Found this pretty fascinating, and I ain't so good with numbers, so any budding statisticians should have a field day with this:
http://www.dailykos.com/story/2010/6...in-plain-sight
Kos had been commissioning R2K polls since the 2008 election season, but recently dropped them after Nate Silver at fivethirtyeight.com basically determined that they suck. (For what it's worth, the R2K polls conducted for the site were always significantly more favorable for Obama than the national polling average.) Not long after, three independent statisticians contacted Kos with deep concerns about R2K's internal polling data. And a few hours ago, Kos just published the exhaustive and damning report on their results linked above.
The short version is that the anomalies noticed would take longer to naturally occur than the heat death of the universe. The easiest one to wrap my head around was the fact that the crosstab splits (M / F) were almost always either both even or both odd numbers, when there is no earthly reason why that should be the case for two independent numbers. It is like flipping heads on a coin 776 out of 778 times, which has a one in 10^228 chance of happening.
So Kos' polls for the last year and a half are bunk, as are the R2K-cited polls in the book he's about to publish, as are likely every poll R2K conducted for any number of other publications. (They've been around a while and were generally well-regarded.) Needless to say, Daily Kos is suing.
It's also worth noting that R2K is/was the most transparent pollster in publishing their internal numbers, which is why Kos picked them in the first place, and is where the (alleged but almost certain) fraud was discovered.
http://www.dailykos.com/story/2010/6...in-plain-sight
Kos had been commissioning R2K polls since the 2008 election season, but recently dropped them after Nate Silver at fivethirtyeight.com basically determined that they suck. (For what it's worth, the R2K polls conducted for the site were always significantly more favorable for Obama than the national polling average.) Not long after, three independent statisticians contacted Kos with deep concerns about R2K's internal polling data. And a few hours ago, Kos just published the exhaustive and damning report on their results linked above.
The short version is that the anomalies noticed would take longer to naturally occur than the heat death of the universe. The easiest one to wrap my head around was the fact that the crosstab splits (M / F) were almost always either both even or both odd numbers, when there is no earthly reason why that should be the case for two independent numbers. It is like flipping heads on a coin 776 out of 778 times, which has a one in 10^228 chance of happening.
So Kos' polls for the last year and a half are bunk, as are the R2K-cited polls in the book he's about to publish, as are likely every poll R2K conducted for any number of other publications. (They've been around a while and were generally well-regarded.) Needless to say, Daily Kos is suing.
It's also worth noting that R2K is/was the most transparent pollster in publishing their internal numbers, which is why Kos picked them in the first place, and is where the (alleged but almost certain) fraud was discovered.
