Reply #197: Yes. I haven't been forgetting [View All]

[Febble]

and mentioned that specifically in a few posts. This is what I mean by getting the units of analysis right.

You need to have a confidence level that not more than a (small) proportion of votes have been miscounted:

• On a machine


• In a precinct


• In a county


• In a state

And the appropriate criteria (and therefore the statistical power required) is not immediately obvious to me.

In particular, at machine level there needs to be a reasonable MoE, because no counting method is perfect, and you don't want people faking it to avoid fuss. But as with the exit polls, the MoE is not everything - all machines could be within their MoE but if more swung the count one way than the other, that might also be statistically unlikely, and suspicious.

Also - I don't yet know how this audit is accomplished - are the votes run through the machines that counted them, as a check on the machine? Or through a different machine as a check on the vote? I hope the latter, though it seems the machines also need to be audited, and an MoE established by finding out the range of values it produces on several runs for a known number of votes.

But then I also have absolutely no idea how you audit a DRE - do you hand-count curly bits of paper? Or are they not going to be curly?

(edited to fix formatting)