I think we shouldn't actually use machines, but until the rest of the country comes around that far, here's what I've thought about voting machines.
Voting Machines are an obvious problem in recent U.S. elections. There are also a few points widely accepted:
There is some debate between the merits of Optical Scan machines which process a voter's paper ballot and Direct Recording Electronic (DRE) machines which record a vote through a computer interface.
|Technological Maturity||Established Technology. Such systems have been in use for many years.||New Technology. Most have been installed to replace antiquated systems after the 2000 election.|
Not all the bugs have been worked out yet.
|Voter Error||Stray marks on the ballot can confuse the scanner. The machine should reject such a ballot immediately, requiring the voter to try again on a clean ballot.||The dynamic computer based interface cannot create a ballot that it cannot count.|
|Machine Error||Typically 1-2% based on hand recounts. Probably due to noise in the mechanical and optical processes.||Existing machines have miscounted by thousands of votes. A well built machine would have no error.|
|Cost||One machine per polling place. $1000 to $4000.||One machine per polling booth. $1000 to $4000 each.|
In the 2004 election there were long lines at polling places that could not afford many DRE voting machines. Some people probably did not vote because of the lines.
|Accessibility||Paper ballots are not easily used by blind people. HAVA requires that people be able to vote unassisted.||The computer interface can have an auditory mode that reads the ballot to the voter, records and reads back the ballot.|
At this time, Optical Scan is in general the better solution. My primary concerns are cost and reliability. The current generation of DRE machines are too expensive and too unreliable. There should also be one HAVA compliant Optical Scan ballot printing machine at each polling place.
I would recommend hand counting all the votes and dispensing with machines entirely except for the number of issues we face on many ballots. I had 8 candidate elections and 18 direct democracy initiatives on my ballot. Perhaps each one of those should just chip in for the cost of doing business on the ballot. If we can accept that it will take human ballot counters a week or two and have more patience than the media circus is in the habit of showing then this may yet be ok. It might not even be a big cost change. A $4000 voting machine is worth perhaps 600 hours of human ballot counting time. How many votes will the machine count before it wears out and how does that compare to the number of votes that could be counted by people for the same price?
I worked a polling place on November 2nd, 2004. We processed just over 700 ballots through our optical scan machine. There were 26 separate races and issues on the ballot. 700 * 26 = 18200 issues counted. I estimate that someone decent at data entry can count an issue in 2-4 seconds. So, 36400 to 72800 seconds, or 10.1 to 20.2 hours of human counting time. You can double or triple that if you want multiple people counting every ballot to check each other's work. At 10-60 hours per election, $7 per hour, that's $70 to $420 of human labor to count the votes at that precinct. A $4000 voting machine has to last at least 9 elections, and as many as 50, without breaking down to beat the cost effectiveness of human vote counting.
See my Voting Machine vs Human Counting TCO calculator.