Finding and Using Census Data

Finding and Using Census Data An Informational Session Presented for: 2015 CityMatCH Annual Urban MCH Leadership Conference Tuesday September 29, 2015 David Drozd, M.S. Center for Public Affairs Research University of Nebraska at Omaha Lead Agency of Nebraska State Data Center Please Like 402-554-2132 ddrozd@unomaha.edu www.unomaha.edu/cpar www.facebook.com/unocpar Nebraska State Data Center CPAR @ UNO 1

Why Census Data It Can Help Determine Health Outcomes

A look at some determinant related maps 3

Terminology & Definitions Decennial Census: the Census ; headcount taken once every 10 years; mandated by Constitution for equal representation Short form: basic decennial Census questionnaire sent to most households in 2000 and all households in 2010; records simple demographics like age, gender, race, housing tenure (own/rent) Long form: detailed decennial Census questionnaire sent to a sample of about 1 in 6 households in the 2000 Census but not used in the 2010 Census (replaced by ACS); recorded socio-economic and detailed housing data American Community Survey (ACS): continuous sample survey (forms sent every month) compiled to provide current annual data; nearly identical to long form allowed 2010 Census to be short form only (simply a headcount); has about 40 questions: data crosstabulated 2015 ACS (and in future years) continues monthly surveying even though 2010 Census is finished, assuming it is funded by Congress Nebraska State Data Center CPAR @ UNO 4

Census Datafiles Summary File 1 (SF 1): Used for both 2000 and 2010 basic demographic data from the short form: age, gender, race/ethnicity, housing tenure (own vs. rent), etc. Summary File 3 (SF 3): Not part of 2010 Census, replaced by ACS datasets; has detailed socio-economic and housing data from the 2000 long form; also has figures for SF 1 items (age, race) but they are based on a weighted sample, not the official counts BE CAREFUL You must go to the right source to get correct data! ACS Datasets: contain data for ACS variables; are based upon the timeframe over which the data was collected 1-year: 2013 3-year: 2011-2013 5-year: 2009-2013 Do NOT Compare ACS datasets to SF 1 from the decennial census Compare SF 1 from 2000 to: SF 1 from 2010 Compare SF 3 from 2000 to: ACS datasets Nebraska State Data Center CPAR @ UNO 5

ACS Data are Released Based Upon the Population of the Geographic Unit Time Period of Data Annual Estimates (2005, 2006, 2013) 3-Year Aggregates (2005-2007 2011-2013) Population Threshold for Data Release 65,000 or more 20,000 or more Nebraska Areas with Data State; Omaha and Lincoln; Douglas, Lancaster, Sarpy Counties; Omaha, Lincoln, Millard Schools All of the above plus regional centers like Kearney, Norfolk, etc. 5-Year Aggregates (2005-2009 2009-2013) No threshold All areas--counties, cities, townships, census tracts, zip codes, school districts, legislative districts, etc. 3-Year Aggregates will no longer be produced!! (2011-13 is last 3-year dataset) Tip: when citing sources, be specific (use table numbers - helps future replication): U.S. Census Bureau, 2013 American Community Survey 1-Year Estimates (S1701) U.S. Census Bureau, 2009-2013 American Community Survey 5-Year Estimates (S1701) Nebraska State Data Center CPAR @ UNO 6

Point Estimates, MOE and Confidence Intervals The ACS data provide point estimates for various characteristics. ACS data also include a margin of error (MOE) for finding a lower and upper bound. (e.g. 64 ± 3) Why? The ACS is a sample and subject to sampling error. Is the ACS data representative of the entire population? Census 2000 long form also a sample 1 in 6 sampling rate with full follow-up made sampling error small and MOE was not released. Adding and subtracting the MOE to/from the point estimate creates a range called a confidence interval. Example: 2009-13 NE fertility rate: 64 3; 64 + 3; so the range or interval is 61 to 67 ACS displays the MOE for a 90% confidence interval. The bounds tell us that we are 90% confident that the figure for the entire population would be in this range if all households were surveyed. Nebraska State Data Center CPAR @ UNO 7

