Download:
pdf |
pdf“NATIONAL SURVEY OF MORTGAGE ORIGINATIONS"
OMB NUMBER 2590-0012
SUPPORTING STATEMENT
The Federal Housing Finance Agency (FHFA or the Agency) is seeking approval for a three-year
extension of the Paperwork Reduction Act (PRA) clearance for the National Survey of Mortgage
Originations (NSMO). OMB has assigned the NSMO control number 2590-0012, which is due
to expire on April 30, 2020. The NSMO is a recurring quarterly survey of individuals who have
recently obtained a loan secured by a first mortgage on single-family residential property. The
survey questionnaire is sent to a representative sample of approximately 6,000 recent mortgage
borrowers each calendar quarter and consists of 96 multiple choice and short answer questions
designed to obtain information about borrowers’ experiences in choosing and in taking out a
mortgage. The NSMO is sponsored by FHFA and is one component of the National Mortgage
Database Program, an ongoing joint effort of FHFA and the Consumer Financial Protection
Bureau (CFPB) (for PRA purposes, the NSMO is sponsored by FHFA). A copy of the survey
questionnaire sent out in the first quarter of 2020 is included as Attachment 1. 1 FHFA is also
seeking clearance to pretest future iterations of the survey questionnaire and related materials
from time to time using focus groups.
A. JUSTIFICATION
1. Circumstances necessitating the collection of information
The NSMO is a component of the “National Mortgage Database” (NMDB) Program which is a
joint effort of FHFA and the Consumer Financial Protection Bureau (CFPB). The NMDB
Program is designed to satisfy the Congressionally-mandated requirements of section 1324(c) of
the Federal Housing Enterprises Financial Safety and Soundness Act. 2 Section 1324(c) requires
that FHFA conduct a monthly survey to collect data on the characteristics of individual prime
and subprime mortgages, and on the borrowers and properties associated with those mortgages,
in order to enable it to prepare a detailed annual report on the mortgage market activities of the
Federal National Mortgage Association (Fannie Mae) and the Federal Home Loan Mortgage
Corporation (Freddie Mac) for review by the appropriate Congressional oversight committees.
1
In addition, copies of the questionnaire in both English and Spanish can be accessed at http://www.fhfa.gov/nsmo.
2
12 U.S.C. 4544(c).
1
Section 1324(c) also authorizes and requires FHFA to compile a database of otherwise
unavailable residential mortgage market information and to make that information available to
the public in a timely fashion.
As a means of fulfilling those and other statutory requirements, as well as to support
policymaking and research regarding the residential mortgage markets, FHFA and CFPB jointly
established the NMDB Program in 2012. The Program is designed to provide comprehensive
information about the U.S. mortgage market and has three primary components: (1) the NMDB;
(2) the NSMO; and (3) the American Survey of Mortgage Borrowers (ASMB).
The NMDB is a de-identified loan-level database of closed-end first-lien residential mortgage
loans that is representative of the market as a whole, contains detailed loan-level information on
the terms and performance of the mortgages and the characteristics of the associated borrowers
and properties, is continually updated, has an historical component dating back to 1998, and
provides a sampling frame for surveys to collect additional information. The core data in the
NMDB are drawn from a random 1-in-20 sample of all closed-end first-lien mortgage files
outstanding at any time between January 1998 and the present in the files of Experian, one of the
three national credit repositories. A random 1-in-20 sample of mortgages newly reported to
Experian is added each quarter.
The NMDB draws additional information on mortgages in the NMDB datasets from other
existing sources, including the Home Mortgage Disclosure Act (HMDA) data that are
maintained by the Federal Financial Institutions Examination Council (FFIEC), property
valuation models, and administrative data files maintained by Fannie Mae and Freddie Mac and
by federal agencies.
The NSMO was developed to complement the NMDB by providing critical and timely
information on newly originated mortgages directly from the borrower. Such information—not
available from other sources— provides information on the borrowers’ experiences with the
mortgage origination process. In particular, the survey questionnaire is designed to elicit directly
from mortgage borrowers information on the characteristics of the borrowers and on their
experiences in finding and obtaining a mortgage loan, including: their mortgage shopping
behavior; their mortgage closing experiences; their expectations regarding house price
appreciation; and critical financial and other life events affecting their households, such as
unemployment, large medical expenses, or divorce. The survey questions do not focus on the
terms of the borrowers’ mortgage loans because these fields are available in the Experian data.
However, the NSMO collects a limited amount of information on each respondent’s mortgage to
verify that the Experian records and survey responses pertain to the same mortgage.
The NSMO has been conducted quarterly since 2014, with the most recent wave—wave 25
(Quarter 1 of 2020)—having been mailed on February 3, 2020. Each wave of the NSMO is sent
to the primary borrowers on about 6,000 mortgage loans, which are drawn from a simple random
sample of the 80,000 to 100,000 newly originated mortgage loans that are added to the National
Mortgage Database from the Experian files each quarter (at present, this represents an
approximately 1-in-15 sample of loans added to the National Mortgage Database and an
2
approximately 1-in-300 sample of all mortgage loan originations). By contract with FHFA, the
conduct of the NSMO is administered through Experian, which has subcontracted the survey
administration through a competitive process to Westat, a nationally recognized survey vendor. 3
Westat also carries out the pre-testing of the survey materials.
FHFA also obtains data from a separate survey, the ASMB, which is also based on the NMDB.
While the NSMO solicits information on newly originated mortgages the ASMB focuses on
borrowers’ experience with maintaining existing mortgages. This includes their experience
maintaining mortgages under financial stress, their experience in soliciting financial assistance,
their success in accessing federally sponsored programs designed to assist them, and, where
applicable, any challenges they may have had in terminating a mortgage loan. 4
2. Use of data
FHFA views the NMDB Program as a whole, including the NSMO, as the monthly “survey” that
is required by section 1324 of the Safety and Soundness Act. Core inputs to the NMDB, such as
a regular refresh of the Experian data, occur monthly, though NSMO itself does not. In
combination with the other information in the NMDB, the information obtained through the
NSMO is used to prepare the report to Congress on the mortgage market activities of Fannie Mae
and Freddie Mac that FHFA is required to submit under section 1324, as well as for research and
analysis by FHFA and CFPB in support of their regulatory and supervisory responsibilities
related to the residential mortgage markets. The NSMO is especially critical in ensuring that the
NMDB contains uniquely comprehensive information on the range of nontraditional and
subprime mortgage products being offered, the methods by which these mortgages are being
marketed and the characteristics—and particularly the creditworthiness—of borrowers for these
types of loans.
