The concern over punch card voting machines is not just partisan chadder. Their shortcomings may lead to biases more significant that you would have thought.
A number of things can go wrong with the punch card vote system.
When a voter punches the square for a particular candidate, it is very important to remove the chad that are punched out because they can make a difference in a systematic way that favors one candidate over another. The way this can happen has several primary causes:
How does this happen? It all starts with how the punch cards are created. When a punch card is created, a die with sharp blades arranged in a special gridded pattern descends on it as it pauses in its motion over a flat support bench or table. The punch holes are arranged in a rectangular grid, but the punch holes are not exactly identical, although very close. The die blades cut down into the card face and perforate or nearly perforate the edges of each individual punch hole. The die blades for an individual hole are arranged usually such that the corners of a square are the parts left attached. The tolerances on the manufacture of the die, irregularities in the surface of the support bench, and non-uniformities of the paper thickness (each only a micron or so different) prevent each blade from scoring an individual hole on one part of a ballot card the same as the hole on a different part of the card. Paper thickness irregularities usually do not produce systematic behavior when a hole is punched out, but table irregularities and die non-uniformities can.
An important result, however, is that if the same die is used to produce a deck of cards (and it is), then punched holes at a given position will be more similar across cards than holes at different positions on the same card.
When the voter takes her or his card, inserts it into the ballot holder and begins punching with the stylus, the manner in which the chad breaks free of its attached corners depends upon how the stylus descends into the hole (straight or crooked, twisting or non-rotating), and the relative width and thickness of each of the attached corners on each individual punch-hole on the card. If the stylus comes down the ballot machine guide for a particular candidate's hole and contacts the surface of the punch card hole in the same manner each time, then the way in which a chad breaks loose will systematically behave the same even though different voters are punching that candidate's name. Quite often the guides on the ballot machine are made such that the stylus can only come down in a near vertical fashion. The point of contact with the card will be different for each voter since each voter will place the card in a slightly different location within the machine. Nevertheless, if the amount of unscored card at the corners of each chad are quite different, the chad can have a preferred way in which it becomes separated.
For example, for the voters that vote for Candidate A in a particular race, when then punch his name, the chad may systematically tend to separate from the card at the upper left and upper right corners with the other two corners attached. The chad may swing free and hang on at the other two corners unless the ballot is cleaned off by hand after removal from the voting machine. Candidate B, however, might have his chad break free at the bottom right and bottom left corners because his chad hole was scored by a different set of blades on a different part of the support table. Candidate C might very well have the chad separate at three corners (say, for example, both bottom corners and one upper corner) since one corner of the attachment is much thicker than the others.
When the card makes its way from the voting booth to the vote tallying machine and is fed in to it, if the chad has not been cleaned from the card, then a chad hanging from the corners may have its fate determined by which corners are attached. Chad hanging for one candidate may be swept away as the card passes through the vote tallying machine, but chad hanging for another candidate may flip back over right into the hole from which it came!
For the example used earlier, if the card is fed in "top" first, then the chad hanging from Candidate A's punched hole (which separated at upper right and upper left) is always brushed and the card moved in a way that tends to prevent it from sliding or flopping back into its original hole. Candidate B's card, however, could have his chad swept up and flip back into the hole if corners hung on. What was once a vote for Candidate B now registers as an undervote! Candidate C is not invulnerable to chad restoration either. Since his chad only hung on by one corner, it doesn't need to have some part of the card reading mechanism sweep up and flip it back into its original hole--all it needs is to be slid and pushed back in!
The net result is that voters who do not clean the chad off their punch ballot can actually have their vote treated differently depending upon who they voted for in a systematic way that favors one candidate over another. In simplest terms, a voter that leaves his chad hanging for Candidate A has his vote treated differently than the voter that leaves his/her chad hanging for Candidate B or C!
There are several ways of detecting a vote using a punch card ballot. One type of machine uses a physical probe (usually a thin stiff wire) that descends from above one side of the card toward a sensor on the other side of the card. The sensor on the other side can be a physical one which senses the pressure of the probe, or it can be an electrical contact which senses the flow of current from the contacting wire that is held at a different potential. Another type of mechanism can sense a vote without a pressure sensor by determining whether the probe was able to descend into the hole or not. None of these systems in widespread use any more.
The most common system uses an array of LEDs and light-sensors. As the card passes over the array, a vote is sensed if a light sensitive diode can "see" a light emitting diode on the other side of the card. If the chad is in place, no light, no vote for that candidate. If the light gets through, tally a vote for that candidate.
LEDs age with use and steadily decline in light output. Likewise, light sensitive diodes become less responsive. LEDs and light sensing diodes must be replaced when they reach such a stage else they can incorrectly sense whether a chad was punched out or not. Electronics can also become faulty, intermittent, or even fail to register signals sent by the sensor-LED system
One symptom of such failures in voting machines is a high number of undervotes or overvotes. A vote is often registered in these LED-sensor systems by light passing through the hole and a signal is sent through the electronics which then records a vote cast for a particular candiate. If light is sensed through more than one hole for a given race, then an overvote is recorded. If the sensor or LED has degraded to the point it does not consistently "see", then an undervote can be registered.
