STATISTICAL INFORMATION FROM ADMINISTRATIVE RECORDS

STATISTICAL INFORMATION FROM ADMINISTRATIVE RECORDS

IN THE FEDERAL TAX SYSTEM

Tom Petska, Internal Revenue Service

P.O. Box 2608, Washington DC 20013-2608

Key words: System of taxation; administrative

records; statistical operations; microsimulation

modeling.

The tax return information system is designed

to facilitate administration of the Federal tax laws.

Under this system, taxpayers report their financial

activities, calculate their tax liabilities, and

forward a remittance or apply for a refund from

the IRS. Although this means of compiling

information differs from that of survey-based

statistical organizations, the Statistics of Income

(SOI) function has a similar mission -- to collect

and process data so that they become meaningful

information and to disseminate this information to

its customers and users [1].

Tax returns are filed and administratively

processed at IRS regional sites, and selected

information from all returns is compiled in a

computerized master file system. Statistics

comprising the SOI studies are generally based on

stratified probability samples of tax or information

returns. As returns are processed into the Master

File System, they are sampled based on such

criteria as gross income, assets, profits, industry,

or presence of supplemental forms and schedules.

The sampled returns are then earmarked for

additional data abstraction and editing.

While SOI projects are loosely referred to as

"administrative records studies," they are clearly

more than that -- they are, in fact, hybrids of

information from administrative and statistical

processing. This paper reports on some

experiences of using administrative data for

analytical and statistical purposes.

In the first section of the paper, a history and

operational overview of the SOI function are

outlined.

The second section highlights

differences between IRS Master File compliance

data and the statistical information in SOI studies.

The third section examines several case studies in

which true administrative data, without benefit of

supplemental statistical processing, were the

primary source of information for research

purposes. In the last section, some conclusions

about the usefulness of administrative data are

presented.

Overview of IRS Tax Statistics Operations

This section briefly describes the history,

customers, and operations of the SOI statistical

system.

Origins of the SOI Function -- The modern U.S.

income tax was enacted in 1913 with the passage

of the sixteenth amendment to the U.S.

Constitution. Subsequently, the Revenue Act of

1916 required the annual publication of statistics,

establishing a role for the SOI function. Despite

many revisions to the tax law, the original

requirement of that Act continues today.

Specifically, the current Internal Revenue Code

states that "The Secretary (of the Treasury) shall

prepare and publish not less than annually

statistics reasonably available with respect to the

operations of the internal revenue laws."

The SOI Division conducts statistical studies

on the operations of the tax laws with respect to

individuals, corporations, partnerships, sole

proprietorships, estates, nonprofit organizations,

and trusts, as well as specialized studies covering

both inbound and outbound international activities.

The annual budget of the SOI program is

currently about $40 million. While this amount

represents a very small portion of the total IRS

resources, SOI is a leading Federal statistical

organization. SOI's budget covers designing the

various projects; overseeing and participating in

the statistical processing necessary to accomplish

them; and tabulating, documenting, publishing, or

otherwise disseminating results.

Customers of SOI Data -- The mandate for the

SOI program is a responsibility required of the

Secretary of the Treasury, and SOI's primary

customer is the Treasury Department's Office of

Tax Analysis (OTA). Another primary customer is

OTA's legislative counterpart, the Congressional

Joint Committee on Taxation (JCT). OTA and JCT

use the microdata files produced by SOI as their

primary source of information for analysis. In both

agencies, microsimulation modeling is employed

using SOI data as the primary database for tax

policy analysis and revenue projections. The SOI

data are also sometimes matched with other data to

build comprehensive data bases that can be used in

estimating the overall impact of tax law changes

and their effects on tax collections.

Although the bulk of SOI's resources are

focused on the statistical needs of OTA and JCT,

SOI has many more customers. The Department of

Commerce's Bureau of Economic Analysis (BEA)

is a significant user of SOI data for estimating

components in the National Income and Product

Accounts related to individuals, corporations,

partnerships, and sole proprietorships. The Census

Bureau, also in the Department of Commerce, is

another significant data user; however, its needs

(unlike those of OTA, JCT, and BEA) are

primarily met using IRS Master File "population"

data, rather than data from the lower-volume, but

content-rich, SOI samples. Other parts of the IRS

are customers for SOI data and use it in a wide

variety of analyses. Finally, the general public

(including academics, "think tanks," accounting

firms, and interested citizens) is another important

customer of SOI.

SOI Products and Services -- Statistics of

Income information is made publicly available

through both printed publications and electronic

media. The Statistics of Income (SOI) Bulletin is

published quarterly, with each issue containing

four to eight articles and data releases of recently

completed studies, as well as historical tables

covering a variety of subject matter, from Treasury

Department tax collections to taxpayer assistance

and tax return projections [2]. SOI also produces

separate annual "complete reports" on individual

and corporation income tax returns, which contain

more comprehensive data than those published in

the Bulletin [3,4]. The Corporation Source Book

is also published annually, presenting detailed

income statement, balance sheet, and tax data by

industry and asset size [5].

