In the post, Aaron used a proof-of-concept model surrounding the displacement of NFL fans using image-recognition and geo-located posts. At the time, Aaron had to make due with a .gif file I had crudely put together (you can see it half way down the article), but—while there was nothing inherently wrong with it—I knew we could do more.
I’ve challenged myself to create something a little better using interactive visualization of D3.js. I’ve been working to demonstrate the value of interactive visuals when combined with large data sets, and this particular scenario was a perfect example.
What Exactly Am I Looking At?
Using a combination of image-recognition software, geotagged data and social buzz, the W2O Group analytics team determined which NFL fans were “most vocal” online when traveling with their teams to away games. With a focus on social media photo content, the team leveraged Sysomos Discover to analyze Instagram data for the entire 20-week NFL season for all 32 teams, using geotagged image-recognition software to tag every post that contained a team logo. The data was split into teams by week and included one “overall” view for each, generating a total of 672 visualisations. We also overlaid the map with the location of each team’s stadium and synched the data with the official schedule so that that the model highlighted the correct location of every week, BYE week and eliminations. Finally, we dynamically labelled the location where each game was taking place.
So when you select a team (either by clicking one of the helmets, or through the dropdown on the top right), you will be taken to that team’s week 1 data. Here you can see who they were playing, where, and a label where the most fans were found, which was almost always the venue of the game. Please note that the label with the number of fans represents only those who were tagged in the EXACT location of the stadium, not in the peripheral metropolitan area. You can hit play and cycle through all the weeks, select weeks at your leisure or simply go to ALL in the week selection dropdown to view the combined data for a given team.
Finally, you can use your zoom and pan using your mouse—just in case you want to get real up close!
Though there are lots of interesting scenarios to pick—and I encourage you to retrace your favorite team’s fate—we’ve recruited Richard Mather (Analytics Manager, W2O Group) to provide some suggestions worth exploring:
Richard Mather: Analytics Manager, Eagles Fan:
Oakland Raiders’ Fans Are Passionate While the Going is Good…
A dominant team in the 70s and winner of two Super Bowl titles in the 80s, the Oakland Raiders (and their fans) had been starved for a 21st century playoff run. In the last decade, the team amassed a subpar 60-100 record with zero playoff appearances—except in 2016. So as 2016 began to unfold and the Oakland Raiders emerged as playoff contenders, the fans followed.
The Raiders’ fan base was also one of the most travelled audiences of the 2016 season and, while that seems surprising, the phenomenon may be attributed to fans jumping on the playoff bandwagon. Looking at the visualization, Raiders fans’ average attendance for away games rises as the season progresses. The data shows that for every game the Raiders separated themselves from their divisional competition, conversation grew by 218 fans. Unfortunately for the Raiders, their star quarterback got injured in the second to last game of the season, which caused a sharp decline for week 17. Better luck next year!
Feast and Famine
It is not all misery in the NFL, especially for Patriot fans. However, with such a stellar record, winning is less exciting. Looking at the chart below, we see that the Patriots fans, even with a seven-game winning streak to close the season, were not nearly as vocal down the stretch as Raider fans. With a 126-34 record and nine playoff appearances over the past 10 years, it is understandable how Patriot fans may have become a little complacent.
The Effect of Disappointment
With a rookie QB and what looked like a strong defense, the Philadelphia Eagles had high hopes for their 2016 campaign. This turned out to be fool’s gold, and around week 11, their playoff chances began to diminish along with the travel from their fans. After week 11, the average volume of online road game fan support dropped by 15%.
Some Thoughts on Geotagging in General
Geotagging is an incredible technique for stitching online data into the physical world. There is something inherently fascinating about conjuring a concrete location for these data.
Concentration Versus Volume
Sometimes, a particular segment of data and its specifications can obfuscate other portions. Geotagging and analyzing at a location level can help remove bias and group data around a new structure. As we can see from the analysis above, just because a team has a lot of fans, it doesn’t mean they all travel.
Superposition of Data
A lot of additional data can be found on a location basis. The government’s census data and other sources can be superimposed onto geo data and compared alongside it. It is relatively easy to bring up a layer denoting house-prices, salary or ethnicity from census data to enhance our data. Parting from geolocation can provide some pretty powerful demographic detail, if wielded correctly.
Time-Frame and Location Analysis
It goes without saying that geolocation is greatly enriched when juxtaposed with time data. These techniques can be used together for extremely detailed event coverage, effectively geo-fencing an area and monitoring activity (stay tuned for some examples of this in our next blogpost).
Thoughts on Interactive Programmable Visualizations
It is worth highlighting that these types of visualization techniques represent a breakthrough in data representation. Not only are you able to craft an interface through which to render complex datasets that are approachable and attractive, but—through interaction—you allow your audience to chart their own course through the data, making the experience much more personal.
Furthermore, once the initial interface has been cast, appending new data is effortless. It may have taken me hours to put together the initial model for Green Bay, but adding the remaining 672 configurations took a matter of minutes. Take a step back and look at the data acquisition, all done through algorithms and API calls, and you can see how repeating this exact exercise next year could be done in mere moments. A few years ago, this would have been a painstaking process requiring coding and processing in huge quantities, with each additional data set increasing the amount of data required.
Progress in data visualization been so intuitive as to seem effortless, but it is worth reflecting on just how the barrier to achieving usable large-scale data sets continues to lower. Making insightful use of such data, now there’s the rub!
So there you have it. I hope you enjoy messing around with the data set and following your team’s performance through the season. Stay tuned for more interactive charts in the coming weeks!