Major polls forecasted a hung Parliament, underestimating the Tory vote by an average of 4.2 points and overestimating the Labour vote by an average of 2.4. Pollsters changed practices viz. British Polling Council recommendations. Is there now a consensus on :
What they need to do differently to make more accurate predictions for the 2017 election onwards?
A few fault points in polling.
Sampling: most polls used voter rolls and telephones (land line) to reach its samples. It is unlikely to represent the actual voters.
Reweighing: as the samples don't quite represent the actual voters, pollsters reweigh the samples results to mimic the voting. This practice is both controversial and necessary. Rarely disclosed Reweighing is considered a fraud by some and secrete sauce by others.
Voter turn out: tough to predict who is going to show up at the voting booth. Young voters are notorious for this.
Voters not willing to tell the pollsters the truth. Pollsters are generally not well likely. In races where some candidates are disparaged, voters may not be willing to say who they are voting for, or may intentionally mislead the pollsters.
Biased pollsters: the pollsters may be subject to own biases and see the results not fairly, subconsciously or consciously.
Echo chamber: some of the pollsters aren't that good to begin with. So they mimic the "herd" which forms a self fulfilling phenomenon.
Polling business model; the pollsters have a challenging task at hand. They want to be right to be credible. Yet, they want to sell their results to their customers - mostly newspaper, television, ..., mostly left leaning media outlets. You will observe that polling usually favors liberal candidates and the gap narrows towards the voting date.
.....
In sure you can come up with additional reasons as to why polling isn't reliable. That's why we haven't replaced actual voting with polling.
The British Election Study commented on your first question.
In essence, pre-election polls got too large a proportion of responses from voters, hence over estimated the turnout, especially for young people who are more likely to answer surveys but less likely to turn out to vote than other groups.
The British election study had a survey of over 3000 adults which was pretty accurate as opposed to other polls, which suggests that the other polls were not nearly random enough to be accurate. So to make more accurate predictions for the 2017 election the people that they poll need to be selected more randomly.
