Programming Errors Often Blamed : Accurate Vote Tally With Computers Can Be Elusive

When Democrat Buddy MacKay narrowly lost a U.S. Senate seat to Republican Connie Mack in Florida last November, the credibility of computerized vote-counting took a beating.

Mack won by 34,518 votes out of more than 4 million cast in an election in which there were puzzling differences between the total vote for President and for U.S. Senate.

In four populous counties where MacKay was thought to be ahead--Dade (Miami), Hillsborough (Tampa), Palm Beach and Sarasota--almost 200,000 more people voted for President than for the U.S. Senate.

All four counties have the same “Votomatic” punch-card tabulation system used by many California counties, including Los Angeles.

In those four counties more people voted in such races as state treasurer and secretary of state than in the U.S. Senate race, even though the Mack-MacKay contest was a bitter, well-publicized contest.

“That’s peculiar,” veteran elections analyst Richard Scammon observed.

“Something strange happened there,” said Robert W. Flaherty, executive director of News Election Service, a national vote-reporting service owned by the three major television networks and the Associated Press and United Press International wire services.

Flaherty’s service projected MacKay the winner, as did both ABC and CBS.

MacKay and his campaign advisers agreed that “something strange” had happened.

“Our late polls showed us with a 5% to 9% lead,” MacKay said in a telephone interview from his Miami law office. “We knew that would be cut some because (Democratic presidential candidate Michael) Dukakis was sinking into a black hole in Florida, but we thought we had enough of a cushion to win and I think we did have.”

MacKay’s advisers wanted to challenge the results in five counties, adding Broward County (Ft. Lauderdale) to the other four.

However, Florida law leaves recount decisions in the hands of local canvassing boards, which generally approve a new tally only when there is a claim of fraud.

“It’s a real Catch-22 situation,” MacKay said. “You’ve got to show fraud to get a manual recount, but without a manual recount you can’t prove fraud.”

Even when a recount is ordered in a computerized election in Florida, it is usually done on the same system that produced the first results, which one critic said is “like singing the same song twice.”

Of the five counties where MacKay sought a recount, only Palm Beach County agreed. There, a manual check of 10 precincts produced almost the same result as the computerized tally, causing the Democratic candidate to drop his protest in that county.

“It was a helluva way to have your political career end, I can tell you,” MacKay said.

In the aftermath, supervisors of elections in the affected counties concluded that a ballot placement problem led to the sharp vote drop-off.

Confusing Ballot

In several counties the Mack-MacKay race was listed at the bottom of the ballot’s first page, below all of the presidential candidates, including minor Libertarian and New Alliance Party candidates.

“A lot of people simply missed the race,” said David C. Leahy, supervisor of elections in Dade County. “They figured the whole first page was for presidential candidates, so they voted their choice and then flipped the page without ever seeing the Senate race.”

But if that was the reason, there should have been a sharp drop-off in all counties where the Mack-MacKay race appeared on the same page as the presidential candidates.

The ballot layout was almost the same in Pinellas County (St. Petersburg) as it was in Hillsborough County across the bay, yet the drop-off in votes between President and Senate was less than 1% in Pinellas, while in Hillsborough it was 25%.

Discrepancies of this kind have led many observers to speculate that computer programming errors, accidental or intentional, were at least partly responsible for the outcome.

“I think somebody made a mistake in programming,” said Robert J. Naegele, who has been California’s chief consultant on computerized vote-counting for more than 20 years.

In the aftermath of the Mack-MacKay contest, legislation has been introduced to strengthen Florida law by requiring state certification of vote-counting systems, providing for manual recounts of at least 1% of the votes and requiring that “source codes,” the heart of the vote tabulation programs, be deposited with the secretary of state so they can be checked in case of disputed results.

(California already requires state certification, as well as a 1% manual recount, and two bills calling for source codes to be deposited with impartial “escrow agents” are pending in the Legislature.)

The U.S. Senate race in Florida was the latest in a series of flawed elections in which the results were tabulated electronically.

In a report published last year, Roy G. Saltman of the Federal Institute for Computer Sciences and Technology listed numerous recent instances of mistakes in computerized vote counting. In one of these--the 1985 Dallas mayoral election--incumbent Starke Taylor narrowly avoided a runoff in a contest that had several peculiar features.

When the vote-counting computer experienced a power failure on election night, Max Goldblatt, one of Taylor’s opponents, was leading, but when the power came back on, Taylor had mysteriously moved ahead.

When a machine recount was ordered, only 89 out of 250 precincts showed the same totals. Even stranger, in 109 of the 250 precincts, the total number of counted ballots changed between the original tally and the recount.

When the official canvass for this election was published, it contained three different vote totals, evidently because the computer did not correctly tabulate 11 “split precincts” in which some voters lived within the Dallas city limits but others did not.

Legislative efforts to investigate these mysteries were thwarted because the ballots and other documents had been destroyed.

Federal law requires that ballots and other materials be retained for 22 months after an election for President, the U.S. Senate or the House of Representatives, but state laws governing local races vary widely. California law requires that the material be kept for six months after the official result is announced (longer if there is a dispute) but no such requirement covered the 1985 Dallas mayor’s race.

Eventually, the Texas Legislature passed a law mandating, among other things, the certification and testing of vote tabulation systems, manual 1% recounts, clear “audit trails” that can be followed to check disputed results and the deposit of vote-counting source codes with the secretary of state.

