By Nyah Gray and Miriam Celebiler

A link to our website can be found here.

Objectives

The English Premier League is the most-watched football league in the world, garnering an international audience of around 1.4 billion people annually. Its players are some of the highest-paid in the world, but there isn’t a standard salary or easily determinable, consistent reason for disparities. Our goal with our visualizations was to analyze players’ salaries in relation to their attributes and markers of success, and try to see if there were any significant correlations. We hoped that these statistics would reveal surprising patterns or even more surprising outliers, possibly providing insights into how Premier League success is measured, and what contributes to a player’s value. The intended audience for our visualizations is Premier League fans, ones who are familiar with the teams by logo and with the players by name, and would be interested in the sport from not only an entertainment perspective, but from a cultural and business-minded perspective as well. The information is still understandable and accessible to a wide audience, however, and could potentially be of even greater interest to a non-Premier League fan who isn’t aware of the economics around the sport.

We aimed at presenting our visualizations on a simple, easy-to-follow website, which guides viewers through our findings and analyses with text and images. Where necessary, certain visualizations are interactive, allowing for the viewer to isolate relevant data points and become engaged with the data.

Data we used

For this project, we created a data set composed of the top 22 players in each of the 2021-2022 English Premier League teams. Each data point consists of the player’s name, their team, their nationality, the total amount of minutes they played during the season, their Instagram followers as of November 2022, and their gross salary for the entire season. We sourced the basic details of each player from the official Premier League Stats Centre website, and we sourced their salaries from Capology.com, a website that covers football finance. To find each player’s Instagram follower count, we had to search them up by name and manually check that they were the correct, verified account before inputting the data.

The Tools

In the data collection step, we used google sheets to compile our data set. Sourcing information from the internet, we manually inputted data into our spread sheet. 

Once completed, we brainstormed ideas for how we wanted to represent our data using a pen and paper.

We then ingested the data into Tableau desktop, and used the software to assign the correct attribute type to our attributes, fixing Instagram followers from a sting to a whole decimal and nationality from a string to a geographical region. With our dataset, we created our visualizations in Tableau, using online resources to learn new functions on the software. One of these new functions was figuring out how to use the team logos as marks and another was about changing the coloring of the visualizations. Tableau was the ideal choice for us since it is a strong visualization making software that we already learned how to use. Tableau fit our needs perfectly, though we did have some trouble with publishing to Tableau Public.

After we completed our visualizations in Tableau, we prepared a website on Wix as our final product. The website embeds our Tableau visualizations to maintain interactivity and also allows for text that can guide the viewer through our visual story. We chose Wix because it is easy to use, but some of the negatives that came with using it was that it was hard to work on it collaboratively and it was hard to customize.

Analytic Steps

A lot of our analysis of the data, and therefore the types of visualizations that we would include, had to be done through playing around on Tableau Desktop, and seeing what results we could garner from different iterations of scatterplots and bar graphs. We knew that we wanted to analyze the correlation between salary and total minutes of playing time, as well as between salary and Instagram followers, so we made scatterplots pertaining to those attributes. They somewhat revealed the results that we hypothesized, but not explicitly and didn’t reveal any consistent positive correlations. We included these graphs in our final visualization as they are still interesting and a crucial part of the story we are telling. The analysis of Instagram followers especially was a tricky one, since Cristiano Ronaldo’s extremely high follower count makes him an anomaly that makes the rest of the data appear skewed and minuscule. We thought that this phenomenon was very telling about the “success” disparity between players in the EPL, so we included this as part of our story in the final visualization.

We found that the inclusion of each player as a separate data point wasn’t as easily readable or informative as we had hoped, so we created a more comprehensive scatterplot which took the average salary of each team. Initially, they were ordered alphabetically, but then we had the idea to add another attribute to compare this data to, which is the final standings of the teams, which we placed them in manually. 

We found that the most interesting aspect of our data had to do with salary disparities within teams, or the difference in salary between the highest and lowest players, in other words. To analyze this, we created a box and whisker plot which measured the range of salaries within the teams, and we ordered them from largest disparity to smallest disparity.

Design Process

Once we had all of our data collected, we revisited our proposal and set goals for what we wanted to achieve with our visualizations. Our goals were to see if there were any correlations between player salaries and their time played, as well as how their salaries compared to their Instagram follower count.

As mentioned above, we chose to represent this data using scatterplots. Once we began putting our visualizations for these points together we wanted to be able to visually represent the teams of players to see if that had any impact on the results. At first we used color hue to code circle shaped marks, but since we were working with 20 teams the visualization wasn’t that effective. Instead we chose to use the team logos as the marks, as well as a key for which logo matched which team to the right.

We kept the salary on the x-axis to maintain consistency with the visualization that showed salaries compared with Instagram followers. When designing our website we chose a black theme in order to show that our story is not necessarily neutral. It is shocking and dark. With a black website, we wanted to match the color by making our visualizations dark as well. Using the web to figure out how the change the color of the background, different boxes, the text, gridlines, and more, we finally got the visualization’s shading where we wanted it. At that point, some of the logos were not visible against the black background. From our knowledge of opponent process theory as mentioned in Ware (2008), we knew we needed to change the dark blue Tottenham logo to increase visibility. We used the black-white channel by finding a Tottenham logo with a white background to create contrast and fix our problem.

