Plain Talk on Election Audits and Recounts
What’s an election audit? How is it different from a recount?
They’re asking with both interest and confusion, probably because of the just-completed ballot-counting activity in Georgia in November 2020. Of course, there are many explainers about audits and recounts, including this brief if dry one from Verified Voting or the longer more detailed and nuanced articles we posted last week here and here with an eye on Georgia and other prospective recount states.
But for some really plain talk, it’s not so complicated. We can forget about the Georgia situation and simply read on for some plain language about the what, how, and why of audits vs. recounts.
First, one key point: this is about the concepts, explained in basic terms, without going into every last detail about the procedures to actually perform and audit or a recount. So, for example, a really important part of an audit is ensuring that every paper ballot is authentic, and that its “chain of custody” can be verified through documentation. That task can turn into a whole related process of its own, called a “compliance audit.” Those procedures and details are essential to election operations, but not needed to explain the basic concepts. In other words, we’re talking about a ballot audit, not an audit of everything that happened in an election.
What’s an Audit?
We’re talking about a ballot audit, not an audit of everything that happened in an election. A ballot audit is where people manually count a subset of all the ballots, in order to cross-check the work of the computers that scanned and counted ballots. To perform a ballot audit, we must have paper records of the voter’s choices, and voters had to have an opportunity to verify the paper records. And we’re going to compare those to results reported by the voting system. Why? Because…
We don’t trust computers to decide election results!
All the technology is great for quickly processing many ballots with dozens of different counts needed. But it’s just hardware that can have glitches, and software that definitely has bugs, so we can’t blindly assume they work right every time. So in addition to the rapid machine count, our hard working election officials in a few states and counties do a cross-check: they take a random sample of ballots, count them by hand, and compare their hand-count vote tallies to the tallies created by the computers. Most of the time variances are small — and there are virtually always variances — but with the right sample size (let’s skip a bunch details of the statistics mathematics, and just say “right-sized”), election officials can be highly confident that the machine counted election results are correct.
Why the emphasis on election results? Because that’s what counts: who won. Getting an exactly correct count of every vote from every ballot is not the point – and often, it’s not even possible. There are always a few ballots with marks that might or might not be votes, depending on how a human reviewer sees them, and depending how they interpret state election law and local guidance on it.
And since we never trust computers to decide election results, then ballot audits are essential for every election. In an ideal world, that’s what every state would do. But currently, post-election ballot audits are required in only some states. However, in those states where they are required, an audit is done for every election, regardless of whether there was a “close” contest or one with a “small” margin of victory.
In a future when routine audits are done everywhere in the US, then we can trust that national election results are correct, even if the election tech malfunctioned — whether by software bugs, hardware glitches, foreign hackers, or election officials’ “operator errors” of the counting machinery.
What’s a Recount?
Unlike audits, a re-count happens only in some elections.
A re-count happens when an individual contest has election results that are “close” as defined by the state election law. Every state has differences: a different definition of “close;” different triggers for when a re-count is required; different rules for when a re-count may be done; different definitions of “automatic” and other types of recount, and so on. Every state also has different rules about how re-counts are performed (and some states like Georgia forbid hand-counting in re-counts, except in special circumstances).
Besides the difference of “every election” for audits and “only when permitted” for re-counts, here are key differences between recount and audit. First, the how:
- A ballot audit by definition consists of counting ballots by hand (only).
- A recount can be done either by hand, or by re-running ballots through automated ballot-counting computers.
Second, the what:
- A recount is, by definition, a count of every ballot with votes for the close contest.
- A ballot audit typically counts only a random sample of some of the ballots.
That difference is significant. Except in very very close contests, the sample size for an audit can be a very small subset of the ballots – perhaps only in the hundreds, for example. By contrast, in re-counts, all ballots must be re-counted.
Why the Confusion?
I think it gets confusing because in both audits and recounts, ballots that were counted before, are being counted again. But the meaning and purpose of these additional counts are quite different.
- A recount is intended to answer the question, “If I total up the voters’ choices a second time around, will the outcomes still be the same? Will the same candidates be declared winners and losers?” Recounts ensure that election laws are complied with, to give candidates and interested parties the information that they due, in an individual contest in one election.
- An audit is intended to answer the question, “Can I have confidence that the voting system has interpreted voters’ intended choices properly, and that we have not declared an incorrect outcome?” Audits protect us from election tech naming the wrong winner in every election.
