Shakespeare and principal components analysis

Abstract

Abstract The use of principal components analysis (PCA) in literary studies was pioneered by John Burrows. From him it was adopted by the New Oxford Shakespeare project team, who used it to support their controversial attributions of parts of Arden of Faversham to Shakespeare and parts of the Henry VI trilogy to Marlowe. This essay examines a representative sample of the PCA tests done using function words by that team. It observes that, as with their Zeta method, they used the unsound bisector line method of interpretation. By calculating information which they did not disclose, it finds that their decision to discard all but two principal components was not justified by the data they were using, putting into doubt their interpretations of the results. It shows that their test method is vulnerable by design to the making of false attributions and demonstrates this by a new experiment. By a further experiment, it shows that the PCA tests may be telling us that differences between characters are greater than those between authors, at least in the use of function words. It concludes that the PCA tests done by that project cannot be relied on; and, given the already demonstrated unreliability of their Zeta tests, these new findings invalidate all work presented in the influential book Shakespeare, Computers, and the Mystery of Authorship.