Voting Method: Paper Ballots Optically Scanned
From Trust The Vote
[this needs to be revised] [this was transferred from the old wiki] Precinct Count Optical Scan (PCOS) is the name for both a type of computing device used in elections, and a procedure for using the devices. When combined with some amount of hand counting of ballots, PCOS forms an important part of a hybrid ballot tabulation scheme in which there is not requirement for trust in the correctness of either the computing part of tabulation, or the human part of tabulation. Each provides a cross check on the other. However, in almost all cases, not all ballots are cast in precincts, because of remote voting, which ranges from full vote-by-mail (VBM) in Oregon, to tightly limited absentee voting for overseas voters or local voters with a documented hardship that prevents their voting at a precinct. For VBM and absentee voting, ballots are tabulated with central count optical scan (CCOS) devices and procedures.
There are many variations on hybrid tabulation schemes, and there probably isn't a one-size-fits-all procedure. Certainly different states and election jurisdictions have and will continue to have their own procedures. But in all cases, for a particular approach to hybrid tabulation to work well, it is very important to have well-defined procedures for how to conduct the separate tasks of tabulation, audit, re-count (if needed), and canvassing, all of which may be needed to end up at the goal of official elected results being certified.
Though many variations are possible, one fairly common approach is to do a rapid machine count to get preliminary results, followed by an audit to statistically assess the likelihood of errors in the preliminary count. The machine count is done with optical scanners in the precincts, and a scanner for absentee ballots that arrive at the county election office. These counts are aggregated to produce rapid vote totals. Then, to check whether the scanners worked correctly, some ballots are hand counted. Precincts are randomly selected until the group of selected precincts collectively have some percentage of the total votes cast, e.g. 5%, on the theory that if a hand count of 5% agrees with the machine count, then it is sufficiently likely that all the machine counts were similarly in agreement with hand counts. Of course, hand counts are not infallible either, so it often requires some effort to understand the cause when hand and machine counts disagree. After the 5% audit is complete, the results are assessed by some formula that determines the acceptable variance between hand and machine counts. If the variance is too large, more precincts are chosen to expand the size of the statistical sample, and the process repeats.
To some, this process may seem more complicated than simply counting all the ballots by hand -- during the time between an election day and the day that a state officially certifies the election. That is, the machine counts satisfy the public and media hunger for rapid results, with a 100% hand count, compared on a precinct-by-precinct basis with machine counts, should confirm the accuracy of the machine counts, or occasionally identify cases where the state canvassing board (the final arbiters of election results) should do more investigation. However, there is typically reluctance by elections boards to even contemplate full hand counts, first because of the cost, but also because of risks. It may seem counter-intuitive to be cost conscious about getting correct election results, but in fact most election organizations in the U.S. are not well funded; a decision to spend more on hand counting, beyond what the state may fund, will mean reduced spending on other county-funded government programs. But in addition to cost factors, there are risks as well. In many elections, there are dozens of contests, each needing to be hand counted, and each one consisting of multiple passes of counting to detect human error. And given the use of multiple counters per pass, and you've got each ballot getting handled hundreds of times, with risks of them being damaged or obscured or lost, and concerns about potential intentional malfeasance being enabled with such a wide scope of physical access.
Bearing in mind that a recount is different from an audit (a 5% audit, or 100% hand count, is something routinely done, while a recount happens in addition when a contest is too close), just consider the effort of recounting a single state-wide race in the Minnesota 2008 Senate race, the amount of time expended, the controversy, etc. -- and imagine doing something similar for every race at every level. So, for those folks who favor 100% hand counts, the real issue is about election reform - changing the way that elections are funded and conducted, so that hand counts can be done with acceptable cost and benefit.
But regardless of how much hand counting is done in a hybrid system, and regardless of the details of the mix of precinct count and central count, the basic principles are the same: paper ballots allow for counting independent of computers; computers allow for counting independent of error prone or potentially corrupt people; neither is perfect, but the combination, each cross-checking the other, can provide a higher degree of public confidence than either approach alone, especially approaches where machine counts are the only option because there are no paper ballots. Real experts and self-styled pundits will continue to debate the merits of the various approaches, but where public confidence in election results is the main issue, the hybrid approach seems to have the most favor and most promise at present. Whether that favor is justified, only some years of experience will be able to tell.
