1. The American Community Survey Getting Ready for a New Decade of Socio-Economic Data

2. The Topics What is the American Community Survey? ACS Methodology ACS Methodology Using ACS Data Using ACS Data Multi-Year Estimates Multi-Year Estimates Reliability Reliability Accuracy Accuracy The 5-Year Release The 5-Year Release

3. “The ACS is the first nationwide survey that can be used to monitor annual trends in local communities and make valid comparisons among communities in the years between censuses”- Population Reference Bureau American Community Survey - What is it? The biggest change in 60 years in conducting the decennial census The biggest change in 60 years in conducting the decennial census

4. American Community Survey - What is it? A large, continuous demographic survey A large, continuous demographic survey –Produces annual and multi-year estimates of the characteristics of the population and housing –Produces characteristics, not a population count –Produces information for small areas including tracts, block groups and population subgroups - updated every year –Key component of re-engineered 2010 census

5. ACS – What does it do? Implements a national sample of about 3 million addresses in every county in the US and in Puerto Rico Implements a national sample of about 3 million addresses in every county in the US and in Puerto Rico Provide profiles every year for communities of 65,000 or more Provide profiles every year for communities of 65,000 or more Provide 3- to 5-year averages for communities of less than 65,000 population Provide 3- to 5-year averages for communities of less than 65,000 population

6. ACS – How does it work? Questionnaire is mailed every month of every year to a random sample of addresses - Monthly survey size 250,000 Questionnaire is mailed every month of every year to a random sample of addresses - Monthly survey size 250,000 –No address will receive a form more than once every 5 years 3 phases of data collection 3 phases of data collection –Mail-out, mail-back –Computer assisted telephone interviewing for non- response –Computer assisted personal interviewing of a sample of the remaining unanswered forms (1-in-3 sample).

7. ACS – How does it work? Response is required by Title 13 Response is required by Title 13 Respondents must answer all questions Respondents must answer all questions Administered by the Regional Offices Administered by the Regional Offices

8. ACS - When Will Data Be Available? Now, Single-Year Estimates for population areas larger than 65,000 Now, Single-Year Estimates for population areas larger than 65,000 Now, Multi-Year Estimates for population areas larger than 20,000 Now, Multi-Year Estimates for population areas larger than 20,000 In December 2010: In December 2010: –Smaller geographies will be available in five- year moving averages (every year!) –Other kinds of geographies will become available (Census Tracts, etc)

10. ACS Data Product Release Schedule Data products are released in the year following the single-year or multi-year period in which data are collected.

11. Pennsylvania Coverage Areas How many areas in Pennsylvania are covered now (3-year estimates)? How many areas in Pennsylvania are covered now (3-year estimates)? 61 Counties 61 Counties 20 Places 20 Places 79 Townships and Boroughs 79 Townships and Boroughs 197 School Districts 197 School Districts PaSDC Map of the Month – March 2009

12. ACS Current Coverage

13. ACS Data for Counties: Pennsylvania Counties by Size Population 65,000+ Philadelphia, Allegheny, Montgomery, Bucks, Delaware, Lancaster, Chester, York, Berks, Westmoreland, Lehigh, Luzerne, Northampton, Erie, Dauphin, Cumberland, Lackawanna, Washington, Butler, Beaver, Monroe, Schuylkill, Cambria, Fayette, Centre, Franklin, Lebanon, Blair, Mercer, Lycoming, Adams, Lawrence, Northumberland, Crawford, Indiana, Clearfield, Somerset, Armstrong, Columbia Population 20,000 to 64,999 Carbon, Bradford, Pike, Venango, Wayne, Bedford, Mifflin, Huntingdon, Jefferson, Perry, McKean, Union, Susquehanna, Warren, Tioga, Greene, Clarion, Snyder, Clinton, Elk, Wyoming, Juniata Population Less Than 20,000 Montour, Potter, Fulton, Forest, Sullivan, Cameron

