5 Dec 2016
Polls were not wrong with Brexit when undecideds are included
Polls often just give the proportion of those voting for or against, ignoring the undecided (indeed trying to force them to decide). Plotting the poll results leading up to brexit with the % leave vs. the % undecided gives a better picture (above). The red dot is the final result which is pretty consistent with the general trend. The stay vs. undecided graph is very similar (below).
This suggests (but does not prove of course) that if everybody were forced to vote Brexit would have been narrowly rejected. Indeed if the fitted trend line were true, if the undecided (those eligible but not voting) were reduced to 20% (from 27%) the result might have been against Brexit.
Artificially forcing people to decide or ignoring the undecided in polls may not give the most accurate picture. For completeness here is are the poll results (just remain % - leave % adjusted for known biases in online and telephone polls) over time.
If you plot the % undecided in poll sequence order you seem to get a trend downwards, but that is not reflected in the final result (Red dot).