The author draws conclusions based on a sample dataset of 4000 submissions, whereas currently there are 2,962,457 submissions (stories, askHN and showHN combined). Seems a bit risky at best, unless the sample was designed in a very clever way.