14. Comparing ACS and Other Sources Decennial Census vs. ACS Decennial Census counts people Decennial Census counts people ACS shows their characteristics ACS shows their characteristics ACS sample size is smaller – 1 in 8 ACS sample size is smaller – 1 in 8 ACS data is supplied with “margin of error” columns showing the upper and lower bounds of the sampling error around the estimates ACS data is supplied with “margin of error” columns showing the upper and lower bounds of the sampling error around the estimates

15. Reference Period Reference Period –The Census uses April 1 as the base for most questions. Some questions use 1999 as a reference year. –ACS reference varies based on which month the form is mailed. Comparing ACS and Other Sources Decennial Census vs. ACS

16. Residency Rules Residency Rules –Decennial Census  Usual place of residence is the standard. Self- identification of where you live most of the year. –ACS  Current residence, counted at the sampled address if living or staying there for more than TWO MONTHS. Comparing ACS and Other Sources Decennial Census vs. ACS

17. Most comparisons are fine Most comparisons are fine Compare With Caution Compare With Caution –Income and Poverty –Veteran Status – Service-Connected Disability –Disability Use comparison chart – some areas are “compare with caution” Use comparison chart – some areas are “compare with caution” Comparing ACS and Other Sources ACS 2007 vs. ACS 2008

18. Comparability of ACS data

19. Types of Data Summary File 3, long-form type data will be available Summary File 3, long-form type data will be available –Educational attainment, poverty, migration, commuter patterns, disability, housing values, income and earnings data (and much more)

20. ACS Multi-Year Estimates

21. 2006 - 2008 ACS Multi-Year Estimates 3-year estimates are now available for areas with populations of 20,000 and greater 3-year estimates are now available for areas with populations of 20,000 and greater 3-year estimates are based on data collected in 2006, 2007 and 2008. 3-year estimates are based on data collected in 2006, 2007 and 2008.

22. Multi-Year Estimates Period estimates describe the average characteristics over a specified period (1-year estimates are also period estimates). Period estimates describe the average characteristics over a specified period (1-year estimates are also period estimates). A multi-year estimate is a period estimate that encompasses more than one calendar year. A multi-year estimate is a period estimate that encompasses more than one calendar year. Both the 3-year and 5-year estimates from the ACS are Multi-Year Estimates. Both the 3-year and 5-year estimates from the ACS are Multi-Year Estimates. 1-year estimates 3-year estimates 5-year estimates 65,000+ people XXX 20,000+ people XX Less than 20,000 people X

23. Multi-Year Estimates: FAQ’s How do I label a multi-year estimate? How do I label a multi-year estimate? –ACS estimates based on data collected from 2006- 2008 should not be labeled “2007” or “2008” estimates (In 2005-2007, the poverty rate was…). When should I use a multi-year estimate? When should I use a multi-year estimate? –No one-year estimate is available –Margins of error for one-year estimates are larger than required –Analyzing data for small population groups (tables may be suppressed)

24. Multi-Year Estimates: Comparing Different Years How can I use multi-year estimates to make comparisons? How can I use multi-year estimates to make comparisons? –Only compare the same type of estimate  1-year estimate to other 1-year estimates, 3-year estimate to other 3-year estimates, and 5-year estimate to other 5-year estimates –When comparing across time periods for the same geographic area, compare non-overlapping periods  The 2005-2007 estimate should be compared to the 2008- 2010 estimate because they do not represent any overlapping data.

25. ACS Accuracy and Methodology

26. Reliability ACS Sample size is smaller than Census 2000, so reliability is an issue. ACS Sample size is smaller than Census 2000, so reliability is an issue. –Studies indicate 1.75x the sampling error of Census 2000 Be careful about data items: the more detailed your characteristic, the higher level of geography you should use. Be careful about data items: the more detailed your characteristic, the higher level of geography you should use. Collapsed tables help increase reliability. Collapsed tables help increase reliability.

