Wednesday, October 9, 2019

Decisions in visualization should be based on reasons

The New York Times's David Leonhardt has a column titled The Rich Really Do Pay Lower Taxes Than You, meaning a lower effective tax rate. The column is based on a recent book, and it contains a graph that reminded me of something I wrote a while ago about line charts not being appropriate just for time-series data.

The graph displays the flattening of tax rates in the past seven decades. In the 1950s households in the lower income groups paid an effective rate of ~20%, while the richest were hit by a total tax bill of ~70%.

Today the picture is quite different: the line is almost flat, and the top 400 households pay ~23% while the lowest decile of households pays a higher rate of (I think) ~27%.

Responses to Leonhardt's social media postings about the article have been intense: people on the left claim that the flattening of the tax rate curve is unjust (Leonhardt agrees,) while conservatives say that the current tax rate distribution is fair, as everyone should pay more or less the same percentage of their income as taxes. In How Charts Lie I wrote that charts are rarely the last word in discussions about relevant issues, but they can certainly inform them. This chart is a great example of that.

A few words about the design of the graph itself: RJ Andrews, author of Info We Trust, sent me some intriguing suggestions. He worries about the evenly spaced tick marks on the X-axis of the graph. RJ proposes to subdivide the graph into three sections to separate the top 1% and the top 400 households from the other 99%. Here's a quick sketch he drew:


I liked the original NYT graphic, and it didn't bother me much that the X-axis tick marks are evenly spaced even if the last three tick marks don't correspond to income deciles. But I also like RJ's idea. Stuart Thompson reminded me that they tried an alternative design for a similar type of graph in a previous article—and I also like it!


I see good reasons to justify different design solutions in a case like this, so I'm torn. Perhaps this is a good topic for a class discussion or even for a little research experiment: does it matter more to be geometrically accurate? Or is it better to preserve continuity and maybe increase clarity, as the NYT did, by showing all data into a single chart that magnifies the upper echelons of the income distribution? Or could it be that none of this matters, and that all these design alternatives are equally effective and persuasive? Are we facing a distinction without a difference?