My initial judgement is that a "computer rescan" would have less chance of being biased and less change of making mistakes than a "hand recount", and so the "hand recount" can only introduce more uncertainty and a worse quality result. Is there any reason I should think otherwise? And if the result is not qualitatively better, then what? 2 out of 3?
This article covers the details of how Georgia handles this process, and what Georgia law says about hand recounts. In particular, it says
Georgia law only provides for recounts by hand if a certain number of machines are unavailable or a court order is granted. In Georgia, a candidate can request a recount if the margin of the race is within .5 percent.
To my knowledge, none of those requirements, other than the .5 percent threshold to request a recount, is met to date, meaning that the recount would not be done by hand. I have heard nothing indicating there are any problems with the machines, though part of their vote verification process includes an audit to make sure the machines are, in fact, functioning properly. So we'd have to wait for that to know what situation their machines are actually in. A court order is hard to predict. Many of the litigation claims being made in the current nationwide debacle have been thrown out as unsubstantiated already, but there's no telling how a court might interpret the particulars of the situation in Georgia.
But Georgia's Secretary of State appears to be invoking the power to conduct a "risk-limiting audit". Normally such an audit is done on a random selection of ballots to verify the machines are producing appropriate results, and is a normal part of the vote counting process. The SoS seems to have broad authority on the audit by statute, and in this case he is going so far as to insist the audit be a hand recount of every ballot cast in the state. He has stated this as necessary due to the close margin of the vote and the great national importance of the vote. Moreover, he has stated his opinion that Trump could still request and receive a recount after the audit is completed, if he still trails by at most .5% after the audit completes, since the "audit" is legally distinguishable from a "recount". Such a recount would be done with the scanning machines.
So, beyond that...
First and foremost, people still trust people more than computers. While true that computers and machines, when used and designed properly, are much more reliable at almost any task they can do in reasonable timescales (and are frequently faster to boot), the public perception remains concerned about major errors that result from the inevitable involvement of humans. This is not without reason: someone (multiple someones, usually) had to build and design the machine, program the computer, feed in the ballots at some stage, etc. If someone does those steps incorrectly—almost always by accident, or having been given poor instructions, but politicians like to claim there's some nefarious partisan with a cartoonish plot to foil—systemic errors, potentially very large ones, are introduced. So there is a feeling that you can't just trust the machines and computers alone, you have to perform some independent checks. Which are already done during the normal vote counts, but again the popular perception is that if a recount is being done then you have to go above and beyond to make sure the machines aren't messing with us.
Second, and related to the first, there is the "observability" factor: people believe they can stand by and independently verify that everything is being done correctly and accurately if they can see it all happening before them, in human timescales. That we can in principle independently verify the mathematical and technical unassailability of the machines and computers is, again, not something the common person has much trust in. In the cases where the counting machines are proprietary commercial bits of tech this can be justified, as the company may actively prevent us from such a verification in order to protect their product. What verification we do have of it is still pretty strong, however, as these vote counts are all run with multiple failsafes at every level. Nevertheless, it seems easy to sow distrust in any process that doesn't involve a bunch of observable humans doing things out in the open.
Third, this falls into the apparent Trump/Republican strategy of trying to slow down the vote results and sow distrust in the vote to give them more time to produce a narrative to "explain" Trump's loss (as being totally not his fault, and totally not related to his behavior and actions and failures) and undercut a Biden presidency, and hype his accomplishments; or worse yet try to enable some sort of legal shenanigans whereby the vote results are ignored (because they aren't certified and submitted in time) and Trump gets installed for another term. While there is nothing I am aware of to suggest this is a motivating factor here by those making the decision—other than what we know of and from Trump and his campaign and administration and the general obeisance of Republicans to Trump—it is assuredly not unwelcome to those attempting to execute, or hoping for the success of, such a strategy.
Money for Trump
Sometimes things are motivated by money, rather than legal or technical reasons.
Recounts are covered by section 21-2-495 of Georgia law. Part (a) covers individual "precincts where paper ballots or scanning ballots" are used. Part (b) covers individual "precincts where voting machines have been used". Part (c) covers how adjustments to the count are handled. Finally, part (d) covers cases that involve multiple precincts, including federal elections. In the latter case, the law authorizes the Secretary of State to make a decision, but part (d) states that local election superintendents -- not the Secretary of State -- decide the manner of recount. Simply put, the hand recount is not required by law.
