I absolutely agree with the main point of the OP. Nevertheless, it should be noted that in the article referenced as an example (parenthetically, how did that link get through the mods?), the survey was of a sample of 365 tweets. I would think that this is too low a sample from which to draw conclusive results.
