Paper or Silica?
In the wake of the voting catastrophes in Maryland during the September 12th primaries, Governor Ehrlich is now calling for a return to paper ballots for the November 7th General Election. This is a bold and applaudable move. The Diebold solution that the State of Maryland paid over 100 mil for failed not only the laugh test, but the personnel failure test, and the hardware test as well. With machines rebooting during voting, registering a vote when none was cast, and leaving no verifiable paper trail behind, this is a system that can’t stand for trustable elections.
What I propose is this: Scantron. Everyone remembers these machines, albeit with a bit of horror, from testing in High School, but they provide everything that a ballot needs: Machine Readability, Paper Backup, Hand-countability, easy manufacture, easy reading. This is the perfect ballot and we’re ignoring it for a touch screen system because it looks flashy?
We have abandoned the paper ballot to our own peril, Governor Ehrlich knows it, you know it, let’s do something about it.
Actually, I work with scantron sheets on a daily basis (I work for an org that administers tests), and scantron sheets are not a good solution. I couldn’t even list the many number of ways in which the scantron system could get messed up between the candidate Joe Voter chooses in his head and the candidate that actually gets the vote.
You couldn’t just send a pile of scantron sheets through a scanner and compile the votes. When we score tests, there are at least a dozen confirmation points that a human being is needed for.
It can’t be worse than a system that can fail at any point, register votes when none were cast, and leave no human verifiable paper trail, Robis. Scantron isn’t entirely machine-dependent, and represents a system that scales across a city, is human verifiable, and measures intent that touchscreens cannot.
But it doesn’t do any of those things, Tom. That scantron are not machine-dependent means it takes as long to scan as it would to count votes manually. One error on one sheet can cause scoring times to increase threefold. This also applies to the scantron being human-verifiable; the range of possible errors is so wide that any verification procedures will make scantron no more time or cost effective than hand-counting.
Also, scantron does allow for measuring intent in most cases. The most common error in the scantron system is a person filling in two choices–which is comparable to the hanging chad. Yes, in cases in which the choice might not be clear to the scanner, a human eye can look and see if a clear choice is made, but that’s about it. To do so, you need a human looking over all of the scan sheets that have been kicked out—and that takes us right back to scantron being no more efficient than hand-counting.
Add to that the fact that scantron systems require a very precise answer. You know why the teacher was so anal retentive about the instructions? Fill out the bubble completely with a #2 pencil, don’t make any stray marks, erase completely, etc.? Because all of those things cause a scantron scanner to kick out the sheet, or void out the choice. When you consider how confusing many people find even the paper ballots, suddenly you have a system that just begs for errors.
Finally, the mechanism of scanning sheets is not foolproof either, and in fact if a problem arises, it can be completely camouflaged. A few months ago, ETS got in trouble for underscoring hundreds of thousands of tests. The reason for the underscoring was the higher than average humidity levels caused scanning problems that were not detected. Because of the high volume of tests that were being scored, that underscoring went unnoticed for months. And that’s just one example.
Scantron is ok for testing, in which a score will be sent back to a student who can call into question the score if it doesn’t seem right. It’s not so good for a system in which once the vote is counted there is return result for the voter to make sure that their vote was counted correctly. In short, scantron systems combine the worst of technology with the worst of human error.
So, where is the solution? There’s got to be an optical reader technology that can work with ballots that are also human readable. With no paper to go back to, can we ever be sure that recounts are done right?
Scantron problems are a mac guffin here. The field of optical scanning is so far advanced from what those things used that talking about their limitations is like talking about phone inadequacies by bringing up party lines. The only reason they continue to be used anywhere is that they’re old, cheap technology.
In many of the non-psychotic voting counties the systems use a fairly standard, heavyweight paper and markers. They’re even MORE readable
As far as “To do so, you need a human looking over all of the scan sheets that have been kicked out—and that takes us right back to scantron being no more efficient than hand-counting.” that just makes no sense. Of course it’s more efficient – by your own description that human looks at the sheets which have been kicked out, which means LESS than the total number of sheets – only the flawed ones need to be hand checked.
Any system that uses paper votes and machine scanning is inherently more verifiable than the Diebold-style machine voting because you don’t have to scan EVERYTHING to detect fraud. Any statisticians in the house who’d like to chime in on detection of fraud through random sampling?
Pull 0.1% of the paper ballots and compare your results to the distribution in the final count. YOU CAN’T DO THAT WITH ELECTRONIC VOTING, even with the ‘recipt’ talk that runs around. There’s no way to verify that the receipt you got matches what was recorded. The creation of two separate paths for the data virtually insures it’s impossible to reconcile the votes and the receipts aren’t designed to be robust enough to withstand hand counting.
If you or others really want a machine to accept the input from the voter because you don’t trust human fingers on paper then there’s a simple solution: make the machine produce a human readable & machine scannable piece of paper. Anyone who has used TurboTax to prepare their return and sent in a paper copy knows it can be done, and if you look at the bottom of your checks you’ll see characters that have been printed by a machine to be read by a machine.
In the case of checks that’s a magnetic ink, but it doesn’t have to be. If that kiosk produces something with consistent characters then it can be scanned… AFTER the voter holds it in their democratic little hand and sees yes, this says the exact same thing I punched into the system.
While I was very happy with the quality of my computer science education, at no point did anyone explain to me how to examine the bits in one of those Diebolds to verify they recorded what I meant.