However, a hand recount will take longer than a machine recount. According to yesterday's PBS Newshour, Trump is using this time to raise money:
Well, President Trump's refusal and his — Trump campaign's refusal to acknowledge that Joe Biden is the president-elect and to continue to file lawsuit after lawsuit comes down to two big things, politics and money.
On the money side, they are raising boatloads of money, thousands of dollars every day by sending out next messages and e-mails, messaging that they need to keep up the fight, that they want to have money for an election defense fund.
Furthermore, only 40% of the money being raised is allocated to challenging the election. The remaining 60% goes to the "Save America" PAC which is essentially a slush fund which can be used for any current or future political activity:
But when you look at the money being raised, it tells a story. So, to put up for folks, 60 percent of that money goes to Save America PAC. It's a political action committee. And then, after that, if someone donates thousands of dollars more, they then — that money then goes to the Trump recount account.
Now, this is important. This kind of account, the political action committee, it faces fewer restrictions on how the money is spent. Unlike candidate campaign accounts, this can — the money that can be raised here can be spent on personal expenses.
Trump can even spend the money on himself:
Critics say that this is really a slush fund and that it can go to funding President Trump's lavish lifestyle.
The longer the election outcome is delayed, the more money Trump can raise to line his own pocket, and the pockets of family members ("consultants").
(Worth mentioning that if Trump loses, he can keep up this scheme for the rest of his life as a perpetual "run for a second term." But as soon as he wins a second term, this scheme is finished.)
Update: The New York Times confirms this story:
Initially, 60 percent of donations were going to pay off debts that the Trump campaign had accumulated. But as of Tuesday, that had shifted.
Now, 60 percent of donations are earmarked for Save America, which was registered with the Federal Election Commission on Monday by Bradley T. Crate. Mr. Crate also serves as Mr. Trump’s campaign treasurer.
Only after a donor gives more than $5,000 does any of the money go to the recount account that Mr. Trump set up.
ABC News has also posted a similar story about the Save America PAC.
Regarding why hand count ballots are more secure and trustworthy, I would introduce you first to the concept of a black boxes vs white boxes- and also the self contradictory attributes sought for in a supposed "free and fair election".
Think of a black box almost as something which you have absolutely no idea what the internals are. you have no idea what the machine or system does precisely- there is a ton of PRESUMPTION and assumptions of what it does but you ultimately don't know. Voting systems unfortunately are largely black boxes- and worse than that, they are very rarely tested. In a white box you can see the innards and how it works etc.
What are the attributes sought for in an election ballot? You want it to be secure & trustworthy yet anonymous. You do NOT want anyone who isn't authorized to be able to cast votes - or else the entire system may become corrupted. And simultaneously you want anonymity which is the opposite of security, accountability and trust. The two attributes of security, accountability, and trust vs anonymity are entirely contradictory because the whole notion of a secure and trustworthy system is that there is recourse, and definitive logs to ensure proper function and basically know whom to blame when CHTF.
As computerphile points out, most voters would never ever be able to understand the innards of an electronic election voting system nor would they really care to (in general most are as economists say, "rationally ignorant"). Much of these electronic voting systems are proprietary and secretive also. Thus it is a huge black box ... with a ton of questionable faith in them. And we already know how relatively easy it is to hack these vulnerable systems. Every year hackers hack them at conferences for fun as practically a joke. As vulnerable as the election process already is, we can at least track down and hold person accountable (for certain misappropriations) which is practically impossible to do for an electronic system. So hand ballots MAY provide what is called transparency.
When dealing with anything related to security or for trustworthy systems, transparency and accountability are absolutely paramount. It is precisely the lack of these attributes or trying to have contradictory anonymity and total security and trust that causes a lot of election problems.
I think the question in itself might be an XY-problem.
It doesn't matter that much if counting by hand is better or worse. What matters is, that the only way to detect an unforseen systematic error in any of the counts is by using two independent methods and buying two sets of machines with different working principles is a bit expensive.
And there doesn't have to be any malice involved. Think of the famous hanging chads or just a set of misprinted ballots with a bad alignment between human and machine readable parts that throws of the scanner. A second run with the same equipment won't find those.
Also keep in mind that its not like they will throw all ballots onto a single pile before recounting. They can count them in the same smaller units (personally I'd use stacks of about a hundred, but I have no insight into their methods) they counted them in before. So even if there is a human element of error, there is never the need to do a full third count of all votes, just to have another closer look at the few stacks that actually produce a different count.