Multi-year Aggregate Estimates: Advantages The aggregates provide data for more geographies Annual data for NE: 3 counties (Douglas, Lancaster, Sarpy) 3-yr aggregate: 17 counties; 5-yr aggregate: all 93 counties For areas that get annual data, the aggregates based on more completed surveys will be more accurate and have smaller margins of error (MOE) Especially important for sub-groups data for specific age groups or racial/ethnic groups General rule: whenever you split a stat (e.g. education by age; poverty by race, etc.) use the longest timeframe Will help reduce variability in year-to-year figures Some FactFinder tables are prepared but not released annually because of inaccuracy concerns some of those tables will now have data released Nebraska State Data Center CPAR @ UNO 8

Geography Size continuum: large to small, or small to large Legal/Administrative vs. Census/Statistical Nation State County City Township Block Block Group Census Tract Zip Code PUMA Metro Area (MSA) Division Region The West North Central Division of the United States Nebraska State Data Center CPAR @ UNO 9

FactFinder Address Search Page & Results 10

Block 2022 in Block Group 2, Census Tract 47 Douglas County, NE This is a single block they are building blocks for larger geographies. Blocks are bounded by physical features like roads or streams. 11

Block 2022 in Block Group 2, Census Tract 47 Douglas County, NE A group of usually about 15-30 blocks comprise a block group. 12

Block 2022 in Block Group 2, Census Tract 47 Douglas County, NE Two or more block groups comprise a census tract, a key unit for small-area analysis. Census tracts typically have ~4,000 people and nest within counties. 13

Relationship between Data/Datafiles and Geography For blocks, you can only get SF 1 or Redistricting data from Decennial Confidentiality reasons: you d know certain incomes on your block (also sampling/error issues) Block groups have most SF 3 data; only meant as building blocks for larger custom geographies in the ACS No ancestry or unmarried partner data Census Tracts are a key geography as they are one of the smallest geographies to have most all data compiled (both Decennial & ACS) PCT (population items) and HCT (housing) tables in decennial files Now have 5-yr ACS zip code data; like tracts they are a good unit of analysis for neighborhoods but typically larger than tracts so more accurate/less variability Public Use Microdata Areas (PUMAs) are combined census tracts that contain at least 100,000 people and are the smallest geography for the PUMS files; important in ACS as > 65,000 person threshold! Nebraska State Data Center CPAR @ UNO 14

PUMA geography State Maps for PUMA boundaries can be viewed on the internet http://www.census.gov/geo/maps-data/maps/reference.html Nebraska has 14 PUMA areas (4 in Douglas Co., 2 in Lancaster Co.) ACS data used 2000 based PUMAs from 2005-2011; 2012 and future years use updated 2010 based PUMA boundaries The 2010 PUMAs are nearly identical: a couple tracts changed in Douglas and Lancaster counties; Sarpy Co. is now its own PUMA PUMA geography has more importance in the ACS Data are released annually since PUMAs exceed the annual ACS population threshold of 65,000 Not many geographies in low populated states get annual data; PUMAs are the smallest annual geography for making statewide comparisons Breaks core metro counties into smaller areas and gives proxies for smaller geographies that are similar to the larger PUMA area Northeast NE puma a proxy for Pierce Co; Southeast Douglas for South Omaha Nebraska State Data Center CPAR @ UNO 15

16

Are you ready to rumble?? Geography VS. Data (A Duel) There is usually a tradeoff between available information and the geography for which it is available Can be due to confidentiality restraints More specific information may require expanding to a larger geography (or a longer timeframe) Even if the data is provided, you must ask whether the data is reliable for that geography? Are the figures based on a small number of cases (e.g. specific small racial group) where the sampling error could be large? Sampling error occurs if the figures would be different if different households/people were included in the sample. ACS data always lists the margin of error (90% confidence). Nebraska State Data Center CPAR @ UNO 17

Example analysis using Census Tracts and 2010 Census data note the categories: round numbers with about an equal number of tracts in each

Example using Zip Codes and the Foreign Born variable to display the location of the Immigrant Population; Foreign Born only available on the ACS Split abovebelow the county average.

