The past matters: eliminating the pro-Labour bias in British opinion polls

Opinion polls in Britain display a persistent tendency to overestimate Labour's share of the vote in the ballot box. This appears to arise from failure to secure a politically representative sample and to estimate accurately who will actually vote. We argue that, despite some potential pitfalls, polls based on fresh cross-section samples have to engage in weighting by recalled past vote choice and recalled past turnout in order to overcome these problems.