Computer Makes $850-Million Error in Social Security

In a major computer snafu, the Social Security Administration has determined that about 700,000 Americans have been shortchanged out of more than $850 million in retirement benefits since 1972, agency officials disclosed Thursday.

Although the agency recognized two years ago that there was a glitch ticking in its computers, only recently did it determine the full extent and complexity of the problem.

It is already making back payments of nearly $400 million to about 402,000 retirees. Social Security officials now estimate that another 295,000 as-yet-unknown recipients, all of whom continued to work after they began receiving benefits, are owed back benefits of another $450 million.

“We think that the average amount of back benefits is going to be about $1,500 per person,” said D. Dean Mesterharm, deputy Social Security commissioner for systems. “We need to do the right thing. So we have decided to correct this down to the last dollar.”

The agency hopes to identify all of the 295,000 still-unknown retirees during the next year, but Mesterharm acknowledged that officials may not meet that deadline because of the complexity of the problem.

The fix will be too late for many retirees. More than 57,500 of the first group of 400,000 are dead, having missed out on benefits of $34.8 million. The Social Security Administration does not know how many of the 295,000 retirees are no longer alive.

The agency is having difficulty locating survivors, and in many cases no survivors exist to accept the money. Under federal law, retirees are not entitled to back interest on any of the benefits.

The software glitch involved the formula used to calculate benefits for individuals who continued to work after they began collecting Social Security benefits--a small portion of the 44 million Americans who will collect a total of $350 billion this year.

As a result of the glitch, some Social Security recipients who continued working did not get any credit in their benefit calculation for their post-retirement income.

The error in the formula was coded into the software system in 1972 and went undetected until late 1994, when the agency’s inspector general and its office of integrity review discovered the error.

Programmers have written a new software code to make sure the error does not happen again.

Phil Gambino, Social Security press secretary, said the problem affected only a tiny fraction of the retirees who have post-retirement income. “I don’t want to needlessly alarm every beneficiary who worked past retirement,” he said.

Nonetheless, the Social Security glitch, without doubt, ranks among the most serious computing errors in government history, when measured by the number of people and the amount of money involved.

Mesterharm said the glitch reflected problems that existed in the 1970s, not today. Although Social Security is credited with operating a fairly competent computer system, Gambino said Congress had reduced computer modernization spending below the planned level in each of the last five years.

Evelyn Morton, a legislative analyst with the American Assn. of Retired Persons, said that neither her organization nor Congress had looked into the software problem. After the initial problem was disclosed, “nobody ever asked what happened after that.”

Retirees hold mixed views of the Social Security Administration’s customer service, Morton said.

“For an individual with routine requests, there is a high level of satisfaction,” she said, “but if you have a request that needs some attention, then people say Social Security didn’t do a good job.”

The 295,000 retirees owed back payments make up about 0.7% of the 44 million Social Security beneficiaries. Although the average back payment will be $1,500, some retirees could get as much as $2,000 and others might receive only a few hundred dollars. Agency officials said they would not know the range until they compute the back benefits.

The software glitch that skewed computations for 22 years has left the agency with a shockingly complex technical problem. The basic software system is capable of making automatic adjustments in benefits only as far back as four years, Mesterharm said.

The Social Security benefit formula is exceptionally complex, vastly more so than federal income tax computations. It makes dozens of distinctions among individuals--for example, workers with railroad income are subject to different rules from those that apply to other workers. Benefit calculations are particularly convoluted for retirees who have had several spouses.

Another complicating factor is that Congress has changed the Social Security benefit formula 16 times since 1973 and each year’s worth of back benefits must be calculated with the formula that was in effect that year.

The Social Security code, the body of laws and regulations that govern the system, takes up 47 thick binders that stretch out about 20 feet along a bookshelf.

To get out the checks and track the wages of every working American requires one of the largest computer systems in the world, occupying two floors of a massive complex in suburban Baltimore. The array of computers can perform about 2.2 trillion computations per second and requires about 2,800 employees to operate.

At first, it looked as if the problem would be easy to work out if the agency simply ignored individuals who had post-retirement earnings of less than $2,200 in any past year--based on the assumption that the glitch had minimal impact on such individuals.

As the agency examined that assumption, however, it turned out to be dead wrong. Even small amounts of post-retirement earnings could have a major impact on later retirement benefits, Mesterharm said.

The agency had hoped to automate the job. So it asked its top benefit-calculation experts to list the steps they would take to determine back benefits manually. The agency’s computer specialists then attempted to mimic that process with software.

However, the agency found that more than 100 different categories of retirees existed, and even within those categories every individual’s benefit calculation involved unique characteristics.

In its initial attack on the problem, the agency had to review the files of 700,000 Americans, which yielded the finding that 400,000 individuals had been shortchanged. The more recent review enlarged that by the additional 295,000 individuals.

Despite substantial effort, none of the software systems has succeeded in fully automating the process of identifying the individuals and calculating their back benefits. Thus the agency is facing a massive workload in doing the job manually. Only a few of the agency’s employees know how to do the calculations without a computer.