Visualization Critique and Redesign

by Rafe Batchelor

Through a small-scale survey of my peers in our Human Computer Interaction course at Bucknell, the majority of individuals indicated that the best way of visualizing information would either be in the form of a bar graph or a pie chart, as opposed to other, less traditional means of bubble charts or stacked bar charts.

However, the use-cases of a pie chart and bar graph are far from interchangeable; data presented clearly in the form of a bar graph might be unreadable, and frankly, detestable in the form of a pie chart. It is this exact lack of translation between such mediums for visualization that I hope to investigate throughout this article.

What makes a good visualization?

A good visualization relies on an assortment of factors including the veracity of not only the data itself, but the representation of such data; the degree of clarity to which the data is presented; and the intentions of the visualization itself. In regards to the veracity and representation of data, a visualization fails if proportions and scaling is skewed in such a manner that the agenda of the publishing organization is subliminally conveyed. Take for instance an example that we covered in class:

Here, we see a radically different scaling on each of the intersecting lines with a decline of roughly one million cancer screenings, as opposed to an increase of only thirty-thousand abortions over the allotted period. Through this misrepresentation of axis scaling, the audience is led to infer an unpresent correlation between data points, which is precisely the intention of this visualization’s creators. A more accurate depiction of these values would appear as:

where clearly, the rapid decline of cancer screening and prevention services has no clear correlation to the subtle increase in abortions over this span.

However, bias and agenda make only one side of poor visualizations. The other falls on the side of interpretability and aesthetics; if an individual must struggle and pour considerate effort into understanding the significance of the visualization, then the visualization has failed. It is often the case that a visualization struggles to convey a large amount of data involved in a readable manner, as labels and backgrounds and colors become jumbled together. The idea of maximizing the data-to-ink ratio is a general rule of thumb for producing a visualization that captures the desired amount of information, but in a minimalist, barebones, or simplistic fashion. However, maximizing this ratio won’t necessarily lead to a clear graph; rather, it’ll help one avoid the traps that come when attempting to improve clarity in a display.

The particular instance of poor design shown below is a prime example of a failure to apply this principle. This chart displays all countries on earth by their land area.

Pie Chart

So you might be thinking, “Okay, that’s pretty hard to read. But what if I zoom in?”

Well, here you go!

Better? Not only does the larger pie chart on the left require an actual clump of labels to fit enough countries, but it also requires its own, second pie chart to display the remainder of the countries that couldn’t fit in that clump. This second pie chart even has its own clump of labels as well!

Now obviously a pie chart is not the best way to communicate this information. I’d like to believe that this visualization almost exists to convey that exact idea. So let’s take a look at where this breaks down.

The underlying data that this graphic relies upon is land area by country, published by Wikipedia; for now, we can assume that there would be little to no bias or intentional skewing of such data for the sake of conveying some political agenda. I chose to express critique over this visualization because its failures are purely aesthetic. We breach the principles of a maximized data-to-ink ratio clearly by way of having two giant pie charts, each with hundreds of labels that literally cannot be read by the observer. That’s not to say that this visualization has included unnecessary junk, or chartjunk. Indeed, this chart maintains the usage of a limited, unordered hues-based color palette of clearly discernible colors — something that typically improves the contrast and overall digestibility of a visualization.

So the issue at large is: how can we convey a large data set with unique dictionary entries in a way that is clean, readable, and by no means overwhelming? What if we were to display this data in the form of the other most suggested means of visualizing information: a bar chart.

Bar Chart

This bar chart is also taken from Wikipedia. Better right? Well, this is only 18 countries, out of the near 200 available for display. If we were to continue adding countries to this graph, we’d end up with a visualization as unwieldy as the last; imagine the pages of scrolling and the sheer width required to even display a bar for the countries that are below 0.1% the size of Russia. While we would potentially have a more readable chart, where labels could, in theory, be clearly discerned, there would be no way of displaying this chart in any kind of media due to its size. So while we may have improved in the areas of readability that the pie chart lacked, at least the pie chart can fit in the article! Perhaps, in this instance, a bit more creativity is required; perhaps a bar graph and pie chart aren’t going to cut in when a large set of unique properties must be conveyed.

So where can we go from here?

Perhaps a world map of countries would do us better?

Here we see a map where every country has received a unique color encoding.

However, this doesn’t yet convey a clear idea of how these countries are related by size; at this point it is simply a map.

Now let’s add a legend sorted by country size. I’ve only included the 18 countries displayed in the bar graph above:

Now combining these ingredients we get:

Now clearly this map is unfinished, as the legend displays only the 18 largest countries shown in the bar graph, and the colors on the legend are not yet correlated with the color encoding displayed on the map. Additionally, this visualization requires that the viewer manually observe a color of interest and match it to that location on the map, requiring some prior knowledge of geography if they don’t want to be searching for the smaller countries for a long period of time.

However, this method, in theory, does preserve the elements that the bar and pie charts were trying to convey — elements of a sorted list of countries by area, as well as the actual sizes of these countries. Sure, anyone can look at a map and see that country X is larger than country Y; yet by adding the element of a sorted, color coded legend, the viewer has the ability to make observations about country sizes that may have previously gone unnoticed. Perhaps more manipulation can be done to the legend itself to convey a sense of relative sizes by means of adding numeric information pertaining to area in square kilometers. Other functions to be included in the future should allow the user to have some control over which countries they’d like to compare; perhaps a separate panel that overlays the silhouetted borders of two or more countries:

For example, here we’ve placed the silhouette of France within the silhouette of the United States, allowing us to directly compare relative sizes. Granting the user the ability to choose which countries they’d like to compare by means of overlaying silhouettes could be a useful feature that targets why an individual might be looking at such a visualization in the first place.

Visualization Fun

Here’s a visualization that promotes some correlation between two elements that, on initial thought, one might think are uncorrelated.

What do you think?

Data is Personal, Evan Peck: https://medium.com/multiple-views-visualization-research-explained/data-is-personal-what-we-learned-from-42-interviews-in-rural-america-93539f25836d

Look at Data, Kieran Healy: https://socviz.co/lookatdata.html