That sounds important, right? “Every”? That’s why the whole country is moving to modern paper-ballot processes and technology, plus audits. The progress is slow, but when we get there, we will be protected from TWO very important problems:
- Election technology naming a wrong winner; and
- Disinformation and misinformation about election technology being used to name the wrong winner.
Since #2 is so corrosive that it can undermine democracy, and given our experience in 2020… let’s pick up the pace on getting to nationwide ballot audits!
 The Georgia situation was confusing and not typical of either audits or recounts. It wasn’t a recount (although it involved hand counting) in the sense of Georgia state election law (and despite the Secretary of State’s claim that it was a re-canvass, recount, and audit all in one). And it wasn’t a typical audit either, because election officials ended up hand-counting 100% of paper ballots, instead of just a sample.
 To be very specific about “ballot” – this means a paper record of the voter’s intent, that the voter had the opportunity to verify; that is, check for any mistakes in capturing their intent. For a hand-marked ballot, that’s obvious, but for voting-machine-printed ballots it’s essential that the voter have the chance to examine it for mistakes, before casting it.
 In the very close Georgia 2020 presidential contest, given the margins, counties would have needed to review about 30% of all ballots, for a total of 1.5 million ballots. Because the required batch of ballots would have been so large, instead of hunting down millions of specific ballots from many storage boxes, the state chose instead to simply hand-tally all the ballots. But that was unusual. In most elections, there is no one contest with such razor thin margins, and the sample size can be quite small. You can read more about that in this article from last week.
Game of Margins—Part II
Following our previous post about the recount process in general, and in particular what it normally would mean in Georgia, this article is an “explainer” on the effort begun recently to “audit” the Georgia election per a directive that Georgia Secretary of State Brad Raffensperger announced this week (and which we try to sort out below). The results are due by next Wednesday. Let’s first reset the situation.
Georgia on Our Minds
Georgia has been watched for years, as a cornucopia of issues related to election administration, election technology, and yes, allegations of voter suppression. We’ve written and commented extensively on the technology issues, for example in this paper last year.
Importantly, there is no evidence that calls into question the legitimacy of the results that the Georgia election officials have reported. However, Georgia Secretary of State Raffensperger has essentially intervened prior to the election certification, to engineer a complete hand “recount” of the ballots (more on those quotation marks in a moment), by relying on the state’s provision for conducting a post election audit.
Core to the OSET Institute’s mission is advocacy and support for evidence-based elections. However, if elections are to be properly audited they must be carefully planned and executed so that:
- Election officials can follow detailed procedures in accordance with laws and administrative rules; and
- Voters can transparently understand what is going on and properly assess whether post-election audits and recounts show meaningful evidence to bolster public confidence.
- And finally, audits must be sufficiently well-organized and rigorous that they do not potentially risk becoming yet another “political football” for partisans to argue over; the whole point of a post-election audit is to produce clear evidence that reduces uncertainty – not to give politicians a fresh set of new “irregularities” to argue about. In the current context, this is especially important because Georgia’s 159 counties are being asked to perform a daunting and labor-intensive task on a very tight time frame, and hand counts are generally less reliable than machine counts; so some discrepancies are likely to emerge.
Next, let’s sort through the Secretary Raffensperger’s press conference earlier this week.
- Undoubtedly Secretary Raffensperger’s press conference raised more questions than answers—we’re in a position to say so because we engaged in these topics every day.
- Raffensperger announced that Georgia would perform a “hand count” of approximately five million ballots in the presidential race; and stated, “it will be a full hand recount… it will be an audit… and a recount and a re-canvassing, all in one.” No. No, it won’t. More on why below.
- There is no reason to believe Georgia is doing anything nefarious, although the Secretary is not helpfully explaining to the public exactly what the State is, and is not, doing, (At worst, it could be costly and introduce new uncertainty; but that’s different than being malicious or misguided.)
- The problem is that Secretary Raffensperger is using phrases such as “risk-limiting audit,” (RLA) and words like “recount,” and at times it appears as if he is using them interchangeably. That’s confusing to the public, and the distinctions matter.
- What’s more, the “hand recount” (or, as the state has recently starting saying, a “re-tallying” – a term which only further muddies the waters) audit activities related to the presidential race do not appear to align with Georgia state law and election procedures, which further adds to the confusion.
- Counties across the state had until today, Friday, November 13th, to certify election results. Then, the Secretary of State’s Office must certify the state’s results by Friday, Nov. 20. And that is why the window for this “audit/hand-count” is today until 11:59 PM next Wednesday, November, 18th.
