Next, on When Graphs Attack…
We’re back for another episode of bad statistics, bad data visualization, and the bad conclusions you arrive at from it. This time from The Economist.
Specifically, their graph of the day:
Their conclusions from this graph?
“Americans work longer hours: theirs is the only rich country that does not give any statutory paid holiday…This work ethic may in turn help explain America’s material wealth. Even adjusting for purchasing-power parity, America generates more wealth per person than all but a handful of mainly oil-rich economies such as Norway.”
Beyond their incomplete data (where is the rest of the world?) and the question of whether or not naive correlations of X and Y variables is worth while, does their data – handpicked, lovingly displayed as it is, actually support their conclusion? Sadly no.
Eyeballing their data, I put it into JMP 8.0 (SAS Institute, Cary NC) and fit a simple linear regression model. Here’s the plot we get, with its line of best fit and a 95% confidence interval:
See that nice flat line? That’s an utter lack of correlation between the number of days off and per-person GDP. None. The number of mandated days off explains a whopping 0.4% of the variability between countries. With a p-value of 0.80 – about as far from statistical significance that you can get. In short, they have nothing to do with each other.
Its sad when even your hand-picked statistics can’t support your point.
Filed under: General | 2 Comments