Periodically, special compendiums of research

and analysis, covering topics such as nonprofit

organizations, estate taxation and personal wealth,

international business activities, and partnerships,

are produced. Research articles documenting

technological and methodological changes in SOI

programs and other related statistical uses of

administrative records are also published in a

series of reports [6].

The IRS World Wide Web site provides users

an easy option for accessing SOI data [7]. At

present, 65,000 files are downloaded monthly

from the Tax_Stats portion of this site. While SOI

manages Tax_Stats in order to make tax-related

data on individuals, corporations, and other

entities available to the public, we are also the

conduit for releasing other IRS information,

including the Internal Revenue Service Data Book

(containing tax collections and other tax

administration data), tax return projections, and

microdata records of exempt organizations [8].

Statistical Operations -- This section highlights

statistical operations and procedures utilized in the

development and implementation of many of the

SOI statistical studies.

§

Sample Design and Selection -- U.S. tax

returns are filed and administratively

processed at one of ten IRS regional sites,

called "submission processing centers." Once

processed, IRS compiles selected information

from most return forms into a computerized

Master File System, which is the informational

backbone of the agency. Most SOI operations

begin by sampling returns from the Master

File System; the Master File offers a sampling

frame that enables use of sophisticated and

efficient sample designs.

Statistics compiled for the SOI studies are

generally based on stratified probability

samples of tax or information returns. As

returns are processed into the Master File

System, they are assigned to sampling classes

(strata), based on criteria such as size of

income or assets (or other measures of

economic importance), industrial activity,

accounting period, or the presence of certain

supplemental forms or schedules.

Each taxpayer, whether an individual or a

business, has a unique number, the Social

Security Number (SSN) for individuals or the

Employer Identification Number (EIN) for

businesses.

These

unique

Taxpayer

Identification Numbers (TIN's) are used as the

seed for a pseudo-random number (using an

algorithm, this is a transform of the TIN),

which, along with the sampling strata,

determines whether a given return is to be

selected for the SOI sample [9]. The

probability of a return being designated for the

SOI sample depends on the sampling rate

prescribed for its sample class or stratum and

may range from a fraction of 1 percent to 100

percent.

§

Data Capture Techniques -- After sampling,

the relatively few data items pulled

electronically from the Master File System are

substantially augmented with additional items

key-entered from hardcopies of taxpayers'

returns. Statistical abstracting can take as little

as a few minutes for a simple return, to as long

as several days for a large corporate return.

SOI has built a network of midrange enterprise

servers in selected IRS submission processing

centers that are dedicated to SOI statistical

processing. The processing system uses online

transaction processing, so that all data capture

operations are completed in a single pass. One

editor is responsible for ensuring the validity

of all data processing for a given return.

Due to substantial penalties for misreporting,

the income and expenditure data reported on

tax returns have proven to be more reliable

than comparable survey data. Even so, SOI

employees go to great lengths to protect

against nonsampling errors, such as those due

to

taxpayer

reporting

variations

or

inconsistencies, or data processing errors. In

order that final statistics are consistent and

reliable, SOI economists develop extensive

online tests, and error resolution procedures

that are applied to each sampled return. The

tests and correction procedures are based on

the structure of the tax laws and forms,

generally accepted accounting principles, and

the

improbability

of

various

data

combinations.

Editors in submission processing centers and

SOI economists "statistically edit" data items

in order to make each sampled return

internally consistent. Missing data problems

arise, albeit infrequently (under 1 percent of

the time). Missing items can be obtained

through direct contact with taxpayers, or be

estimated through imputations based on other

return data, prior-year data for the same

taxpayer, or same-year data from a

"statistically similar" return.

§

Weighting and Estimation -- As noted

above, the probability with which a return is

selected for inclusion in an SOI sample

depends on the sampling rate prescribed for

the stratum in which it is classified. Weights

are computed by dividing the population count

of returns filed for a given stratum by the

count of sample returns for that same stratum.

"Weights" are used to adjust for the various

sampling rates used--the lower the rate, the

larger the weight.

The data on each return in a stratum are

multiplied by the weight assigned for the

given stratum. To produce the tabulated

estimates, as shown in the SOI Bulletin and

other publications, weighted data are summed

to produce statistical totals.

Of over 200 million tax returns processed each

year for administrative purposes, only about

half a million are sampled for the various SOI

programs. However, since sampling rates

generally increase with increases in the size of

financial amounts (such as income or assets),

the returns in the samples are, on average,

disproportionately larger and more complex

than those in the master files, which include

the population of returns. Thus, in comparison

to IRS administrative processing, which

captures 100 percent of the tax returns but

with limited item content, SOI programs

collectively represent a smaller volume, but

with a proportionately higher fraction of

complex returns and with greater item content.

