cover one omitted part of the travel hypothesis from the last posting. Maybe the effects of travel aren’t just from travel but from the accumulated effect of being on the road. I took every game by each team and characterized it by its place in the current road trip or homestand. Thus, for example, a team might be in the opening game of a homestand (which would be 1 for the home team) and the 4th game of a road trip. That’s a little confusing, but the following graph is somewhat interesting:
I have truncated the graph at 11 game homestands and 11 game road trips, because there really isn’t enough data to say anything at longer periods than that. On the far left, we see that the first day of a homestand has just over a 54 percent win rate for the home team (we saw that result before.) The first game of a road trip has just under a 54 percent home winning rate (ie a 46 percent road winning rate).
But the next 5 games are at least slightly interesting. Games 2,3,4 and 5 of a homestand are much closer to 53.5 percent, although this is compensated for by game 6 of a homestand which jumps to 55 percent.
While none of these are really significant, it is a sign that something might be going on (although not of gigantic quantitative significance) at the start of a homestand.
Furthermore, at the start of a road trip, after the second day, visiting teams slightly lower the home field advantage on days 3-5, with the most pronounced effect on day 3, which is often the last game of an initial road series, but will also fairly often be the first game of a road series against a second team.
These effects are weak, too weak to make travel much of an explanation of anything, but are (at least to me) interesting nonetheless. It does suggest that if you wanted to bet on a home team, your best bet is the sixth game of a homestand which is also the sixth game of the visitor’s road trip. The best time to bet on a road team is the third day of a road trip, but it’s still not much of a bet.