27. What is Sampling Error? Definition The uncertainty associated with an estimate that is based on data gathered from a sample of the population rather than the full population

28. Sampling Error is Related to Sample Size The larger the sample size, the smaller the uncertainty or sampling error The larger the sample size, the smaller the uncertainty or sampling error Combining ACS data from multiple years increases sample size and reduces sampling error Combining ACS data from multiple years increases sample size and reduces sampling error All sample surveys have sampling error – including decennial census long-form data All sample surveys have sampling error – including decennial census long-form data

29. How are Measures of Sampling Error Used? To indicate the statistical reliability and usability of estimates To indicate the statistical reliability and usability of estimates To make comparisons between estimates To make comparisons between estimates To conduct tests of statistical significance To conduct tests of statistical significance To help users draw appropriate conclusions about data To help users draw appropriate conclusions about data

30. Measures Associated with Sampling Error Margin of Error (MOE) Margin of Error (MOE) Confidence Interval (CI) Confidence Interval (CI) Coefficient of Variation (CV) Coefficient of Variation (CV)

31. Margin of Error (MOE) Definition - A measure of the precision of an estimate at a given level of confidence (90%, 95%, 99%) - Confidence level of a MOE - MOEs at the 90% confidence level for all published ACS estimates

32. Margin of Error (MOE) Formula MOE = +/-1.645 x SE (90% level) Values for other confidence levels 95% = 1.960 99% = 2.576

33. Interpreting Margin of Error Indicates that a data user can be 90 percent certain that the estimate and the population value differ by no more than the value of the MOE Indicates that a data user can be 90 percent certain that the estimate and the population value differ by no more than the value of the MOE MOE can help data users assess the reliability of an estimate MOE can help data users assess the reliability of an estimate MOE can help data users avoid misinterpreting small differences between estimates as significant MOE can help data users avoid misinterpreting small differences between estimates as significant

34. Interpreting Margin of Error Example for Baltimore City: 52.1% Percent of males who have never married 1.7% Margin of Error –Indicates 90 percent chance that the estimate of 52.1% and the population value differ by no more than 1.7% –Size of MOE relative to size of estimate

35. Deriving New Margins of Error For sums: the square root of the sum of the squares of the margins of error: For sums: the square root of the sum of the squares of the margins of error:

36. Deriving New Margins of Error For Proportions (or Percents), but not ratios: the square root of: the square of the MOE of the numerator, minus the square of the new proportion, multiplied by the square of the margin of error of the denominator; all divided by the denominator of the proportion. For Proportions (or Percents), but not ratios: the square root of: the square of the MOE of the numerator, minus the square of the new proportion, multiplied by the square of the margin of error of the denominator; all divided by the denominator of the proportion.

37. Deriving New Margins of Error Where: MOE num : Margin of Error of the numerator MOE den : Margin of Error of the Denominator X num : Numerator X den : Denominator

38. Confidence Interval Definition A range that is expected to contain the population value of the characteristic with a known probability.

39. Confidence Interval Calculation Example for Baltimore City 52.1% – 1.7% = 50.4% 52.1% – 1.7% = 50.4% 52.1% + 1.7% = 53.8% 52.1% + 1.7% = 53.8% Confidence Interval = 50.4% to 53.8%

40. Confidence Interval Interpretation We can be 90 percent certain that the confidence interval from 50.4% to 53.8% contains the population value of never married males 15 years and older in Baltimore City We can be 90 percent certain that the confidence interval from 50.4% to 53.8% contains the population value of never married males 15 years and older in Baltimore City Useful to display confidence intervals Useful to display confidence intervals

41. Displaying Confidence Intervals

42. Coefficient of Variation (CV) Definition The relative amount of sampling error associated with a sample estimate Formula CV = SE / Estimate * 100%

