Over the last two weeks, there has been much discussion of the recounts requested by the Green Party in Michigan, Pennsylvania, and Wisconsin. Some of this request is based on discrepancies in results depending upon the type of voting technology use, but others think that those discrepancies can be explained by the demographics of the counties. Historically, the gap in the states in which the recounts are being requested is larger than the typical swing from a recount.
What these requests do demonstrate is the need for a regular and public audit process for the election. Many states do have an audit or verification process, but it needs to be public. (One of the few states that do have such a process is Arizona which not only does an audit but requires the counties to submit the results to the state and the state publishes that information on-line.) Saying that there should be an audit does not however define what a proper audit should do.
First, a proper audit begins with the voting technology. Today, there are two basic forms of voting technology. (Some small jurisdictions still use paper ballots, but most have moved to some automated system.) My own preference is that the default technology should be the optic-scan ballot. My preference comes from the simple fact that, on the optic-scan ballot, the voter records their own vote. Any mistakes on the ballot are made by the voter, and that ballot can be looked at during the audit to determine the voter’s intent. The only thing that the equipment does is count the vote.
The alternative technology are touch-screen devices. The primary concern with touch-screen devices is that because the device both records and counts the vote a machine error can occur in either recording the vote or counting the vote. One way to reduce this concern is to require a print-out of the proposed vote that the voter can review before finalizing the vote. (While better than nothing, such a system requires the voter to actually look at the printout which voters may or may not do.) A printout also allows the vote to be reconstructed if the device suffers a fatal crash before the vote totals are downloaded and to compare to the vote totals during an audit.
Second, the goal of an audit is to verify that the vote counting technology worked properly. The best way to do this is by a public hand recount by the canvassing board. Because a full recount is too time consuming (and the very purpose of vote counting technology is to speed up the counting process), an audit is not a full recount. Instead, done properly, it is a recount of randomly select precincts in randomly selected races. It is important that the selection of precincts and races be truly random (i.e. publicly using a random process to select the precincts and races) rather than determined in advance by the election authority. If the election authority chooses the precincts and races in advance, a dishonest election authority would know which counts had to be honest and which ones could be skewed. Additionally, the precincts selected should represent a large enough percentage of the total precincts in the jurisdiction. (My own preference would be a minimum of three precincts or 2% of the total number of precincts whichever is greater.)
Third, after the precincts and races have been selected, step one is determining that the number of ballots cast matched what the precinct records indicate. That is, if the precinct records from precinct A shows 200 voters appeared on election day to vote, the canvassers need to verify that they have 200 ballots, not 210 ballots.
Fourth, after verifying the correct number of ballots, the canvassers publicly hand recount the selected races from that precinct. The recount results are then compared to the numbers as shown by the automated counting system. There will be some differences because there will be some ballots that were clearly not properly marked but do reveal voter intent. There will also be some ballots that are sort of marked that the machine might have or not counted. During the count, somebody should keep track of the number of ballots that are not “machine countable” so that the canvassing board can track how much of the change is due to ballots that could not be counted by machine and whether there is any machine error.
There should be two separate standards for the audit. One standard would be the precinct machine error standard. If the counts — either in an individual race or cumulatively go above those standards — then there is a potential problem with that precincts machine. In that circumstance, all of the races need to be recounted and another precinct added to the audit.
The other is the jurisdiction-wide total error. If the randomly selected precincts as a whole show enough of a change in the vote totals — for example 1% change in any candidate’s vote total — that race needs to be recounted in all precincts and, if the total change for all of the races is above a certain threshold, all races need to be recounted. (This possibility is why the canvass and audit process needs to start early after election day.)
Fifth, each county should be required to report the results of their audit to the state, noting the races and precincts that had errors in the election day count and whether the canvassing board could determine the reason for the errors. The State should post the audit results on-line as received.
I believe that most election authorities in this country run clean elections. They may make decisions about the registration process and precinct and early voting locations through their partisan bias, but very few are trying to cook the books or produce fraudulent vote totals. While there are always conspiracy theorists (on both sides) who will believe that there must be vote fraud or some form of behind the scenes shady operations to explain why the election turned out as it did, the public as a whole should have confidence that the vote counts are accurate. A strong election audit system during the canvassing process would provide that assurance to the general public.
As noted at the start, it will be the rare close race — one that probably would go to a full-fledged recount anyway — where the audit may show enough change to call the initial results into question. In most cases, audits will show that the initial counts are “good enough” for all but the closest races — errors of less than 1% in most precincts and most races. While I wish the race was close enough for a recount to make a difference, even in Michigan, the margin is probably too big for a recount to change things. However, I would like it if these states had a good audit system to assure everyone that the apparent differences between certain counties reflect the differences between the opinion of the voters and not the machines used in those counties. Such an audit (or even a good recount) may not be possible because some of these counties apparently use touch screen machines without a paper trail. While it is unlikely that the new Administration will make aiding local election authorities a high priority, every jurisdiction should have an auditable election process.