This week, our Digital Methods course examined Data Visualization and how it is used in Humanities research. While it is common practice to create bar charts or plots of data points quickly and easily (often in programs such as excel), researchers often overlook the inherent biases in their data or are unaware of how the visualization of said data can shape the meaning for viewers and/or readers. Therefore it is important to be cognizant of what your data is, how your data can or should be represented, and how it can be interpreted.
One of our readings this week, by Braun (2018), titled “Critically Engaging with Data Visualization through an Information Literacy Framework” provided a clear set of guidelines that I found extremely informative. First, the article really broke down how we approach data, as data looks different to different people and varying definitions of data can inform how data is approached. And if we look at data in different ways, we inherently understand it individually and may recognize divergent relationships and reach inconsistent conclusions. Going back to one of my earlier blogs, this really reinforces the idea that data isn’t everything and models aren’t always perfect – because we, as humans, can’t help but impose our own views and opinions on them.
To help combat this, Braun (2018) introduces a few rules to follow. First, remember that data represented in a chart or graph should illustrate the actual patterns of the data – not patterns introduced by the choice of model. Next, simplicity is key. While researchers can add more information to a model that is useful and demonstrative to readers, there is a certain point where any thing extra takes away from the meaning of the model. Next, patterns can be observed at all levels, both through close readings or distant reading, but it is truly up to the insight of the researcher as to what patterns are important enough to illustrate. Therefore, bias of each individual researcher can really play into data analysis, model creation, and reader interpretation. The researcher has to be careful about what message they would really like to broadcast about a particular dataset or project.
To support his claim, Braun (2018) describes an instance where a group of individuals were asked to create a model for a set of data using a “Choose Your Own Adventure” style framework. Choices included selecting a type of model to display the data (line chart, bar chart, scatterplot, etc.) and how the models should be altered (axis length or inversion, etc.). At the end, users could examine how trends in their data looked exaggerated or minimized purely based on the model choices made. And this exercise really shows how it doesn’t matter what data is used, but just the choices in how it is visualized.
This had made me think more about how I have approached data visualization in my own research (mostly course research papers or posters at this point in my career). It makes me wonder if good practice might be to always make a varying set of graphs and then have a third party observer look over my work – see what patterns or conclusions they are observing and making to better understand how my research is being conveyed to a public audience. Because it’s so easy to get tunnel vision when working with a subject you know very well. Recently, I went back to review my master’s thesis work and realized that some aspects that had initially been so clear to me were not well laid out in my manuscript – I was to close to my research that I didn’t understand how to explain it to someone else. All in all, it really doesn’t matter how well you know a project – what matters are the choices you make to display that data.
