Published online by Cambridge University Press: 02 September 2013
In a recent paper in this journal, Butler and Van Peek (1990) advocated analyzing patterns of electoral change in the USA with the swing statistic—the average change in the share of the vote won by two parties contesting successive elections. The generality of their case was countered by Rose (1991), who showed that most liberal democracies do not have two-party systems; thus the swing statistic, which involves comparing the performance of two parties only, conceals more than it reveals of the pattern of electoral change in most situations. His preference was for the separate study of the “ups and downs” of each party.
In a riposte to Rose, Gibson (1992) argued for the superiority of swing over the single party measures. His case was built on a curious argument, however. On the basis of goodness-of-fit statistics, he showed that a combination of independent variables relating to the characteristics of constituencies in Greater London predicted variations in swing better than they predicted variations in the performance of individual parties. Such analyses in no way indicate the superiority of swing as a measure of electoral change, however; they merely demonstrate that swing is more closely correlated to some independent variables than is another measure of change. Gibson's argument is a case of a spurious correlation.