Today in Bad Infographs

The Washington Post has a cool infograph showing the results of a racial Implicit Association Test by state.  I have a couple of problems with it.

First, the population sample is composed of people who went out of their way to take the test on the website, which is a long way of spelling “the population sample is invalid”.  Second, the volume of use of implicit association in psychology seems to be driven more by a compelling story of what it means and psychologists’ sheer joy at having a thing they can measure, in a short lab test, with numbers and everything, than any actual evidence that the story is true.

But those are objections I would bring up to any presentation of this data.  What is bothering me about this infograph in particular is a new and exciting variant on misleading axis choice.  The Post chose blue to represent a lower IAT score (which, if implicit association’s claims are correct, means being less racist) and red to be a higher one. The lowest values are a darker blue, the higher a darker red, with a neutral value being white.  At this point, blue and red are so tightly associated with Democrat/liberal and Republican/conservative that I think using them for anything else is manipulative.    But having white as a middle color also strikes me as weird.  Wouldn’t an even mix of the two be purple?

Worse, the “neutral” color does not rest on a score of 0, because that is the lowest score on the IAT.  Instead white represents a score of 0.402, which is almost but not quite the middle of the range of state averages (0.341-0.456) It was chosen because it was the IAT score of the median state, Michigan.  The overall effect is that a casual reading of the infograph would lead to conclude southeast and eastern states are racist and New England and northwest states are anti-racist.  In fact, if you treat IAT score as a quantity, the most racist state is about 33% and 10 percentage points more racist than the least racist one.    We don’t know what that corresponds to in actual behavior- does 10pp translate to a 1% difference in likelihood of hiring a black person, or 50%?- but that makes the color choices more misleading, not less.