Administrative Data for Statistical Purposes

Most of what is familiar to users of SOI data

are the data products from traditional SOI sample

studies. However, there is another side to SOI, that

of a user and developer of data from the IRS

Master File System.

As noted above, SOI has, for many years,

selected samples from the Master File System, but

use of these data as program content is less

frequent. In this section, important examples of the

use of administrative tax data from the U.S. and

other systems are described, and commentary is

offered on its current value and potential for

additional analytical purposes. In IRS terms, such

data files are referred to as the Individual or

Business Returns Transactions Files (i.e., IRTF

and BRTF), key inputs to the Master File System.

SOI’s traditional data editing processes are

labor-intensive. Few SOI studies have been

accomplished without substantial manual data

editing and abstraction. However, there is a

reluctance to change this, since the final weighted

estimates and overall quality of data are so heavily

dependent on the accuracy of the data editing

process.

In general, the need for manual abstraction

and editing is dependent upon two issues:

•

•

Is the statistical item content (e.g., the

financial data) included in the population files

adequate for analytical and revenue estimation

purposes?

Does the level of data complexity adversely

impact an acceptable level of data quality for

the return type or population being studied?

Concerning item content, the issue is fairly

straightforward. If program-critical data were

available on the return form or attached schedules

or worksheets, but not abstracted into the Master

File System, the only alternative would be to

obtain copies of these schedules and abstract these

data.

On the issue of complexity, the situation is

less clear. In the U.S., many relatively lowincome individual income returns are quite simple,

with only a limited number of income types and

other taxpayer-reported characteristics. For such

cases, an automated or high-level system of

reviewing outliers and imposing error corrections

to the sampled file cases would in all likelihood

yield reasonable results.

But as complexity increases, this may not be

an acceptable means of data editing. For example,

large U.S. corporation returns, often with

multinational operations, are extremely complex,

and SOI staff spend hundreds of hours reviewing

and correcting these data, even after the initial

abstraction and editing have taken place. These

often come about as “referrals” to professional

staff for closer scrutiny. In extreme cases, where

the return complexity is so high or the data

reported are incomplete, SOI staff correspond with

the taxpayer, which is time-consuming and often a

source of delay in project completion. But our

view has been that, in such cases, it is essential.

Many statistical agencies in the U.S. and

elsewhere have begun to develop automated data

edit systems, building an “artificial intelligence”

knowledge base for data editing. Most applications

have started with fairly simple item content where

the editing relationships could be specified to

handle the majority of cases. However, before

such routines can become the norm for studies

with more complex data content, such as most

business tax returns, extensive additional study of

data relationships and taxpayer reporting

tendencies is needed.

Administrative Data in Statistical Studies

This section provides brief descriptions of

experiences of using true “administrative data” for

statistical or analytical purposes.

Administrative Data in SOI Programs -- For

many years, SOI has, in addition to sampling,

brought in program content data from the

Individual and Business Returns Transactions

Files. Currently, such data are essential parts of

record content for SOI sampled returns in the

Individual,

Corporation,

and

Partnership

programs.

This process was initiated many years ago from

the perspective that it would save resources from

re-keying these data, but it has not been without its

problems. Data editing for the Master File System

not only differs from SOI statistical processing,

but SOI has had very little control in these

processes.

Still, with many years of experience in using

these data as part of SOI program content, subjectmatter analysts have devised ways of making these

data useful and of high quality so that the

processing efficiencies envisioned years ago have

generally been realized.

Sub-national Economic Estimates -- The SOI

samples are not reliable below the national level,

which is a conscious decision based on priorities

and budgets, since sample sizes would have to be

substantially inflated.

Thus, sub-national

estimates require data from the master file system

since they account for the entire population of

returns.

Presently, State-level estimates are

produced for Individuals, Proprietorships, and

Partnerships, and ZIP Code data are also produced

for Individuals. In general, these programs have

been largely successful.

The data on individual income and taxes by

State, which are derived from the IRTF data, are

published annually in the Spring SOI Bulletin and

also made available on the SOI Internet site.

Migration Data -- In a joint venture with the

Census Bureau, IRS Individual Master File data

have been appended with geographic codes for use

in Census State and county statistics programs.

These data have been tabulated by State and

county, and net migration has been calculated.

These data are made available to the public from

SOI’s Statistical Information Services (SIS) office.

SOI Fiduciary Study -- Recently, a study of

Fiduciaries was completed solely from the

population data on the filing of Forms 1041,

Fiduciary Returns, in the IRS Business Returns

Transactions File.

This study included the

population of returns and developed edit rules to

ensure reasonableness and consistency of data at

the "micro" level. This study has since been

published in the SOI Bulletin [10].

