Page headlines
Montana’s 2008 recount season concluded yesterday with recounts in state house districts eight and 58. Earlier, Jefferson County recounted ballots in a bond issue election. None of the recounts, each conducted by hand, changed the original outcome of the election. But not all is well.
There were small changes in the final numbers in all three elections. In HD-8, winner Cheryl Steenson’s total decreased by two votes, while defeated Republican incumbent Craig Witte’s total increased by three votes. In HD-58, the winner, incumbent Republican Krayton Kerns gained two votes while Doc Woerner, his Democratic challenger, lost two votes. And in the Jefferson County recount, the law enforcement bond issue, which had passed by one vote out of more than 6,000, one more Yes vote, and one less No vote, were found. The change rates were on the order of one-tenth to five-hundredths of a percent; small, but significant in contests separated by less than ten votes.
In the absence of fraud or serious mistakes, this is what we should expect when votes are counted by optical scan tabulators. There is a high probability of small changes, a low — very low — probability of large changes.
We should not, however, take much comfort in this. In Montana, a candidate or citizen can ask for a recount only if the difference in the vote totals in the final canvass is one-half percent or less. If the difference is greater, a court must order a recount.
That may sound reasonable, but it creates a situation in which fraud or mistakes cannot be easily detected, or even detected at all. If a tabulator is fraudulently or incompetently programmed, the wrong candidate could win by a small margin, but a margin greater than one-half percent. A wider margin in a race thought to be close might raise eyebrows, but a small — and this is the key fact — but unreviewable margin might come as a mild surprise without inciting suspicion.
Montana has no safeguards against this.
Counts by automatic tabulator are secret counts. A hundred officials and citizens may be gathered around the machine, but the decision to recognize a mark next to a candidate as a vote for that candidate is executed inside the machine, beyond the vision of human beings. The officials and observers watch the machine, but unlike the situation during a hand count of ballots, they cannot verify whether the machine recorded the vote correctly.
California, Minnesota, and several other states, recognizing the potential for incorrect machine counts, require a post-voting hand count audit of a random sample (usually five percent) of the precincts. The precincts usually are chosen after the polls close, but before the counting starts, and the choices are kept secret from the counters until the counting is concluded. If the audit detects an error rate for which there is a sufficient probability of a reversal of the election, a hand recount of the election is required. The principle is the same as in a financial auditor’s sampling of accounts. But as yet, there’s not much agreement on an operational definition of “sufficient probability of a reversal”.
A pre-voting test of a tabulator only confirms that the tabulator passed the pre-voting test. It cannot ensure that every vote will be counted correctly. It will flag some problems, but it cannot flag any problems that develop during the course of the voting. For example, a build-up of paper fibers or dust late in the day could cause votes not to be recorded.
But not even a formal post-election audit will uncover all problems. Only a public count of the ballots — a hand count of paper ballots — can provide a sufficiently high level of assurance that an election was free of error and fraud.