The verdict is in. This year's 401 contested US House seats resulted in closer electoral outcomes than the 400 seats that were up for grabs in the 2004 election. This means larger audits would have been necessary this year than in 2004 to actually confirm the results independently of electronic vote counting software.

For the following calculations, we assumed a 99% probability of miscount detection (confidence level), 400 precincts per Congressional District on average and the possibility of up to a 20% undetected vote shift per precinct.

Based on these assumptions, in 2004, 57 outcomes (14.3%) of the 400 contests required more than a 2% random audit of precinct totals to be confirmed. This year, using the same assumptions, 99 races (24.7% of the 401 contests) would require more than a 2% audit to confirm their outcomes.

While the above assumptions are not necessarily etched in stone (and do not take precinct size variations into account), it's clear that a small fixed percentage audit is even a worse idea now than it was two years ago.

The "If Margin %" columns indicate the difference in percentage points between the candidates with the most votes and the runners up in each race that determine whether or not an outcome can be confirmed with a given Audit Percentage. Note that smaller audits can only confirm races with larger margins.

Using any of the three Audit Percentages shown (2%, 5% or 10%), the larger number of races with narrower margins this year results in more unconfirmed outcomes than in 2004. In 2006, only 302 outcomes can be confirmed with the 2% audit compared to 343 outcomes in 2004.

From the above table, we can easily see that fixed-percentage audits are not acceptable.

On the other hand, having to audit 33 out of 401 races (8.2%) with more than a 10% hand count is quite feasible, even if a few of these races require full recounts.

For more information on post-election auditing see:
http://josephhall.org/nqb2/index.php/2006/11/26/pea_biblio