Do Experts or Collective Intelligence Write with More Bias? Evidence from Encyclopædia Britannica and Wikipedia.


Working Paper by Shane Greenstein and Feng Zhu.  Which source of information contains greater bias and slant—text written by an expert or that constructed via collective intelligence? Do the costs of acquiring, storing, displaying and revising information shape those differences? We evaluate these questions empirically by examining slanted and biased phrases in content on US political issues from two sources — Encyclopædia Britannica and Wikipedia. Our overall slant measure is less (more) than zero when an article leans towards Democrat (Republican) viewpoints, while bias is the absolute value of the slant. Using a matched sample of pairs of articles from Britannica and Wikipedia, we show that, overall, Wikipedia articles are more slanted towards Democrat than Britannica articles, as well as more biased. Slanted Wikipedia articles tend to become less biased than Britannica articles on the same topic as they become substantially revised, and the bias on a per word basis hardly differs between the sources. These results have implications for the segregation of readers in online sources and the allocation of editorial resources in online sources using collective intelligence…Key concepts include:

  • The costs of producing, storing, and distributing knowledge shape different biases and slants in the collective intelligence (Wikipedia) and the expert-based model (Britannica).
  • Many of the differences between Wikipedia and Britannica arise because Wikipedia faces insignificant storage, production, and distribution costs. This leads to longer articles with greater coverage of more points of view. The number of revisions of Wikipedia articles results in more neutral point of view. In the best cases, it reduces slant and bias to a negligible difference with an expert-based model.
  • As the world moves from reliance on expert-based production of knowledge to collectively-produced intelligence, it is unwise to blindly trust the properties of knowledge produced by the crowd. Their slants and biases are not widely appreciated, nor are the properties of the production model as yet fully understood.”…(More)