by Frank Schnittger
Thu Nov 6th, 2008 at 11:00:14 AM EST
We've had a lot of debates here about possible Bradley effects, Shy Tory effects , refusals (to be interviewed), Cell-phone only households, early voting, likely voter models, and differential ground games and GOTV campaigns on the likely accuracy of opinion polls. Certainly some of these phenomena may have effected the accuracy of individual polls which have been widely divergent on occasions.
However when you aggregate thousands of polls into one database the results seem to be altogether more accurate. fivethirtyeight.com alone has the results of over 2 Million interviews in its database and trends can be observed where consistent methodologies have been applied even if there remains some statistical noise and systemic bias in particular types of polls.
Pollster.co has just published a comparison of the provisional actual election results with the last opinion polls taken just before the election.
The biggest "mistakes" were in North Dakota and Arizona where there was relatively little polling in the first place. There may also have been a slight last minute swing back to McCain in the most Republican leaning states after the last Opinion polls were taken. Certainly, it is noticeable in the table above, that the more Democratic leaning swing states voted more Democratic than the last opinion polls, and ditto for the Republicans.
It is also important to note that we are dealing with provisional and incomplete vote counts in many cases - with absentee, provisional, and perhaps even early votes yet to be counted. The provisional vote tallies for Georgia and Alaska, in particular, seem suspiciously low (much lower than 2004) with many votes apparently not yet counted or perhaps even having gone "missing". If found, these may yet effect the important Senate votes there and tighten the Georgia Presidential vote considerably.
In general, Obama may have outperformed the Democratic Party with congressional Democratic pick-ups to date (Senate +6, House +19) very disappointing compared to earlier hopes. Democrats may yet pick up Senate Seats in Alaska (after absentee votes are counted), Minnesota (after a recount) and Georgia (after a run-off election in December) and still reach their filibuster proof target of 60. But it will be very tight and winning all three looks like a long-shot.
So perhaps we can now finally put the "Bradley Effect" debate to bed. I hold my hand up. I predicted a 10+ Obama margin of victory based on what I thought were systemic polling errors due to under-sampling of cell-phone only households and a failure of many "likely voter" models to adequately account for the Obama ground game and the level of youth and minority enthusiasm for his candidacy. Perhaps I was just plain wrong. Perhaps the long lines (apparently disproportionately in Obama leaning areas) suppressed his vote somewhat. Perhaps the likelihood of his victory suppressed his vote in later time-zone states. Perhaps there was a last minute swing to McCain by voters worried about an Obama Presidency combined with strong Democratic majorities in House and Senate. Perhaps the early votes haven't been counted in Georgia and he actually won the State or came much closer than reported.
Fortunately his margins were sufficient to overcome any suspicions about the fairness and integrity of the process. Fortunately, the emergence of sites like Pollster.com and Fivethirtyeight.com have reduced the ability of partisan and dodgy polls to influence the narrative of the campaign. Fortunately the emergence of the blogosphere has challenged the dominance of the MSM in framing the political discourse.
My thanks to all of you who have read, recommended and commented on this series of diaries. I've certainly enjoyed following and taking part in the debate. Lets hope the European Parliament Elections in June can generate similar interest and have as great an impact on the future direction of the EU.