Census shock result of flawed estimates

How do 121,000 people disappear into thin air?

That, in essence, is what happened between 2009, when the U.S. Census Bureau made its latest population estimate for the city of Atlanta, and 2010, when the census performed its actual once-a-decade count.

It’s a huge deal, because it threatens to rob the city and Fulton County of millions of dollars and some of their political muscle. Even the state could take a financial hit.

Most federal programs — for transportation, health care, policing and social services — allocate money based partly on population figures. Every person counted by the census translates to $2,000 a year in federal funds, according to an analysis by the U.S. Conference of Mayors.

At that rate, Atlanta’s phantom people took $242 million a year with them when they vanished, in addition to some of Atlanta and Fulton County’s seats in Congress and the Legislature.

“Fulton County and the cities within the county could lose millions of dollars that are needed to serve our citizens,” the chairman of the County Commission, John Eaves, said in a written statement.

The sources of the discrepancy are complex and still very much in dispute. But an Atlanta Journal-Constitution investigation identified several factors that clearly played a part.

Some responsibility rests with the city and county themselves, because they worked to drive up the census estimates in 2005 and 2006. They succeeded because the Census Bureau failed to push back. In addition, the census estimates understated the impact of the Great Recession on Atlanta and Fulton County.

“We now know that we can improve that process of estimating the population,” U.S. Census Director Robert Groves said in an AJC editorial board meeting last week.

The gap between the 2009 census estimate of Atlanta’s population (541,000) and the count (420,000) was 22 percent — the worst discrepancy among the nation’s major cities. Only one city, Detroit, had a similar discrepancy. No others were nearly as far off the mark.

Not surprisingly, the estimate and the count for Fulton County were similarly out of whack: The count found 113,000 fewer people than the census had estimated in 2009, a difference of 11 percent. Statewide, the 2010 count was 142,000 lower than the 2009 estimate; census officials attribute most of that to Atlanta’s gap.

Elected leaders are acutely sensitive to population trends, said Doug Bachtel, a professor of housing and consumer economics at the University of Georgia. They like to see the numbers go up, up, up — and they’re not shy about hiring their own experts and exerting political pressure to make it happen.

“Any time you tell a local public official that you’re not going to get as much money as you thought you were, you’ve got an argument on your hands,” Bachtel said.

A tale of two methods

That argument is raging right now between the Census Bureau and demographers for the city and county, who suggest that the 2010 census count missed tens of thousands of people. Both city and county officials are weighing formal challenges to the count.

“We continue to take a hard look at the data,” Mayor Kasim Reed said this week in a written statement.

But the Census Bureau stands by the count, which is its constitutionally mandated reason for existing. The count gathers information directly from each household, either through a mailed questionnaire or an in-person interview.

In contrast, all estimates are approximations, based on certain assumptions as well as on data such as birth and death rates, migration patterns and housing data.

In Atlanta’s case, census officials say, some of their assumptions — for instance, that the incidence of vacant housing stayed the same throughout the decade — were simply wrong. Some independent demographers agree.

“If you go through the worst recession since the Great Depression, and a mortgage crisis, it’s going to have a big impact,” said Atlanta Regional Commission research chief Mike Alexander. “I think we all underestimated how severe the impact was going to be.”

Bachtel, at UGA, distrusts the estimates so much that he refused to use them on his Georgia Facts website, sticking with 2000 count figures and taking much heat for it.

“That’s the problem,” Bachtel said. “There’s lies, there’s damn lies, and then there’s estimates.”

But in every other city, the count and the estimates agree more closely. Which raises the question again: Why Atlanta?

How we got here

It all started back in 2005 with a number that Fulton County’s in-house demographer refused to believe.

The Census Bureau estimated that Fulton County’s population in 2004 was a little more than 814,000. That was 8 percent below the previous year’s estimate.

Demographer Alexander “Sandy” Speer, who had just been hired by the county, said he didn’t see how that was possible, given the number of apartments, condos and houses springing up around the county. He filled out an 18-page packet of Census Bureau forms to challenge the estimate.

The data he supplied, expressly allowed by the census as valid evidence, dealt with building permits, demolitions, mobile homes and estimated populations of college dorms, jails, nursing homes, rehab centers and military barracks.

Six weeks later the Census Bureau revised its estimate upward to the number he had argued for: more than 905,000.

Speer e-mailed then-County Manager Tom Andrews with the good news, saying: “Now let’s see if this results in any increases in grant money.”

“Good work!” Andrews e-mailed back.

Andrews, who now works for a firm that recruits government executives, said the county was certainly interested in “optimizing” its grant funding. But he said Speer did not inflate the numbers. “Sandy was just an absolute stickler for accuracy,” he said.

Greg Harper, a demographer in the Census Bureau’s Population Division, said there is no way to prove Speer was wrong. “Maybe the challenge result was closer to the truth at the time,” he said. “It’s really impossible to go back and reconstruct what the actual truth was.”

Fulton appealed again in 2006, pushing the census estimate up to more than 934,000.