- And one more thing, to be very clear:
whatever comes out of this hand recount will not change or affect the results of the presidential election.
President-elect Joe Biden has 306 electoral votes as of today, and losing Georgia’s 16 electoral votes would have no impact on the overall outcome.
However, this is a fast-moving story, and details are subject to change. Although questions remain, if the Georgia SOS disseminates additional details (and we hope he will), we will update “the state of the state” through social media channels.
To be sure, as we discussed in Part I of this “game of margins,” recounts rarely change outcomes, especially when the number of votes in question grows beyond several hundred ballots. Joe Biden sits on a 14,000 vote margin.
Some Basic Terms
Exact definitions of what constitutes recounts and audits vary from state to state, according to law and rules, but the general concepts are as follows (and to avoid this becoming a legal treatise, we acknowledge up front that we’re going to stay with basic principles and avoid the inevitable legal nuances—this is simply to help the reader grasp the situation). Here are three terms you’ll want to basically understand:
- Recount: Re-tabulating the votes on existing paper ballots to see if the outcomes/results are different the second time around. A “machine recount” uses automated voting system equipment to re-scan ballots and re-tabulate results; a “hand recount” relies on human review of voter choices on the paper to manually tabulate results.
- A recount is intended to answer the question, “If we total up the voters’ choices a second time around, will the outcomes still be the same? Will the same candidates be declared winners and losers?”
- Risk-limiting audit: A human review of a sample of ballots, and a comparison of the results to the voting system’s reported results for the same sample, to detect whether an incorrect outcome was reported. The size of the sample depends on the total number of ballots cast; the margins in the race; and the jurisdiction’s own tolerance for how much “risk” it will set for the possibility of reporting an incorrect outcome. The smaller the risk, the greater the sample size of ballots that must be reviewed with human eyes.
- A risk-limiting audit (RLA) is intended to answer the question, “Can we have confidence that the voting system has interpreted and tabulated voters’ intended choices properly, and that we have not declared an incorrect outcome?” (Notice how that’s a different question from the term, “recount.”)
- Dr. Philip Stark of U.C. Berkeley, the inventor of risk-limiting audits, offers this important observation. When many of us lay people hear about “samples” and “audits” we may think of old classroom terms like “statistically significant” and the like. Importantly, none of that applies here. In fact, the ballot samples are not “statistically significant” or “statistically insignificant.” The sample needs to be random. The size of the initial sample is irrelevant for an actual RLA: all that matters is when one stops auditing. And here’s the point: a genuine RLA does not stop until (and unless) there is convincing statistical evidence that the reported results are right. If they don’t find that evidence, they keep going… eventually to a full hand count. That’s the mechanism that limits the risk that a wrong reported outcome will become final, by replacing the reported outcome with the outcome according to the full hand count.
- Most important to this discussion, Dr. Stark makes the point that Georgia’s regulations make the “RLA” not an actual risk limiting audit, because they do not have the statutory ability to correct wrong outcomes. Their audits cannot limit the risk that a wrong reported outcome will be certified (for many reasons, but as Philip notes, that’s an obvious one).
- Canvass: Before certifying results, the canvass is the process of officially confirming every valid ballot cast and counted—absentee, early voting, Election Day, provisional, challenged, and ballots cast by military and overseas voters.
So with those three terms in mind and their high-level descriptions, let’s return to the Secretary’s soup of terms, from the press conference and from related statements.
Disambiguating the Georgia Secretary’s Comments
That six-syllable fancy word means, “sorting out and clarifying.” And we need to do that because the first question is,
Why is the Georgia Secretary of State mashing up the terms “recount” and “audit”?
Sometimes it appears as if Secretary Raffensperger does not completely understand the difference between a recount and an audit, or is being too imprecise with words. For example, an official SOS tweet on Nov 11 stated,
“According to law Georgia passed in 2019—Georgia is required have a state-wide hand Risk Limiting Audit (RLA), trigging [sic] a recount.”
This appears to be incorrect. Strictly speaking, RLAs and recounts are distinct animals under Georgia regulations, and an RLA would not typically “trigger” a recount.
