A look at how the final polls fared, with all the vote in and... roughly counted. I can think of plenty of ways to measure pollster error in a multi-candidate race, and it's possible other measures are subtly different, but here taking the rmse of the Sanders - rest of top 4. 
A clear pattern, I suppose: online & live interview with full universe doing well; live phone with a constrained universe of voters (like last two primaries, in case of neighborhood research).
As an aside, I do think poll aggregators could have excluded some of these polls who didn't call the full universe of eligible voters, and we'd be better off. It's not like they'd add a poll of "White Wisconsin voters" to the average, even if they're >85% of the electorate
And some of these pollsters were calling nowhere near 85% of the electorate.
