Reply #138: ah, heck, I don't need that table gimmick (grin) [View All]

[OnTheOtherHand]

Yes, your approach of a 2% random sample plus an additional precinct in any county that was previously missed would work pretty well.

The counterintuitive thing here, I guess, is the same thing that is counterintuitive about polling in general: the effectiveness of the audit depends a lot more on the sample size than on the percentage of the population sampled. So, a 2% audit works better when sampling 400 out of 20,000 precincts than when sampling 140 out of 7000, because 400 is a lot more than 140. (And a 2% audit works a lot worse in smaller jurisdictions.)

You cited, for a 2% audit of 7000 precincts, a 98.65% probability of catching at least one corrupt precinct if 3% are corrupt, 75.86% if 1% are corrupt, 50.78% if 0.5% are corrupt, and 13.19% if 0.1% are corrupt. For 20,000 precincts, the numbers would be 99.99+% at 3% corrupt; 99.97% at 2% corrupt; 98.3% at 1% corrupt; 86.8% at 0.5% corrupt; 33.3% at 0.1% corrupt.

I think all those numbers indicate a very high likelihood of detecting a sneaky theft pattern, which would be something like 10% of votes in maybe 10% of precincts in order to alter the outcome by 1%. At that point, I am more worried about recourse when anomalies are found, and/or recourse for screechingly blatant anomalies that don't happen to turn up in the 2% (or whatever) sample. No way, no how can a sample do all the work. (And of course chain of custody has to be secure, not that it ever is. Heck, I don't see hand-counted paper as a panacea either.)

I haven't tried to figure out precinct counts etc. for House districts, but presumably a higher audit proportion makes sense, depending on the level of confidence one wants.