The Secretary’s frequent sliding between “recount,” “audit,” and “hand count” is confusing (even for us) because Georgia sets forth specific rules for each, and Secretary Raffensperger’s announcement appears different from how most anticipated a “risk-limiting audit” in Georgia would be conducted. Let me break that down for you in three points:
- This year, Georgia was supposed to conduct a “pilot” or “dry run” audit for the first time, in accordance with Georgia legislation HB 316. Until the recent press conference, the Georgia Secretary had signaled for several days that the race selected for the RLA would not be the presidential contest, and that it was likely going to be a down-ballot race, which would require a small sample size. Most observers and election officials were expecting to “get familiar” with RLA procedures in a far less contentious race, with a smaller number of ballots, and—most importantly—the pilot RLA was not anticipated to amount to a “full hand count.”
- At the November 11th press conference, however, Secretary Raffensperger changed his position and declared that the presidential race would be the statewide race to be used for the RLA. That is where the situation quickly grew confusing.
- Although the Secretary did not explain this, others have pointed out that to conduct a proper RLA in the presidential race, with the current margins, as I mentioned in Part I of this discussion, mathematical calculations reveal that a sample of approximately 1.5 million ballots out of approximately 5 million ballots would need to undergo human review by audit teams. And more importantly, according to RLA procedures, those “random” ballots would need to be individually identified and removed from boxes and boxes of scanned ballots—essentially like finding needles in a haystack. Needless to say, that would significantly add to labor time. So it appears that the SOS concluded that it would simply be easier to hand count all ballots, rather than doing the extra work of “hunting for” and “extracting” over a million randomly-selected specific ballots.
That leads us to the next question:
Why is the Georgia Secretary of State calling for the hand count (and not a candidate, or an election official)?
Let’s examine that in three more points:
- Pursuant to Georgia law, if the margin is within 0.5%, then the trailing candidate can call for a recount, or an election official may call for a recount.
- Georgia law specifies that “recounts” be performed by re-tabulating ballots through automated voting machines, but not by a hand count. A “hand count” is authorized only if such voting machines repeatedly fail in pre-recount testing and the failures cannot be resolved; or a hand recount can be required by a court order.
- None of those conditions apply in the present situation: all results have not been certified; we are unaware of the Trump campaign having officially requested a recount (although that is also unclear); and finally, automated machine technology is not being used to re-count the ballots.
So what is going on?
My answer is that it appears that the Georgia Secretary came up with a novel and unanticipated approach:
Use the legal requirements to conduct an RLA as a justification to also require a hand-count of the most contentious race, by setting the “risk limit” for the audit to zero percent – which results in a full hand count of all the ballots.
Remember that the lower the risk limit, the larger the required sample set of ballots, so “zero risk” means effectively a full hand count. Therefore, the full hand-count is the RLA, and for the same reason, what looks like a recount doesn’t need to comply with “typical” Georgia laws and regulations governing recounts, because it’s not, legally/technically speaking, a Georgia “recount.”
I believe this also answers why Secretary Raffensperger announced the decision to hand-count the presidential race, rather than having the request come from a candidate, or a county election official.
OK, with that, let’s look at the next issue in this confusion:
What is meant by “full hand count? Is it actually a hand count?
Performing an audit of the presidential race on every ballot is happening based on a manual human review, and it appears that machines will play no significant role, if at all.
Since Georgia apparently will not use automated scanners for tabulation (which, again, would normally be used under typical Georgia recount requirements) then human eyes will count voter choices in the following manner:
- Traditional format absentee ballots (with columns and ovals) will be counted through manual examination of the marked ovals to determine intent of choice.
- Summary-format in-person ballots printed by Georgia’s Dominion ballot marking devices (BMDs) will be counted by manual examination of the human-readable text of voter choices. (Paper records from ballot marking devices contain a combination of human-readable text and an obfuscated QR code that encodes voter choices for purposes of automated scanning).
That leads to another question:
What are the issues/controversies associated with doing a hand-count?
Generally speaking, the concern with Secretary Raffensperger’s imprecise and lazy use of these terms and laws is that it may result in insufficiently rigorous audit procedures.
Furthermore, even worse, an insufficiently rigorous process could lead to additional discrepancies that partisan campaigns could use to sow confusion, and to cast doubt on election results for which there is no credible evidence of fraud or other ballot counting irregularities.
Whether one labels the process of hand-counting the presidential race in five million ballots an “RLA” or (non-legally speaking) a “recount,” the fact is that the Office of the Georgia Secretary of State has been relatively quiet on many important protocols that are required to protect the integrity of such a process, including:
- “Chain of custody” requirements (the related term is a “compliance audit”);
- Detailed provisions for “watchers” or “observers,” to allow transparency while ensuring that ballots are only physically handled by authorized personnel;
- Any guidance on Georgia statewide standards for determining voter intent on hand-marked ballots.
