I don't think there was any manipulation (at least in the more relied upon polls like rasmussen, etc. ) honestly I just think that their methodology fundamentally sucked in some way or another. Likely because it couldn't get "clean" results or didn't take a weighting in terms of voter turnout, etc. They might have even had a high sample right but failed to ask the right questions to determine things like voter turnout by race/ethnicity or political designation.
One fun example of shit methodology problems- you take 2 women walking down a street and a pollster stops them but they stop together; and one is an independent and the other one is liberal. One of them hates shitlery and thinks she is a piece of waste, but doesn't want to start an argument with her friend, or admit that she's actually 110% going to vote for trump. So she lies to the pollster to save face with her friend, etc, because she doesn't want to get into an argument with her. You can't get accuracy in polling without complete privacy. So at the end of that "stop" you now have "2 people voting for Hillary" when in reality one of them definitely isn't. So now you have just collected contaminated data. I could go on and on for 3 pages of ways that polling data is likely contaminated.
I also think that polling is likely to get worse as time progresses because of political threats, and the general deterioration of civil discourse, etc, people are going to be more and more afraid to go on the record about who they are actually voting for.
I think a cool exit poll question would have been an anonymously collected whopper- "Did you lie to any of your friends or family about who you were actually voting for because of felt political pressure?"
It would not surprise me in the least if that number is a HUUUUUUGE number of votes, with most of the crossover being on the trump side of the fence.
-Mike