Improving the front page

There are a number of inaccuracies in the front page. Moreover, if the approach were followed as described, it would have big security gaps. For instance, picking the contests to audit based on the margin, then picking the risk limit, is an invitation to use RLAs as a fig leaf, rather than to detect and correct problems.

I think it’s crucial to distinguish between the true margin and the reported margin in discussing these issues.

That’s in part because the description of how the sample size depends on the margin is a source of confusion–evident in comments in last weekend’s VRTF conference. The sample size depends on the method of auditing, the risk limit, the true margin (or true diluted margin, as appropriate), and either (i) for comparison audits, the discrepancies observed [which involves the true errors and the luck of the draw] and (ii) for polling audits, the luck of the draw as well. If every ballot was tabulated correctly, then the sample size for a comparison audit is predictable. Even if every ballot was tabulated correctly, a polling audit has a random sample size.

The INITIAL sample size is generally a guess of where the audit will be able to stop, based on some assumption about either (i) the tabulation errors, for comparison audits or (ii) the true margin, for polling audits.

But the FINAL sample size depends entirely on the evidence. The audit does not stop until either there has been a full hand count or the audit has found sufficiently strong evidence that a full hand count would produce the reported outcome.

1 Like