Moving from the Census to the American Community Survey Richard Lycan Population Research Center Portland State University North American Cartographic Information Society October, 2008

PRCNACIS PRC – Who we are ► An applied research center:  College of Urban and Public Affairs  Portland State University ► The Oregon link with the U.S. Census Bureau for:  Population Estimates  Census State Data Center ► Courses in applied demography ► Contract demographic research ► Staffing a mix of demographers, geographers, and planners ► George Hough, PRC Director, usually with me to answer the tough questions

PRCNACIS Outline for Presentation ► Some background and history on the Census of Population and Housing and the American Community Survey (ACS) ► What demographers and other scholars are saying about the pros and cons of the ACS ► Examples showing use of data from the ACS ► Sources for more information

PRCNACIS Part 1 – Introduction to the Census and the ACS ► There will be no 2010 Census long form data on income, poverty, housing values, and dozens of other socio-economic characteristics of persons, households, and housing units ► Instead, these data are becoming available from the American Community Survey (ACS) on an annual basis

PRCNACIS A Very Short History of the Census ► Article 1, Section 2 of the U.S. Constitution requires that a census of the population be conducted every ten years so that the representatives in Congress and direct taxes might be apportioned. ► Federal marshals conduct the first census by going door-to-door through the 13 states plus the districts of Maine, Vermont, Kentucky, and the Southwest Territory (Tennessee). ► Statistical sampling techniques are introduced, which allow the Census Bureau to create a “long form” answered by only a subset of the population, the source for topics such as income and poverty. ► Mail surveys, computers, the internet, wars, Elvis, TIGER ► 1990’s – Initial planning for “continuous measurement” approach used in ACS ► Last use of the “long form” in the Decennial Census. To be replaced by the annual American Community Survey (ACS) ► Initial public release of the ACS data. ► 2010 – Release of the ACS data for all geographies down to block group level

PRCNACIS An Oversimplified View of the 2000 Census ► Build master address file (MAF) of all residential addresses in US ► Refine and finalize the questionnaire ► Build the maps and geography files ► Test the census – survey, tabulate, evaluate ► Hire thousands of temporary workers ► Collect the data: mail-out, field follow-up ► Tabulate the results: make many adjustments, create summary tables ► Publish the results: print, CD, Internet ► Get ready for the 2010 Census

PRCNACIS The Paired Surveys ► Since 1940 there have been two key components to the Decennial Census  The “short form” survey is comprised of approximately 7 items asked of every household, the results sometimes known as Summary File 1  The “long form” survey comprised of approximately 40 items asked of about 1 in 6 households, the results sometimes known as Summary File 3  The two components are linked in that data from the “short form” are used to adjust results from the “long form” by age, race, housing units, etc. No “long form” in 2010

PRCNACIS How the ACS Collects Data ► Surveys approximately 250,000 households each month ► Uses professional survey staff rather than temporary hires as in the Decennial Census ► Uses a Master Address File to attempt to identify all households in the US. Also gathers data from group quarters residents ► Uses a mail-out census form with follow-up telephone and field interviews ► The survey contains 55 questions for persons and 30 for housing units, about same as long form 2000 census

PRCNACIS Tabulation and Publication of the ACS ► Similar to the 2000 Census  Tables are similar to those for the sample data (SF3) in the 2000 Census  Geographies are similar to those in the 2000 Census data – states, counties, cities, tracts, school districts, zip code areas, and many others, but no block level data. There will be Public Use Micro- sample (PUMS) data ► Differs from the 2000 Census  Data will become available sooner after collection  Data will be available on an annual basis  Some data will be available for a single year but much will only be published for 3 and 5 year averages  There will be no paired “short form” data to use to adjust the ACS survey results

PRCNACIS Geographies for the ACS ► The geographies for the ACS are basically the same as for the 2000 Census  The hierarchy of states-counties-tracts-block groups  Metro areas, cities, census designated places  Other geographies – congressional districts, zip code areas, school districts, TAZ’s, rural areas of 60,000+ population ► Some boundaries change: political jurisdictions annually, school districts every two years ► Block groups generally the smallest geography available, no block level data ► Reference maps in PDF format and boundary files in SHP format on

PRCNACIS The Release Schedule ► One year, three year, and five year data ► Data now available for larger areas such as counties, larger school districts, PUMS areas ► Issues of statistical reliability and confidentiality limit detail ► Multi-year data a solution Note that only the beginning and end years differ in the data

PRCNACIS For Example: ACS Data For Oregon School Districts ► Now – Single year data only available for districts 65,000 and over ► Nov – three year data for districts 20,000 and over ► Late 2010 – five year data for all districts, plus 3 year and 1 year data

PRCNACIS Part 2 – What Demographers and Others Are Saying about the ACS

PRCNACIS Pay Attention to Sampling Errors ► ACS sample smaller: For five year data the ACS sample is smaller than that for the Census long form ► We have sinned: Sampling errors a problem with previous census data, but we tended to ignore, especially in mapping of the data ► You can’t ignore it: The ACS provides explicit data on confidence limits whereas the 2000 census provided calculating formulas

PRCNACIS Data Will Be Better for Large Areas, Poorer For Small Areas ► Annual data: At the state, county, large school district level publication of annual data will allow tracking of changes ► Better surveys: The use of professional surveyors and careful use of telephone and field follow-ups will result in more complete questionnaires and less “imputation” for missing entries ► Larger sampling errors: For small areas such as census tracts data will only be available as five year averages. Sampling errors may be large