43. Coefficient of Variation (CV) Example for Baltimore City Estimate = 52.1% of never married males Standard Error = 1.033% CV = SE / Estimate * 100% CV = 1.033% / 52.1% * 100% CV = 1.98%

44. Interpreting Coefficients of Variation Size of the CV Size of the CV In Baltimore City example, the CV is small (< 2%) indicating this is a reliable estimate In Baltimore City example, the CV is small (< 2%) indicating this is a reliable estimate No hard-and-fast rules about the size of CVs No hard-and-fast rules about the size of CVs Caution for proportions close to zero Caution for proportions close to zero

45. Conducting Tests of Statistical Significance Formula where is the critical value for the desired confidence level where is the critical value for the desired confidence level for 90% confidence level = 1.645 for 90% confidence level = 1.645 Z CL

46. The 5-Year ACS Release

47. 5-Year American Community Survey Data The first 5-year ACS data will be released in December, 2010. The first 5-year ACS data will be released in December, 2010. This is the first 5-year data release in ACS history This is the first 5-year data release in ACS history –It is also the first updated data since 2000 for thousands of geographies. Not all data will be available from AFF. Not all data will be available from AFF.

48. 5-Year Details The first 5-year ACS products will cover 2005 through 2009. The first 5-year ACS products will cover 2005 through 2009. Data will be available for every state, county, municipality, and for census tracts. (Block Groups : Limited availability) Data will be available for every state, county, municipality, and for census tracts. (Block Groups : Limited availability) Some tables, and all block group data, will be excluded from American FactFinder. Some tables, and all block group data, will be excluded from American FactFinder.

49. What Will Be Included? Of the 930 tables in ACS: Of the 930 tables in ACS: –616 tables will be available from AFF (and in Summary File downloads). –294 tables will only be available from Summary File downloads. –20 Tables will have geographic limits (state and above; county and above, etc) in AFF. All block group data will be excluded from AFF. All block group data will be excluded from AFF.

50. Subject Tables, Profiles Subject Tables and Data Profiles will be released for: Subject Tables and Data Profiles will be released for: –States / Counties / MSAs –Congressional Districts –State Legislative Districts –School Districts –Some others Selected Population Profiles (SPP) will not be included in the 2005-2009 release. Selected Population Profiles (SPP) will not be included in the 2005-2009 release.

51. Notable Geographies in the 5-Year Release State Legislative Districts State Legislative Districts –Including county breakouts School Districts School Districts Zip Code Tabulation Areas will not be included in the 2005-2009 release. Zip Code Tabulation Areas will not be included in the 2005-2009 release.

52. Geographic Definitions The first 5-year data (2005-2009) will use the Census 2000-era geographies and population controls, not Census 2010. The first 5-year data (2005-2009) will use the Census 2000-era geographies and population controls, not Census 2010. 2011’s 5-Year data release (2006-2010) will shift to using Census 2010 geographies and population controls. 2011’s 5-Year data release (2006-2010) will shift to using Census 2010 geographies and population controls. –Other geographies like ZCTAs will become available in 2011, as well.

53. In Conclusion ACS deals with characteristics of the population, not a pop count ACS deals with characteristics of the population, not a pop count The 2010 Decennial Census will deal with coverage The 2010 Decennial Census will deal with coverage Pay attention to reliability and comparability between data sources Pay attention to reliability and comparability between data sources 5-Year Data will be Released in December 5-Year Data will be Released in December Not all 5-Year tables will be on AFF. Not all 5-Year tables will be on AFF. Pay attention to reliability and comparability between data sources Pay attention to reliability and comparability between data sources

54. For More Information E-mail us at:PaSDC@psu.edu E-mail us at:PaSDC@psu.edu American Community Survey Web site (www.census.gov/acs/www) American Community Survey Web site (www.census.gov/acs/www) Call us at: 1-717-948-6336 Call us at: 1-717-948-6336 Subscribe to the ACS Alert Subscribe to the ACS Alert