Counting on new census approach
About this time every decade, the U.S. begins to hanker for the next great national self-portrait. That old still life that told us who we were -- magnificent as it once was in scope and detail -- has grown dusty and is ready for the closet.
It’s the time when the Census Bureau begins mustering the army of enumerators and tabulators who will reach across the country to tally us up in all our dynamism and diversity.
But in the age of the information superhighway, who expects to wait until 2010 to know how many grown-ups in the neighborhood have moved back in with Mom and Dad?
So today, doing as much as a tradition-bound institution can to embrace the future, the Census Bureau is doing something different: premiering what amounts to Census, the Movie.
At 9:01 PST tonight, census officials will release to the public a giant collection of data that will offer the first geographically detailed look at the new, annual American Community Survey. (Locate the “American Fact Finder” site through an Internet search engine and then follow the links to the American Community Survey.) The change marks a turning point for professional demographers and amateurs alike, promising a much fresher look at social trends but introducing nettlesome limitations.
“The good news is we’re now going to have data every year,” said William H. Frey, demographer for the Brookings Institution in Washington, D.C. “Ten years is a long time to wait in a society that moves as fast as ours does.”
The downside is that instead of a single, massive sample providing a snapshot of information clear down to individual neighborhoods, the survey will be a rolling average of samples taken each year.
The Census Bureau, which has been previewing the survey with smaller releases over the last three years, has compared the change to switching from a photograph to a video. Ken Hodges, chief demographer for the firm Nielsen Claritas, sees it more as time exposure.
“Instead of a snapshot of a guy running across the street, you’d see a photo of a guy in the street and it looks like he is running,” Hodges said. “You see this blur running across the street. It might look a little like a guy.” The blur, he said, is analogous to what demographic changes look like when averaged over several years.
The rolling estimates present an obstacle for demographers including Paul Ong, professor of urban planning, social welfare and Asian American studies at UCLA.
High on his agenda for studies of poverty at the neighborhood level in Los Angeles are questions about highly volatile data such as income, unemployment and housing costs. With a blurrier sample, “that information is not there,” Ong said. “That’s a problem.”
The timing of the current release, during the sharpest economic downturn in decades, highlights the drawbacks of averaged data.
Averaging over time “only works if things aren’t changing very much,” said Dowell Myers, professor of urban planning and demography at USC. “When you have a steep trend, you want to know where things are changing now. You don’t want to know where it was changing two years ago.”
As Census Bureau director Steve H. Murdock points out, the treasure trove of self-discovery that the census has become -- how much money we make, who our ancestors were, how much time we spend in our cars, what race we identify with, whether our kids go to private or public school -- is incidental to the original purpose of the census.
The Constitution mandates that once every 10 years the government count the population to apportion electoral districts. That once-a-decade count will continue. What has changed is the way the census will query Americans about all those other subjects that, since 1940, have been detailed in a questionnaire known as the long form.
Those questions are not just for curiosity. They provide essential data for distributing money from federal and state programs. “In general, we say $300 billion of federal funds per year is allocated on the basis of census data,” Murdock said. “Every question is required by some part of the federal establishment for the administration of programs.”
Responding to the social mores of the times, recent censuses have captured how racial attitudes are changing -- not much -- and probed tentatively into the bedroom, telling us how many Americans consider themselves domestic partners with someone of the same sex.
In the past, the census sent the long form to about 15% of American households. It used that sample to provide detailed estimates on demographic facts every 10 years. But, going back to the 1990s, government’s need for more up-to-date information was becoming painfully evident.
The answer was continuous sampling. A full-time staff now sends out long-form questionnaires to 250,000 households each month and follows up to get as many as possible returned. As the years go by, those samples will pile up into an increasingly rich view of the nation’s continuous change.
But even averaged over five years, the annual survey includes fewer people than the former long-form sample. That exacerbates the problem statisticians know as standard error. Estimates are derived mathematically from a fraction of the population. The precision of the result depends on how large the sample is.
For questions that concern the country as a whole, there’s no problem. But sampling error can bedevil the search for answers about small areas or uncommon characteristics -- an individual neighborhood or small city or a group such as single men heading households that include children under 18.
“If you’re looking at big-picture stuff, [the American Community Survey] is clearly better,” said John Blodgett, programmer and data analyst at the University of Missouri’s Office of Social & Economic Data Analysis.
“If we want to know how old we are, or how the Hispanics are doing, or how many unmarried partners are living together, that’s fine,” he said. “How we’re slipping is what’s going on in Little Korea over those five years. That’s been lost.”
Over the last three years, when the new survey was being previewed with annual releases, the problem has been particularly acute. The census could provide data only on places with populations of 65,000 or more. That meant no data on many small Southern California cities and 18 entire rural counties.
With today’s release of the three-year average, data will provide information on places as small as 20,000 in population. But the data will still treat even much larger cities, including Los Angeles, as undivided wholes. Demographers will have to wait until December 2010 for the next advance in the survey -- a rolling five-year average that will paint a picture down to the level of the census tract, essentially a neighborhood.
A rolling five-year average will follow each December after that.
Blodgett, like other demographers, said a larger sample size would make the census data more useful by reducing the margin of error. Murdock, the census director, said he would like a larger sample size but that budget constraints made it unlikely.
And so, until the next great turning point in the census, the palm readers of academia, business and the news media will still have work to do converting this imperfect treasure into the information that people want.
“I think it’s going to take a decade, maybe more, for people to figure out the nuances and the opportunities that these new data provide,” said Frey of the Brookings Institution.
Frey said it took demographers decades to fully tease out the tapestry of the long form, and it will be no different with the community survey.
“I don’t think everybody is going to be able to mine it for its full value immediately,” Frey said.
“It’s going to take time, and the analysts are going to have to lead the way.”