Daniel Gayo-Avello has strong feelings about the unpromising record of efforts to predict election results using Twitter data.
“No, you cannot predict elections with Twitter”
And shows that, despite various claims by academics in computer science, no-one seems to have actually tried. I don’t know whether the spelling mistake in this sentence – “In all probability this is the paper which started all the fuzz regarding predicting elections using Twitter” – is deliberate or accidental, but either way, it’s delicious. The paper (correctly) points to incumbency as a better baseline than the one used in the most of the papers, but there is a largish literature among political scientists about election prediction too, which the author (and other computer scientists) seem largely unaware of, and which might provide a better (and more demanding) set of baselines for computer scientists to test their own models against. Perhaps as computational social science gets going, we’ll see more convergence of the literatures …
[Cross-posted at The Monkey Cage]