Upload
sheryl-barnett
View
215
Download
0
Embed Size (px)
Citation preview
DRAFT - not for publication
Nonresponse Bias Analysis in a Survey of Banks
Carl Ramirez
U.S. Government Accountability [email protected]
DRAFT - not for publication
Overview
• Describe NR bias analysis techniques and results in one survey of small, minority-owned financial institutions.
• Make observations about NR bias analyses in small population surveys
• Describe emerging NR bias analysis practices at one US Federal research agency
DRAFT - not for publication
Survey of 196 Minority-Owned Banks
• Census of all institutions defined by at least one regulator as minority-owned as of March 2006.
• Web survey, mail/fax options upon request• President/CEO was targeted respondent• 76% unit response rate (AAPOR RR2).• Key estimates: Awareness of and attendance at
regulator programs, rating of regulator efforts to preserve minority ownership, financial outlook
DRAFT - not for publication
Survey Fieldwork
• Field period: March 14 to April 28 – 6 weeks & 3 days• Contacts:
– Precontact call in Feb. to determine eligibility and get email
– Prenotification email in early March
– Survey cover email March 14
– 2 email NR followups
– National Bankers Association endorsement contacts
– Multiple phone NR followups by program analysts
– Final reminder and closeout emails in late April
DRAFT - not for publication
NR Bias Analysis Typology(Groves & Brick)
• Compare survey estimates to other benchmarks
• Compare R’s to NR’s on auxiliary variables
(frame, external data, fieldwork observations, etc.)
• Examine variation within respondents (subgroups,
converted nonrespondents, early/late responders)
• Alter weighting adjustment
DRAFT - not for publication
Unique Aspects of Establishment Survey NR Bias Analyses
• More longitudinal sample designs = previous
response benchmark opportunities
• Richer administrative data = more benchmarks
and survey variable correlates
• Smaller populations and subgroups = potential
for examining individual nonrespondents
DRAFT - not for publication
NR Bias Evaluation Methods Applied to Survey of Banks
• Compare respondents to sample on key variables
• Compare response rates of subgroups defined by frame
information, related to survey variables of interest
• Level of Effort (Time of Return)
• Subsample of nonrespondents converted with high-effort
followup
DRAFT - not for publication
Respondents vs. Entire Sample on key auxiliary variable: Regulator
R’s Sample
FDIC 54% 55%
Federal Reserve 11 11
OCC 23 22
OTS 12 11
DRAFT - not for publication
Respondents vs. Sample: Minority Type
R’s Sample
African American 26% 24
Native American 10 10
Asian 39 34
Hispanic 17 24
DRAFT - not for publication
Respondents vs. Sample: Size
R’s Sample
< $100 Million assets 43% 41%
$101-$300 M 32 32
$301-$500 M 12 9
$501M - $10 billion 11 15
DRAFT - not for publication
Differences in Response Rate by Regulator of Bank
R’s NR’s
Federal Reserve 82% 18
OTS 82% 18
OCC 77% 23
FDIC 73% 27
DRAFT - not for publication
Differences in Response Rate by Minority Type
R’s NR’s
African American 81% 19
Native American 80% 20
Asian 76% 24
Hispanic 70% 30
DRAFT - not for publication
Differences in Response Rate by Profitability
R’s NR’s
Low return on assets 81% 19
Medium 75% 25
High 73% 27
DRAFT - not for publication
Level of Effort Analysis(Time of Return)
Key Estimates Early
(n=51)
Middle
(84)
Late
(9)
Received regulator mailings on minority program
48% 50 56
Aware of FDIC minority bank web page
55% 51 44
Attended FDIC minority roundtable or conference
45% 43 56
Used regulator’s technical assistance program
25% 27 56
DRAFT - not for publication
Analysis of Respondents with other NR Characteristics
• Comparison of low-salience (low response
propensity) respondents to all other respondents:
12 banks not considered by primary regulator as
“minority owned” but so designated by one of
the other 3 regulators.
• Removing these respondents changed key
estimates by 1-2% , none by more than 4-5%
DRAFT - not for publication
Micro-Level Examination of Nonrespondents
• Small sample allowed in-depth study of
individual banks not responding
• 4 explicit refusals – known reasons appeared
largely unrelated to key measures
• Personal contacts for NR followup yields known
reason for delay
DRAFT - not for publication
Emerging GAO practice on NR Bias Analysis?
• Groves & Brick typology
• Suggests hierarchy of methods on 2 dimensions:
– Prefer use of data related to survey variable over
comparison of subgroups or time of return analysis
– Prefer methods using data available for entire
population over portion of population or sample
• If conduct analyses, describe in report