I have no idea what USA Today is thinking with this graphic. They seem to have taken all the polling data from 2008 for Obama vs. McCain and fit them to some arbitrary function. Never mind the convention bumps or that the lead has switched a few times, what they show is two lines with very little motion, from January to October. For a contrast, look at the supertracker at fivethirtyeight.com. There, the lines help you to see the motion in the polls, such as the sharp changes near the conventions. The USA Today lines simply make it harder to look at the graph.
What gets me about this, though, isn’t the ridiculously bad math at one of the nations top selling newspapers. It’s the caption that appears if you hover over the line for, say, Obama:
This smoothed trend line is mathematically fit to the polling data for Barack Obama, like an average of the blue dots. At least three polls required.
It’s like they’d heard about something called “math” and thought it might be useful, so they broke out a textbooks and discovered that you can draw a line through a bunch of dots, and that there’s a formula to tell you the best way to do it. You certainly can, but just because you’ve invoked the power of mathematics math doesn’t mean that your result has anything to do with reality. The phrasing of the caption makes it clear they want us to believe that they’ve done something rigorous and scientific. Ugh. The point of mathematical modeling is find patterns among the noise; to help us see the underlying structure in messy data. All USA Today has done is to draw two distracting lines which further obscure any interesting patterns that might be present.