http://www.wheresthepaper.org/CACM_YaleStudy.htm

Teresa Hommel

12/3/04

Must Evote Machines Be 100% Accurate?

Comment on "Small Vote Manipulations can Swing Elections"

If an evote machine shifts a few votes from one candidate to another, only a careful audit would discover it. And if discovered, so what?

Elections are not expected to be perfect. If 200 ballots are cast on a machine, and the machine tally is five votes different from a hand-count tally of voter-verified paper ballots, some people might be curious but few would be suspicious. Who would demand a statewide recount? Except in rare cases, five votes cannot change the outcome of a statewide election.

But what if those five votes are the tip of an iceberg? If the state has 40,000 evote machines, and each one shifts five votes, the recipient gains an advantage of 400,000 votes. (200,000 votes are switched. Each vote switched means that the voter’s choice loses one, and the recipient gains one, causing a net advantage of two votes to the recipient.)

Would you like your statewide recount now?

Computers used in business are audited continuously, and 100% accuracy is required at all times. Computers used in elections should be held to the same standard, or you can kiss your democracy goodbye.

Our ideas about “close elections” are based on the realities of paper and punch-card ballots or mechanical lever voting machines. With our old equipment, switching votes in every precinct requires vast armies of corrupt insiders. With evote equipment, it requires one person with modest skill and a few minutes access to the master copy of the software before the election, or to the central tabulating equipment afterward.

Charges of computer fraud in our November 2 can never be proved or disproved because the evote computers either cannot or will not be audited. Democracy cannot survive such misuse of technology.

A study by four Yale students has shown that altering only a single vote per machine would have changed the electoral college outcome of the 2000 election. Changing only two votes per machine would have flipped the results for four states.

DiFranco et al, "Small Vote Manipulations Can Swing Elections," Communications of the ACM, Vol. 47, Issue 10, October 2004, pages 43-45,

http://doi.acm.org/10.1145/1022594.1022621

The article is available here Copyright ACM, posted by permission.

An abstract of the article is available here

The ACM Portal to the article is here.

Applying the study to New York State.

1. Assume that polls are open from 6 AM to 9 PM. This is 15 hours or 900 minutes.

2. Assume that 2 evote machines are needed to replace each lever machine.

Evote machines are sold to handle 175-200 voters per day, but in Florida in early voting and on 11/2/04, voters averaged 10 minutes each (6 per hour). In 2002 in some areas in Florida, the ballot was long and many voters required 25 minutes.

New York's legal time limit per voter is 3 minutes, allowing 300 voters per 15 hour day if each voter takes the full three minutes. Usually each voter is faster, taking between one and two minutes.

3. Voters per machine per 15-hour day (900 minutes).

Column 1 is "minutes per voter"

Column 2 is "how many voters can vote in 900 minutes"

1 --- 900

2 --- 450

3 --- 300

4 --- 225

5 --- 180

6 --- 150

7 --- 128

8 --- 112

9 --- 100

10 --- 90

4. New York has about 20,000 lever machines. If replaced by 2 evote machines each, that gives 40,000 evote machines.

5. Vote Switching, assuming 40,000 evote machines, and a statewide candidate such as Governor or US Senator.

Column 1 is "Votes switched per machine"

Column 2 is "net advantage to recipient"

1 ---- 80,000

2 --- 160,000

3 --- 240,000

4 --- 320,000

5 --- 400,000

6 --- 480,000

7 --- 560,000

8 --- 640,000

9 --- 720,000

10 -- 800,000

6. If an evote machine with a voter-verified paper audit trail handles 200 voters, and a hand count of the ballots shows 5 votes switched, that can indicate that statewide 200,000 votes were switched, giving an advantage to one candidate of 400,000 votes.

Will a 5-vote error on one machine trigger a statewide recount? Or will everyone say, "Even if we give the candidate 5 votes extra, that would not change the statewide race which was won by a 400,000 vote margin?"

If computers are used in voting, because of the ease of vote switching like this on a statewide basis (via manipulation of the original copy of the software), audits would have to show 100% accuracy in order to eliminate the possibility of this kind of error or fraud.