While PPP's final polls are generally quite accurate, they do occasionally miss the mark. See WA Sen 2010, NV Sen 2010, IL Gov 2010, and their entire last 2010 Alaska poll. In each case they called the wrong winner with an error of at least 6 percentage points. They've had some other large errors that didn't call the wrong winner: MO 2012 Sen (12 points too R), CO 2010 Gov (10 points too "R"), ME 2010 Gov (10 points too R), NH 2010 Sen (8 points too D), ME 2012 Sen (8 points too R), MI 2012 Sen (8 points too R), NH 2012 Gov (8 points too R), etc.
Among polls where PPP missed by 6+ points since 2010, 80% were biased against Democrats/Independents and 20% were biased against Republicans. Of the 21 polls biased against Republicans, 76% were off by 4 points or fewer; of the 45 polls biased against Democrats, 60% were off by 4 points or fewer.
In all Senate, Governor, and Presidential contests from 2010 on, PPP calls the incorrect winner with their last poll 9.9% of the time. Excluding Presidential contests, that rises to 13.7% of the time. Of the 24 races where the results was between +5/-5, PPP missed the winner 25% of the time. The only incorrect call favoring the Republicans was 2010 Florida Governor.
Going back to 2010 (71 polls), the mean final PPP poll missed by 4 points, the median by 3 points. The mean and median bias was 2 points too Republican. Going back through 2008 (102 polls), the mean final PPP poll missed by 3.6 points, the median still by 3 points. The mean bias was 1.6 points too Republican, the median 1.5 points too Republican.
While they do better than other pollsters, they are hardly infallible.
One thing it shows is that they definitely don't have a Democratic bias.