- Procedures to detect and resolve potential human error during the hand-count (which is common).
And there are several other concerns, many involving unclear legal justifications for some of the activities being proposed. The Coalition of Good Governance enumerated many of these in a letter that was sent to all county officials in Georgia this week.
In addition to concerns over a scarcity of well-documented statewide procedures for this process, there is one more controversy:
With respect to the machine-marked paper records generated by ballot marking devices (BMDs), even if human eyes manually review votes by reading only the human-readable text of choices, can those paper records rightly be called a trusted and auditable record of voter intent that was verified by the voter?
Dr. Stark has explained in many forums that when voter’s choices on paper records are machine-marked (i.e., in other words an imperfect complex electronic computing device is inserted between the voter and the act of recording a choice) and furthermore, given research indicating that most voters do not take time to review and verify the machine-marked record, then it is problematic to assume that the BMD records are “auditable” records of voter intent.
This last controversial point is not going to be solved in the present moment. However, it remains a concern. There is little choice other than to note the controversy, which creates additional uncertainty, and to let Georgia election officials move forward with the assets they do currently have. Improving technology and procedures to achieve better and better evidence-based elections is a race that never ends.
And finally, to wrap up this confusing exercise, here’s where the rubber meets the road:
If the audit reveals discrepancies as compared to results that election officials originally reported, how does the audit impact the overall results that the Georgia SOS will certify on November 20th?
On this all-important question, the Georgia Secretary of State appears to have fully waded into unclear legal territory. According to Georgia laws, audits alone cannot change an election result in Georgia; so it appears as if the 100% hand-count, which looks like a recount, but which is not a recount, cannot change the election. But the Secretary of State has not clarified what significance the resulting numbers from the audit will have. Some observers have suggested that it seems reasonable that the results from a full hand-count would become the results that are certified, but the legal basis for doing so is unclear. And that leads to a somewhat uneven set of concluding thoughts, in terms of assessing the value of this entire exercise.
A Few Closing Thoughts
To begin, I probably cannot make this point enough given the climate, but…
There is no credible evidence to undermine the legitimacy of Georgia elections results.
Second, whatever questions remain about this process, the fact that election verification has taken center stage in the highly visible venue of Georgia, with all its controversial voting history is a good thing. Even this confusing, uneven, and fitful affair showcases the importance of paper trails, audits, and the imperative of evidence-based elections. To put it candidly, it makes the point that if Georgia—for all its unfortunate history and problems—can take steps toward higher-integrity elections, then any of the states can do so.
Third, time and again, we’ve observed that public confidence in elections is a product of communication and transparency. Secretary Raffensperger has room to improve in this regard. We remain hopeful in the days ahead that his office has an opportunity to clarify matters. I believe this is important because the audit, or recount, or “re-tally” (or however one wants to characterize it) is very unlikely to alter reported results as they stand today. This final point deserves extra emphasis.
Given the fact that the re-counting of the presidential race in Georgia will not change the outcome of the presidential election, it is reasonable to set a high bar for precision, because in addition to the upsides noted above, very real risk also remains for damaging downsides.
In order for this costly and time consuming effort to be determined “worth it,” the execution and transparency of the process will need to be provably on par with best practices. If not, this exercise could amount, at best, to a waste of time, taxpayer dollars, and charades that fails to convince anyone of anything. Or, even worse, a sloppy process could potentially introduce “new cuts” of discrepancies for partisans to fight over. As stated earlier, the whole point of a post-election audit is to produce clear evidence that reduces uncertainty – not to give politicians a fresh set of new “irregularities” to battle over. In an environment where disinformation is rampant, the last thing America needs is new opportunity for fear, uncertainty and doubt (“FUD”) to reduce trust in the vote.
We’re hoping that in the days ahead, Secretary Raffensperger runs a tighter ship than we’ve seen so far, and we are wishing him and Georgia election officials success and transparency in this high-stakes Game of Margins.
 That “sorting” begins by noting that the NY Times article cite-linked above about the Secretary’s announcement deserves some clarification of its own, which we attempt to do so with this article; notably: what is happening in Georgia is not a recount (although its not really a genuine audit either—yes, its kind of complicated which is why we’re trying to sort it out for you here.)
 Alternatively put, so voters can understand whether such activities (when poorly executed) might be unnecessary and costly forms of “election integrity theater” that amount to little more than public relations exercises.
