Data Visualization in R and Python

13Correlation Graphics and Cluster Maps

Correlation graphics are a family of graphics aimed at showing the possible statistical correlation between variables. With respect to case studies discussed in previous sections, for instance, we may want to know which is the correlation between the hour of day or the month with bike thefts in Berlin. From the statistical correlation index is then possible to analyze the possible cause–effect relationship between two variables. For example, is it true that thefts happen more frequently in certain hours of the day or in certain months? Intuitively we might be tempted to answer positively, but intuition often fails us when correlation is inquired and it is not rare to end up misleading pure chance with causality or imagining a direct correlation between two events when instead they are correlated with a third one (e.g., seasonal phenomena), somehow hidden or ignored.

Data science and statistics have a long history of mistakes of this sort, seeing correlation where there is none because finding causes for an effect is a desire deeply buried into the human nature or, sometimes, just the most convenient answer. For this reason, when analyzing data, one should be conscious of this always looming risk and proceed with extreme caution before stating the presence of causality. Data visualization, as a language for communicating knowledge from data, could also easily mislead an observer, either inadvertently or due to voluntary manipulations, ...