Depending upon the construction of the ballot reading machine, different LEDs may be used (in some cases, MUST be used) to read different holes. For example, if the reader uses a 2-dimensional array of LEDs & sensors to read a card, each hole is read by a different sensor. If a 1-dimensional array is used, then holes in either a column or row are read by different LED-sensors. A faulty LED or sensor can affect the tallying results in one race or in multiple races depending upon how the ballot names are laid out on the card.
Is it reasonable to expect intermittent failures? Let me relay a direct experience of mine with intermittent failure of LEDs or sensors or the electronics which reads them which had a direct impact on my life and hundreds of others.
I work for Computer Sciences Corporation at the Space Telescope Science Institute in Baltimore, Maryland as a science planner for the Hubble Space Telescope (HST). In the summer of 2000, one of HST's main 'cameras', WFPC2, began to experience an intermittent malfunction in its shutter mechanism. The computer in the camera measures whether the shutters (WFPC2 has two shutters) are open or closed just before and just after an observation to verify that the motors which move the shutter blades have indeed moved them to the correct position. Indications at first were that occasionally one of the shutters would remain closed when it was supposed to be open.
This problem became more frequent as the months wore on and the project decided that it was potentially so serious that it was better to shut down the camera and try to test if something was occurring that could damage the instrument, even though that meant that observations that had been waiting for 4 years would be missed! Eventually, it was decided the likely problem was with aging of either the electronics (radiation damage) or the sensor-LEDs, and a flight software patch was uplinked to the spacecraft which allowed more time between internal activities and more time to read the LED-sensors. Details can be gleaned at the HST safing page which may or may not be visible to outside users.
Is it possible to cheat with such vote tallying machines? I do not know for certain, however, below is the verbatim content of a relevant email message from a (anonymous until he gives his permission) friend who has kept track of electoral procedures in his district and state (Nevada):
The Clark County elections are all electronic now. There are no paper ballots to count. They go through a big formality of giving you a paper tag after you sign the book, but it doesn't go anywhere, and the results are electronically recorded into the voting machine. I'm not sure how they are tabulated after the voting is over. There was a fellow named Steve Miller, a former City Councilman, who explained how the voters could be defrauded with the old ballot machines and the new machines too. With the old ballot cards, using the electronic counters, there were at least two ways to cheat. First was the use of a test feature in the electronic ballot counter; a card was punched in a certain way, not in the normal spots and submitted with the normal ballots. This card would be read by the voting machine as a "test" and it could be used to add or subtract votes or reset the machine to zero. All on one punched card, slick eh? The second method was to order extra ballots for an election, and then save the ballots for a few years , then discard them before using the same COLOR ballot again say 8 - 12 years later. This actually happened here in Vegas. Miller and another fellow found discarded old' ballots discarded in a dumpster that were the same color as the current election. They had the wrong date on them, but the date was only on the part the voter keeps, not on the counted part. Amazing. It doesn't take a rocket scientist , or a polisci major, to figure out how a couple of volunteers could switch the contents of a couple of boxes on the way to the count. All it would take was a phone call on how many punched ballots were needed. Many times these boxes were put in someone's trunk and driven away. Now its all in cyberspace, Trust Me. We all know computers don't lie. There are people who have programmed slot machine EPROM's to pay out on certain coin combinations, so it is not inconceivable, that code could be hidden in these voting machines. Who knows what Easter Eggs are hidden in the code? Can you imagine the power to "force" an election your way? It would be a secret worth killing for, not that any of our politicians would do such a thing.
Such cheating with punch card ballots can be detected with manual examination of the ballots or by random shuffling of the ballots with a machine recount. If the vote tallies the same, then no such 'special cards' were inserted. If the count is different, however, well, it could be dislodged hanging chad or . . . could it be someone who knows the reset and increment codes on vote ballots?
Manual recounts are open to manipulation, but even machine recounts cannot be relied upon with absolute certainty. Whoever you voted for or whatever your political allegiance, think before you criticize someone else's arguments about the propriety of recounts. There are valid fairness and legal points that both sides of the 2000 U.S. Presidential Election have put forth, but much of that seems lost in the partisanship and rhetoric. The media has not yet asserted itself broadly on the side of presenting the facts of how these machines could not accurately produce a count (weren't you concerned when the machine recounts didn't match exactly?).
It is our misfortune that voting laws and regulations have not been enacted in such a way as to take into account a fact well recognized in the sciences: whenever you measure something, there is a margin of error or a margin of uncertainty in the result. When we vote, the same principle should be applied. But how to do it?
How about we leave the office vacant for four years and see what happens? After all, by the rules whoever wins is likely to be within the uncertainty of how we count the ballots. Since there was no clear winner, why should we be stuck with either one?