Data Analysis is Never Quixotic
Sunday, November 1st, 2009I will admit that I occasionally show strong signs of favoritism to a particular news outlet. No, not Fox News. My weakness is for NPR. Fortunately they screw up just often enough to snap me back to the jaded, cynical place I call home. This afternoon I was using my extra hour from the fall-back (did you remember to set your clocks, people living in the United States, except for Arizona?) and the fact that the weather is nice on Nov. 1 in Cleveland to sit on my porch and drink iced tea (it’s not actually warm, I’m just crazy) while listening to the archives of my favorite NPR shows from the week. Then, on the front page, a blurb for this article hit me:
One [poll] released by Franklin and Marshall College last week shows just 40 percent of voters in Pennsylvania believe the president is doing a good or excellent job, versus 59 percent who grade his performance as fair or poor.
There is some important data missing, here, I think. They’ve split the positive responses, and lumped one of them with the one negative response. So we actually have no way of telling if people are overall dissatisfied! Maybe people are very polarized and mostly lumped in the excellent and poor opinion categories, and there’s a reason to talk about the widespread dissatisfaction. On the other hand, maybe everyone save a few have a positive outlook. Since fair essentially means adequate it seems wrong to call it a negative opinion in opposition to viral enthusiasm. So if most people in the second group thought the administration was fair, then the statistic as reported is misleading.
Unlike Don Quixote, if you set out into the data looking for windmills, you will find them.