Day 2:

1
Day 2

• Data Collection and Sample Design

2
DATA COLLECTION AND SAMPLE DESIGN I. What
was the overall goal for the samples in the 4
cities? II. How did the cities accomplish
their goals? III. What are the analysis issues
emanating from the sample design? IV.
SUMMARY
3
I. What was the overall goal for the samples in
the 4 cities?

• Sufficient sample sizes for race/ethnic and
  income groups
• Race-matching
• Survey in multiple languages

4
II. How did the cities accomplish their goals?

• Detroit Survey Research Center, University of
  Michigan
• Atlanta Mathematica Policy Research Institute
• Boston Center for Survey Research, UMASS Boston
• L.A. Institute for Social Science Research, UCLA
• Boston Center for Survey Research, UMASS Boston

5
Geographic Boundaries for the Metropolitan Areas
Comprising the MCSUI Household Survey
Samples Atlanta Clayton, Cobb, DeKalb,
Douglas, Fayette, Fulton, Gwinnett, Henry, and
Rockdale Counties (913,292 HUs) Boston
Boston-Lawrence-Salem MA-NH Consolidated
Metropolitan Area, Massachusetts portion only
(1,440,078 HUs) Detroit Macomb, Oakland, and
Wayne Counties (1,540,237 HUs) L.A. Los Angeles
County (2,989,552 HUs)
6
Sampling Design
Multi-stage stratified, clustered
area-probability design
7
Sampling Procedure

• Define groups to be over-represented in sample
• All cities African American households low
  income households
• Boston and L.A. Latino households
• L.A. Korean, Japanese, and Chinese household

8
Sampling Procedure (cont.)

• Define target sample sizes
• approximately 800 households from each
  race/ethnic group for final sample
• Sample disproportionately from areas with
  concentrated target race/ethnic groups and
  low-income households

9
Stages of Sample Selection

• Divide metro area into geographic segments
• census tracts, blocks, or contiguous blocks with
  a minimum number of occupied housing units
• Group segments into strata (geographic and/or
  demographic)
• More stages...select sample

10
Distribution of Cases Across Cities Detroit A
tlanta Boston L.A. TOTAL TOTAL 1543 1528 1820 40
25 8916 Whites 728 642 585
835 2790 Blacks 741 824
443 1103 3111 Latino/as 30 30
703 1020 1783 Asians 12 23
34 1055 1124 Other 32 9 55
12 108 Computed from MCSUI 4-city
file. Hispanics are excluded from White, Black,
Asian, and Other categories.
11
B. Stratification of geographic areas
(segments) Note different strata definitions
in each city 1-4 STRATUM
Detroit Atlanta Boston Black (gt70) Black
poverty Black All other Black
non-poverty Hispanic White poverty White
low-income White non-poverty White high-inc.
12
L.A. Japanese low poverty Korean low
poverty Chinese low poverty Chinese high
poverty Black low poverty Black medium
poverty Black high poverty Hispanic low
poverty Hispanic medium poverty Hispanic high
poverty Asian low poverty Asian medium
poverty Mixed low poverty Mixed medium
poverty Mixed high poverty
13
B. Stratification of geographic areas
(segments) Note different strata definitions
in each city
Detroit Atlanta Boston gt 70 Black gt 50
Black, gt 50 Black gt 40
poverty All other gt 50 Black, gt 50
Latino lt 40 poverty gt 50White,
gt 40 poverty gt 50 White,
lt 20 poverty gt 50 White, lt 40
poverty gt 50 White, gt 20 poverty No
racial/ethnic majority (mixed)
14
L.A. gt10 Japanese, lt 20 poverty gt 10
Korean, lt 20 poverty gt 10 Korean, gt 40
poverty gt10 Chinese, lt 20 poverty
15
L.A. (cont.) gt 50 Black, lt 20 poverty gt 50
Black, 20- 40 poverty gt 50 Black, gt 40
poverty
16
L.A. (cont.) gt 50 Latino, lt 20 poverty gt
50 Latino, 20- 40 poverty gt 50
Latino, high poverty
17
L.A. (cont.) gt 10 Asian, lt 20 poverty gt 10
Asian, 20- 40 poverty Mixed, lt 20
poverty Mixed, 20- 40 poverty Mixed, gt 40
poverty
18
Finding and Training Interviewers

• Detroit DAS students and SRC interviewers
• Other cities recruitment through local media
  organizations
• newspapers, radio, student employment agencies,
  community centers, churches
• Intensive Training

19
The Interview

• Introductory Letter
• Compensation
• Detroit 10 towards end of survey
• Atlanta tote bag
• Boston 5
• L.A. 10

20
The Interview (cont.)

• Verification and Corrections
• phone calls to respondents by field supervisors
  or interviewers
• EX L.A., 16 interviewers dismissed after
  problems discovered (fabricating an interview
  changing the household roster, etc.)

21
Race-Matching of respondents who were
interviewed by someone of their same race
22
Overall response rates
23
III. What are the analysis issues emanating from
the sample design?
Weights in multi-city file 1-6. WPSTHHDE
Expansion Household Weight 1-7.
WPSTPERE Expansion Person Weight (See sample
design reports from each city for a description
of how the weights were calculated).
24
Clustering Effects

• Clustering creates samples of individuals with
  similarities within clusters (sample is not iid
  sample affects error terms)
• Must control for design effects in multivariate
  analyses
• SVYREG depvar indepvars PSU (ucluster) STRATA
  (ustratum)

25
SUMMARY

• Professional survey organizations
• Carefully selected sample
• Careful interview procedures and verification
• Multiple checks and data cleaning
• Variation in sample design across cities