FR3066_20220729_omb_B

Supporting Statement Part B for the
Federal Reserve Payments Study
(FR 3066; OMB No. 7100-0351)
Summary
For all information collections that involve surveys or require a statistical methodology,
the Board of Governors of the Federal Reserve System (Board) is required to provide a complete
justification and explanation of the use of such a methodology. For collections that employ
surveys without such a methodology, the Board should be prepared to justify its decision not to
use statistical methods in any case where such methods might reduce burden or improve
accuracy of results.
Background
The FR 3066a and FR 3066b are part of the latest iteration of the Federal Reserve
Payments Study (FRPS), which has been a collaborative effort of the Federal Reserve Bank of
Atlanta (FRB Atlanta) and the Board since 2000. The FRPS originated from a Federal Reserve
System-wide effort to improve the measurement and public availability of information on
volumes and trends in checks and other noncash payments. The FRPS filled a significant gap in
quantitative information on U.S. noncash payments by providing a reliable and transparent nonmandatory approach to surveying payment institutions, constructing U.S. domestic total
estimates from the survey data, and publishing them. The focus of the surveys has adapted to the
substantial evolution and growth in emerging and innovative payment types over time, as well as
the refreshed strategic direction of Federal Reserve Financial Services. The strategic direction
includes a focus on meeting the evolving needs of payment system users for end -to-end payment
speed, efficiency, and security, while remaining true to a longstanding financial services mission
to foster the integrity, efficiency, and accessibility of the U.S. payment system. The Retail
Payments Risk Forum (RPRF) at FRB Atlanta and the Payment System Studies sectio n at the
Board jointly conduct the study using experienced contractors that engage respondents and
collect and deliver survey data according to the survey design requirements set by the Federal
Reserve.
Surveys in previous years received robust industry support and participation, and the
aggregate estimates produced from the survey data are widely cited in academic working papers,
journal articles, and industry publications, reported in the media, and used by the public,
industry, and policy makers as a quantitative aggregate benchmark of noncash payments and
cash withdrawal and deposit activity in the United States. As the noncash payments system
grows larger and more complex, the Board expects the data collected under the FRPS to play a
crucial role in objectively maintaining and updating quantitative information on the U.S. noncash
payments system. The information collected through the FRPS is not available from other
sources.

Universe and Respondent Selection
FR 3066a
The FR 3066a collects the number and value of noncash payments, cash withdrawals and
deposits, third-party payments fraud, and related information from a nationally representative
sample of commercial banks, savings institutions, and credit unions. Administrative data on the
types and sizes of the population of insured depository institutions is available in reports filed
with the Federal Reserve. After consolidating affiliates, the 2021 population consisted of 9,580
independently operated institutions at the highest holding company level with non-zero
transaction deposits.
The sample size of 3,800 institutions is the same as the sample size in the previous
triennial (2019) version of the survey. Sample stratification and selection methods follow
classical and innovative techniques based on the state of the art of the literature on business
survey methods. As in 2019, the 2022 triennial version of the survey is administered using a
complex planned-missing-data design with 11 questionnaire versions allowing shorter
questionnaires for smaller institutions. 1 The allocation of institutions to size strata has been
updated for 2022 due to lessons learned from analysis of the 2019 survey outcomes. To account
for the increased concentration of the financial industry and to improve the expected precision of
total estimates, the size of the certainty group of the largest institutions is 1,665 for 2022
compared with 1,750 for 2019.2 The remaining 2,135 institutions were selected at random with
probabilities declining with size. The response rate for 2019 was 36 percent, and is expected to
be similar in 2022.
FR 3066b
The FR 3066b is designed as a census. The Federal Reserve would work with a
contractor to identify the final list of networks, processors, and issuers from which to collect
data. Estimation of national aggregate payment volumes from the survey is based on developing
a complete population frame of all relevant organizations (approximately 230) and requesting
data from each. There are 17 different surveys, and respondents only provide information in the
survey forms applicable to their organizations. For the 16 non-transit surveys, the survey
response population is not large enough to employ formal statistical methods to useful effect. In
cases where a response is not returned, the missing items would need to be imputed using
publicly available information and analysis of data from similar organizations that did provide
data. In such cases, expertise and heuristic methods are employed to estimate the missing data.
Totals are constructed by summing the reported and estimated data. The 2019 triennial survey
1

This method has been used since 2016. Analysis of the outcome of the 2016 planned missing data survey design
compared with the 2013 full survey design is discussed in Geoffrey Gerdes and Xuemei (May) Liu, “Improving
Response Quality with Planned Missing Data: An Application to a Survey of Bank s” in The Econometrics of
Complex Survey Data: Theory and Applications, Advances in Econometrics, Volume 39, 2019, pp 237-58.
2
This substantial increase in the certainty sample size is primarily the result of the lifting of an arbitrary restriction
on the number of institutions sampled with certainty that was imposed on the optimization routines used to allocate
the sample in 2016 and previous years. The change is expected to improve the precision of total estimates for a
given sample size. The change is likely to reduce the amount of information received from smaller institutions, but
the reduction should have a relatively minor affect on study goals.

