Comparing the average difference between the final result and what the pollster predicts, IBD/TIPP has been the best for three straight presidential elections
Nate Silver concluded that the IBD/TIPP poll was the most accurate in 2012 based on analyzing the poll's results over the last three weeks, rather than just the final result. This makes much more sense because pollsters often tend to herd toward a collective result at the end, throwing off an assessment of their actual polling methods.
Some analysts had criticized the labeling of IBD/TIPP as the most accurate in 2012 because of it's final result, but looking in aggregate at a wider sample, it's clear they deserved the label. Apparently other pollsters don't like it when you limit their ability to pad results for their chosen candidate before they get serious about the final result.