PRCNACIS Weighting of the Sample ► Need to inflate sample: If you sample 1 in 10 households, the results need to be inflated by a factor of roughly 10 (more complicated than this) ► The long and the short of it: The Census long form data could be weighted from the short form data by households, population, race, age, sex ► Weighting weaker for ACS: The ACS data will be weighted from various Census Bureau estimates

PRCNACIS Some Suggestions on Using ACS Data – Linda Gage ► Don’t try to analyze all of the data at once, even if you use all of the items. ► Concentrate on the data items that you already use in your work. ► Don’t assume the Census data are more accurate than ACS. ► Compare Census and ACS data to administrative records that you have available. ► Consider whether the data make sense. ► Learn to use and provide the standard errors provided with the ACS. ► Share what you learn about using the ACS with the Census Bureau and other professionals.

PRCNACIS Part 3 – Examples Using the ACS

PRCNACIS Example 1 – Viewing Tabular Data from the 2005 ACS ► The next data are for Public Use Micro-Sample Areas (PUMAs) and school districts with populations 65,000 and greater ► PUMA’s have more than 65,000 persons and single year data are available for all of them. ► However, sampling errors can be large. Data for sub-county areas Several counties combined to Reach 100,000 threshold

PRCNACIS Data for Oregon School Districts ► The “margin of error” is for the 90% confidence level. This is the type of error reporting used in the ACS. These data are for a PUMA, many of which have over 100,000 population ► Note the range for kindergarten enrollment of 1,046 – 2,068

PRCNACIS Or, consider the variability in enrollment data for school districts ► The values in red show estimates that have a margin of error at least 20% of the estimate, then at least 10%

PRCNACIS Example 2 - Analysis of change using census tract level data ► Housing tenure – Large universe, modest change from year to year. We examine the change from 2001 to 2003 using ACS five year data for and ► Citizenship – Small universe (foreign born) and modest change from year to year. We use the same data as above.

PRCNACIS Housing Tenure Change - Census Tracts ► A large universe. Overlapping time periods. ► % in 2001 ► % in 2003 ► Varies from county mean ► % Change 60% significance 80% significance 90% significance 95% significance Tracts with values significantly different than the county average of 42.7% There remain a number of tracts for which change over time is statistically significant and they appear not to be random in space.

PRCNACIS Housing Tenure – Grid Map Generalization ► Choropleth mapping ► Allocation to census block centroids ► Grid map generalization  0.25 sq mi  0.50 sq mi  1.00 sq mi  1.50 sq mi The highly generalized form of the contours may lend itself to easier verbalization of the spatial patterns. How would one assess statistical significance?

PRCNACIS Change in Foreign Born Who Are Citizens – Census Tracts ► A smaller universe than for housing. Overlapping time periods. ► % in 2001 ► % in 2003 ► Varies from county mean? ► % Change 60% significance 80% significance 90% significance 95% significance Significantly different from county percent of 37.4 who are citizens This doesn’t look good! Only a few tracts show change that is statistically significant, fewer than one would expect by chance. I would not publish this map.

PRCNACIS Example 3 – The Published Tables Are But One Representation of Reality ► The following example shows the large sampling variability in the block group data using % rental households of all single family households. ► The universe of households was recreated from the 100% data from the ► 20 random samples were drawn on a 1 in 6 basis. ► Choropleth maps were draw for each sample? ► Which map is the right one, the best representation of reality?

PRCNACIS Here are 20 “might have been” samples for 1990 SF3 sample data ► The following sequence shows 20 possible sets of data that might have been obtained in the sampling process. ► Note the considerable local variation but also the persistence of some of the high values ► The lesson: Be cautious about interpreting small area sample census data. Which of these 20 representations of reality is closest to the truth?

PRCNACIS ► On running 100’s of sampling simulations for row houses and condos we think we came up with an image of Elvis ► Can you see an image of Elvis in this map? ► Beware: you can get some strange geographical coincidences in maps of sample data No telling what you can find in some weird sample ?? ?

PRCNACIS Suggestions on mapping with ACS data ► Reliability - Inform your readers that the map is based on sample data and the reality might vary. Where you can, provide quantitative measures of error. ► Period averages - Inform your readers when the data are for a 3 or 5 year period, and what this means in the particular context. ► Aggregate, Aggregate - Reduce standard errors by aggregating over longer time periods or for larger geographies. Avoid mapping with block group level data. ► Broaden class intervals - Look at the standard errors in the data when setting class intervals so as not to give a false impression of precision. ► Educate your clients - Let them know the limits of the ACS data for mapping. Say “no” to ill conceived requests. ► Share - Share what you learn about mapping with the ACS with colleagues. Provide feedback to the Census Bureau.

PRCNACIS Where does one get the ACS data? ► ACS : ► American Factfinder: ► FTP Site:

PRCNACIS Additional Resources ► US Census Bureau ACS: ► Population Reference Bureau: The American Community Survey, ► National Academy of Sciences: Using the American Community Survey: Benefits and Challenges, forthcoming in paperback, ► Population Research and Policy Review (2006), No 25: A Special Issue on the ACS: ► Missouri Census Data Center: Ten Things to Know about The ACS: ► Portland Demographic Trends CD – Copies available, see presenter.

PRCNACIS The End Richard Lycan Population Research Center Portland State University Portland, Oregon,