 The Secretary’s press conference and subsequent commentary from Georgia election officials initially added to the confusion on what “hand-count” means. After the press conference, for example, the Secretary of State’s Chief Operating Officer Gabriel Sterling made comments to the effect that, after paper ballots are manually placed into stacks in order to audit them, they would be run through automated scanners—but only to count the raw quantity of ballots (e.g., confirm that there are 50 ballots in the stack), but not to count voters’ choices in the presidential race. Initially, Sterling’s remarks led some to suggest that Georgia will not actually be performing a “hand count.” However, after the SOS backtracked further on the use of scanners at all, that appears not to be the case.
 There are some “training materials” associated with Georgia’s statewide audit materials, supplied by a third-party vendor that is assisting the State. However, they can rightly be characterized as relatively “high-level” and lacking in details. See this article.
About the Author
Gregory Miller is the co-founder and Chief Operating Officer of the 501(c)(3) OSET Institute, Inc. He also serves as a senior election infrastructure analyst. A computer engineer/scientist and a (nonpracticing) intellectual property lawyer, Mr. Miller has spent nearly a decade immersing himself in election law and technology policy. He recently published an essay on New America’s Commons explaining why election technology is public interest technology and a Theory of Change essay about the Institute’s mission.
Game of Margins
[Updated: Nov 11, 2020 5:15pm]
[Editor Note: As this is going to publication (posting), the Secretary of State of Georgia has announced that he will direct a complete audit of all 5 million ballots in Georgia, which is essentially going to be conducted as a hand recount. Stay tune for an updated posting to explain what that means. The balance of this article remains viable for a layperson brief on how recounts work.]
Last Saturday, November 7th, the media called the 2020 U.S. presidential election in favor of former Vice President Joseph R. Biden Jr..
It remains to be seen exactly how things will unfold, but President Trump has raised the prospect of recounts in one ore more states, and at least one Secretary of State (in Georgia) has indicated the same. A few days ago, President Trump’s campaign stated it will request a recount in Wisconsin, where unofficial totals show Trump trailing by approximately 20,000 popular votes. President Trump also intends to initiate law suits in a few other states where he believes there is cause for recounts.
As the final votes continue to trickle in from the remaining polling precincts around the country, the margin of victory for President-elect Biden continues to widen in Pennsylvania, Nevada, and Arizona, with the prospect of a recount in any of these states unlikely. Thus, with the Georgia Secretary of State’s pronouncement this week, Georgia may be the only state to proceed with a recount (where at the time of this writing, President-elect Biden leads the President by more than 14,000 popular votes in Georgia and more significantly, Biden does not need Georgia’s 16 electoral votes to win the election). While that may not have any impact on the presidential race and its certified results, the outcome of a recount could be a barometer to evaluate Georgia’s two Senate runoff elections this coming January 5th, 2021.
A cornerstone to trustworthy elections is a recount process to ensure accurate outcomes. Likewise, trust in the recount process is essential. So, below I offer an explanation for how recounts are conducted in just the states of most potential consequence in the coming weeks.
Because it’s easy to get bogged down in the ‘legalese’ and partisan arguments about the vote count, recounts, and related regulations, below I try to provide you with a layman’s guide to the Georgia recount process, which would not differ greatly from most other recount procedures around the country. We also offer some basic guidelines about some of the other states President Trump appears poised to challenge.
The major differences between states is in their margins for automatic recount triggers, and the agency of candidates when it comes to filing for a recount (some states, like Pennsylvania and Arizona, may only conduct a recount if there is a close enough vote margin between the top candidates, as I observe further below).
Here is a quick overview of facts. Our colleagues at FairVote produced a November 2020 report by Deb Otis, Declan Alvidrez, and Austin Bartola examining statewide election recounts outcomes and practices in the United States. This builds on prior versions of the report, and they used a compilation of data across the 20 years of 2000 to 2019 including:
- The frequency of recounts;
- How often they changed outcomes;
- How much vote totals changed, and finally
- How those numbers varied with the size of the electorate.
Their newest version builds on the same findings made in the 2016 edition, with the four most important findings in 2016 remaining true today:
- Statewide recounts are rare.
- Outcome reversals are even more rare.
- Margin shifts in recounts are small.
- Margin shifts are smaller and recounts even more rare in larger electorates.
The Georgia Recount Process
Before I delve into the recount basics for Georgia, this game of margins is playing at full speed (as our Editor notes at the top of this post). So, I must first observe that:
What is about to happen in Georgia is not a recount at all, at least not strictly speaking under Georgia law.
