Wednesday, May 6, 2015

Not telling the truth with charts

Yesterday Seth Godin blogged about Telling the truth with charts, but he didn’t really. He began with the second chart from an article by Jordan Weissmann on The Decline of the American Book Lover  in The Atlantic last January. He whined that it presented color coded data from three years (Gallup  1978 and 1990, and Pew 2014) shown in reverse chronological order, and five sets of percentages for number of books read (none, 1 to 5, 6 to 10, 11 or more, and no answer). Seth claimed the 1990 Gallup poll told us nothing, omitted it, and also dropped the no answer category. Then he left out the 1 to 5, and 6 to 10 books read categories. What was left is a comparison of 1978 and 2014 data for number of books read (none, or 11 or more). He showed the following chart, which has serious problems of both form and content:

The form problem is his use of red and green colors, which I have called Christmas Camouflage because some color blind people will see both as olive drab, as shown above in the Vischeck Deuteranope Simulation view at the right.

The content problem is a change of scale between the results for zero books and heavy readers (more than 11 books). For zero books the left bar represents 8%, while for heavy readers the right bar with the same height represents 28%. Also, the left chart runs downward from zero while the right chart runs upward from a different origin. For heavy readers the left bar (for 42%) should be 1.5 times the height for 28% but instead is 2.5 times that height. An honest chart with the same scale and origin reveals that there still is a higher percentage of heavy readers (28%) than non-readers (23%) and it looks like this:

Does the 1990 Gallup Poll tell us nothing? No, it tells us something useful - that the percentages were changing from those the measured in 1978, and so the difference for 2014 is not from Pew and Gallup just phrasing their questions differently. Here is another honest chart including that data:

No comments: