Exploring constituency-level estimates for the 2017 British general election

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.