That perhaps explains why so many have complained about the confusing words that Secretary Raffensberger has been using lately.
Legally speaking (as a non-practicing lawyer) in terms of Georgia law, the “audit” that is about to be conducted in Georgia does not meet the requirements of a recount (as outlined below). Actually, it’s packaged as what we call in this neck of the woods a “risk limiting audit where the risk is set to zero.” That means rather than creating a selection of ballots to review, the “selection set” is the entire population of ballots — some 5+ million.
The great statistician in this space, Philip Stark describes it well. Let me unpack that a bit here and note that we’ll soon post another explainer on audits and Georgia, so I do not want to drift into a rabbit hole on this. But…
The current vote count for Georgia today shows Joe Biden winning the election by ~14,000 votes. That’s a margin of 0.3%. Three tenths of one percent is tight in this game of margins. When a risk-limiting audit is performed, the number of ballots reviewed is based on the margin. The tighter the margin, the larger the sample size needs to be, and conversely the wider the margin the fewer ballots required to be examined. It turns out that in Georgia that .3% means that over 1.5M ballots would be required. Ouch. So, it would be less work to simply use every single ballot than going through the ordained process for ballot selection to acquire 1.5M of the 5M+ total.
So, what is being “pitched” or “positioned” as an “audit” while bypassing the normal audit procedures is effectively a re-tally, by hand. I’ll defer any more on this little hot mess to a future article (watch this space) and return you to the point here… a simple explainer on how recounts are actually supposed to work.
Back to recounts…
In Georgia, a recount can be initiated by three different mechanisms: a close vote margin, a candidate-initiation, or election official-initiation. Sparing you the legalese and raw regulatory code, here’s what that means: 
1. Close Vote Margin
- When the difference between the number of popular votes received by candidates is less than 0.5 percent, the losing candidate can request a recount of the popular votes cast within two days after the certification of results.
- This is what is occurring right now; as of November 8th, the margin was 10,353 votes, or about 0.21 percent.
- If the losing candidate feels there was a discrepancy or error in the vote count, they may ask the Secretary of State for a recount. It’s at the discretion of the Secretary to grant their request and carry it out.
- The candidate may choose which precinct’s ballots to recount, and how many.
3. Election Official-Initiated
- The superintendent of elections (a localized election supervisor) in any precinct where paper ballots are used may order a recount if there appears to be a discrepancy or error in the results of a referendum or constitutional amendment, or if there is a vote margin of less than 0.5 percent.
Once a recount is requested and scheduled by the Secretary of State alongside relevant election superintendents, it occurs by passing all ballots cast for a specific race through an optical ballot scanner (a machine that counts paper ballots). Georgia Secretary of State Raffensperger had earlier indicated that in each of Georgia’s counties, they would scan paper ballots through a centralized high-speed scanner, which is integrated with voting system software. Traditional-format paper ballots from absentee/by-mail and summary-format QR code ballots from in-person ballot marking devices will all be re-scanned for tabulation.
- Preparation. Before recounts can begin, the ballot scanning machines must go through a testing process to ensure that they are counting votes accurately. This testing is conducted by running a batch of ballots from the relevant precincts through the scanner being tested, alongside a contemporaneous hand-count. If the numbers match, that machine is usable for the recount. Otherwise, more testing would be required to understand and resolve the discrepancy. If it cannot be resolved, the machine is not used for the recount. If there are no scanning machines that pass this test (which is highly unlikely), then the recount is conducted exclusively by hand-counting the ballots. A hand-count may also occur if mandated by a court order.
- Observation. During the recounting process, the public legally has access to the proceedings, but may not touch any of the ballots or scanning machines in use.
- Audit Controls. Additionally, ballot containers must be unsealed and scanned one at a time, with a clear paper trail maintaining details on when each box was opened.
- Overvotes. If there are ballots scanned that contain an overvote (meaning the scanner picked up on a ballot that may have marked both Trump and Biden), the ballot is set aside for a bipartisan team to examine and rule as a majority on the voter’s intent. A duplicate ballot is created (and clearly labeled as a recount duplicate) to be scanned in the original ballot’s place (which is stored for integrity purposes).
- Damaged Ballots. The same inspection process occurs if a ballot is too damaged or otherwise defective to be run through the machine.
- Reporting. When all is complete, the election superintendent in charge of the recount will create a printout of the new results. If this happens to differ from the originally certified numbers, then the original results are corrected in accordance with the new tally, and the election results are finalized. Recount complete.
