Supporting Statement Part B - Debt Collection Survey 10-14-14(CLEAN)

CONSUMER FINANCIAL PROTECTION BUREAU
INFORMATION COLLECTION REQUEST
SUPPORTING STATEMENT – PART B
DEBT COLLECTION SURVEY FROM THE CONSUMER CREDIT PANEL
(OMB CONTROL NUMBER: 3170-XXXX))

1. Respondent Universe and Selection Methods
The Bureau recently acquired a nationally representative sample of de-identified
consumer credit records (the “Consumer Credit Panel” or “CCP”) from one of the three
national credit reporting agencies. This sample is comprised of a 1-in-48 random sample
of all credit records drawn from the credit reporting agency’s archive from December
2012. The resulting sample includes approximately 5 million de-identified credit records,
representing the universe of approximately 240 million credit records. At the end of each
calendar quarter, the CFPB receives updated credit records for these sampled consumers
(if available) and a 1-in-48 sample of credit records that were newly created since the
previous quarter. This sampling process was designed to provide an on-going panel of
credit records that remains representative of the universe of credit records at each point in
time. The contract with the credit reporting agency also allows for testing to verify that
the panel remains representative.
The de-identified credit records that the CFPB receives carefully exclude any
direct identifying information in order to maintain the anonymity and protect the privacy
of sampled consumers. The records include information about the credit accounts, such
as mortgages and credit cards, that are included in each consumer’s credit record (though
the identity of the creditor is excluded from the information the CFPB receives). Also
included is information about non-credit-related debts that have been reported by thirdparty collection agencies, monetary-related public records (like tax liens and bankruptcy
filings), and details of any credit record inquiries made by lenders in response to an
application for credit. The credit information in the CCP is used to monitor conditions in
consumer credit markets, to study consumer behavior regarding credit, to evaluate the
Page 1 of 8

effects of consumer regulations, or to address other issues in support of the Bureau’s
research, monitoring, and supervisory missions. 1
The target population for the survey is U.S. borrowers, more specifically,
individuals in the United States with credit records at the time of the survey. We plan to
select a stratified random sample with twelve strata (see Table 1), using credit records in
the most recent available quarter of the CCP (expected to be the third quarter of 2014) as
a sampling frame. Survey weights will reflect this sample design.
The proposed survey largely focuses on the experiences of borrowers with recent
debt collection experience. Only a fraction of collections, however, are reported to credit
bureaus. In particular, some third-party collections of non-credit-related debts (for
example, medical bills or utility bills) are reported, but other collections generally are not.
Most—and likely the vast majority of—collections of credit-related debts such as student
loans, credit card bills, and vehicle loans are not reported to credit bureaus. The credit
bureau data is thus an imperfect frame for identifying persons with recent debt collection
experience. However, drawing a sample from the general population to identify such
perons would be prohibitively expensive.
Table 1. Estimated number of records in CCP, number of total credit records, sampling
fraction, and approximate sample size by stratum
New 60-day delinquency
on credit-related debts
in past year

New reported collections of
non-credit-related debts in past year
Any medical

Nonmedical only

Total

Neither

Student loan only
CCP records
Credit records
Sampling fraction
Approx. sample size

862

1,239

4,198

6,299

41,376

59,472

201,504

302,352

0.67%

0.67%

0.67%

0.67%

6

8

28

42

27,502

37,399

109,573

174,474

Credit card only
CCP records
1

Similar datasets are used by the Federal Reserve Board and the Office of the Comptroller of the Currency, so the
CFPB is not unique among regulatory agencies in collecting this type of information in order to inform its
policymaking.
Page 2 of 8

Credit records

1,320,096

1,795,152

5,259,504

8,374,752

0.67%

0.67%

0.67%

0.67%

185

252

738

1,175

27,066

34,093

75,808

136,967

1,299,168

1,636,464

3,638,784

6,574,416

0.67%

0.67%

0.67%

0.67%

182

230

510

922

315,177

362,267

4,198,839

4,876,283

15,128,496

17,388,816

201,544,272

234,061,584

Sampling fraction

0.67%

0.67%

0.08%

0.16%

Approx. sample size

2,122

2,439

3,300

7,861

370,607

434,998

4,388,418

5,194,023

17,789,136

20,879,904

210,644,064

249,313,104

0.67%

0.67%

0.10%

0.19%

Sampling fraction
Approx. sample size
Other + multiple
CCP records
Credit records
Sampling fraction
Approx. sample size
None
CCP records
Credit records

Total
CCP records
Credit records
Sampling fraction

Approx. sample size
2,495
2,929
4,576
10,000
Note: Estimated number of CCP records based on CCP data for 2014Q2. Estimated number of credit
records is equal to 48 times the estimated number of CCP records.