My initial judgement is that a "computer rescan" would have less chance of being biased and less change of making mistakes than a "hand recount"
Your initial judgement is probably correct. Since there seems to be no proof that a machine rescan is less accurate than a human recount, it would make sense that a verified (and audited) machine count would introduce less error than human vision-based count.
That position may well rile the advocates of 'but the vote counting machines can be hacked' contingent. To those I would pose the question: Have there been any instances where election results have been manipulated by hacking optical scanners. (Just to be clear, I'm not asking about demonstrations)
As to the "black box" explanation, the manner in which optical scanners work is not really mysterious, if the voter really cared to find out. Granted the popular bias may be to be distrustful of those things we don't attempt to understand ( viz: Is the earth really flat ? , Does the earth rotate around the sun?). But going down that rabbit hole leads people to not trust much of our world (viz: gasoline pumps, internal combustion engines, banking systems, EKG machines, etc.)
Historically you'd get ballots filled out wrong so that a machine couldn't count them, but a person could. For an optical scanner this would include an X or circle (instead of filling in the bubble), or a small mark on one but a complete mark on another; or even both marked, one crossed out, and a hand-written "this one" and an arrow.
That doesn't seem like it would amount to much, but mlk alludes to the old punch-out ballots. The error-rate with those was around 5%. Voters who did everything right could have messed-up ballots, like a big buldge where the perforation wasn't done properly (and some voters wrote on them, too). People used punch-cards for other things and knew they were buggy and sometimes needed to be examined by hand.
Because electronic voting machines have security vulnerabilities.
Commercially-Available Voting System Hardware Used in the U.S. Remains Vulnerable to Attack
And once again, Voting Village participants were able to find new ways, or replicate previously published methods, of compromising every one of the devices in the room in ways that could alter stored vote tallies, change ballots displayed to voters, or alter the internal software that controls the machines. In many cases, the DEF CON participants tested equipment they had no prior knowledge of or experience with, and worked with any tools they could find - in a challenging setting with far fewer resources (and far less time) than a professional lab (or even the most casual attacker) would typically have.
In most cases, vulnerabilities could be exploited under election conditions surreptitiously by means of exposed external interfaces accessible to voters or precinct poll workers (or to any other individual with brief physical access to the machines). In particular, many vectors for so called "Advanced Persistent Threat (APT)" attacks continue to be found or replicated. This means that an attack that could compromise an entire jurisdiction could be injected in any of multiple places during the lifetime of the system.
As disturbing as this outcome is, we note that it is at this point an unsurprising result. It is well known that current voting systems, like any hardware and software running on conventional general-purpose platforms can be compromised in practice. However, it is notable - and especially disappointing - that many of the specific vulnerabilities reported over a decade earlier (in the California and Ohio studies, for example), are still present in these systems today.
A hand recount can be watched, as it is a more spread out, and slower process. This increases trust in the process, especially for non tech savvy voters. Even as a programmer myself, without access to the source code of every single tabulation and scantron software used, why should I trust a tally generated by said software.
If a system provides one result, you dont check the validity of the system by using the same process. You need independent verification. Multiple processes that yield the same result is better evidence of an outcome than one process run over and over.
While in theory, a machine scan should be both more reliable, and faster, its unknown if source code is provided to all counties or if they have anyone who could understand it anyway. The fear is that a malicious programmer (or an unforeseen glitch, but I feel like scantrons are pretty simple) could bias a machine to occasionally add or remove votes for a candidate they want.
If tallies are live updated or samples are randomly checked by two independent programs on different machines, I think that would be a better solution. If machine A running program A1 saw a vote for x, but machine B running program B1 saw a vote for y, that would be a fast and more robust way to protect against software that occasionally adds to, or removes from certain candidates, with humans none the wiser as tallies are not updated one at a time.
I have gotten to be very cynical about the Republican party's support of Trump.
Georgia Secretary of State Brad Raffensperger is a Republican.
A recount is mandated by Ga law since the % difference is less than 0.5%.
A hand recount will be slow thus delaying the official results for as long as possible.
This gives President Trump more time to say that massive voter fraud took place.
EDIT Fair point that I should have been more on point to the OP's question. I think that trying to decide why a hand count is "better" is the wrong question. To me the more pertinent question is: Why did Georgia Secretary of State Brad Raffensperger propose a hand count? I think he did it for purely political reasons to support the president. The deciding factor to me wasn't that a hand count was "better," but that a hand count would take longer.
Software is not always flawless (citation needed). Of course there is human error too, but a hand recount would likely catch any "systemic" problems like the one linked below.