Fordham University political scientist Costas Panagopoulos has compared the last pre-election national poll from each of the pollsters available in the Pollster data to the estimated national popular vote. He writes:
For all the derision directed toward pre-election polling, the final poll estimates were not far off from the actual nationwide vote shares for the two candidates. On average, preelection polls from 28 public polling organizations projected a Democratic advantage of 1.1 percentage points on Election Day, which is only about 1 point away from the current estimate of a 2.2-point Obama margin in the national popular vote (Obama 50.3% versus Romney 48.1%).
Following the procedures proposed by Martin, Traugott and Kennedy to assess poll accuracy, I analyze poll estimates from these 28 polling organizations.
Most (22) polls overestimated Romney support, although some only slightly, while 6 overestimated Obama support.Most (22) polls overestimated Romney support, while six (6) overestimated Obama strength, but none of the 28 national preelection polls I examined had a significant partisan bias.
the data above a simpler quantity than what Panagopoulos calculated, which conveys the thrust of his analysis. If the overall picture changes dramatically as the votes are counted, we will report back. I also hope to have something on the state polls soon.
(Errata: This post has been corrected to account for my misunderstanding of what Panagopoulos had done. The data in my chart are not the basis of his analysis that I quoted above.)