Quick note about bivariate choropleths

Responding to a quick Twitter prompt that required more than the character limit.

Walkability Index overlaid with Median Building Age in Miami-Fort Lauderdale-West Palm Beach. I didn’t know what the relationship between these datasets looked like in this area so identifying a phenomena between the two, sight unseen, would have presented a real challenge for me.

We actually have a long-standing internal debate about bivariate maps along the lines of Andy’s thinking (“I’m not sure the challenge to the reader of a bivariate map is ever worth it.” vs. “It is when the phenomenon is complex enough in the right way.”).

As context for our argument, both sides accept that the purpose of these choropleth maps is identification of phenomena and selection of features (and the communication of those phenomena and features once found). The #neverbivariate camp usually frames their argument as “Bivariate maps are made out of a desire to filter on another attribute. With multi-attribute filtering, you can identify exactly the geographic features and phenomena you’re looking for without the trouble of selecting, configuring, and interpreting a bivariate map.” #teambivariate will usually respond along the following lines, “Sure, that’s what these maps do, but what if the criteria you want to filter on is hard (maybe even impossible) for the analyst to express? Or, what if they don’t know what the filter should be yet because they need to look at the attributes overlaid first so they can start with an intuitive (informed by a visual, in this case) understanding of the two attributes?”

In summary — Andy’s criticism of the interpretation challenge of bivariates is valid, I’m not completely confident there are 2-variable phenomena that necessitate a bivariate map, but it will continue to be an option we offer to our users who want to analyze and communicate visually in that manner. One of our current major goals is to differentiate analysis/identification workflow from the artifact used to aggregate, report, and communicate the findings of that workflow and in doing so, we hope to reduce the necessity of a complex visualization (bivariate choropleths) in communication — we’re cognizant that bivariate choropleths might not be the every audience’s thing.

Bonus points for our bivariates: we have assembled a library of workable color schemes for the user — there are bivariates that die before they get out of the door because of the difficulty in choosing complementary colors. Additionally, we supply correlation as supplemental information when creating/examining a bivariate and hope this can put users on a track to identifying a visual/intuitive correlation in the attributes they’re examining.