What I learned about data visualisation
I went into my MSc in Human-Centred Systems with a lot of assumptions and a healthy amount of scepticism about data visualisations. I was a zealot at the church of Microsoft Excel, believing that data was self-explanatory and charts were a cursory addition to reports and presentations. If people can’t interpret data correctly, that was their problem. Right?
I was fortunate enough to take a class in data visualisation from Jo Wood, who would also go on to supervise my dissertation. His pragmatic approach to teaching helped me understand the theories of data visualisation and how to implement them. My experiences learning from him and researching my dissertation taught me four valuable lessons:
1. Not all sciences are created equal
2. The main benefit of data visualisations is efficiency
3. Data visualisation can help us embrace shades of grey
4. Visual biases can drive cognitive biases
I’ll discuss each of these lessons in turn.
Not all sciences are created equal
The scientific method has given us a structure for investigating cause and effect relationships: come up with a hypothesis, design an experiment and gather information. However, this process has resulted in findings that have been called into question. One study found that there was a large discrepancy in research hypotheses supported between disciplines: 90% in psychology but 70% in space science. Another study found that only 24% of highly cited clinical research studies remained unchallenged by subsequent studies.
Figure 1 — Papers reporting support for the tested hypothesis by scientific discipline (Daniele Fanelli)
Two factors help explain these discrepancies. The first is the extent to which humans are the subjects of studies. We are hugely complex species and, as much as researchers try, cannot control for the vast array of variables that could impact a research study. The second factor is flexibility within a discipline. The more freedoms researchers have to design experiments and analyse outcomes, the more opportunities they have to “hack” their results to support novel findings that get them published in prestigious journals. We thus should consider the extent to which humans are studied and how much freedom researchers exercised when conducting studies before accepting the results of a research study.
The main benefit of data visualisation is efficiency
One day after class, I approached Jo to ask him something to the effect of “why do we bother using data visualisations when we can get to the same outcome using statistical models or numbers in a table?” He agreed that other methods allow people to come to the same conclusions but made the point that data visualisation was the most efficient way of guiding users to those conclusions. Traditional methods of data presentation require us to use our prefrontal cortex but data visualisations are handled by the visual cortex, which processes information much faster. In an age where we are inundated with information, anything that allows us to digest more information faster is becoming more and more valuable.
Data visualisation can help us embrace shades of grey
When faced with complexity, we have a tendency to frame situations as having discrete, often binary, outcomes. However, forcing situations into false dichotomies gets in the way of more nuanced, insightful interpretations of information. This occurs in academic research papers where hypotheses are argued to be either “true” or “false” depending on whether they meet an arbitrary threshold. Researchers and decision makers can be lulled into a false of confidence when the presentation of information insinuates a degree of certainty that is not actually representative of the data. If we allow users to “see” the data, we help users move away from harmful dichotomous thinking and embrace shades of grey.
Figure 2 — Anscombe’s Quartet is the quintessential example of visualisation adding value to statistical properties (Wikipedia)
Visual biases can drive cognitive biases
Data visualisation designers need to be acutely aware that design choices can greatly impact how users interpret information and ultimately make decisions. This is one of the great paradoxes of data visualisation: whilst they are created to objectively encode quantities to visual properties (lengths, widths, colours, etc.), they can lead to cognitive bias because visual properties are perceived differently in our minds. Steven’s Psychological Power Law shows us that there are different relationships between stimulus and their perceived intensity. So whilst we accurately perceive line length, we perceive changes in colour to be stronger than they actually are but perceive changes in area to be lower than they actually are.
Figure 3 — Steven’s Psychophysical Power Law highlights discrepancies between the intensity of real world stimuli and how we perceive the intensity of those stimuli in our minds (Tamara Munzer)
Data visualisation is a unique discipline that exists at the intersection of psychology, design and data science. When they are applied correctly, they can help us distinguish the signal from an increasingly noisy world.