Because most debt collections cannot be directly identified in credit bureau data,
the sampling strata will be defined based on: i) reported third-party collections for noncredit-related debts (for example, medical bills or utility bills) in the past year, and; ii)
past-due payments on credit-related debts that were past due by 60 days or more in the
past year. We intend to oversample credit records with recent third-party collections for
non-credit-related debts and those with past-due payments. As shown in the third row of
the cells for each of the nine strata, the sampling fraction for records with neither a thirdparty non-credit-related collection nor a delinquency will be about 8.5 times lower than
the rate for other credit records (0.08 percent compared with 0.67 percent). Among the
remaining strata, proportionate selection will be used. Systematic selection will be used
to select records within each of the strata at the specified sampling fraction.
We expect an overall response rate of about 30 percent, but we are not aware of
any data that indicate how nonresponse might vary across these strata. More generally, to
Page 3 of 8

the best of our knowledge, no survey like this of debt collections has previously been
attempted. The closest effort in approach is likely the pilot of the “continuation survey”
of the National Survey of Mortgage Borrowers conducted by Freddie Mac. The
continuation survey was sent to mortgage borrowers with an oversampling of consumers
who were delinquent on their mortgages. Like the survey the CFPB seeks PRA approval
to conduct, Freddie Mac’s survey was a mail survey sent to consumers who were
randomly selected from a sample of credit records. The continuation survey oversampled
mortgage borrowers who were delinquent on a mortgage and had a response rate of 32
percent. We expect a lower overall response rate to this survey because we are also
surveying consumers who do not have mortgages and who therefore may be more mobile
and harder to contact.

2. Information Collection Procedures
As described in the question above, we plan to randomly select approximately
10,000 credit records from the CCP. Unique record locators assigned to these credit
records will be sent to the credit reporting agency along with a survey instrument that has
been designed by CFPB staff. The credit reporting agency will identify the consumers
associated with each sampled credit record, and will mail the survey instrument. The
field period will be eight weeks and will include up to four first-class mailings (Table 2).
All sampled consumers will receive an initial mailing that includes a cover letter
introducing the survey along with a paper questionnaire, postage-paid return envelope,
and five dollar cash incentive. A reminder letter will be sent one week later (week 2).
Table 2. Summary of contact and incentive protocol
Week

Letter

Additional content

Incentive

Recipients

1

Cover letter

Questionnaire and postage-paid
return envelope

$5 bill

All

2

1 reminder

5

2 reminder

st

nd

All
Questionnaire and postage-paid

$5 bill

Nonrespondents
Page 4 of 8

return envelope
7

rd

3 reminder

Nonrespondents

In week five, consumers who have not yet completed the survey or opted out will
receive a second reminder letter with a replacement questionnaire, postage-paid return
envelope, and five dollar incentive. A final reminder letter will be sent to remaining
nonrespondents in week seven. The planned contact and incentive protocols follow those
used by the National Survey of Mortgage Borrowers.
The mail survey will include instructions for completing the survey online, and
the survey website will be hosted on servers owned and operated by the credit reporting
agency’s subcontractor. The credit reporting agency’s subcontractor will receive the
responses that come in either by mail or online, strip off any direct identifying PII that
may have been included in the consumer’s response, and send the de-identified
information to the CFPB. This process, which has been successfully pilot tested for the
National Survey of Mortgage Borrowers, allows the CFPB to survey consumers without
revealing direct identifying PII directly to the Bureau.
3. Methods to Maximize Response Rates and Address Issues of Non-Response
Obtaining sufficiently high response rates is a challenge for any survey. The
CFPB intends to include a cash incentive with each survey to boost response rates.
Additionally, the Bureau will incorporate lessons learned by the National Mortgage
Database (NMDB) 2 pilot team, comprised of staff from the CFPB, Freddie Mac, and the
Federal Housing Finance Agency, which has pilot-tested a similar methodology for new
or delinquent mortgage borrowers.
The extensive information from the de-identified credit records for both
respondents and nonrespondents will provide a strong basis for investigating potential
2

