Over the weekend Drew Conway posted about a data analysis project he’d just completed called Shades of Time. Very briefly, he took a dataset of Time magazine covers from 1923 to March 2012, then used some Python libraries to identify the faces in the covers and identify the skin tone of each face. The result is a really great interactive visualization implemented in d3.js.

From looking at this data, Drew, with some caveats, observes that “it does appear that the variance in skin tones have [sic] changed over time, and in fact the tones are getting darker.” He also notes that there are more faces on covers in later years.

Why I don’t believe it

There’s no real statistical testing done here–no formal quantification how skin-tone representation on covers is changing over time. Instead, I think he’s drawing his conclusion on the vizualization alone, especially the scatterplot in the bottom panel that seems to show more darker tones appearing later in the date (starting in the 70’s, the skin-tone dispersion in his data starts to increase).

He notes that there are difficulties in both identifying faces and skin tones. After going through his ...