# Imagine All the People:

> Briefs, arguments, decisions, and more.

URL: https://www.frixlaw.com/law-library/documents/agency%3Airs%3A6dc5004f6676fdb9

## Record

- **Collection:** Agency decision
- **Document type:** Agency decision

## Text

Imagine All the People:
Using Tax Data to Build a Representative Sample of the US Population
Peter J. Brady
Steven Bass
Investment Company Institute*
1401 H Street N.W.
Washington, DC 20005
pbrady@ici.org
Draft: April 24, 2023
Abstract
This paper documents our method of building a representative sample of the US population
from tax data, the first step in larger research project measuring changes in the amount and
composition of income over the life cycle. We supplement tax return data—which allows us to
identify filers and dependent nonfilers—with information return data—which allows us to
identify non-dependent nonfilers. The share of the population who are nonfilers increases with
age, making their inclusion crucial for measuring elderly income. Our work builds on the
existing literature, especially Cilke (2014) and Lurie and Pearce (2021). The data captures a
larger share of the population than most previous studies using tax data because, following
Lurie and Pearce (2021), we include in our sample individuals identified only on Form 1095,
which reports health insurance coverage. Innovations in our method include sampling
individuals rather than tax returns and using the individual rather than the tax return or family
as the unit of observation. Our method allows us to examine individual income by single year
of age—which is difficult, if not impossible, to do for tax returns or families. Inclusive of both
filers and nonfilers, roughly 95 percent of the population older than 26 in 2016 has income.
Among those with income, median income follows a hump shape over the lifecycle, peaking at
$41,000 per individual at age 46. We find that tax data produce a population estimate similar to
that of the U.S. Census Bureau (Census) and that the two estimates track closely by age. Our
examination of 2010 and 2016 data show that, across years, differences between tax data and
Census estimates are more correlated with birth year than they are by age.

* This research was conducted as part of the Statistics of Income Joint Research Program. Views presented are those
of the authors and do not necessarily represent the views of the Internal Revenue Service or the views of the
Investment Company Institute or its members. We thank Kevin Pierce for his assistance with this project.

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

1. Introduction
This paper explains our method of building a representative sample of the US population
from tax data. The construction of this data is the first step in a larger project measuring
changes in the amount and composition of income over the life cycle. We document our method
both for readers of our broader research, so they understand how we derive our estimates, and
for other researchers analyzing tax data, so that they can use or improve upon our methods.
Our method of estimating the population from tax data builds on the work of many at the
US Department of the Treasury Office of Tax Analysis (OTA), the Joint Committee on Taxation
(JCT), and the Internal Revenue Service Statistics of Income Division (SOI)—in particular Cilke
(2014) and Lurie and Pearce (2021).1 The information returns we use to identify nonfilers are
largely the same as those used in Cilke (2014), with the addition of information returns created
after 2010. The most important of these new information returns is the Form 1095 series
(inclusive of Forms 1095-A, 1095-B, and 1095-C) which reports health insurance coverage and is
the focus of the analysis in Lurie and Pearce (2021).2
The goal of this research is to use tax data to create independent estimates of the US
population and their income, with both the population of interest and the measure of income
based on the rules of the federal income tax. Our population of interest is generally individuals
who, if required to file a tax return, would file a Form 1040, and those individuals’ dependents.
This would include US residents, but it would also include members of the US armed forces
stationed overseas and US citizens or resident aliens living outside the US. Our measure of
income generally follows the definition of income in the Internal Revenue Code. The only
exceptions are that we include some items defined in the tax code but excluded from taxable

Early work trying to measure the nonfiling population include Cilke (1998), Sailer and Weber (1998), Mortenson et
al. (2009), and Lawrence et al. (2011). Examples of other recent studies using tax data to represent the US population
include Saez (2016) and Larrimore et al. (2019).
1

We wish to acknowledge and thank Ithai Lurie and James Pearce for providing us with the data they processed and
analyzed in Lurie and Pearce (2021). Form 1095 data are complex and require careful processing to be usable. See
Lurie and Pearce (2021) for a description of how they identified individuals with health insurance coverage using
Form 1095.
2