In November 2018, FHFA and CFPB released the first loan-level dataset collected through the
NSMO for public use. The first release contained data for mortgages originated in 2013 through
2017. An updated version of the dataset that added 2017 mortgage originations was released on
February 20, 2020. This dataset provides a resource for research and analysis by federal
agencies, by Fannie Mae and Freddie Mac, and by academics and other interested parties outside
of the government. 5
3
The Fair Credit Reporting Act, 15 U.S.C. § 1681 et seq., requires that the survey process, because it utilizes
borrower names and addresses drawn from credit reporting agency records, must be administered through the credit
repository (in this case Experian) in order to maintain consumer privacy.
4
OMB has assigned the ASMB control no. 2590-0015, which expired on July 31, 2019. The ASMB was conducted
annually from 2016 through 2018, but was not conducted in 2019. FHFA expects to conduct the survey again in late
2020.
5
The February 2020 NSMO public use dataset can be accessed here: https://www.fhfa.gov/nsmodata.
3
FHFA is also seeking OMB approval to continue to conduct cognitive pre-testing of the survey
materials. The Agency uses information collected through that process to assist in drafting and
modifying the survey questions and instructions, as well as the related communications, to read
in the way that will be most readily understood by the survey respondents and that will be most
likely to elicit usable responses. Such information is also used to help the Agency decide on how
best to organize and format the survey questionnaires.
3. Use of information technology
The NSMO uses machine-readable paper questionnaires in English and also gives recipients the
option of completing the survey online in either English or Spanish. At first, approximately one
third of the overall surveys were completed online, but the share of online surveys has
approached 50 percent in recent waves. Completed paper questionnaires are scanned and the
responses are automatically uploaded into the electronic National Mortgage Database.
4. Efforts to identify duplication
As explained above, the majority of data included in the National Mortgage Database is drawn
from existing sources—primarily, the consumer credit database maintained by Experian; the
HMDA data released by FFIEC; and administrative data in the possession of FHFA, its regulated
entities, and other federal agencies. As described under Item #1, the NSMO questionnaire is
designed to obtain critical and timely information that is not available from existing sources.
The survey obtains this information directly from borrowers, who are likely to be the most
reliable and accessible—and, in some cases, the only—source for this information.
5. Impact on small entities
The information collection will not have a significant economic impact on a substantial number
of small entities. The survey recipients are individuals only.
6. Consequences of less frequent collection and obstacles to burden reduction
Section 1324 of the Safety and Soundness Act requires that FHFA undertake a survey of
mortgage markets on a monthly basis. 6 While the performance data on existing mortgages in the
National Mortgage Database is pulled from the Experian database on a monthly basis, newly
originated mortgages are added to the National Mortgage Database on a quarterly basis. The
NSMO questionnaires are sent to a random sample of borrowers that originated their mortgages
in the year and quarter that corresponds to the quarterly draws of newly originated mortgages
from the Experian database. One important purpose of the survey is to monitor loan origination
trends. While monthly housing surveys would provide the optimal feedback regarding these
trends, FHFA believes that quarterly surveys are sufficient.
6
See 12 U.S.C. § 4544(c).
4
7. Circumstances requiring special information collection
There are no special circumstances that require FHFA to conduct the information collection in a
manner inconsistent with the guidelines provided in this Item 7.
8. Solicitation of comments on information collection
In accordance with the requirements of 5 CFR 1320.8(d), FHFA published a request for public
comments regarding this information collection in the Federal Register on December 10, 2019. 7
The 60-day comment period closed on February 10, 2020. FHFA received no comment letters.
9. Provision of payments or gifts to respondents
Until recently, survey recipients received a cash payment of five dollars as an inducement to
complete and return the NSMO questionnaire. Recipients who failed to respond to the first two
survey solicitations also received an additional cash inducement of five dollars. Recent waves of
the survey, however, have shown a slow but steady decline in the response rate, a problem facing
many other surveys like NSMO. In response, FHFA has been experimenting with the cash
incentives.
In wave 22, one half of the usual 6,000 borrowers were randomly selected to receive a ten-dollar
cash incentive with the first mailing instead of the typical five-dollar incentive. In wave 25, all
borrowers were sent an initial incentive of ten dollars. In terms of the second incentive, one
random half of the non-respondents were sent the normal five-dollar cash incentive as in
previous waves and the other random half were sent a letter informing them that they will be sent
a twenty-dollar incentive upon completion of the survey. In wave 26 (to be sent in Quarter 2 of
2020), one random half of the borrowers will receive an initial incentive of five dollars and the
other random half will receive an initial incentive of ten dollars, but all non-respondents will
receive a letter informing them that they will be sent a twenty-dollar incentive upon completion
of the survey. Data collected during these experiments will be used to make decisions about the
NSMO methodology in 2020.
Each cognitive pre-testing participant may receive approximately fifty dollars as an incentive
payment.
10. Assurance of confidentiality
With respect to the confidentiality of survey responses, the cover letter that accompanies each
NSMO questionnaire contains the following statement:
This survey is voluntary, and we ask that you not identify yourself in any way
when you return your questionnaire in the enclosed postage-paid return
7
See 84 FR 67447 (Dec. 10, 2019). A copy of the 60-day Notice is included as Attachment 2.
5
envelope. The code numbers on the survey are there to aid in processing and
keep track of returned surveys. No names or other identifying information is
ever included in the data.
The questionnaire itself contains a statement, required by the Privacy Act, 8 informing recipients
that “[s]ubmission of the survey authorizes FHFA to collect the information provided and to
disclose it as set forth” in the current System of Records Notice (SORN) for the National
Mortgage Database. 9 The questionnaire also instructs recipients not to include their names or
addresses when completing the questionnaire.