Census Concepts Householder: simply the first person on the census form Is NOT necessarily the head of the household (that s old terminology) Supposed to be the person in whose name the residence is owned or rented ; that s usually but not always the person who fills out the form Can be male or female; householder is somewhat arbitrary but makes a big difference for how families and by extension items like poverty are defined Relationship: how other persons living in the household are related or connected to the householder Spouse, biological/adopted/step child, parent-in-law, roommate, unmarried partner, foster child, etc. A household with 2 or more related individuals is a family household One person households are not a family by definition, and thus are excluded from variables like median family income (but are included in median household income) Unmarried partners are not a family by definition If a child lives in the household of unmarried partners but is from a prior relationship, it is a function of who fills out the form (is the householder) for whether that is a family or not IT GETS COMPLEX impacts items like poverty Nebraska State Data Center CPAR @ UNO 20

Householder Example: Susan and her child Emily Live with Paul Notes: Susan and Paul are dating but not married; Paul is not the father of Emily If Susan is the householder If Paul is the householder Emily is related to Susan as biological child and this is a family household; Paul is unmarried partner & not part of the family Emily is an other nonrelative and Susan is unmarried partner; this is NOT a family household In both cases Paul s income is NOT included in Susan and Emily s poverty calculation; if they were married it would be.

Census uses 2 separate questions on race/ethnicity. First is a person Hispanic or not? Second, which race(s) do people identify with (White, Black, Asian, etc.)? How does CPAR typically analyze race/ethnicity? We usually list totals for Hispanic/Latino, and then all races ALONE as being non-hispanic (NH): White NH; Black NH; etc. This avoids double counting and the sum of all groups when combined with 2+ races NH will equal the total population (percents sum to 100%) Race ALONE means that only one race was stated on the form Many groups/agencies and grants ask for simply White, Black, etc. and Hispanic by definition many Hispanics in this case would be represented twice (their race is often listed as White or Other) You heard it here first the 2020 census will likely use a combined race/ethnicity question! It will get similar results, other race category will be selected less frequently, & will have areas to write in ancestry/origin (so we d get data on African and European countries/origins in addition to Asians, Native American, & Hispanic) Also can evaluate race for those ALONE or IN COMBINATION This includes those who selected that race specifically (alone) as well as those who selected that race and another one (2 or more races) Double counts the people selecting 2+ races (doesn t sum to total pop. or 100%) Might use more often as lessens effect of multi-race births always being minority Nebraska State Data Center CPAR @ UNO 22

Possible Determinants of Health CDC suggested items from the Census Bureau Income/Poverty; Education Levels; Family/Household Structure; Foreign Born; Language; Vacant Housing Units; Home Ownership (Tenure); Unemployment Census is the only source of sub-county (city, zip code, tract) employment data Specific Examples from Healthy People 2020 http://www.healthypeople.gov/2020/topics-objectives/topic/socialdeterminants-health/objectives Census has Data Profiles to quickly see the most commonly used statistics (DP1 from 2010 Census; DP2 (Social), DP3 (Economic), DP4 (Housing), DP5 (Demographics) from ACS) Census Subject Tables provide fast access to data on one subject often is split by age, gender, race, etc. Recommend using 5-yr datasets whenever data is split (subject tables) Census also provides detailed tables for download/analysis Nebraska State Data Center CPAR @ UNO 23

Example Demographic Determinants Americans with the least education are more than twice as likely to die from cancer as those with the most education Cancer Death Rates Continue Drop; Disparities Among Some Groups Seen Newsmax Health, Friday June 17, 2011 (no longer online, hard copy only) Male Smokers (from same article): 31% of men with 12 or fewer years of education 12% of college graduates 5% of those with advanced degrees U.S.-born Hispanics had higher prevalences of obesity, hypertension, smoking, heart disease, and cancer than foreign-born Hispanics: 30% higher, 40%, 72%, 89%, and 93%, respectively Vital Signs: Leading Causes of Death, Prevalence of Diseases and Risk Factors, and Use of Health Services among Hispanics in the U.S. CDC May 2015 http://tinyurl.com/onvr3tw Older male life expectancy gains since ~ 1980 Upper ½ of incomes: + 6 yrs; Lower ½: + 1.3 yrs Income Gap, Meet the Longevity Gap NY Times 3-15-14 http://tinyurl.com/obwwff9 Nebraska State Data Center CPAR @ UNO 24