Brady and Bass

1

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

income—such as tax-exempt interest and Roth-IRA distributions—and exclude others that are
included in taxable income—such as Roth contributions.
Our method of estimating the population differs from the typical approach in that we
create a sample of individuals rather than a sample of tax returns. Our full sample consists of
three subsamples—one for filers (primary or secondary taxpayers listed on a return), one for
dependent nonfilers, and one for non-dependent nonfilers. Multiple individuals from a single
tax return would be included in the sample only if they independently meet our sampling
criterion.
We can take this sampling approach because the individual—rather than the tax return,
family, or household—is our unit of analysis. There is no need to link primary taxpayers to
secondary taxpayers, and no need to link taxpayers to the dependents they claim. All income
reported on tax returns is allocated to filers—the primary taxpayer in the case of non-joint
returns and the primary and secondary taxpayers in the case of joint returns. Dependents have
income only if they file their own return or have income reported on information returns. This
approach allows us to tabulate individual income by single year of age.
Similar to the findings in the existing literature, filers and dependent nonfilers identified on
tax returns represent the bulk (87 percent) of the population.3 Like Lurie and Pearce (2021),
however, the use of Form 1095 does allow us to identify more non-dependent nonfilers than is
typical.
Although non-dependent nonfilers are not a large part of the overall population, their
importance increases with age, making their inclusion crucial for measuring elderly income.
From about five percent of the population through age 16, non-dependent nonfilers increase to
11 percent by age 24, 16 percent by age 60, and 30 percent by age 80.
Inclusive of both filers and nonfilers, the share of the adults with income remains fairly
steady by age, while median income follows a hump shape over the life cycle. The share of the
population with income averages 94 percent from age 27 through 61 and it averages 96 percent

3

See, for example, Cilke (2014) and Larrimore et al. (2019).

Brady and Bass

2

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

from age 62 through 99. Among those with income, median income increases rapidly early in
life, peaks at $41,000 at age 46, and then declines with age—to $38,000 at age 60 and $34,000 at
age 70.
To get a sense of how well the tax data capture the US population, we compare our
estimates to those of the U.S. Census Bureau (Census). Before we make the comparison, we the
adjust our data to better conform to the Census population concept. Consistent with existing
research, we find the two population estimates are similar and track closely by age. We also find
some age ranges over which the two estimates diverge noticeably, but the ages differ from those
noted in earlier studies. Replicating the results of the previous studies, we find that differences
between tax data and Census estimates are more consistent over time by birth year than they
are by age.
The paper is organized as follows. Section 2 describes the data we analyze. Section 3
defines our unit of analysis, measure of income, and population of interest. Section 4 describes
our method of constructing our representative sample of the population. Section 5 examines the
composition and income of the population by single year of age. Section 6 compares our
estimates to Census population estimates. Section 7 summarizes our results and concludes the
paper.

2. Description of the Data
This study uses Internal Revenue Service (IRS) administrative tax data from tax year 2016.
These data include information from both federal individual income tax returns filed by
taxpayers and information returns issued by third parties and sent to both taxpayers and the
IRS. We also incorporate Social Security Administration (SSA) data on gender, date of birth, and
date of death (if applicable).

Brady and Bass

3

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

In general, US citizens and resident aliens4 are required to file a Form 1040 (inclusive of
Form 1040, Form 1040-A, and Form 1040-EZ) if their gross income is above a certain threshold.5
The filing requirement applies even if a US citizen or resident alien lives outside the US during
the tax year.6 An exception to this rule is US citizens or resident aliens living in US possessions,
who generally file a return with their territory’s tax authority rather than with the IRS.7
US citizens and resident aliens with gross income below the filing thresholds may also file.
Regardless of gross income, filing is required for individuals who received any advance earned
income tax credit (EITC), who owe payroll taxes that were not withheld, or who owe special
taxes (such as the alternative minimum tax). In addition, although not required to file,
individuals with gross income below the thresholds can only receive refunds of withheld taxes
or refundable tax credit payments if they file a return.
Although they would not file a Form 1040, residents of US possessions and nonresident
aliens may file a different type of tax return with the IRS. Residents of US possessions with selfemployment earnings generally must file a Form 1040-SS or Form 1040-PR and, if necessary,
pay self-employment tax.8 Nonresident aliens who were engaged in a trade or business in the
US or who had US-source income are required to file a Form 1040-NR.

