Richard Florida has created quite a brand for himself with the notion that there is a creative class and that the presence of such is the source of all things good in a virtuous circle. While I see the appeal of this argument to a certain type of ‘knowledge worker’ it has always struck me as somewhat tautological. Worse, it has also always struck me as self-serving and elitist in a way that is somehow more insulting than economic definitions of elitism.
Now Dr.Florida has weighed in on the upcoming elections with an analysis over at The Atlantic. Here’s a surprise–the election results will reflect the presence of the Creative Class. You can see in the nifty chart that the correlations between economic factors and the polling numbers are weaker than the correlations between creative class attributes and both poll data and 2008 election results.
Here’s the problem. I’m no PhD but I know enough about statistics to know a lurking problem when I see one. It’s right there at the bottom of the chart. The x-axis measures the Pearson correlation. But like many statistics, Pearson’s r was developed to explain natural phenomena. The ‘book’ on it is that it is “sensitive only to a linear relationship between two variables.” The caution is always to use it only with ratio level data.
I can hear the protests already. All these data are ratio based you say. Sure they are. Take the Gay Index. I’m not sure how that’s computed and many indexes are constructed with a true zero. But why not use a population measure other than the fact that we don’t do something distasteful such as making gay people register as such?
Or look at the religion attribute, We’re told it’s based on the Gallup data for the “Is religion important in your everyday life?” question. There are alternate, reputable sources that ask about attending religious services. These are less prone to error from people wanting to satisfy a social norm. Also, they are rooted in behavior and counting which makes them better candidates for correlation analysis.
I’d have prefered this analysis to use an alternative measure of association. I think most probably Lambda would be a better choice because almost all of the creative class attributes strike me as nominal variables that somehow, probably through the miracle of proportions, have been turned into numeric variables. I don’t want to presuppose why r was selected but I have my suspicions.
Mostly, though, I think this is case of fitting the data to the world view. For some reason, we never see data from Florida that suggests the creative class can’t explain whatever he’s writing about. Is it expectation bias? Or confirmatory bias? Take your pick but one can’t help thinking there’s some cognitive bias at work here.
Or maybe it’s just delivering the Florida brand. I understand this is his intellectual progeny. But Adam Smith said science is the great antidote to enthusiasm. For me, this whole creative class thing needs a lot more science and a little less enthusiasm.