Case Study: The Triumphs and Pitfalls of Data Visualization as Narrative

A project by Anthony Bushong.

Introduction

This case study seeks to evaluate the process of creating data visualizations in order to outline the strengths and weaknesses of using graphics to provide narratives for data sets. Using the data viz platforms Many Eyes and Tableau, data sets from a gaming platform in Professor Susan Lohmann’s Political Science 115D course at UCLA will be used for evaluation and experimentation. In this class, students participated in online surveys and games in competition for gaming points. Their profiles, though anonymous, did record personal traits such as political orientation, race/ethnicity, gender and other categories that allow for the data to be examined for patterns and provide for arguments to tell a story. For the purposes of this study, Many Eyes and Tableau will be used to tell these narratives. Both their abilities as a software and the nature of narrative through data visualization will be discussed.

The Coin-Tossing Game 10x

In this game, students were told to flip a coin times. If the coin landed on Heads, students would receive 10 points towards their profile. If the coin landed on Tails, they would receive 0 points. The probability of getting heads on a coin flip is 50 percent, and the probability of getting tails on a coin flip is 50 percent. However, the results from the students did not reflect that probability. Examining the idea of what In this game, the user’s profiles gathered their race/ethnicity, political orientation, religion and gender.

ABCSTableau-1Above is Figure 1.1. Perhaps the most basic of visualizations, this graphic was created in Tableau and effectively tells a story from the game. By breaking down the students into political orientation and race/ethnicity within those political orientations, the probability of each group flipping heads according to the data collected in the Coin Tossing Game 10x is displayed. It is interesting to note that moderately liberal asians were closest to the probability of flipping heads at 53.53 %, compared to libertarian asians flipping heads 78.57% of the time. However, the narrative that is told within this infographic is narrow while the aesthetics are plain. Why specifically would one want to gather data on these specific groups? Tableau gives the user the option to simply create a table, but what benefit comes with that over simply creating the table in Excel?

ABCSTableau-2

Figure 1.2 addresses the issue of poor aesthetics, but perhaps over simplifies the question of narrow data by providing a very generic look at how each race/ethnicity reported their Coin Toss results. It is easy to see that the “asian” category reported heads 62.2% of the time while the “white” category reported heads 71.912% of the time. Tableau allows for not only arrangement of the x-axis by their respective weight in the y-axis, but also allows for a easy access to labeling data. This is issue of flexibility is one that escapes Many Eyes’ interface.

ABCSMany-Eyes-1

In Figure 1.3 is a graphic made in Many Eyes. In this graphic, the same data is being shown. However, there are issues with using Many Eyes to interpret your data. 1.) by pasting .csv files into an open text box rather than directly uploading the dataset, there were a couple rows that were missing from the final dataset. This reflects a lack of compatibility and accuracy. 2.) In this y-axis, the groups were automatically ordered by their appearance in the spreadsheet. Many Eyes is not friendly for users who want to make manual changes within their platform; rather the user must work around Many Eyes’ interface to get some resemblance of what they want the graphic to look like.

ABCSTableau-3

Figure 1.4 above shows the exact same data as Figure 1.1, except now we are provided with an easy to read visualization. This table shows the strengths of visualization as narrative, we can quickly read all the data at once versus having to refer to the numerical value of data. This table quickly shows each group’s probability of reporting heads, and is easy to read. However, when the big picture within visualizations are successful, then smaller issues such as making these groups stand out with things like color come into play. If color is going to be used, how will each shade be decided upon? How do you keep color from distracting the audience from what the data set is trying to say? These seemingly arbitrary decisions actually play into how the graphic is read and received. Figure 1.5 below displays this same graph with color brought into the mix. Which is more effective and why?

ABCSTableau-5

 

The Case of the Speluncean Explorers

In this game, students were exposed to a fictional tale in which five explorers were trapped without food, under the impression that they would starve to death, four of the explorers killed and consumed the fifth explorer. Students were then to decide the fate of these explorers, choosing between community service, one year in prison, ten years in prison, twenty five years in prison, life with parole, life without parole, death, or no punishment. In this data set, the students’ gender, race, religion and political orientation were all recorded in addition to their verdict.

ABCSTableau-9

Figure 2.1 displays a heat map that shows students’ race/ethnicity and their corresponding verdict. This heat map, while showing the mark of each square next to it, is not very effective in that it does not quickly deliver data to the viewers’ eye. Though perhaps more original and maybe aesthetically pleasing than a simple bar graph, it is nowhere nearly as effective as giving the user a quick report of the data, forcing the viewer to parse through the information themselves. Nevertheless, it is interesting to see that the “white” category were much more apt to giving no punishment than other major categories, at least according to this info graphic. Despite the heat map, “middle eastern” students were much more likely to give no punishment, with 6 out of 9 giving none. This heat map misses out on that.

ABCSTableau-7

Figure 2.2 displays a variant on a bar graph, and delivers information quick to the eye. With colors organized the same in each bar, we are able to tell which political parties comprise each verdict. As we see, moderate liberals comprise most of the “life with parole” verdicts, while zero conservatives took part in giving the explorers community service.

Competitive Public Goods Game with Jets and Sharks

In this game, students would be randomly placed into teams of ten. Each student begins with 10 points, and without knowing who their teammates are, must decide if they want to keep those points or share them with the group. They can either keep 10 points, 5 points or 0 points. If they give points to the group, those points are then multiplied by 5 (i.e., if someone gave all of their ten points up, the team would receive 100 points). Each team’s final score will comprise of the number of points their members gave to the team. This final score will be randomly pitted against the final score of another team. The losing team’s members will get no points except the points they kept for themselves. This game reflects the ability to coordinate with a large group of people versus the desire to act selfishly in order to still receive points should the team lose. Students reported their donation or lack thereof as well as their political party and gender.

ABCSTableau-10

Figure 3.1 shows how data visualization can quickly and effectively compare data between two groups, in this case, “male” and “female”. Here, we see how much the average male student gave compared to how much the average female gave to the group. However, it is dangerous to use one visualization to show two things — by pairing the data for points given and points kept, two things that shouldn’t necessarily be displayed side by side, the graphic can be confusing. Nevertheless, this visualization still displays the similarities between the actions of male and female students.

ABCSTableau-11

Figure 3.2 uses the same side by side graphic, this time comparing how much winning democrats and republican students gave compared to how much losing democrat and republican students gave. What is interesting to note here is that giving points perhaps was not as much as a partisan issue after all. Rather, there were republicans who donated just as much as democrats. Even so, this graphic fails to let us know how many republicans won and how many democrats won. This demonstrates the importance of always knowing the sample size that you are working with, as simple bar graphs can be very misleading.

ABCSTableau-13

Figure 3.3 shows the same data as 3.2. This graphic, however visually pleasing it may be, is far less effective, begging the question: what is more important, style or function? It must be established that data should come before the aesthetics of the visualization.

ABCSTableau-14

Figure 3.4 exposes the flaw of 3.2 and 3.3, for they are misleading in that they make it seem as if there were the same amount of democrats and republicans. Nevertheless, there were more winning democrats than losing democrats, while there were more losing republicans than winning republicans. This displays the need to use and juxtapose several graphics together rather than using one graphic that tells a narrative that might not necessarily be true.

(2013, Anthony Bushong)

Copyright © 2014 - All Rights Reserved