Section 1324 of the Safety and Soundness Act authorizes FHFA to modify the mortgage data
released to the public as necessary to ensure that it contains no “representation of information
that permits the identity of a borrower to which the information relates to be reasonably inferred
by either direct or indirect means.” 10 For each sampled loan and its associated borrower(s),
Experian provides its survey subcontractor, Westat, with the identifying information it needs to
administer the survey. However, the data on borrowers and loans that is accessible to FHFA,
CFPB, and any other authorized user of the National Mortgage Database, including data obtained
through the NSMO, does not include any direct identifying information such as borrowers’
names, addresses, or Social Security numbers or the name of any financial institution.
Westat mails a survey questionnaire to the borrower(s) on each sampled mortgage loan at the
property address associated with that mortgage. It then uses an encrypted key to track the
surveys so that it can compile and maintain the survey opt-out list and identify non-responders to
whom it must send follow-up correspondence. All returned questionnaires and any nondelivered mail are sent directly to Westat, not to FHFA, CFPB, or Experian. To maintain the deidentified nature of the data and the confidentiality of the survey responses, Westat purges all
responses of any identifying information before providing the collected information to FHFA’s
National Mortgage Database Program staff for further processing (which is described in Part B
of this Supporting Statement).
Similarly, while Westat knows the identity of the cognitive pre-testing participants, that
information is not conveyed to FHFA and is not included in the National Mortgage Database in
any form.
8
5 U.S.C. § 552a.
9
See 80 FR 52275 (Aug. 28, 2015); 81 FR 95595 (Dec. 28, 2016). Copies of the current SORN for the National
Mortgage Database and a subsequent revision are included as Attachment 3.
10
See 12 U.S.C. §§ 4544(c)(3), (4).
6
11. Questions of a sensitive nature
Of the 96 questions on the most recent iteration of the survey questionnaire, approximately 20-25
might be considered to be of a sensitive nature by particular borrowers. Questions that FHFA
has identified as potentially sensitive include those requesting information on loan terms,
purchase price, household income and other sources of funds, employment status, level of
education, age, sex, race, ethnicity, and marital status.
Each of those questions is designed to elicit information that FHFA is required by statute to
collect or that is otherwise essential to fulfilling the purposes of the NSMO and the National
Mortgage Database Program as a whole. While FHFA understands that some survey recipients
will be reluctant to answer questions about these potentially sensitive topics, the Agency believes
that others will look upon doing so as an opportunity to express themselves about issues of
concern to them.
12. Estimates of the hour burden of the information collection
This information collection comprises two components: (1) conducting the survey; and (2) pretesting survey questionnaires and related materials through the use of cognitive testing. FHFA
estimates that the total annualized hour burden imposed upon members of the public by this
information collection will be 12,030 hours: 12,000 hours associated with conducting the survey
and 30 hours associated with pre-testing the survey materials. Because the survey recipients and
cognitive testing participants are individuals only, there are no hourly costs associated with the
burden estimates. The overall burden estimates are based on the following calculations:
1) Conducting the Survey
The estimated annualized hour burden associated with conducting the NSMO is 12,000 hours.
The NSMO questionnaire will be sent to 6,000 recipients quarterly. Although, based on
historical experience, the Agency expects that only 20 to 30 percent of those surveys will be
returned, it has assumed that all of the surveys will be returned for purposes of this burden
calculation. Based on the reported experience of respondents to prior NSMO questionnaires,
FHFA estimates that it will take each respondent 30 minutes to complete the survey, including
the gathering of necessary materials to respond to the questions.
Recipients read and complete survey questionnaire and return the completed form to the
survey subcontractor:
•
•
•
•
•
Completion time per recipient:
Survey mail-outs annually:
Recipients per survey:
Total recipients annually:
Total hours annually:
0.5 hours
4
6,000
24,000
12,000 hours
7
2) Pre-Testing of Survey Materials
The estimated annualized hour burden associated with the pre-testing of the survey materials is
30 hours.
Selected individuals participate in cognitive testing to pre-test the survey questionnaire and
related materials:
•
•
•
Time per participant:
Total participants annually:
Total hours annually:
1 hour
30
30 hours
13. Estimated total annualized cost burden to respondents
FHFA has not identified any costs to respondents.
14. Estimated cost to the federal government
The estimated annual burden to the federal government is $732,000 and 400 hours, calculated as
follows:
FHFA analyst embeds NSMO data into a query-based electronic database and carries out
data cleaning, imputation, and non-response bias weighting:
•
•
•
•
•
Processing time per survey:
Total surveys annually:
Total hours:
Hourly rate:
Total cost:
100 hours
4
400
$80 (includes salary, benefits, and overhead)
$32,000
In addition, approximately $700,000 will be paid annually to the contractor hired to conduct the
surveys. Of this, approximately $250,000 will be attributable to the cash incentive payments to
survey recipients; approximately $250,000 will be for printing and assembly costs;
approximately $100,000 will be for postage costs; and approximately $100,000 will be for other
fixed costs.
$32,000 (hourly cost) + $700,000 (paid to subcontractor) = $732,000.
15. Reasons for change in burden
The estimated burden has not changed.
8
16. Plans for tabulation, statistical analysis and publication
On April 18, 2018, FHFA and CFPB released aggregate data from the 6,285 responses pertaining
to borrowers who obtained a mortgage in 2016 and whose loans were reported to Experian. 11
The data provide an overview of the mortgage market and borrowers’ experiences in 2016.
These unweighted responses were about one-third of the sample drawn from mortgages
originated in 2016. On February 20, 2020, FHFA and CFPB released additional loan-level data
for public use collected through the NSMO. 12 The analytical techniques used in preparing the
data are discussed in detail in Part B of this Supporting Statement.
Going forward, FHFA and CFPB plan to publish similar data annually, with each presenting
NSMO data pertaining to borrowers who obtained a mortgage during a particular calendar year.
The agencies expect to release reports on NSMO data for loans originated in 2018 and
subsequent years approximately 24 months after the end of the calendar year to which the data
pertains.
NSMO data releases and technical documentation are made available on FHFA’s public website
at: https://www.fhfa.gov/nsmodata.
17. If seeking approval to not display the expiration date for OMB approval of the
information collection, explain the reasons why display would be inappropriate
FHFA will display the expiration date for OMB approval.
18. Explain each exception to the topics of the certification statement identified in
“certification for paperwork reduction act submission.”
