The Monkey Cage

(Be forewarned: a dorky post to follow…)

Via CQ, I came across this Gallup map this week, based on these data. The map captures the percentage who answered when asked “Is religion an important part of your daily life?”

The map is explicitly designed to depict relative differences. That’s interesting enough, but two things struck me as misleading. First, I think the casual viewer tends to gloss over the relative nature of the differences. People look at the map and say, “Wow, Mississippi is really religious, but Washington isn’t.” And then we’re back in the usual caricatures of red and blue states, etc., etc. Promoting this misperception is the fact that more religious states are coded dark, as if they’re “full” of something. But less religious states are colored very lightly, as if they’re “empty.” (It doesn’t help that Gallup’s map fails to specify what counts as “most” religious, “more” religious, and so on.)

But a closer look at the data shows that in almost every state, nearly 50% or more of respondents said that religion was an important part of their life. So all states are at least somewhat religious. What happens if we do a second map that plots (1) absolute levels and (2) takes account of the full range of responses (from 0-100%)?

This map is a bit ugly. Apologies. I’m just getting the hang of Stata’s spmap command. Here, I picked 20-point increments to demarcate different colors. This shows that even in the least religious states, there’s plenty of that old-time religion. No states are shaded with the lightest colors.

But now a different problem emerges, one that affects both maps. The color gradations distort the underlying data. It looks as if something is qualitatively different when two states are different colors, even when the quantitative differences are small. In the Gallup map, Arizona and New Mexico are different colors, but only 5 points separate them. Mississippi and Alabama are the same color, but 5 points also separate them. In my map, Mississippi and Alabama look really religious, but in fact only 2-3 points separate them from South Carolina or Tennessee.

Here’s a dot plot with the original data:

In reality, there aren’t any clear points at which to “cut” the data. There is certainly a difference between Vermont and Mississippi, but any cutpoints would create distinctions between states that are essentially equal in their apparent religiosity.

I’m open to further thoughts on this. These seem to me potential problems in mapping lots of different quantitative data.