Reply #97: actually, no part of my argument hinges on the MoE [View All]

[OnTheOtherHand]

Everyone here agrees that the discrepancy between the exit polls and the official returns can't be explained by random chance alone. This business about "hav(ing) to keep those MoEs as wide as the holland tunnel" is quite obviously wrong.

But analytically it is essential to distinguish between MoE and accuracy. Sampling error is just one form of error, and you like to talk as if it is the only kind.

And analytically it is useful, at least for the sake of credibility, to make a good-faith effort to calculate the MoEs correctly. Disregarding the design effect fails that test.

Your post seems to assert that E/M's table implies a design effect multiplier of 1.0, but it clearly does not. Look at it harder.

Since the design effect multiplier is incorporated directly into the MoE, it is confusing to say that the factors you cite "make up for" clustering. The MoE refers only to sampling error, which is why it incorporates a design effect correction.

If we wish to assess the plausible magnitude of non-sampling error, a good approach might be to take a dispassionate look at recorded errors in other exit polls. Merely asserting that exit polls are more accurate than telephone polls doesn't help much (especially if you never actually assess how accurate the telephone polls are).