The Emotional Highs and Lows of the NFL Season
By Sean Taylor on Thursday, January 30, 2014 at 10:03am
Sports fans—and perhaps football fans in particular—know that supporting a team over an entire season can be an emotional roller coaster. There are moments of excitement, pure joy, and crushing defeat that are all core components of any true fan's experience. It is a fascinating part of human culture that we care so much about our sports teams and let their performances affect our emotions so strongly.
A great thing about Facebook is that it provides a place to freely share our emotions with our friends through status updates. Collectively, these status updates can tell a story about how the millions of NFL fans on Facebook experience the highs and lows of a football season. To measure fan emotion, we anonymously tagged millions of status updates over the course of the season as pertaining to specific NFL teams and then counted the number of positive and negative emotion words (from a predefined list) in each post. The ratio of positive to negative emotion words provides an intuitive and useful measure of how people are feeling about their teams.
To validate our sentiment measure, we looked at all of the games over the whole season to see if it aligns with the outcome of games. This plot shows the average sentiment paths for posts about winners and losers of games in the 2013 NFL season:
Win or lose, fans start using fewer positive words soon after the game begins. On average, posts about winning teams have more positive sentiment after the game than before it, while posts about losing teams exhibit a dip in sentiment. We can also see that people talking about winning teams generally use more positive words before and during the game—probably because winning teams were expected to perform better and usually end up playing better than their opponents. However these lines only show what happens on average. Serious fans know that games can cause a lot more volatility in their happiness! If we look at the recent AFC and NFC championship games, we can see all the characteristic fluctuations that we expect from an NFL game:
The NFC championship was more tightly contested, and was decided on the last play of the game. Seahawks fans had to wait to celebrate their victory until the very end. The AFC championship plot tells a very different story. The Broncos led for the entire game and their fans posted much more positive status updates during and after the game.
Perhaps more interesting than within-game emotion is the emotion that NFL team fans experience over the course of the season. We can compute a 7-day rolling average sentiment measure and see a 5-month story unfold about the two Super Bowl teams' seasons .
This plot shows how even fans of successful teams—each losing only three times in the regular season—go through a range of emotion over the course of an NFL season (see the high quality version). Losses drop the average sentiment significantly while big wins cause fans to post much more positively about the teams. The bars on the bottom indicate relative volume of words. Important and close games tend generate a lot of posts, but so do one-sided games where one team substantially outscores the other. Really important games have a longer-term effect, causing a big bump in team discussion the following day.
Fan sentiment seems to be a good indicator of how well a team is doing throughout the season, as well as during games. But can it predict the outcome of games? The first plot provides a clue. Fans of the team that eventually wins tend to be slightly more positive before the game than fans of the team that loses. This is probably because the best teams have more positive fans and also tend to win their games. In the playoffs so far, the team with the best pre-game sentiment (two hours before each game) has won nine out of ten times. However, when we used ratio of pre-game sentiment scores between the two teams as a predictor for the regular season, we measured an AUC of 0.65, indicating that it's not generally useful (beyond say, their win-loss records). That little caveat notwithstanding, when we look at the Facebook user sentiment about the upcoming Super Bowl matchup, a favorite emerges:
Update: See the story of your team this season! Plots of season-long sentiment for all eight NFL divisions are available here:
Thanks to the entire Facebook Data Science team, most of who provided valuable input on these analyses. Plots were produced using the excellent ggplot2 library for R, by Hadley Wickham.
A great thing about Facebook is that it provides a place to freely share our emotions with our friends through status updates. Collectively, these status updates can tell a story about how the millions of NFL fans on Facebook experience the highs and lows of a football season. To measure fan emotion, we anonymously tagged millions of status updates over the course of the season as pertaining to specific NFL teams and then counted the number of positive and negative emotion words (from a predefined list) in each post. The ratio of positive to negative emotion words provides an intuitive and useful measure of how people are feeling about their teams.
To validate our sentiment measure, we looked at all of the games over the whole season to see if it aligns with the outcome of games. This plot shows the average sentiment paths for posts about winners and losers of games in the 2013 NFL season:
Win or lose, fans start using fewer positive words soon after the game begins. On average, posts about winning teams have more positive sentiment after the game than before it, while posts about losing teams exhibit a dip in sentiment. We can also see that people talking about winning teams generally use more positive words before and during the game—probably because winning teams were expected to perform better and usually end up playing better than their opponents. However these lines only show what happens on average. Serious fans know that games can cause a lot more volatility in their happiness! If we look at the recent AFC and NFC championship games, we can see all the characteristic fluctuations that we expect from an NFL game:
The NFC championship was more tightly contested, and was decided on the last play of the game. Seahawks fans had to wait to celebrate their victory until the very end. The AFC championship plot tells a very different story. The Broncos led for the entire game and their fans posted much more positive status updates during and after the game.
Perhaps more interesting than within-game emotion is the emotion that NFL team fans experience over the course of the season. We can compute a 7-day rolling average sentiment measure and see a 5-month story unfold about the two Super Bowl teams' seasons .
This plot shows how even fans of successful teams—each losing only three times in the regular season—go through a range of emotion over the course of an NFL season (see the high quality version). Losses drop the average sentiment significantly while big wins cause fans to post much more positively about the teams. The bars on the bottom indicate relative volume of words. Important and close games tend generate a lot of posts, but so do one-sided games where one team substantially outscores the other. Really important games have a longer-term effect, causing a big bump in team discussion the following day.
Fan sentiment seems to be a good indicator of how well a team is doing throughout the season, as well as during games. But can it predict the outcome of games? The first plot provides a clue. Fans of the team that eventually wins tend to be slightly more positive before the game than fans of the team that loses. This is probably because the best teams have more positive fans and also tend to win their games. In the playoffs so far, the team with the best pre-game sentiment (two hours before each game) has won nine out of ten times. However, when we used ratio of pre-game sentiment scores between the two teams as a predictor for the regular season, we measured an AUC of 0.65, indicating that it's not generally useful (beyond say, their win-loss records). That little caveat notwithstanding, when we look at the Facebook user sentiment about the upcoming Super Bowl matchup, a favorite emerges:
Update: See the story of your team this season! Plots of season-long sentiment for all eight NFL divisions are available here:
Thanks to the entire Facebook Data Science team, most of who provided valuable input on these analyses. Plots were produced using the excellent ggplot2 library for R, by Hadley Wickham.