There are no exceptions to the topics of the certification statement identified in the “Certification
for Paperwork Reduction Act Submission.”
11
The April 2018 release is available at:
https://www.fhfa.gov/PolicyProgramsResearch/Programs/Documents/NMDB-technical-report_6pt0_041818.pdf
12
The February 2020 release is available at: https://www.fhfa.gov/DataTools/Downloads/Documents/NSMOPublic-Use-Files/NSMO-Technical-Documentation-20200220.pdf.
9
B. COLLECTIONS OF INFORMATION INVOLVING STATISTICAL METHODS
Question 1. Describe (including a numerical estimate) the potential respondent
universe and any sampling or other respondent selection methods to be used. Data on
the number of entities (e.g., establishments, State and local government units,
households, or persons) in the universe covered by the collection and in the
corresponding sample are to be provided in tabular form for the universe as a whole
and for each of the strata in the proposed sample. Indicate expected response rates
for the collection as a whole. If the collection had been conducted previously, include
the actual response rate achieved during the last collection.
The NSMO represents a universe of 6 to 9 million recently originated closed-end first-lien
mortgage loans on single-family residential property that are reported to Experian annually. A 1in-20 simple random sample of those mortgages are added to the National Mortgage Database
each quarter, resulting in approximately 80,000 to 100,000 loans being added quarterly and
320,000 to 400,000 loans being added annually. For each of the twelve quarterly waves of the
NSMO that have been mailed to date, FHFA has selected an approximately 1-in-15 random
sample of loans from one or more recent quarterly updates of the National Mortgage Database.
This represents an approximately 1-in-300 sample of the universe.
For each quarterly survey FHFA randomly selects a sample of approximately 6,000 mortgage
loans from those reported to Experian during the preceding quarter and newly added to the
National Mortgage Database, with the additional conditions that the loans must have been
reported to Experian within a year of origination and that the borrowers must not have been
selected for an earlier NSMO survey. As of this writing, the NSMO survey is currently on wave
25 (mailed on February 3, 2020). Table 1 shows the survey field periods to date.
Table 1. Survey Field Periods
Wave
Survey Field Dates
Calendar Quarter
Surveys Mailed
1
April to June 2014
2014 Quarter 1
15,000
2
June to August 2014
2014 Quarter 2
3,000
3
August to November 2014
2014 Quarter 3
5,992
4
November 2014 to February 2015
2014 Quarter 4
5,795
5
February to May 2015
2015 Quarter 1
5,925
6
May to August 2015
2015 Quarter 2
4,428
7
August to November 2015
2015 Quarter 3
7,352
8
November 2015 to February 2016
2015 Quarter 4
5,913
9
February to May 2016
2016 Quarter 1
5,907
10
10
May to August 2016
2016 Quarter 2
5,885
11
August to November 2016
2016 Quarter 3
5,904
12
November 2016 to February 2017
2016 Quarter 4
5,919
13
February to May 2017
2017 Quarter 1
5,910
14
May to August 2017
2017 Quarter 2
5,804
15
August to November 2017
2017 Quarter 3
5,809
16
November 2017 to February 2018
2017 Quarter 4
5,707
17
February to May 2018
2018 Quarter 1
5,755
18
April to July 2018
2018 Quarter 2
5,773
19
August to November 2018
2018 Quarter 3
5,759
20
November 2018 to February 2019
2018 Quarter 4
5,770
21
February to May 2019
2019 Quarter 1
5,746
22
May to August 2019
2019 Quarter 2
5,720
23
August to November 2019
2019 Quarter 3
5,737
24
November 2019 to February 2020
2019 Quarter 4
5,676
25
February to May 2020
2020 Quarter 1
5,698
Total
151,884
As shown in Table 2, NSMO typically samples about 6,000 new mortgage originations each
quarter. Over 23 waves for which FHFA has received data from Experian, nearly 31 percent of
the surveys mailed to borrowers of sampled mortgages were completed and 65 percent of
completed surveys were received by mail.
Table 2. Survey Samples and Returns
Surveys Completed
Average
Sampling
Weight
Surveys
Mailed
Postal
NonDelivery
Surveys
Delivered
1
464.21
15,000
218
2
296.14
3,000
3
280.96
4
263.63
Wave
OptOut
Total
By Mail
Online
English
Online
Spanish
14,782
5,793
4,410
1,360
23
169
37
2,963
1,076
858
214
4
31
5,992
110
5,882
2,073
1,534
524
15
40
5,795
86
5,709
2,020
1,496
514
10
53
11
5
247.32
5,925
126
5,799
2,089
1,567
520
2
39
6
238.92
4,428
38
4,390
1,489
1,133
353
3
31
7
296.64
7,352
147
7,205
2,494
1,744
744
6
39
8
326.97
5,913
99
5,814
1,899
1,305
587
7
24
9
292.31
5,907
155
5,752
1,824
1,230
584
10
42
10
253.27
5,885
98
5,787
1,765
1,148
607
10
36
11
278.27
5,904
172
5,732
1,733
1,097
627
9
21
12
343.76
5,919
167
5,752
1,778
1,078
687
13
18
13
363.21
5,910
127
5,783
1,885
1,197
675
13
32
14
318.55
5,804
107
5,697
1,681
1,085
588
8
21
15
270.61
5,809
136
5,673
1,537
765
760
12
24
16
305.24
5,707
164
5,543
1,507
757
738
12
26
17
304.31
5,755
112
5,643
1,647
879
762
6
45
18
262.93
5,773
163
5,610
1,536
812
711
13
32
19
266.84
5,759
242
5,517
1,464
760
695
9
29
20
284.5
5,770
206
5,564
1,396
762
627
7
11
21
266.12
5,746
251
5,495
1,511
777
719
15
17
22
213.35
5,720
219
5,501
1,405
757
630
18
29
23
262.92
5,737
235
5,502
1,236
647
579
10
29
Total
303.25
140,510
3415
137,095
42,838
27,798
14,805
235
838
100.0%
2.4%
97.6%
30.5%
19.8%
10.5%
0.2%
0.6%
100.0%
64.9%
34.6%
0.5%
NA
Percent of Mailed
Surveys
Percent of Completed Surveys
In 2014, the first year of the survey, a modified version was conducted for the first three waves
in April, June, and September. Wave 1 (April) included a sample of 15,000 mortgages. This
was a catch-up period to cover mortgages originated in 2013 and newly reported to Experian in
the archives for June, September, and December 2013. Wave 2 (June) included 3,000 surveys
and was for mortgages that were originated in 2013 and newly reported to Experian between
January and March 2014. For wave 3 (August), Westat mailed out 6,000 surveys representing
mortgages that were originated in 2013 and reported to Experian between March and June 2014
within a year of origination as well as any mortgages originated in 2014 and reported to Experian
between January and June 2014.
