2014 SDC and CIC Annual Training Conference: Accessing ACS PUMS Data Tim Gilbert U.S. Census Bureau April 2, 2014

2014 SDC and CIC Annual Training Conference:

Accessing ACS PUMS Data

Tim GilbertU.S. Census Bureau

April 2, 2014

Outline

Fundamentals of PUMS Data

Geography and the PUMS

Accessing PUMS Data

Multiple Vintage Variables in PUMS

Working with multiple vintage variables

Considerations with PUMS

Resources

2

3

Summary Data and Microdata What’s the Difference?

Summary data are predefined tables for specific geographic areas (states, counties, etc.)

In the ACS microdata, the basic unit is an individual housing unit or person

4

Public Use

Microdata

Sample

anonymized, downloadable

records of individual people

a representative sample of the population

What are PUMS data?

5

PUMS Overview

PUMS sample is a subsample of ACS interviews, one percent of all US households

PUMS is a “weighted” sample Weighting variables must be used in analysis

A set of two files - housing units and persons

ACS produces 1-, 3-, and 5-year PUMS files

Available as SAS files and CSV files, or via DataFerrett

6

Why Use PUMS?

Data needed for a tabulation or a specific universe not supported by standard ACS tables (e.g., population groups by single year of age)

Statistical analysis required to understand relationships between economic, demographic or housing variables (e.g., correlation analysis)

Can create new measures using multiple variables or other people in household (spouse’s occupation, same-sex couples, number of kids)

7

Types of PUMS Files Released

We release 3 new PUMS files every year 1 year PUMS (example: 2012 1-year PUMS)

December 2013

3-year PUMS (example: 2010-2012 3-year PUMS) February 2014

5-year PUMS (example: 2008-2012 5-year PUMS) March 2014

8

Modifications to Multiyear PUMS

Multiyear PUMS have the same cases and geography as their component 1-year files

How are multiyear PUMS different from single year? Weights are produced using latest population estimate

“vintages” Dollar amounts are standardized

Why use the multiyear PUMS files? For studying small groups, where more cases are needed When analysis is also making use of multiyear summary

data

Outline



Accessing PUMS Data




Resources

9

10

Limited Geographic Detail

Geographic identifiers are region, division, state, Public Use Microdata Area (PUMA)

PUMAs can be used to identify geographic areas of 100,000+

PUMS is not designed for statistical analysis of small geographic areas

11

Public Use Microdata Area (PUMA)

Defined after each census by the states in coordination with the Census Bureau’s Geography Division http://www.census.gov/geo/puma/puma2010.html Redefined PUMAs for 2012 PUMS files DY 2012 multiyear files have dual PUMA vintages

Large enough to meet disclosure avoidance requirements

PUMAs are identified by a five-digit number, unique within each state

Author

Will Missouri SDC update to include both 2000 and 2010 PUMAs?

12

PUMA Reference Mapshttp://www.census.gov/geo/maps-data/maps/reference.html

13

Interactive PUMA Mapshttp://tigerweb.geo.census.gov/tigerwebmain/tigerweb_main.html

Outline



Accessing PUMS Data




Resources

14

15

American FactFinderhttp://www.census.gov/acs/www/data_documentation/pums_data/

16

American FactFinder

17

American FactFinder

18

PUMS on FTP sitewww2.census.gov

19


20


21

DataFerretthttp://dataferrett.census.gov/

Outline



Accessing PUMS Data




Resources

22

23

What are multiple vintage variables?

2010-2012 3-Year PUMS and 2008-2012 5-Year PUMS contain variables with multiple vintages

Multiple vintage variables have differing sets of values for different years within the same multi-year file

24

Multiple Vintage Variables in PUMS2010-2012 ACS 3-Year PUMS 2008-2012 ACS 5-Year PUMS

2010&2011 2012 2008&2009 2010&2011 2012

ANC1P05ANC2P05CITWP05 LANP05MARHYP05 MIGSP05OCCP10POBP05 POWSP05 RAC2P05RAC3P05 SOCP10YOEP05

ANC1P12ANC2P12CITWP12LANP12MARHYP12MIGSP12OCCP12POBP12POWSP12RAC2P12RAC3P12SOCP12YOEP12

ANC1P05ANC2P05CITWP05LANP05MARHYP05MIGSP05OCCP02POBP05 POWSP05 RAC2P05 RAC3P05SOCP00YOEP05

ANC1P05ANC2P05CITWP05LANP05MARHYP05MIGSP05OCCP10POBP05 POWSP05 RAC2P05 RAC3P05SOCP10YOEP05

ANC1P12ANC2P12CITWP12LANP12MARHYP12MIGSP12OCCP12 POBP12POWSP12RAC2P12RAC3P12SOCP12YOEP12

Outline



Accessing PUMS Data




Resources

25

26

PUMS Documentation

PUMS ReadMe

List of variables with multiple vintages

PUMS Data Dictionary

Variable names, descriptions, and values

Accuracy of the PUMS

Information about working with multiple vintage variables

http://www.census.gov/acs/www/data_documentation/pums_documentation/

27

1. Verify variable has multiple vintages from PUMS ReadMe – Marital History

Using Multiple Vintage Variables

28

2. Look up differences between vintages in data dictionary


29


3. Recode and combine vintages to create one variable

If MARHYP05 less than or equal to 1932

OR

If MARHYP12 equals 1932

THEN

MARHYP (derived variable) equals 1932

30

1. Look up PUMA variable vintages in data dictionary

Using Multiple Vintage PUMAs

31


2. Look up PUMA in Missouri State Data Center’s MABLE/Geocorr12 at http://mcdc.missouri.edu/websas/geocorr12.html

32


3. Find corresponding PUMAs across vintages

33


4. Combine the PUMA vintages across years

If ST equals 26 and PUMA00 equals 03806

OR

If ST equals 26 and PUMA10 equals 03204

THEN

PUMA (derived variable) equals XXXXX

Outline



Accessing PUMS Data




Resources

34

35

Analyzing PUMS Data

National level files must be concatenated See PUMS ReadMe

Use SERIALNO variable to merge housing and person records to create complete file See PUMS ReadMe


36

Types of PUMS Weights

PUMS household weights (wgtp) must be used to produce housing unit estimates

PUMS person weights (pwgtp) must be used to produce population estimates

PUMS replicate weights (wgtp1 – wgtp80 and pwgtp1 – pwgtp80) are used for calculating standard errors

37

Estimating Variance with PUMS

Problem: PUMS is not a simple random sample Stratified samples with complex weighting Sample drawn at household level (i.e., not a simple

random sample of individuals)

Solutions: Use weighting variable and a “design factor” Use weighting variable and 80 “replicate weights”

See Accuracy of the PUMS


Outline



Accessing PUMS Data




Resources

38

39


40

Accuracy of the PUMS

41

http://www.acsdatausers.org/

42

Contact Information

ACS/PRCS website:www.census.gov/acs

ACS User Support:

[email protected]

Questions?

http://www.census.gov/acs

mailto:[email protected]

Documents

2014 SDC and CIC Annual Training Conference: Accessing ACS PUMS Data Tim Gilbert U.S. Census Bureau April 2, 2014