Household Surveys
ACS – CPS - AHS
INFO 7470 / ECON 8500
Warren A. Brown
University of Georgia
February 22, 2011
1
2
3
4
5
CES Discussion Papers
6
7
8
9
10
Outline
•
•
•
•
•
•
Purpose
Target Population
Sampling Frame
Data Collection
Non-Response
Missing Data
•
•
•
•
•
•
Weighting
Sampling Error
PUF / IUF Data
Access to Reports
Research Questions
Resources
American Community Survey
•Sponsor: Census Bureau
•Collector: Census Bureau
•Purpose: “The American Community Survey (ACS)
is an ongoing survey that provides data every year -giving communities the current information they need to
plan investments and services. Information from the
survey generates data that help determine how more
than $400 billion in federal and state funds are
distributed each year.”
12
ACS: Purpose
•
•
•
•
•
Key Component for Re-Engineered Census
Replacing the Decennial Census “Long-Form”
Continuous measurement rather than snapshot
Meet federal legislative and program needs
Other stakeholders in state and local
government and private sectors
• Provide annual data on demographic, social,
economic and housing characteristics.
• Improve the Federal Statistical System
13
Advantages of the ACS
• Timeliness
• Comparability
• Reliability
• Numerous data products
14
ACS: Target Population
• Current residence not “Usual
Residence”
– “two-month rule”
• Household population in 2005
• Total (HH + GQ) in 2006 and ….
• GQ residency is “de facto”
15
ACS: Sampling Frame
• Master Address File (MAF)
– Official inventory of known living quarters
– Linked to TIGER
• Housing Units
– Based on Census 2000 MAF and updates from the
USPS’ Delivery Sequence File
• Group Quarters
– … and updates from the administrative records and
the FSCPE
– Excluded from ACS are domestic violence shelters,
soup kitchens, commercial maritime vessels,…
16
ACS: Design of the Sample
• Annual Sample Size of 3 million addresses
• Series of Monthly Samples of 250,000
addresses
• HU sample in each of the 3,141 Counties
• Areas with smaller populations sampled at
higher rates than those with larger populations
• HU Address sampling rate set by Block based
on entity (municipality, school district, tract)
• Final sampling rate varies between 1.6% and
10%
• No HU address can be sampled more than once
in 5 years
17
ACS: Questionnaire
• Content designed to meet the needs of
federal government agencies
• 21 housing and 48 population
questions
• Household Respondent provides
responses for all other residents of the
household
• Householder or Reference Person is
“Person 1”
18
ACS: Data Collection
• HU addresses by three modes
– Mailout of paper questionnaire in 1st month
– Telephone (CATI) non-response follow-up
in 2nd month
– Personal visit (CAPI) non-response followup in 3rd month to a sub-sample
• GQ
– Personal visit within 6 weeks of sample
selection
19
Data Collection Timetable
20
Non-Response Follow Up
• 1st Month – Mailed Questionnaire
• 2nd Month – Attempt at Telephone Interview
• 3rd Month – Sub-Sample of NonRespondents
– 67% unmailable addresses
– 50% low response tracts
– 33% high response tracts
ACS: Response Rates
• Participation is mandatory
• Sample addresses eligible for
interviewing:
– 51% Mail response
– 9% CATI non-response follow-up
– 38% CAPI non-response follow-up
– 2% Non-interview
ACS: Missing Data
Item non-response or failed edit check
• FEFU – failed-edit follow-up
– More than 5 household members
– Critical questions not answered
– 33% of mail return questionnaires in 2005
• Imputation
– Assignment used other data on
respondent
– Allocation – hot deck procedure
– Flags variables
23
ACS: Weighting
• Person and Housing Unit weights
• Three stages
– Probability of selection (initial sampling rate)
– Adjust for non-response
– Control to Population Estimates
• Sum the weights
– Person weights for person characteristics
– HU weights for family, household or housing unit
characteristics
• Householders = Households ???
24
ACS: Weighty Issues
•
•
•
•
ACS 2005 for United States
288,378,137 Persons in Households
111,090,617 Households
114,763,475 Householders
Average Household Size
– 2.60 based on Households
– 2.51 based on Householders
ACS: Sampling Error
• Publishing margins of error in tables
• More sample less error
• 1, 3, and 5 year estimates and
associated sampling error
• Sampling error for small area data is
proving to be a problem
26
Complex Sample Surveys and Weights
• Weight - numeric variable expressing the number of
housing units or people that an individual microdata
record represents
• Sum of the housing unit and person weights for a
geographic area is equal to the estimate of the total
number of housing units and people in that area
• Values for weights vary
– Different Probability of Selection
– Differential Response Rate
– Control to Population Estimates
27
Alternatives for Estimating
Standard Error
• Design Factor
• Replicate Weights
• Ignore
28
Alternatives for Estimating
Standard Error
• Design Factor
– Design factors are factors to multiply times the
standard error of a simple random sample.
– Easier to use than the replicate weights
• Replicate Weights
– Generally, more accurate
– Somewhat more work than design factors
29
Estimate of Sampling Error Using
Design Factors
30
Design Factors
Characteristic
Design Factor
Age
1.1
Employment Status
1.2
Person Income
1.6
* West Virginia is 1.0 for Age and 1.5 for Person Income
31
Estimate of Standard Error Using
Replicate Weights
80
4
2
X r  X 
SE 

80 r 1
where:
– X is the estimate formed from the PUMS
weight
– Xr is the estimate formed from the rth
replicate weight.
32
PUF / IUF
•
•
•
•
•
•
1% Sample
PUMA
Top Coding
Collapsed Categories
Perturbation
No Administrative
•
•
•
•
•
•
2.5% Sample
Block/Tract/Place
Full Distribution
Full Distribution
Actual Ages
Details of Collection
33
34
ACS: Research Questions
Resources
•
•
•
•
Census Bureau
IPUMS
Michigan Population Studies Center
National Academy Sciences-CNSTAT
36
ACS on www.census.gov
37
38
39
40
http://www.psc.isr.umich.edu/dis/acs/dataanalysis/UsingReplicateWeights.html
41
NAS-CNSTAT ACS Reports
2001, CNSTAT’s
core sponsors
***********
2007, Census
Bureau
2008,
NSF/SRS