The National Mortgage Database (OMB No. 2590-0012) is a joint effort of FHFA and the CFPB and is a uniquely
comprehensive database of information on the residential mortgage origination market. A key purpose of the NMDB
is to make accessible accurate, comprehensive information for monitoring the residential mortgage market. The need
for the NMDB grew out of the subprime mortgage crisis that began in the United States in 2007. The subprime
crisis, and the financial crisis that followed, revealed the extent to which regulators lacked sufficient data to foresee
and adequately respond to a crisis in the mortgage markets.
Page 5 of 8

nonresponse bias relative to the CCP. The data will, for example, permit us to examine
differential rates of nonresponse correlated with credit score, dollar amounts of various
types of credit lines and demographics and geographic information. Based on this
analysis, the CFPB will also construct survey weights so that the survey will be
representative of American consumers with credit records (i.e., consumers who appear in
the CCP or credit bureau data). The nonresponse bias analyses will be of benefit to other
federal agencies using samples based on commercially-available administrative data.
For survey respondents, the CFPB will additionally be able to compare the selfreported demographic information in the survey (for example, education, age, and marital
status) to the auxiliary demographic information included in the credit-record database.
This comparison may shed light on the reliability of such auxiliary data and, thus, provide
information that may be valuable to government researchers and others that rely on such
data when direct measures of these characteristics are unavailable. Further, while the
questions are not identical, comparing the answers of survey respondents to related credit
and debt information in the CCP will shed further light on the quality of self-report data
relative to the credit-record database. This analysis will be of use to other federal
agencies considering supplementing or replacing survey data with data from
commercially-available administrative sources.
Finally, as noted below, the CFPB is considering examining the effects of
alternative versions of the required Privacy Act statement on survey nonresponse.

4. Testing of Procedures or Methods
The CFPB plans to test experimental instruments on a small scale prior to their use in
full-scale experiments. These techniques are meant to reduce the total public burden of the
information collection by ensuring that the large-scale information collection is optimized.
In addition, the CFPB intends to examine whether differences in the length and content of
the Privacy Act statement affect survey nonresponse. The CFPB understands that there has
been limited research on this and related questions. The proposed experiment is based on

Page 6 of 8

findings from the informed consent literature suggesting that shorter statements may both
increase consent rates and knowledge of the consent process. 3,4 Our working hypothesis is
that the longer Privacy Act statement may deter people from participating in the survey. A
counter-hypothesis is that a longer statement reassures potential respondents, increasing
responses rates.
The analysis would be based on a split-sample experiment that will randomize across
two versions of the Privacy Act statement of differing length and content. The version of
the Privacy Act statement for a given sample case would be identical in the paper and web
questionnaires.
To mitigate the potential effects of this experiment on overall survey response rates
in the event there are sizable differences in response rates, the CFPB expects to assign 80
percent of sample cases the shorter version, which the CFPB believes is more likely to
produce a higher response rate. With an overall sample size of 10,000, and with the unequal
allocation described above, we have sufficient power to detect a difference in response rates
of about 2.9 percentage points with .90 confidence (2-sided) and .80 power, assuming a
response rate of 30 percent for one of the groups.
The primary outcome of interest would be differences in (unit) response rates
across the two groups. The CFPB will also examine differences in item nonresponse,
particularly with respect to questions that address sensitive topics (for example, income, race
and ethnicity, or reasons for not paying a debt in collection). Results from this experiment
may provide evidence about the effects of information related to the confidentiality and the
uses of survey responses that may help other entities, particularly other government
agencies, that collect financial or other sensitive information through similar surveys.
5. Contact Information for Statistical Aspects of the Design.
Ken Brevoort (Supervisory Economist, Office of Research, CFPB)
3

Tait AR, Voepel-Lewis T, Nair VN, Narisetty NN, Fagerlin A. (2013), “Informing the Uninformed: Optimizing
the Consent Message Using a Fractional Factorial Design.” JAMA Pediatr. 167 (7): 640-646.
doi:10.1001/jamapediatrics.2013.1385. Error! Main Document Only.
4
Das M, Couper MP (2014), “Optimizing Opt-Out Consent for Record Linkage.” Journal of Official Statistics, 30
(3): 479-497, http://dx.doi.org/10.2478/JOS-2014-0030.
Page 7 of 8

Brian Bucks (Economist, Office of Research, CFPB)
Mick Couper (Research Professor, University of Michigan and Joint Program in
Survey Methodology)

Page 8 of 8