The Power of Visualization
Our visual system starts its training pretty much when we are born, learning to detect objects, recognize faces, see motion, and so forth. Not surprisingly, most people are thus much better at seeing things in graphics than in a big table of numbers – which relies on reading the numbers, doing mental comparisons, etc, much higher order functions trained at a later age. But representation of data in a visual form (maps, charts, graphs, etc.) must be reliable. In particular, this means that structures we observe in the visualization should readily correspond to patterns in the data. In fact, this idea lies behind one of Edward Tufte’s principles for good data visualization design: for example, when we use a measuring tape to find the lengths of bars in a bar chart, this should correspond directly to the data value that the bar represents.
But how well can we actually see patterns? Or more specifically, how well can we tell whether one spatial pattern is stronger than another?
We published a paper entitled “Map LineUps: effects of spatial structure on graphical inference” in IEEE Transactions on Visualization and Computer Graphics that looks into precisely this question.
Moran’s I is a formula that allows us to measure the strength of spatial patterns - as shown above. We worked to determine a “just noticeable difference” – in other words, how big the difference in Moran’s I needs to be, before people can reliably tell that one pattern is stronger than another.
We do this in particular for spatial patterns in choropleth maps, and show that the power of visualization varies as the strength of pattern changes. We observe similar patterns as were noted for other types of patterns and visualization methods.