UPDATE: The recount process, at least for early voting ballots, includes more examination by human eyes than it seemed to me at first, though it remains true that many, if not most, of the ballots not showing a vote for the AZ-02 race will be recounted by the same scanning equipment used on election night but reconfigured to count only the Barber-McSally race. Unless there are recount procedures not mentioned in the Election Procedures Manual, we will not know if machine-recounted undervotes contained valid votes.
The Election Procedures Manual does include instructions for duplicating early voting ballots that were returned by the machine as unreadable. The procedures for duplicating ballots apply only to early voting ballots because, as the manual states, a voter voting at a polling place on election day "has the option of correcting his or her ballot at the polling place, and those ballots will have been read by the precinct-based tabulation unit" (p. 177). There appears to be no manual examination of early voting ballots that did not register a vote for a given race, but were not returned by the machine as unreadable (for example, the machine will return as unreadable ballots marked entirely by circling candidates names rather than filling in the ovals, or ballots that include marks on the timing marks). More as I learn it. - MM
The recount of the House race in AZ-02 will begin this week. Democrat Ron Barber trails Republican Martha McSally by 161 votes. Arizona election hands don't expect the numbers to move that much, because previous recounts in Arizona haven't changed results even in races were as close as this year's race in AZ-02.
It might be different if Arizona election law allowed for truly useful recounts.
There are, by my count, 2782 ballots cast in AZ-02 that did not register a vote for the Congressional race - way more than the tiny margin of victory. (See here for results and turnout statistics for Pima County, and here for results and turnout statistics for Cochise County. All of the district is contained in these two counties.)
Nearly 2800 ballots with no vote for the Congressional race, and 161 votes separating the apparent winner and runner-up...good thing Arizona has a liberal standard for determining voter intent on recounted ballots: votes marked in a way that makes them unreadable by ballot scanners, but that show clear voter intent (e.g., a circled name, or a clearly attempted erasure of a previous mark), are to be counted in a recount (see Chapter 12, Sections XIV and XVII (pages 212 and 214-15) of the state Election Procedures Manual, which has the force of law).
But here's the kicker: it's all but certain that only five percent of the precincts will be hand-counted. The other 95% of precincts will be counted by the same equipment that did not read valid votes on the 2800 "residual" ballots in the initial count.
The state standards for determining voter intent in recounts will mean nothing for the majority of ballots in AZ-02.
Here's a summary of the recount process in Arizona. State election law requires a random selection and hand count of 5% of the precincts. An expanded hand recount occurs only if two hand counts of the selected precincts differ from the electronic tallies by a number equal to or greater than a "designated margin" chosen by each county's vote verification committee prior to the election (see here; this statute, section 16-602, controls the conduct of both routine post-election audits and hand counts in recounts). However, it's unlikely that there will be enough ballots not showing a vote for the House race in a 5% sample of precincts to trigger an expanded hand count (5% of the 2782 residual ballots is only 139 ballots, which does not translate to 5% of precincts, but the number offers a ballpark figure) .
Experts on electronic voting systems, like computer scientist Douglas Jones, have long called for all undervotes and overvotes in recounted elections to be counted by hand when their number exceeds the margin of victory [this sentence has been edited since the original posting to include the phrase "when their number exceeds the margin of victory" - careless oversight on my part]. Jones stated in a 2010 affidavit:
Even where such [rigorous, "real-world" pre-election] testing is present, I recommend that ballots containing overvotes or that scan as blank be routinely examined by human eyes on the first count.
In case such an examination is not feasible on the first count, I believe that a hand recount of undervoted and overvoted ballots will result in a substantially better assessment of the will of the electorate if the number of undervotes plus overvotes in a race exceeds the margin of victory [p.11 of the linked affidavit].
There's also the question of whether the software used to determine the initial count functioned correctly. The 5% sample is useful for discovering widespread bugs or malware, but vote miscounts could be concentrated in a small number of precincts. That's why
risk-limiting audits, which use a statistical method to provide a known, pre-specified chance of discovering error, are now the gold standard for verifying electoral outcomes. A risk-limiting audit of a race as close as AZ-02 would likely require a full hand count.
Contrast the recount theater in AZ-02 to the upcoming recount of Measure 92 in Oregon, where all ballots will be hand-counted.
There is a real risk that the country's hodgepodge of recount laws, not to mention the still-widespread use of unrecountable electronic voting machines, will create an ugly mess someday. (As of 2012, 25% of the electorate used paperless electronic voting machines that do not permit a meaningful recount.) We could have faced a mess in 2014: if control of the Senate had come down to a recount in Kansas, we would have been looking at a processing of "recounting" votes cast on combination of machines with no paper trail, machines that offer voters a verifiable paper trail but that many voters do not check, and optically scanned paper ballots. If Senate control had come down to a recount in Iowa, where at least all votes are cast on paper ballots, the decision of whether to count ballots by hand, machine, or a combination of the two methods would have been left to 99 county recount boards.
There are proposals to fix this state of affairs. The Voter Empowerment Act, the benchmark of Democratic election reform proposals, would require voter-marked paper ballots for every vote cast in Federal elections, and mandate that all recounts of Federal elections be hand-counted. The VEA's language on voting equipment and recounts is identical to Rush Holt's Voter Confidence and Increased Accessibility Act. Of course, Rush Holt, who in the past has been able to muster some Republican support for his bill, is retiring, and the Voter Empowerment Act, assuming it is reintroduced, is going nowhere in the upcoming Congress. It may take another Florida to get Congress to pass a law making recounts of Federal elections meaningful and fair.
Many thanks to Verified Voting and Citizens for Election Integrity Minnesota, who provide much of the data used in this diary.
UPDATE: Via email, it's been called to my attention that AZ law does require that automatic tabulating equipment used in recounts utilize "programs" different from the programs that were used in the initial count (see here). However, the interpretation of the law is such that counties use the same equipment used on election night, but configure the equipment to count only the recounted race. Changing the "programming" in this manner does not make the detection of machine-unreadable but legally valid votes any more likely.