12
Wave 4, mailed in November 2014, was the first sample that is comparable to subsequent
surveys. It comprised sample mortgages newly reported to Experian in the most recent quarter
(July to September 2014) that was reported within a year of origination. It is also the first wave
for which Experian eliminated potential sample cases deemed to not have legitimate addresses or
names prior to mailing. Other than slight changes to two questions, the questionnaire was
unchanged from prior waves. This same questionnaire was used for wave 5.
Initial analysis of data from the first four waves of the survey suggested that respondents may
have frequently misunderstood or misinterpreted some of the questions, prompting major
revisions to the questionnaire for part of wave 6 and all of wave 7 (users should be aware of
these interpretation inconsistencies when using data from the earlier waves). For wave 6,
surveys for mortgages that were originated in 2014 were mailed on the established schedule and
using the original questionnaire; surveys for mortgages originated in 2015 were held back to be
mailed with wave 7, using the new questionnaire.
Wave 7 consisted of three samples drawn independently. The first were those 1,236 respondents
selected for wave 6 with loans originated in 2015. The second were 4,981 respondents with
mortgages newly reported to Experian between April and June 2015 (the normal quarterly
sample frame). Finally, a special sample of 1,142 borrowers residing in “remote rural” counties
as defined using a USDA criterion with 2014 loan originations reporting to Experian within a
year of origination was selected. Each subsample was assigned a different sample weight. All
subsequent waves of the survey have included only the regular sample, mailed on-schedule.
Returned questionnaires and online responses were evaluated to determine the set of usable
responses. Table 3 summarizes the results of this analysis through the 21 waves which have
been completely processed and indicates the four criteria for rejecting a completed questionnaire.
Table 3. Usable Survey Responses
Survey
Wave
Returned
Duplicate
or
Ineligible
Answered
No to Q1
Did Not
Finish
Survey
Wrong
Loan
Usable
Weighted
Usable*
1
5,793
84
738
127
216
4,628
6,871,209
2
1,076
15
84
16
38
923
875,467
3
2,073
37
108
36
59
1,833
1,659,752
4
2,020
164
88
46
64
1,658
1,395,466
5
2,089
37
81
46
62
1,863
1,443,963
6
1,489
116
69
29
50
1,225
991,516
7
2,494
65
144
78
98
2,109
2,108,874
8
1,899
42
73
28
59
1,697
1,900,299
13
9
1,824
38
69
27
37
1,653
1,701,989
10
1,765
59
84
40
58
1,524
1,432,246
11
1,733
41
92
38
38
1,524
1,602,998
12
1,778
58
102
49
50
1,519
1,995,025
13
1,885
48
103
52
54
1,628
2,107,377
14
1,681
50
66
52
44
1,469
1,817,181
15
1,537
30
140
78
33
1,256
1,544,244
16
1,507
27
116
70
26
1,268
1,713,007
17
1,647
25
127
64
34
1,397
1,723,003
18
1,536
18
117
60
34
1,307
1,500,016
19
1,464
23
108
61
31
1,241
1,512,449
20
1,396
22
116
66
25
1,167
1,617,383
21
1,511
29
152
82
23
1,225
1,499,320
Total
40,197
1,028
2,777
1,145
1,133
34,114
39,012,784
Percent
of Mailed
Surveys
31.1%
0.8%
2.2%
0.9%
0.9%
26.4%
NA
* The weighted usable total excludes the remote rural sample in wave 7.
The first category of unusable surveys comes from respondents whose sample loans were
ultimately removed from the NMDB after the survey had been executed, either because the loans
were deemed to have duplicate trade lines and did not meet the criteria for remaining in the
NMDB, or because the sample loan was determined to be a second and not a first mortgage. In
some instances, the survey response itself led to the removal, as margin notes or comments
indicated that the loan was a second lien. This was a particular problem in wave 4.
The second category is a “no” response to the first question (Q1). Q1 is used as a screener
question to confirm that the survey respondent took out a mortgage during the reporting period
as suggested by Experian’s records. In wave 1, a surprisingly high number of respondents (738)
said that they had not taken out a mortgage. An analysis of these responses suggests that many
people did not consider a refinance a “new” mortgage. Consequently, in wave 2, the wording of
Q1 was changed to add the phrase “including any mortgage refinances.” With this change, the
share of “no” responses to Q1 decreased from 13 percent to 8 percent.
The third category eliminates breakoffs, defined as questionnaires for which the respondent did
not answer almost all questions from the middle of the survey through the end, or answered less
than 50 percent of the questions overall.
14
The fourth category is for respondents who provided information on the wrong loan. Although
the sampling frame is tied to a particular loan associated with the borrower, the questionnaire
does not refer explicitly to that loan. Instead, respondents who have taken out multiple loans
during the reference period are asked to report on the “most recent,” which, in some instances,
has not been the sample loan. This was a particular problem in wave 1 which, as a “catch up”
survey, had a relatively long reference period. Also, some respondents who have refinanced
their mortgage have reported on the original home purchase mortgage rather than the refinance.
Finally, it appears that in a few instances the survey has been sent to the wrong person, with
answers bearing no resemblance to the sample loan features as characterized by Experian
records. In each of those circumstances the survey response was removed from the data set used
for analysis.
Overall, for the first 21 waves 34,114 usable responses were obtained from 40,197 returned
surveys. This resulted in a usable response rate of 26.4% out of the mailed-out questionnaires.
Question 2. Describe the procedures for the collection of information, including:
• Statistical methodology for stratification and sample selection,
• Estimation procedure,
• Degree of accuracy needed for the purpose described in the justification,
• Unusual problems requiring specialized sampling procedures, and
• Any use of periodic (less frequently than annual) data collection cycles to reduce
burden.
