http://www.pewresearch.org/2016/01/07/can-likely-voter-models-be-improved/

“In recent years, polling has missed the mark in several high-profile elections, drawing particular attention to the difficulties inherent in using surveys to predict election outcomes. These failures typically result from one or more of three causes: biased samples that include an incorrect proportion of each candidate’s supporters; change in voter preferences between the time of the poll and the election; or incorrect forecasts about who will vote. While not a new concern, the third of these – the difficulty of identifying likely voters – may be the most serious, and that is the focus of this study. Election polls face a unique problem in survey research: They are asked to produce a model of a population that does not yet exist at the time the poll is conducted, the future electorate.”

However, on p. 5:

“The analysis presented here suggests that modeling the electorate is likely to continue to vex pollsters, especially if no official record of past voting is available as an input to the models.”