Two other case studies in which IRS Master

File System data were the primary data source

include the following:

IRS Compliance Data Warehouse -- Over the

past six years, the IRS has been developing a

centralized compliance data warehouse (CDW)

from the Master File System for use in tax

compliance studies. The CDW is a large relational

database consisting of population files from the

IRTF and BRTF. The system has been a mixed

success, but current plans are under way to

improve it by adding more data sets and increasing

availability.

Census Bureau Demographic and Economic

Programs -- By law, the Census Bureau has been

provided with annual extracts of individual

population data from the Master File Systems.

Although item content and uses are restricted,

these data have been used effectively in periodic

censuses and other statistical programs.

Finally, two other case studies, both in the

early developmental stages, are mentioned because

of their far-reaching goals of attempting to build a

comprehensive statistical database exclusively

from Master File System data.

Statistical Data Bank -- From the perspective that

"To get what you want, you may have to do it

yourself," SOI has taken the initial steps to acquire

both the data sets and operational systems to build

a data repository exclusively for analytical and

statistical purposes. Like the CDW, the primary

inputs would include data from the IRS Master

File System. However, since it is being designed

exclusively for statistical purposes, improved user

accessibility and compatibility with statistical

software, including the longitudinal data required

for panel studies, are essential ingredients.

South African Tax Statistics – Current and

former SOI staff have had an advisory role with

officials in the South African Department of

Finance and the South African Revenue Service to

develop the capability to do microsimulation

modeling of their individual and business tax

systems. Since resources are extremely limited,

work is under way to ascertain if the population

files of all individual and business tax returns

could be developed as the exclusive source for

statistical analysis and forecasting [11].

Summary and Conclusions

In conclusion, one might ask what do these

administrative data studies have in common and

what are the factors attributing to their success?

Here are some such factors common to many of

these studies:

§

Data were needed without adequate resources

for transcription and editing.

§

Data were needed with a very short time

horizon.

§

Data were needed for the entire population (or

something close) because sub-national

estimates were a primary focus.

§

Longitudinal analysis of individual case

behavior was a primary focus, and sample data

often did not have a high degree of year to

year overlap of the same cases.

§

The administrative data were perceived as a

rich and largely untapped resource that could

yield substantial benefits if properly developed

and used.

Despite SOI’s long and successful history of

sampling and editing data from the tax compliance

“pipeline,” there have thus been many instances,

both within the IRS and elsewhere, where

administrative data have been successfully used

for statistical purposes. This paper has cited a few

such examples with the hope that lessons can be

learned from these successes so that more and

better uses can be developed in the future.

Acknowledgments

Portions of this paper have benefited from

prior overviews of the SOI function written by

Fritz Scheuren and Tom Petska for Business

Economics (1992); the Proceedings of the

National Tax Association (1992); and the Journal

of Official Statistics (1993). Special thanks to

James Dalton, Beth Kilss, Mark Mazur, and Dan

Trevors for their many helpful comments on an

earlier draft of this manuscript, and again to Dan

for prepublication processing and electronic

submission to ASA. Any errors that remain are

the responsibility of the author.

Notes and References

[1] Principles and Practices for a Federal

Statistical Agency, Martin, Margaret E. and

Straff, Miron L. (editors), Committee on

National Statistics, National Academy Press,

1992.

[2] Statistics of Income (SOI) Bulletin, Publication

1136, Internal Revenue Service.

[3] Statistics of Income--1998, Individual Income

Tax Returns, Publication 1304, Internal

Revenue Service, 2001.

[4] Statistics of Income--1998, Corporation

Income Tax Returns, Publication 16, Internal

Revenue Service, 2001.

[5] Source Book of Statistics of Income--1998,

Corporation Income Tax Returns, Publication

1053, Internal Revenue Service, 2001.

[6] Statistics of Income: Turning Administrative

Systems Into Information Systems--1999,

Publication 1299, Internal Revenue Service,

2000.

[7] The SOI Internet website, Tax_Stats, can be

accessed at http://www.irs.gov/tax_stats.

[8] 2000 Data Book, Publication 55B, Internal

Revenue Service, 2001.

[9] The algorithm used for generating the

Taxpayer Identification Number (TIN)

transform generally stays the same from year

to year. This longitudinal character of the

sample design improves the estimates of

change from one year to the next.

[10] Mikow, Jacob M., "Fiduciary Income Tax

Returns," Statistics of Income (SOI) Bulletin,

Winter 2000-2001, Volume 20, Number 3,

Publication 1136, Internal Revenue Service,

2001.

[11] Petska, Tom, "Exporting a Statistical System:

Towards Establishing a Tax Statistics

Function in South Africa," 2000 Proceedings

of the American Statistical Association,

Sections on Social and Government Statistics,

2001.