(Fast-forward to 2010. “What I wanted to do was get over 1 million,” Speer said. He stands by the number he arrived at: 1,047,000, saying, “I did what I thought was an excruciatingly thorough job.” According to the census count, he was high by more than 125,000.)

Because Atlanta and Fulton County overlap, Speer had indirectly helped Atlanta boost its numbers. In 2006 the city used him to file a challenge explicitly on its behalf. That netted the city another 12,000 — icing on the cake.

Harper, of the Census Bureau, acknowledged that virtually every one of the hundreds of challenges filed nationwide in the past decade was approved. So long as a city or county had filled out the worksheets properly, and so long as the data could be verified, the bureau would accept the challenge as correct.

Speer said he was not trying to inflate the numbers.

“I don’t know what they’re used for, as far as funds allocations,” he said. “All I knew is they [the Census Bureau’s original estimates] were wrong, and it was very misleading.”

In Harper’s eyes, however, the bureau got into this mess by accepting Fulton’s and Atlanta’s numbers. “Without the challenges,” he said, “it looks like our estimate would have been pretty close to the census count.”

Fallacy No. 1: Building permits=buildings=people

Speer’s challenges rested largely on the number of building permits issued, a method the Census Bureau sanctioned. But just because a permit is issued, the building doesn’t necessarily get built. And even if it does, there’s no guarantee that people will move into it.

The ARC also relies on housing permits as one element in producing its population estimates. However, its numbers were consistently lower than the county’s, the city’s or the Census Bureau’s.

The difference, Alexander said, is that ARC does field checks on virtually all multifamily units — three units or more — to verify that they are built and occupied.

Speer said he lacks the staff to do that kind of checking. But Alexander said it’s vital.

“You’ve got to verify,” Alexander said.

Fallacy No. 2: Household size is static

One very small variable that can hugely affect the final estimate is the average number of people living under one roof. The 2000 census found that for Atlanta, that number was 2.3. Throughout the decade, census demographers continued to use that number in their calculations, even though various other data suggested that the shape of households was changing.

Atlanta, like many American cities, was attracting more young, single adults. At the same time, school enrollments were dropping in some sections of the city, suggesting that families with children were moving away.

According to the 2010 count, Atlanta’s average household size was 2.1. That small change, multiplied by 225,000 — the number of occupied housing units found by the count — accounts for 45,000 of the phantom people.

Fallacy No. 3: Vacancy rates are static

Another variable that drove the Atlanta estimates was the vacancy rate — the proportion of housing units that were empty. In 2000, that figure was 10 percent. Again, the demographers continued to use that number throughout the decade — through boom and bust, despite piecemeal but dramatic evidence to the contrary.

The 2010 count found that 17.6 percent of housing units in Atlanta were vacant. That meant the estimates overstated the number of occupied units by about 17,000. Multiplied by an average household size of 2.3, that error translates to more than 39,000 of the missing people.

Fallacy No. 4: Tax returns show all migration

Of course the vacancy rate is closely related to how many people are moving into an area versus how many people are leaving it. For that, the census and other demographers rely on data compiled by the Internal Revenue Service, based on our tax returns.

Each year, the IRS compares the home address each taxpayer listed on his or her previous year’s return with the one on the current year’s return. The result is the most detailed picture available of how the population is shifting throughout the country.

Alexander, the ARC’s research chief, said there’s a problem with this data: Low-income people do not file tax returns as regularly as more affluent taxpayers. If it’s primarily low-income people who are leaving an area — which is what happened in Atlanta, according to the 2010 count — the IRS data may not fully reflect that.

U.S. Rep. John Lewis, an Atlanta Democrat, finds it easy to believe that in some low-income neighborhoods, the people were there and the census count simply missed them. On the other hand, he said, he has seen neighborhoods where many houses stand vacant.

“You move around the county, and people are just not there,” he said.

What’s next?

The bureau will start taking challenges to its 2010 count on June 1.

Speer said Fulton may hire Jerome McKibben, owner of McKibben Demographic Research of South Carolina, to handle the county’s challenge. It would cost about $100,000, he said, and the county could partner with Atlanta.

McKibben believes the census count missed 14,000 housing units in the city of Atlanta and 80,000 to 120,000 people. He did his own population projection for Atlanta for the year 2009, using a method that focused on population characteristics rather than housing data. It came in close to the 2009 census estimate.

“That’s pretty good proof” that the census estimate was correct, he said.

However, the census only accepts challenges to the count based on narrow, highly technical issues. Re-counting, or mailing forms to allegedly missed areas, isn’t an option, said census spokeswoman Stacy Vidal.

Meanwhile, Groves said, the bureau continues to investigate how the estimates and the count went so far astray. And he vowed: “We’ll tell you where we screwed up.”

--------------

About this exclusive story

Our reporters analyzed complex data from more than a dozen sources to show how inflated census estimates could cost Atlanta and Fulton County millions of dollars as well as civic bragging rights.