Suchergebnisse

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.

Most opinion polls conducted during British general election campaigns report on each party's estimated national vote share. Although of considerable interest, these data do not put the spotlight on the marginal seats, the constituencies targeted by the parties for intensive canvassing; these are where the contest for a majority in the House of Commons is won and lost. There have been some polls covering those constituencies as a whole, but very few of individual constituencies so there was very little reporting of the outcome for each party in those individual constituencies. That changed with the 2017 general election, when three analysts published estimates on the internet of each party's vote share separately for each constituency and with those data predicted which party would win each seat. This paper explores the veracity of those estimates, finding that although in general terms they accurately represented the relative position of each constituency in the share of each party's votes, nevertheless their estimates of which marginal seats would be won by each were not as accurate. The implications of such polls, especially as their predictive ability is improved, is discussed.