Once again, the ICM “spiral of silence” adjustment proved a very good pointer as to the direction of the error. At +2 in the final poll, it suggested a greater-than average shy Tory factor:
At this point I should remind people that ICM (one of only two surviving pollsters from 1992) still deserve a huge amount of credit for their innovations in the 1990s, without which 2015 would have been significantly worse.
The polling errors at the general election seemed to fit very well with those in mid-term elections:
On the polls vs votes comparison, the non-linear relationship continues:
(Continued, please navigate using numbered tabs…)