If anyone has tremors over the possibility of a Bush v. Gore (election 2000) type showdown, it’s unlikely. Not only were the voting machines in Florida during the 2000 election ripe for ambiguity, the results were far closer and consequential than they are now. In 2000 it was a matter of 537 votes in Florida that determined the presidency. In 2020 for Georgia, there is currently a margin of over 14,000 votes (and between 16,000 and 43,000 votes for Arizona, Nevada, Wisconsin and Pennsylvania respectively.) In fact, it is now a matter of whether Biden will break 300 electoral votes. With that in mind, let us take a quick tour of those remaining four states that could be subject of President Trump’s legal pursuits.
A Summary Look at Four Additional States
In Arizona, the race was called for Biden. Campaigns cannot request a recount, they can only be triggered automatically when there is a margin of less than or equal to 0.1 percentage point. Said differently, in Arizona, state law requires a recount when the margin between the top two candidates is equal to or less than one-tenth of one percent of the total number of votes cast. Biden was awarded Arizona by 16,985 popular votes, which is 0.5194 percent, thereby eliminating the possibility of a recount unless that margin were to change due to remaining provisional ballots to be counted.
In Nevada, the race was called for Biden. Trump’s campaign may request a recount within three working days of the final canvass of popular votes, no matter the margin — however, Trump must immediately pay a deposit to cover the estimated cost of the recount. If Trump were to end up winning the race after a recount, that deposit would be returned. But if the recount shows that Trump did in fact lose Nevada, then his campaign would have to pay the costs of the recount.
In Pennsylvania, the race was called for Biden. There are mandatory recounts for any race that has a margin of 0.5% or less that must be completed within three weeks of an election. The 47,726-vote lead Biden holds as of this writing (slightly more than when the race was called on Saturday) is outside that margin (at 0.7118%). Secretary of State Kathy Boockvar would have been required by law to order a recount if the winning margin was 0.5 percent or less. The recount would need to be ordered by 5 p.m. on November 12th and completed by November 24th. Campaigns can also request recounts, no matter the margin, within five days of canvass completion. And a recount can also be triggered in each county if requested by three voters.
In Wisconsin, the race was called for Biden. Vote margins must be within one percentage point for a campaign to request a recount. There are no automatic recounts in the state, and the campaign that requests it must pay for it. It must be completed within 13 days of the formal recount order. The request must be submitted within three days of the state’s official county canvass. Under state law, a candidate may request a recount if the margin is less than one percent. The request must be made no later than 5 p.m. on the first business day after the state has received final results from the state’s counties. The race was called by a margin of 20,540 votes or 0.634%, which is less than the 1 percent threshold, and thus the motivation for Trump calling for a recount.
However, a successful recount would require the over-turning of at least 20,540 popular votes and for the likelihood of success on that, I return you to our opening comments about the historical performance of recounts as provided by our colleagues at FairVote.
[Editor Note: The author, a nonpracticing IP lawyer and engineer with a passion for election law, acknowledges the research assistance of Dennis Mema, election policy analyst at the OSET Institute, and the editorial eye of Joy London, Associate General Counsel at the OSET Institute.]
 Citizens for Election Integrity—Minnesota, https://ceimn.org/searchable-databases/recount-database/georgia
 For the purposes of this article, certification of an election essentially means that all the cast ballots have been received, and the results are tentatively certified pending a recount or other form of challenge in the form of an election contest. An election contest is a right of action conferred on every candidate to contest the certification of nomination or the certificate of vote as made by the appropriate officials in any election. It is a post-election contest between two competing candidates. Fraud, corruption, or irregularities in regard to the method of holding an election in a division can affect the entire vote. Therefore, an election contest is a special proceeding created by the legislature to provide a remedy for elections tainted by fraud, illegality, or other irregularity. Generally, there are two types of election contests: a) Motion seeking to oust and replace the certified winner; and B) Motion seeking to declare an election void altogether.
 Georgia Election Results, https://www.nytimes.com/interactive/2020/11/03/us/elections/results-georgia.html
 The full test process, which is a bit more complicated, involves drawing 100 ballots altogether (75 in person and 25 absentee) from at least three precincts relevant to the recount, and running them through the scanner being tested alongside a hand count of the same ballots.
 2000 United States presidential election recount in Florida, https://en.wikipedia.org/wiki/2000_United_States_presidential_election_recount_in_Florida