Does Distance Traveled Impact Games?

Apr 6, 2015; Brooklyn, NY, USA; Portland Trail Blazers center Robin Lopez (42) defends against Brooklyn Nets center Brook Lopez (11) during first half at Barclays Center. Mandatory Credit: Noah K. Murray-USA TODAY Sports

On April 6th, the Portland Trail Blazers flew across the country to play a game in Brooklyn. Usually NBA road trips contain several games spread over a (somewhat) travel optimized route, but Portland flew right back home after facing the Nets to play the Timberwolves on the 8th. The April 6th game was a makeup for a weather postponed game[1. We JUST got rid of the snow here in Northern CT.] and the lengthy distance Portland traveled to play this one game quickly sparked discussion on Twitter about how much the distance a team travels to play a game matters. The idea was simple: If a team travels further to play a game, they may be slightly more likely to lose. Makes sense.

To look into this problem, I used game logs from Basketball-Reference and googled the coordinates of each team’s home stadium since the 1985-1986 season. With a location of every stadium, I was then able to calculate the distance in miles for every possible trip on the NBA schedule, and calculate the (straight-line) distance each team traveled to get to the game they were about to play. I Further assumed they were coming directly from the location of last game. I calculated this per-game distance for both the home and away team for every game of every regular season since 1985-86.

The relationship between the distance a team travels to a game and a game’s result is murky at best. For the sake of the analysis, I skipped game logs of the Hornets team that split its home games between Oklahoma City and New Orleans for one season, and filtered out game logs that were among the first 20 games a team played[2. for the sake of a Last 10 games Distance Variable I created]. Splitting my data into ten and nine season timespans left between eight and nine thousand games to be analyzed for each timespan. I evaluated the effect the distance traveled has on a game’s result by making logistic regression models utilizing variables such as the Home Team’s Distance Traveled, the Visiting Team’s Distance Traveled, the Distance the Home Team has traveled for its last 5 games, and the Distance the Visiting Team has traveled for its last 5 games to predict whether or not the home team won the game.

Logistic regression is a statistical methodology that uses numerical, categorical, or binary variables to  predict a binary result (like whether or not a team wins).  The table below represent the results of a logistic regression model predicting a home team win run on each individual variable. Significance levels represent the chance the relationship could be purely random, and thus lower is better. Log odds represent the strength and nature of the relationship. Above 1 indicate a positive relationship, and those below 1 indicate a negative relationship. The bottom and top bound columns represent either side of a 95% confidence interval that the model finds for each relationship. In other words, one can be 95% confident that the log odds are between those two values in the right most columns.

For the most part, these results indicate that the relationship between the distance either team travels to play a game and the home team winning is mostly noise. Eight of the twelve models exhibit relationships that cannot reject the null hypothesis given a 0.05 significance level criteria. Three of the four variables tested are indicated as having both a positive and negative relationship with a home win depending on the year-span evaluated. Seven of the twelve models exhibit confidence intervals that indicate that we cannot be 95% certain that the relationship between the tested variable and a home team win is either negative or positive. There is also little indication of a trend from the oldest year-span to the newest (if for example, traveling for NBA players had gotten noticeably easier). Since, for the most part, the models cannot decide if the relationships are random, positive, or negative, it is relatively safe to assume that there is nothing here.

If there’s evidence here of anything other than noise, these and other modeling results I evaluated seem to indicate that the opposite relationship is more likely to be true (a larger travel distance before a game, the more likely a team is to win), rather than the one discussed during the TrailBlazers/Nets game last Monday. Most likely, this unexpected trend is the result of distance-traveled’s correlation with another, more telling factor.

These results caused me to check my data several times to insure accuracy, and after several double-checks, I can assure you my data is accurate. It can often be fun to play with statistics and uncover the strong, hidden relationships between one feature and another, to get that brief rush that comes with thinking you know something someone doesn’t. It is important to remember that the boring analyses that uncover little or no relationships can be just as important, though they’re not nearly as exciting.

Sorry to disappoint, but unlike altitude and rest, pre-game distance traveled doesn’t appear to be on the list of significantly predictive regular season attributes. It’s more likely that the Blazers lost to an inferior team because they were on an away game without LaMarcus Aldridge.