Other Examples

The Saltman report included these other examples of problem elections:

* In Gwinnett County, Ga., in 1986, the results of a state Senate race were overturned because of tabulation errors resulting from “handling procedures, the ballot puncher, the vote counter, the punched cards’ density, vote position on the ballot card, human error and pure chance,” Saltman wrote.

* In Moline, Ill., a losing candidate for alderman was declared the winner after it was discovered that a malfunctioning timing belt in a card-reading machine had deprived him of 92 votes.

* In Oklahoma County, Okla., in 1986, some machines failed to count up to 10% of the ballots per precinct.

* In Stark County, Ohio, a recount of a close race for county commissioner produced 165 votes more than the original tally and a different winner. But this result was reversed again when it was discovered that the program used in the recount failed to distinguish between voters of different political parties.

Saltman concluded that most of these problems were because of human error, not fraud or computer malfunctions, but that is small comfort to citizens whose votes have been invalidated.

“Most problems attributed to the computer system are simply human errors, not software or hardware errors,” Lance J. Hoffman, professor of electrical engineering and computer sciences at George Washington University, wrote in a separate report published in 1987. “Although fraud was not involved . . . the effect was the same: Ballots were improperly tallied.”

A common source of error in any punch-card tabulation system is “chad”--small pieces of ballot paper that sometimes cling to the underside of the ballot card after the ballot has been punched.

As ballots are handled when they arrive at counting centers, or as they move through the card readers, a piece of chad is sometimes pressed back into the ballot from which it came, or into the next ballot in the deck. When this happens, the card reading machines register no vote for that race.

A second problem is caused when pieces of chad pop out as ballots are being handled by the many people who wrap, sort and inspect the cards before they are counted. This can mean that two votes are shown for the same race, and that kind of “over-voting” causes the computer to invalidate the vote for that candidate.

Chad also can cause the card readers to jam.

Ralph C. Heikkila, assistant registrar-recorder for Los Angeles County, said that only 1% to 2% of votes cast at the polls have chad problems but that about half of the absentee ballots do. That means almost 200,000 ballots cast in the November, 1988, election, could have been affected.

An especially disconcerting aspect of the chad problem is that no two counts of the same election are likely to be the same, which reduces voter confidence in the outcome.

Poorly prepared ballot pages or ballot holders can result in votes being cast for the wrong candidate or issue in a punch-card election. Because Los Angeles County assembles about 45,000 ballot holders for a major election, there are many opportunities for mistakes of this kind.

Experts’ Opinion

Votomatic and other punch-card systems present so many potential problems that some experts have recommended that they be abandoned.

“There is no circumstance under which I would conduct an election with punch cards,” said Michael Shamos, who is one of three computer experts who certify electronic vote tabulation systems for the state of Pennsylvania. “I would much rather hold a town meeting and have people raise their hands--I’d get a much more accurate tally.”

But Heikkila disagreed. “Basically, the computer systems are all good, if people know the procedures and follow them,” he said.

Some errors are caused by rushing to produce results, to meet the demands of the press and of candidates and their supporters.

“They (election officials) are terribly afraid of the media,” federal computer expert Saltman said. “They’re afraid they’ll be criticized if the results come out slowly and sometimes that causes mistakes.”

But Naegele, the California consultant, said, “The longer you wait to get the answer out, the more suspect the answer is.”

Saltman has made several recommendations to improve electronic vote tabulation and to build public confidence in these systems.

‘Hanging Chad’

He proposed that the prescored punch cards used in the Votomatic, the most widely used vote-counting system in the country, be eliminated because of the “hanging chad” problem.

Saltman also stressed the need for complete audit trails--records of what is happening inside the computer at all times--that can be consulted later if results are challenged.

Saltman recommended better election management and tighter “internal controls”--hiring professional internal auditors who can provide election officials, most of whom are not computer specialists, with as much assurance as possible that the systems are working right.

State certification of vote-counting systems also should increase the probability of accurate results.

California and some other states already do such testing, subjecting the hardware to rough handling and to fluctuation in temperature and humidity. The vote-counting software programs are also tested, to make sure they are written in understandable computer language that could be read by others if disputes should arise.

But many states have superficial certification procedures or none at all.

This could change if states adopt the voluntary standards for computerized elections that are about to be issued by the Federal Election Commission. One of these standards calls for new vote tabulation systems to be tested by one of half a dozen or so “independent testing authorities.”

Careful “logic and accuracy” testing before and after each election also should improve the reliability of results.

Other Testing Methods

“Logic” testing determines if the machine is reading the various ballot styles correctly (in a large election jurisdiction there can be hundreds of different ballot styles, as voters are divided into congressional districts, school districts and other subgroups).

“Accuracy” testing should reveal whether the card reading machines, through which the ballots pass at blinding speeds, are functioning properly.

But many election officials “don’t understand what they are testing for and don’t know what to do about a problem when one arises,” Naegele said. “Operating entirely out of ignorance, they see mistakes but do nothing about them. . . . In that case, you might as well not bother testing at all.”

In the end, American elections will probably never produce precise results because they are so decentralized.

That is what the U.S. Constitution intends, granting, as it does, control over elections to the states, which pass most of the authority on to cities, counties and other election jurisdictions.

Unlike most Western democracies, the United States does not attempt a single, official count in presidential elections. The result, said I. A. Lewis, director of the Los Angeles Times Poll, is that a national election “is only the best approximation of what actually happened.”

But the country muddles along--with some excellent, scrupulously honest election officials and some who are incompetent, poorly trained and occasionally dishonest; with vote-counting equipment that sometimes works and sometimes does not; with only partially successful efforts to improve the equipment and reform the system.