When inputting the data early on in our project, we noticed a huge difference in salaries across teams. We tried a bunch of different ways to visualize the gap. First we made a bar graph, but we ran into the same issue with coloring as we did in our earlier visualizations. We instead opted for a scatterplot, using placement and size to emphasize the differences in average annual player salaries across teams. We chose placement and size because using Muzner (2015)’s explanation of the effectiveness principle, average salaries are quantitative and area and placement are two of the most effective ways to show that.

Since our visualizations with all the players are very busy, we needed to use interaction to allow viewers to sort the data by team and be see more details on a player using the tooltip. We chose to have players of the same team highlighted on hover because hovering is one of the most intuitive ways to interact with a visualization and we did not want users to skip over the ability to interact. For our visualization of Instagram followers we chose to include one visualization using Ronaldo’s data and one excluding it on the same dashboard. We linked the interaction on the two visualizations so that the hover for teams linked.

Finally, we made a box and whiskers plot to show income disparity amongst and within teams, we chose a box and whiskers visualization because it was the most effective in showing all of the key differences in salaries by team.

There were also several visualizations we made that we did not end up using in our final story. We tried to make a visualization using some of the other attributes we had in our data set. We first attempted to find a link between position or nationality and salary but we didn’t have enough data to draw conclusions. Then we tried to make a pie chart of the demographic of EPL players. The pie had too many slices to make an effective visualization, so we turned to a chloropleth map instead. The final visualization was not that interesting as most of the players are from the UK. Since the map did not push forward our story either we decided to scrap it.

The Story

Once all of our visualizations had been made, we started thinking about how we wanted to tell our story. A website seemed like the best choice because it allowed us to embed our visualizations and maintain their interactive elements. Our website guides the viewer through four visualizations, prompting thoughts and asking questions throughout the experience.

When we did embed our visualizations onto the website, the color of the logos appeared very overexposed and washed out. We aren’t sure why this happened because the visualizations look the way they do in the screenshots throughout this report when we view them on Tableau.

We started off with a short introduction to get the viewer thinking about our topic. Then, we showed our most complex visualization, showing all the players and how their salary compares to their playing time. From there we guided them to another visualization which shows each player individually, this time comparing salary to popularity. At this point we introduce the idea of stardom and the effect it has on salary. After that, we begin to move away from individuals, showing the difference in average salary between teams. This is where we are starting to move more towards the crux of our project. Our next visualization ties the knot, showing a box and whiskers plot that makes the salary disparity within and amongst teams shocking. We prefaced each visualization with a small description of what is being shown. While each visualization can stand on its own, we wanted the story to flow through words and visuals alike.

We end with a conclusion that summarizes the key takeaways and why this data matters. At the very end of the page, we added a note on where we sourced our data from and how viewers can access our data set. We did this in order to abide by Schwabish and Feng’s “Do No Harm Guide,” and be transparent with our data.

Evaluation of Design

Pros

Our visualizations accomplish what we had aimed for, which is to visualize a dataset in ways that compare attributes pertaining to EPL players that may not have been done before. We made sure to capitalize on effectiveness and expressiveness during our design process and feel that the choice to use the team logos for the individual marks was successful and unique, and makes it easier to digest the data and gain an understanding of it even at first glance. If viewers feel compelled to inspect the data further, they can do so by hovering over the points, and the interactiveness adds an engaging element overall. 

Cons

The cons of our visualizations are that some of the data attributes are inherently messy, even if they are technically accurate. For example, the measurement of a player’s minutes on the field may have been impacted by things like injury, and therefore that dataset must be taken with a grain of salt in terms of how accurately it represents a player’s contributions to the team. There is also the con that the logos we’ve used are somewhat complex in their designs, and might confuse viewers at first glance or if seen from afar. In the box and whisker plot, we decided to use hue to encode the teams, and opted for colours that are most associated with the teams, using the eyedropper tool to make them exact. While this is understandable to a well-versed EPL fan, a lot of the colours are quite similar, but we hoped that the inclusion of labels would counteract this con and make it so that each team was still distinguishable.

Also, a con that only occurred once we published our visualizations is that the colours in the logos appear to be quite faded on our website, which they didn’t appear to be during the design process. The screenshots we’ve included in this blog post have the properly saturated colours.

References

Capology. (2022). 2021-2022 Premier League Player Salaries. [Data file]. Retrieved from https://www.capology.com/uk/premier-league/salaries/2021-2022/?fbclid=IwAR0IqWGxxciPJ8DO7_4TwOhV8SumyChvYncsMw-LapSF4IjIJU9vwwkszto

Football TV Audience Figures – Who Is The Premier League’s Most Watched Club? (2022, 

October 21). Sporting Index. Retrieved December 6, 2022, from https://www.sportingindex.com/spread-betting-blog/premier-league-viewing-figures#:~:text=More%20than%201.4%20billion%20people,League%20followers%2C%20with%20300%20million.

Munzner, T. (2014). Visualization Analysis and Design (1st ed.). A K Peters/CRC Press. https://doi-org.ezproxy.library.ubc.ca/10.1201/b17511 

Premier League. (2022). Premer League Player Stats. [Date file]. Retrieved from https://www.premierleague.com/stats/top/players/mins_played?fbclid=IwAR1cKESBknrnz5eBhF_78n7NC8-m6eacrPk0O_3lEAOfw8StxMhiGIK9lIE

Schwabish, J. & Feng, A. (2021). Do No Harm Guide: Applying Equity Awareness in Data Visualization. Urban Institute.

Ware, C. (2008). Color. Visual Thinking, 65–85. https://doi.org/10.1016/b978-0-12-370896-0.00004-4