2

had a response rate of 79 percent for non-transit surveys. The less-extensive annual supplements
had similar response rates. Similar response rates are expected in the current surveys.
The survey of transit operators is will be confined to a census of the 30 largest
organizations in the U.S. This differs from 2019 and 2016 in which a random sample of the
population of all transit organizations was collected. The response rate for the transit
organization survey is expected to be similar to that of other surveys.
Procedures for Collecting Information
FR 3066a
Using size measures obtained from regulatory reports, the population of depository
institutions is stratified into sub-populations by type and size, and separate samples are drawn
from each, with the sampling rate declining with size. To draw the sample, we use classical
methods for determining sub-population size boundaries and total sample allocations within
types, based on a general goal of minimizing the standard error of the aggregate estimates. The
use of these allocation methods leads to the treatment of sub-populations with the largest
institutions as a census, i.e. each member is sampled with certainty.
The size distribution of U.S. depository institutions is highly skewed, although less than
in many other industries, and far less than is typically the case in other developed countries,
many of which have fewer than a dozen significant deposit taking institutions. Aggregate
estimates are constructed out of the sub-population estimates using a ratio estimator technique,
which, taking advantage of the high covariance between the size measures available from the
population data and the volumes being collected, is substantially more efficient than alternative
estimators that ignore this covariance. The approach has been designed to achieve high precision
across all variables using the size covariate as a proxy. Past surveys have been able to achieve
estimated confidence interval that ranges as low as +/- 3 percent for some variables at the 95
percent level. This kind of precision cannot be achieved across the board, however. Nonetheless,
given the unique data collected the estimates should be considered the best available national
estimates for many items.
Annual supplements are conducted that collect data from the largest 120 institutions.
These typically achieve approximately a 50 percent response rate. Such data are used to
construct rates of change amongst the responding institutions from year-to-year.
FR 3066b
The uniqueness of each participant does not lend these surveys to use of formal statistica l
techniques.

3

Methods to Maximize Response
FR 3066a
A large-scale effort is made to recruit the participation of sampled institutions, the survey
is designed to use language and organizing principles familiar to the institutions, and review and
feedback sessions are designed to ensure the surveyed information addresses payment issues of
interest and relevance to participating institutions. An incentive of a peer report is provided.
Elevated efforts are made to recruit, assist, and accommodate the needs of these
institutions. Effectively, this elevated effort for a select set of institutions is made due to a) the
constraint on increasing the sampling probability of a census and b) the fact that, all else equal, it
is always preferred to expend resources at the margin on obtaining a response from the largest
non-responding institution. In addition, the length of surveys declines with institution size,
reducing the burden on smaller institutions.
The ratio estimator technique, which computes the within sample ratio of each item of
interest with the institution size variable and then “blows up” that ratio to the population size of
each stratum individually implicitly accounts for unit-level non-response as part of the estimation
technique. In addition, an EM-algorithm-based imputation method is used to account for missing
item-level data (which included planned missing as well as unplanned) using correlations
between reported items from peer respondents and logical relationships is used to enforce
adding-up constraints throughout the survey. Past studies of the data have revealed no evidence
of self-selection.
FR 3066b
Information from past responses and public data are used to estimate and validate the
missing items of nonparticipants. Estimation is based on expert judgement in most cases as
formal statistical methods are not robust enough for extremely small samples with highly
heterogeneous subjects.
Testing of Procedures
Each survey builds on lessons learned from previous surveys, and changes from year-toyear are examined for plausibility. In addition, aggregate estimates that come from FR 3066a and
FR 3066b that should match, such as total debit card transactions reported by depository
institutions and card networks, are compared for consistency. Anomalies are investigated,
described, and accounted for before finalizing estimates and are explained in reports.
FR 3066a
Estimation methods have been stable for two decades and improved incrementally when
the opportunity arises. For the national aggregate estimates conducted on a triennial basis, the
joint estimates based on imputed data are compared with independent estimates using only the
reported data. Aggregates are built up from the stratum-level estimates, and any unusual patterns

4

in the data or implausibly high standard errors of estimates are examined for invalid or outlying
response data and adjusted accordingly.
FR 3066b
Information from past responses and public data are used to estimate and validate the
missing items of nonparticipants. Testing is based on expert judgement in most cases as formal
statistical methods are not robust enough for extremely small samples with highly heterogeneous
subjects.

5