Example using Census data as the base population for calculating rates: Suggest using multi-year data around the Decennial Census hard (accurate) count 25

Example map completed online in American FactFinder: while it isn t the easiest to use it can help tell a story fairly quickly with no special software

Map is different when focusing on older persons, for whom disabilities are most prevalent (note that accuracy is reduced/moe increases with more specific data)

Education is a key driver of outcomes; focusing on increasing education does a great deal as it increases employment, income, home ownership, etc.

Fuller screenshot of map with table number/source: Poverty is safe to compare locally but the official rate doesn t account for cost of living differences across U.S.

THE CENSUS BUREAU NOW HAS 30

31

Summary of noncash benefits and additional relevant costs that the SPM factors into the poverty calculation 32

While the overall Official Poverty Measure and SPM values are similar, there are big differences by age group 33

Other Big Differences in the Two Poverty Measure s 2012 Data for the U.S. Category OPM SPM Difference Male 13.7 15.3 + 1.5* Female 16.4 16.7 + 0.3 Married couple 7.5 10.0 + 2.5* Female householder 29.1 28.9-0.2 Male householder 17.9 23.1 + 5.2* New SPM 30.9 18.4-12.5* White, non Hispanic 9.8 10.7 + 0.9* Asian 11.8 16.7 + 4.9* Black 27.3 25.8-1.6* Hispanic 25.8 27.8 + 2.0* * Denotes a change that was statistically different than 0 at the 90% confidence level. Source: Table 1, 2012 Supplemental Poverty Measure Report The inclusion of unmarried partner income has a huge impact. Hispanic poverty is highest in SPM. 34

Map of Official Poverty Rates

Here are the differences between the SPM and OPM; Note the declines in the Plains and South while big increases in red are on the coasts 36

The Upper Plains has the lowest poverty; the deep south not nearly as solid ; southwest stands out with California highest in the country 37

Tip: Compare not just overall totals, but also specific subgroups by themselves to be truly apples-to-apples and lower distribution effects How can Iowa exceed Nebraska overall when education is higher in NE among key groups? Age 25+ High School or More Education Source: Table C15002 series, 2009-13 ACS Category State Percent Rank All Persons Iowa 91.0 # 8 NE 90.5 #11 White non-hisp. NE 94.1 # 7 Iowa 92.7 #17 Black NE 84.4 #20 Iowa 82.1 #35 Hispanic Iowa 56.6 #46 NE 50.8 #51 Nebraska State Data Center CPAR @ UNO Different Population Makeup State % of Population NE 86.1 Iowa 91.5 NE 3.9 Iowa 2.3 Iowa 3.5 NE 6.8 NE has more pop. with low attainment levels. 38

This shouldn t happen now that you know CENSUS BASICS. 39

The End Comments, Questions? Feel free to contact me any time for data assistance.

Reference Slides

Unemployment Rates for Census Tracts: Split above/below the countywide average

Example analysis showing concentration of the Hispanic population in Douglas County, Nebraska (Omaha area)

What Census data allows us to do make comparisons! Nebraska s higher-thanaverage birthrate cuts across all major ethnic and racial groups and just about any other demo-graphic you can think of. ~ Henry Cordes, Omaha World-Herald Category State Rank Hispanic 4 th Black, non Hispanic 7 th White, non Hispanic 9 th With college degree No college degree Foreign born Native born In poverty Not in poverty 3 rd 10 th 2 nd 8 th 5 th 5 th

How Poverty Varies by Single Year of Age (Custom Analysis using ACS Public Use Microdata Sample) 45

Work Experience by Presence of Children Detailed Table B23010 http://factfinder.census.gov/bkmk/table/1.0/en/acs/13_5yr/b23010/0100000us