Data for the NSMO is collected through a single-blind mail survey format. The survey’s sample
selection is discussed earlier in Question 1. The NSMO is a simple, random sample of mortgage
originations and is not stratified. Alternatives designed for stratifying, clustering, or cut-off
samples were not considered. While subpopulations are of interest, the key purpose of the
survey is to gather information on the origination characteristics of the sampled borrowers. No
estimation procedures are needed in interpreting the survey responses. Similarly, no hypotheses
are being tested and no unusual problems exist that require specialized sampling procedures.
Question 3. Describe methods to maximize response rates and to deal with issues of
non-response. The accuracy and reliability of information collected must be shown to
be adequate for intended uses. For collections based on sampling, a special
justification must be provided for any collection that will not yield "reliable" data
that can be generalized to the universe studied.
Methods Used to Maximize Response Rates
The methodology for National Mortgage Database, including the NSMO, was initially based
upon methodology that was developed and tested during a series of three surveys funded and
15
carried out by Freddie Mac between 2010 and 2012. After achieving a response rate of just 12
percent on its initial pilot survey in 2010, Freddie Mac retained Dr. Don A. Dillman of
Washington State University, a leading expert in mail survey methods, to provide input on ways
to maximize response rate. Dr. Dillman provided recommendations in three critical areas: (1)
the execution/implementation of the survey; (2) the communications package; and (3) the
questionnaire content and format. Freddie Mac adopted Dr. Dillman’s recommendations in these
areas and, in the second and third pilot surveys, achieved response rates of 60 and 45 percent,
respectively. FHFA adopted those recommendations for the NSMO and continues to consult
with Dr. Dillman, among others, in developing the survey.
One important recommendation that FHFA adopted for the NSMO was to have four planned
mail contacts with the survey recipients. The first contact consists of the questionnaire, an
upfront monetary incentive (ten dollars in the current wave), and a cover letter. The second
contact, a reminder letter sent to all recipients, occurs in the second week of implementation.
The third contact, in the fifth week, is sent only to non-responders and includes a second
reminder letter, another copy of the questionnaire, and another incentive. The final contact, a
third reminder letter, is sent in the seventh week to non-responders only. A due date for
returning the survey questionnaire is included in the last mailing, which closes the
communication loop with all survey recipients. 13
Recent waves of the survey have shown a slow but steady decline in the response rate, a problem
facing many other surveys like NSMO. During waves 22 through 25, experiments with survey
methodology have been conducted to address the issue of declining response rate. In wave 22,
one half of the usual 6,000 borrowers were randomly selected to receive a ten-dollar cash
incentive with the first mailing instead of the five-dollar incentive the other half received (and
which had been the prevailing rate to that point). In waves 23 and 24, one half of the borrowers
received revised cover and reminder letters while the other half received the originals. (Edits
were made to the revised letters between waves 23 and 24 to boost response rates.)
In wave 25, all borrowers were sent an initial incentive of ten dollars. In terms of the second
incentive, one random half of the non-respondents are being sent the normal five-dollar cash
incentive as in previous waves and the other random half are being sent a letter informing them
that they will be sent a twenty-dollar incentive upon completion of the survey. In wave 26 (to be
sent in Quarter 2 of 2020), one random half of the borrowers will receive an initial incentive of
five dollars and the other random half will receive an initial incentive of ten dollars, but all nonrespondents will receive a letter informing them that they will be sent a twenty-dollar incentive
upon completion of the survey. Data collected during these experiments will be used to make
decisions about the NSMO methodology in 2020.
FHFA also adopted two of Dr. Dillman’s recommendations regarding the communications
package. First, all communications have a friendly tone and reflect a personal and sincere
13
Copies of the four letters (in both English and Spanish) comprising the NSMO communications package used for
the February 2020 survey mailing are included as Attachment 4.
16
request for help. All correspondence is signed by a senior official of both FHFA and CFPB and
includes contact information for authenticity. Second, each questionnaire is mailed in a plain
white envelope so as not be mistaken for “junk mail.”
To further increase response rates, survey recipients are given the option of completing the
survey online in either English or Spanish. The first mailing contains an insert, in both English
and Spanish, which informs recipients of these options and provides the web addresses to access
the appropriate electronic versions of the survey.
Data editing
The survey responses, once delivered to FHFA’s National Mortgage Database Program staff, are
subjected to thorough editing and review. The initial phase consists of standard data editing—
correcting numbers reported in the wrong units, changing answers in responses based on margin
notes and comments, assigning responses for questions with open-ended “other” responses,
dealing with multiple responses to a question that calls for only one response, and deciding how
to handle situations where respondents followed the wrong skip pattern.
One advantage that the NSMO has over other surveys is the availability of credit and
administrative data, much of which appears to be quite reliable. These data can be used to assist
in the editing and imputation process. Four primary sources of such data are available in
processing NSMO:
(1) credit data from Experian on sample loans;
(2) data collected by Experian from other data sources on the survey respondents,
including loan servicers and data companies;
(3) information from matches to administrative loan files (Fannie Mae, Freddie Mac,
Federal Housing Administration, Department of Veteran Affairs, Rural Housing
Services, and Federal Home Loan Banks); and
(4) information for loans that could be matched to HMDA data (available through
calendar year 2018 as of this writing). 14
The credit and administrative data are used to determine which borrower in the Experian data
corresponded to the respondent (and spouse/partner of the respondent) in the survey and to infer
14
Merges with most administrative files are conducted behind a firewall at Experian using borrower name, address,
date of birth and Social Security number to ensure the highest quality match accuracy (neither FHFA nor CFPB staff
ever receive such information). However, merging of the NMDB data with the HMDA data and the Federal Home
Loan Bank loan files have to rely on variables common to both datasets, including the original loan balance, the
opening date of the mortgage and the general location of the property (census tract or state/county) but not property
address or borrower name. Unfortunately, mortgage servicers report the billing address of the mortgage borrowers
to Experian, but this is not necessarily the property address, particularly for mortgages on non-owner-occupied
properties. Those, when converted to a census tract for matching the address, may be incorrect. Thus, HMDA
merges are less accurate than those employing directly identifying information such as name and Social Security
number because the latter are less reliant on address.
