NAPC Conference: Preparing for the Impact of Census Changes on Community Planning

The opening session of the National Association of Planning Councils (NAPC) conference was titled "Preparing for the Impact of Census Changes on Community Planning." [Update: Read about the session in this article in the Austin American-Statesman. ] Will Wynn, mayor of Austin, Texas (where the conference is being held), welcomed the 200+ people packed into the Austin City Hall and spoke of the importance of reliable, timely data for good policy and resource allocation. In growing communities like Austin, the need for updated data is even more crucial.

Judge Sam Biscoe, chair of the Capitol Area Council of Governments, said a few words about the importance of the Census products for the 10-county region.

Vanessa Sarria, of the Community Action Network, introduced the program. Ben Warner, president of the NAPC, thanked the audience, the City of Austin, the Community Action Network, and everyone else who put the program together, then got out of the way so that the real conversation could begin.

Susan Schechter, Chief of the American Community Survey (ACS), spoke first. Her powerpoint presentation will be online shortly, and I'll link to it when it is.

The American Community Survey is a nationwide survey that collects population characteristics and housing information every year, replacing the Census long form that was collected only every 10 years. After testing the survey for a few years, in 2005, the ACS expanded to cover in all counties in the 50 states and the District of Columbia (as well as Puerto Rico, where it's called the Puerto Rico Community Survey.) Other U.S. territories (such as Guam) are not covered by the ACS and will still have the Census long form in 2010.

The Census long form surveyed about 1 out of 6 households (about 16 percent of the total population.) The ACS surveys about 3 million addresses annually, or about 2.2 percent of the population. This means that the ACS has some tradeoffs between reliability and currency (timeliness, not money).

The Census long form data was a point-in-time survey. The ACS instead is a continual survey process. Every month households are surveyed. For annual estimates, twelve months of data are combined. For three or five year estimates, 36-month and 60-month data sets are used.

For areas of 65,000 or more, one-year estimates of population characteristics are available. For areas of 20,000 and up, three-year estimates are available. For areas under 20,000 population, the only estimates available are five-year estimates.

In 2006, the ACS sample was expanded to include the population living in group quarters. Group quarters include nursing homes, correctional facilities, military barracks, and college/university housing among others.

In 2008, the first 3-year estimates were released. The first 5-year estimates will be released in 2010.

In 2008, there was also some major changes in the survey, including data about people with disabilities. Data users should pay close attention to the ACS guidelines for how to use the data and which data can be compared to prior years.

Alfredo Navarro, Assistant Division Chief, ACS Statistical Studies, U.S. Census Bureau, spoke next. He provided more technical information on the ACS (his presentation should also be online shortly.) He recommended a series of publications on how to use the ACS, including a new design and methodology report due out in March.

After this overview, Jim Walker, Executive Director of the Central Texas Indicators Project (and a board member of the Community Indicators Consortium introduced the panel charged with discussing the challenges that arise as a result of the Census changes and how community leaders can prepare for those changes. Panelists included:

• Susan Schechter, Chief, ACS Office
• Alfredo Navarro, Assistant Division Chief, ACS Statistical Studies
• Robert Kominski, Assistant Division Chief, Social Characteristics Housing and Household Economic Statistics Division, Census
• Nicole Scanniello, Coordinator, ACS Organization
• Karl Eschbach, Texas State Demographer and Director of the Texas State Data Center
• Ryan Robinson, Demographer, City of Austin
• Mark Salling, Research Director, The Center for Community Solutions, Cleveland, Ohio
• Sarah Eckhardt, Travis County Commissioner

The panelists raised some of the challenges faced by data users in transitioning to the new data products. The Census data was predictable and understandable, and had clear information available for small areas. While data users understood there was a margin of error with the information, the margin of error was small and relatively constant, and was effectively ignored by policy makers. It was a single source that could be used by researchers and legislators and communities.

Now there are multiple data products, and data users have choices on which data set to use (if they are looking at a geography of over 65,000 or even over 20,000 population -- under 20,000 has no choices). Choices can bring confusion. The margins of error are necessarily bigger. The one, three, and five-year estimates don't match in a given year (they shouldn't, after all, unless the trend lines were static, but that isn't always intuitive.)

Small-area data and subgroup data are less reliable than in the past, because the sample sizes are smaller. Even five-year estimates will only have a sample size of 11 percent of the housing units, lower than the 16 percent from the Census long form. Other knowledge is often needed to interpret the data.

But there are benefits to the new system, not all of them obvious or just related to timeliness. Instead of temporary census workers distributing the long form and following up, a professional staff of researchers does the work full-time. The reduction in non-sampling errors makes up for the reduced-sampling margin of error. (My translation: the Census long form had bad data in it that we all ignored because it was all we had. Having people collect the data who know what they're doing is a Very Good Thing.) My favorite quote from the session, which I will leave unattributed, was this: "We had census tracts with zero answers to some of the questions on population characteristics, so we imputed the data. "Imputed" means we made it up." (Hasty statement from Someone In Charge: "The Census doesn't say that. The Census has a scientific methodology for careful imputations.")

The ACS has a 97 percent response rate, which is pretty good. We were cautioned, however, not to try to compare 3-year estimates in one area to 1-year estimates in another -- the same level of data should be used consistently. We were also reminded that care should be used in comparing three-year or five-year estimates to the following year's three-year estimates (or five-year.) The difference between the 2005-07 and 2006-08 estimates isn't the change from 2007 to 2008; it's the change between 2005 and 2008.

All in all, a surprisingly enjoyable and incredibly informative session. You should have been there.