State Counts on 1% Sample to Detect Fraud

California election officials count heavily on a 1% manual recount of randomly selected precincts to protect against inaccuracy and fraud in election returns that are counted by computer.

The state-mandated 1% hand count is “the last check on the system,” said Deborah Seiler, assistant to the secretary of state for elections and political reform.

“The best thing we have going is that 1% manual recount,” said Ralph C. Heikkila, who is in charge of elections in Los Angeles County.

The county actually counts more than 1% since state law also requires that at least one precinct in each ballot style be counted by hand. In last November’s general election the county used 388 different ballot styles, with voters divided into congressional districts, Assembly districts, school board districts, etc. This meant that 100 precincts, or about 1.5% of the total, were hand counted.

Statisticians Doubtful

But is a 1% or 1.5% manual recount enough to catch error or fraud? Statisticians consulted by The Times were doubtful.

The chances of finding one wrong precinct out of 7,000 (there were about 6,200 in Los Angeles County in November, 1988) would be about 1 in 100, according to Donald Ylvisaker, director of the Division of Statistics at UCLA. But if 1%, or 70, of the precincts were bad, then the prospect of finding one or more of them with a 1% hand count would rise to 1 in 2, Ylvisaker added.

Solomon W. Golomb, professor of electrical engineering and systems engineering at USC, agreed with this calculation.

“If you sample 1% of 7,000 precincts, and there are 70 precincts with problems, the chance that you would pick one up is almost 50-50,” Golomb said.

“If there is systematic, widespread fraud going on, they have a reasonable chance of finding it with a 1% check,” Golomb added. “But if it is limited to a few precincts, the chances are much poorer.”

Both Ylvisaker and Golomb pointed out that sampling 5%, or even 3%, of the precincts, would substantially increase the chances of detecting error or fraud.

Robert J. Naegele, California’s chief consultant on computerized vote tabulation, said he recommended 10 years ago that 10% of the precincts be recounted by hand but the secretary of state’s office said that would be too expensive.

One estimate is that the 1% recount cost about $350,000 statewide in November, 1988. A 3% recount would cost slightly more than $1 million and a 5% recount would cost about $1.5 million.

But Heikkila of Los Angeles County said more than cost is involved. He said a larger sample would not necessarily improve the chances of finding error or fraud. Heikkila pointed out that candidates, campaign advisers, journalists, political analysts, university professors and others “pore over these results long after the election is over.”

“They know the voting patterns,” he added. “If something is off, they’ll find it” faster than a hand recount of any percentage.

1.5% Recount Questioned

But Naegele replied, “even if these people find something, it’s after the fact; the election is over. The time to prevent mistakes is before the election, not after.”

Roy G. Saltman of the federal government’s Institute for Computer Sciences and Technology described a situation in which the 1.5% recount used by Los Angeles and many others would be inadequate.

Saltman said a clever perpetrator of fraud “would select an intermediate number of precincts and make changes small enough” so that neither election officials nor party workers would detect them.

“In that case, you’re dealing with a simple statistical problem,” Saltman said. “How many precincts do you have to check to find a bad one? In most cases, I think 4% or 5% of the precincts would do it . . . but in that kind of situation, 1% is clearly not enough.”