I find myself more and more struggling over defining break points when displaying choropleth (aka thematic) maps to view by others. Does anyone have any suggested references that help guide, both how to choose the type of scale used and the appropriate number of break points? In particular for the number of bins I have only ever seen arguments for a limiting number (e.g. you should not use any more than 5).
To be more specific about what I am looking for, most references I have come across about the subject are similar to the document referenced by julien in this post, and I'm just looking for a more in-depth discussion about the topic.
A few specific use cases I come across frequently (for examples of my struggles);
- When displaying data that has a large right skew, I'm typically hesitant to display an exponential scale. I fear (for the audiences I am typically displaying maps to) this would cause a greater amount of cognitive burden reading the scale and mapping actual attribute values to the colors. Are my fears incorrect? Also for these types of distributions I find it difficult to justify any particular number of bins.
- When displaying many small multiple maps, how do I choose an appropriate scale that allows one to visualize relationships effectively both within and between the small multiples? My de-facto standard when the attribute scales vary to a great extent is to use quintiles within each seperate distribution. Are quintiles too many classifications and creating too great a cognitive burden to compare between the panels? I assume people understand quantile classifications are equivalent to rankings (and thus when classed that way aids in interpreting between panels), is this assumption correct?
I initially wrote a paragraph trying to describe the goals of such maps, but I suspect my goals are pretty typical so it was unneeded. The only thing to clarify again is that these are for viewing by other people (like in reports, publications) and are not really for my own exploratory data analysis (although I would suspect good advice should translate to either). Perhaps a good reference may describe the potential goals of such maps, and trade-offs associated with using different classification schemes. I would be interested in both specific and general references.