17
the loan the respondent had in mind when answering the survey. These data are also useful in
determining if respondents correctly identified their loan as a home purchase loan or a refinance.
Imputation
After editing and cleaning the survey response data, missing responses are imputed using
answers to related questions or statistical models estimated based on credit and administrative
data and answers to other questions in the survey. Imputations are designed to replicate the level
of inherent inconsistencies between related variables in the actual (non-imputed) responses by
the respondents. Actual responses are generally not changed (except in cases where they are
edited as described above). In order to preserve the original responses, the raw responses are
retained with missing responses coded as such. A parallel set of variables (“X” variables) are
constructed where all missing responses are imputed, and necessary responses are edited as
described above. Each instance in which an X variable differs from original responses is
recorded by a shadow variable (“J” variables) that indicates the method and reason why the
change was made. Missing responses typically total about 3 to 5 percent of responses for most
questions and only in a few instances were more than 10 percent. The X variables are not
created when a directly comparable credit or administrative variable is available for all
respondents (e.g., loan amount, loan payment, number of co-signers) as comparable credit or
administrative variables can be used in lieu of survey responses in analysis. Instead, Z variables
are created in their place to indicate whether the respondents answered the question.
The initial set of imputations are based on inferences drawn from patterns of response. Patterns
of missing responses sometimes provides an indication of how the respondent would have
answered if they had taken the time to fill out all answers of a group. For example, one question
reads, “how important were each of the following…” and provides choices of important or not
important. Some respondents only mark “important” for the choices important to them. Other
respondents might only mark choices that are not important. When all answers are in a group
with only one side answered, the other answers are imputed as the opposite choice. For example,
when a respondent only marks choices that are important, the missing questions were imputed as
not important.
The survey skips do not always work for every respondent and some respondents miss the leadin question. The answers to the lead-in question are often imputed based on actual answers to
the follow-up questions. For example, one question reads, “how many different
lenders/mortgage brokers did you end up applying to” and provides options for one to five.
When a respondent chooses one, they skip the next question about reasons they applied to more
than one. If the lead-in question was left blank, any yes answer to the follow-up is considered a
reason to impute that they applied to more than one lender. All “no” answers to the follow-up
questions means that they probably only applied to one lender. When the respondent skips both
a lead-in and follow-up question, both are imputed with one of the imputation models.
Once these inferential imputations are taken care of, statistical models are used to impute the
remaining missing answers. The most common type of question in NSMO provides a simple yes
or no answer. A binomial logistic model provides an estimated probability of a yes answer. For
18
some questions, such as the number of lenders or brokers the respondent seriously considered,
the answers are in a logical order. For these types of questions, an ordered logistic model is used
to determine the probability of each answer. For other questions the order does not matter, and
the answer choices are not related to the previous choice. For these questions, a multinomial
logistic model is used, and the reference group is selected to be the most common answer.
Again, the model produces a probability of each answer response. A random number is drawn
with a different seed for every question and it is then compared to the probability of each
response level. When the random number falls below the cumulative probability of an answer,
that answer is used as the imputed response. This method injects some randomness to the
imputed answers, but the goal is to provide a distribution of imputed answers that mimics the
distribution of the answers where no imputation was necessary.
The dependent variable (𝑦𝑦𝑖𝑖 ) in all the models used is a value for the missing answer. The vector
of characteristics (𝑥𝑥𝑖𝑖 ) can include information from the credit files or answers to survey
questions. Key demographic variables (age, gender, education, ethnicity, and income) are
imputed first. For these variables, high quality administrative data are generally available and
can be used directly to impute a value for the X variable. For example, lender-reported
information provides high quality data on age. Administrative data also provide reliable
information on race, income, and interest rate. HMDA data also provide reliable information on
race, income, and gender.
The initial statistical imputation models first use all the respondents who provided answers using
a standard set of predictors to provide an initial imputation. The models use age, loan amount,
credit score, loan type, education and income level. Once the initial imputation values are
established, the models are enhanced for any predictor that provides a good fit to the models and
these models use actual and imputed values from all respondents. The missing values are
imputed statistically using an iterative process where each subsequent run of the model uses the
actual responses and the imputed responses from the previous run. Iterating in this way ensures
that correlations among the imputed values will better reflect correlations among observations
where responses were available.
The regression runs always start with key variables first. As with the initial imputations, the first
variables imputed are age, loan amount, credit score, loan type, education and income. The next
level covered by the models imputes marital status, race, and ethnicity. The process then moves
on to other questions and often follows the order of the survey instrument for less consequential
questions. Lead-in questions are always imputed before the follow-up question to keep the
follow-up imputations consistent with the lead-in question.
As the recursive models run, the coefficient of each predictor variable in each model is tracked
and compared with values from the previous runs. The recursive runs are only stopped when the
coefficients have settled down with minimal changes in the last few runs. This ensures that the
recursive effect on each model has fed into all the predictions of imputed values and stabilized.
To find the best model for each imputation, the last recursive run is selected, and the actual
response is subtracted from the predicted value of the response. The difference represents the
19
error term or the portion of the probability of a response that was not explained by the predictive
variables. A large matrix of error terms is constructed, and the values are tested for correlation.
Error terms with correlation coefficient of over 0.30 get explored as possible indicators of new
predictor variables. Each year, new predictors are placed into the recursive model and the results
are tested to see if the model improves. With improved models, the recursive runs are restarted
until all the beta coefficients settle down again.
The final imputations rely on a further set of quality control checks. Conditional correlation
tables of model residuals are constructed to identify any additional significant explanatory
variables which may have been left out of individual equations. Further, imputed values of
similar related variables are sometimes adjusted to ensure that the covariances among the
imputed answers mirror that of the non-imputed responses.
Non-Response Weighting
There are several ways calculations based on the NSMO raw survey responses may not be
representative of the population as a whole. First, as shown earlier in Table 2, the survey waves
do not have the same sampling rates. Second, only about one-third of the sampled borrowers
completed the survey. Commonly, in survey sampling, some individuals chosen for the sample
are unwilling or unable to participate in the survey. Non-response bias is the bias that results
when respondents differ systematically from non-respondents. A common method for mitigating
possible non-response bias is to use weights to align the characteristics of respondents and the
population more closely. This is known as “non-response weighting.” Such weights are
generally calculated from statistical models. Specifically, the non-response weights in NSMO
are designed to “blow up” the usable sample to the total surveys mailed less duplicate and
ineligible loans taken out of NMDB.
