(Reposted from downstairs in case GDF is in here now)

re: #254 goddamnedfrank

One weird assumption that polls make is that admittedly bad data can be stacked to produce nominally larger and more trustworthy samples. Look at the CNN/ORC poll, there’s only two age categories, under 50 and over 50. Now the under 50’s have a margin of error of +/- 8.5, which is the very highest their stated methodology allows before they’ll simply not publish the number because it’s too unreliable. Now their over 50 sample has a MOE of +/- 4.5, meaning that they oversampled the hell out of the over 50 crowd vs. everyone younger. However when it comes time to publish results they add all the ages together and claim an overall margin of error of +/- 4. In other words the sheer badness of their sampling of younger voters is forgotten and the overall sample is given an MOE @ a so called 95% certainty that’s nominally better than either of the constituent results.

One thing I’m not getting from these crosstabs:

In the intro, it says the Party ID for the LVs is 32%D, 32%R, 36% I (which is already sketchy).

But in all the questions where they break it down between parties, the Republican MOE is 6.0%, the Independents MOE is at 6.5%, and the Dems MOE is at 7.0%.

So….how does that happen when there are supposedly the same amount of Dems and Repubs, and more indies?

Looks to me like CNN is just really bad at conducting polls.