Often, little is known about survey non-responders, so the statistical models used to construct
non-response weights are quite simplistic. Compared with many other surveys, however, NSMO
has extensive credit and administrative data on both responding and non-responding borrowers
that can be used to estimate non-response weights.
Sample non-response weights are estimated with logistic models separately for each sample
wave and within a wave for loans with a single borrower versus those with multiple borrowers.
The models estimate the probability of getting a usable response for each wave of the survey.
The predictive equations have had pseudo-R-square values ranging from 0.0467 to 0.1654. The
models for joint borrowers do better than those for single borrower. The largest pseudo-Rsquare values were for models estimated on data from wave 20 joint borrowers. Key predictive
variables included are loan amount, borrower age, the income relied upon for underwriting, the
combined loan-to-value ratio, an indicator of whether it was a home purchase or refinance loan,
and the interest rate spread over the prevailing prime interest rate at origination. The models also
control for credit score, for geography using Census Divisions, and for demographic
characteristics on family composition, race, ethnicity, gender, and educational attainment.
20
The model’s predicted probabilities of response were placed into 5 equal groups of 20 percent
each. The average of the response rates from each of these five groups was used to calculate a
response weight as the inverse of these five average rates. Once within-wave sample nonresponse weights are estimated, they are multiplied by the wave sample weight to provide an
overall weight.
Table 4 demonstrates the effect of differential sampling weights for the first 21 waves. Column
one shows the distribution among various demographic and loan categories of the raw survey
responses. Column two provides the distribution using estimated overall weights. Finally,
column three shows the average overall weight for each category.
Table 4. Overall Weights, 2013 - 2017 Originations (Waves 1-21)
Unweighted
Percentage
Weighted
Percentage
Average
Weight
Purchase
48.4%
49.6%
1,169
Refinance
51.6%
50.4%
1,112
100.0%
100.0%
Less than $50,000
2.6%
2.6%
1,106
$50,000 to $99,999
14.2%
14.2%
1,139
$100,000 to $149,999
20.2%
20.4%
1,147
$150,000 to $199,999
17.6%
17.7%
1,142
$200,000 to $249,999
13.1%
12.9%
1,123
$250,000 to $299,999
9.8%
9.5%
1,112
$300,000 to $349,999
6.6%
6.6%
1,128
$350,000 to $399,999
4.9%
4.8%
1,115
$400,000 or more
10.9%
11.4%
1,197
100.0%
100.0%
Less than 15 Years
4.2%
3.7%
1,004
15 Years
16.6%
15.3%
1,050
Between 15 and 30 Years
6.4%
6.4%
1,043
30 Years or More
72.8%
74.6%
1,168
100.0%
100.0%
Less than 75%
37.9%
35.1%
1,055
75% to 79%
11.4%
11.1%
1,109
Loan Category
Loan Size
Mortgage Term to Maturity
Loan to Value (LTV) Ratio
21
80%
9.7%
9.3%
1,095
81% to 89%
10.0%
10.1%
1,155
90% or More
31.0%
34.4%
1,264
100.0%
100.0%
Lower than 620
4.1%
5.8%
1,592
620 to 639
3.3%
4.3%
1,494
640 to 659
5.0%
6.2%
1,431
660 to 679
5.7%
6.6%
1,323
680 to 699
6.4%
7.2%
1,278
700 to 719
7.5%
8.2%
1,246
720 to 739
9.4%
10.0%
1,209
740 or Higher
58.6%
51.8%
1,006
100.0%
100.0%
Respondent Credit Score
Question 4. Describe any tests of procedures or methods to be undertaken. Testing
is encouraged as an effective means of refining collections of information to
minimize burden and improve utility. Tests must be approved if they call for
answers to identical questions from 10 or more respondents. A proposed test or set
of tests may be submitted for approval separately or in combination with the main
collection of information.
FHFA will use information collected from the cognitive testing participants to assist the Agency
in drafting and modifying the survey questions and instructions, as well as the related
communications, to read in the way that will be most readily understood by the survey
respondents and that will be most likely to elicit usable responses. Such information will also be
used to help the Agency decide how best to organize and format the survey questionnaire. A
copy of the most recent version of FHFA’s NSMO cognitive testing guidance document (or
“Talk Track”), which was provided to Westat on January 29, 2020, is included as Attachment 5.
Question 5. Provide the name and telephone number of individuals consulted on
statistical aspects of the design and the name of the agency unit, contractor(s),
grantee(s), or other person(s) who will actually collect and/or analyze the information
for the agency.
The names of and contact information for individual stakeholders from FHFA, CFPB, and
Experian, including those who were consulted on statistical aspects of the design and who will
22
analyze the data, appear in the list included as Attachment 6. FHFA also consulted with the
following:
Dr. Mick P. Couper
Survey Research Center and the Institute for Social Research
University of Michigan
426 Thompson Street
Ann Arbor, MI 48104
(734) 647-3577
Dr. Don A. Dillman
Department of Sociology and the Social & Economic Sciences Research Center
Washington State University
Pullman, WA 99164-4014
(509) 335-1511
The subcontractor hired by Experian to carry out the survey and the cognitive testing is:
Westat
1600 Research Blvd,
Rockville, MD 20850
List of Attachments:
1. NSMO questionnaire for 2020 Q1 (mailed February 3, 2020)
2. 60-day PRA Notice published at 84 FR 67447 (Dec. 10, 2019)
3. National Mortgage Database System of Record Act Notices:
a. SORN published at 80 FR 52275 (Aug. 28, 2015)
b. Revision to SORN published at 81 FR 95595 (Dec. 28, 2016)
4. NSMO communication package (in English and Spanish) for 2020 Q1
5. Cognitive testing “Talk Track” dated January 29, 2020
6. List of National Mortgage Database stakeholders
23
| File Type | application/pdf |
| Author | raudenbushe |
| File Modified | 2020-03-30 |
| File Created | 2020-03-30 |