Resident aliens include individuals with a green card or who had a “substantial presence” in the US—inclusive of
the 50 US states and the District of Columbia. For more information on US income tax treatment of both resident and
nonresident aliens, see Internal Revenue Service (2017a).
4

Gross income includes all income not exempt from tax. It includes both US-source and foreign-source income. For
self-employed individuals providing services, gross income includes the gross receipts of the business. The gross
income filing thresholds vary based on filing status, age, blindness, and whether a parent (or another taxpayer) can
claim the individual as a dependent. For example, the 2016 filing thresholds for non-dependent single individuals
younger than age 65 was $10,350 and the threshold for a non-dependent married couple filing a joint return with
both spouses younger than 65 was $20,700. The threshold was $1,550 higher for single individuals aged 65 or older
and $2,500 higher for joint filers with both spouses aged 65 or older, respectively. For more information on filing
requirements, see Internal Revenue Service (2016b) and Internal Revenue Service (2016c).
5

6

For filing requirement for US citizens and resident aliens living abroad, see Internal Revenue Service (2016a).

US citizens and resident aliens who are bona fide residents of Guam, the US Virgin Islands, and the Northern
Mariana Islands are not required to file a Form 1040 with the IRS. US citizens and resident aliens who are bona fide
residents of American Samoa and Puerto Rico are only required to file a Form 1040 if they received income from a
source outside of the territory. This would include US government employees, as their wages are considered US
source income. For the definition of a bona fide resident and more information on filing requirements for individuals
with income from US possessions, see Internal Revenue Service (2017c).
7

Bona fide residents of Guam, American Samoa, the US Virgin Islands, the Northern Mariana Islands, and Puerto
Rico must file a Form 1040-SS if they have $400 or more of net self-employment earnings and are not required to file
8

Brady and Bass

4

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Tax return data include both information reported directly on Form 1040 and information
reported on associated schedules (such as Schedule SE, which is used to calculate selfemployment tax) and forms (such as Form 6251, which is used to calculate the alternative
minimum tax) filed with the tax return.
In addition to tax return data, we incorporate data from various information returns, which
are issued by third parties and sent to the taxpayer and to the IRS. Information returns allow us
to identify nonfilers and measure their income. Information returns also allow us to allocate
income reported by married couples on joint tax returns to the spouse who received it. And, in
some cases, they provide information not reported on Form 1040—such as detailed codes for
distributions from IRAs, pensions, and annuities.
To identify nonfilers, we use a large number of information returns. Many of the
information returns report income—such as Form W-2, which reports wages, or Form 1099-INT,
which reports interest income. Other information returns report expenses—such as Form 1098,
which reports mortgage interest expense, or Form 1098-T, which reports tuition expense—or
other tax-relevant information—such as Form 1095-B, which reports health insurance coverage.
For a full listing of information returns used in this study and their description, see Appendix
Table A.1.

3. Conceptual Issues
The goal of this research is to illustrate how the tax data can be used to create independent
estimates of the size and income of the US population. As such, we are not attempting to match
concepts used for existing income estimates derived from household surveys. More specifically,
we do not alter the tax data to match the unit of analysis, the measure of income, or the
population of interest used in the official income statistics published by the Census, which are
derived from the Current Population Survey (CPS) Annual Social and Economic Supplement

Form 1040 with the IRS. Residents of Puerto Rico have the option to file a Form 1040-PR (which is in Spanish)
instead. In addition to facilitating tax collection, the information reported on these forms is used by the SSA for
calculating Social Security benefits. For a complete description of who must file Form 1040-SS/1040-PR, see Internal
Revenue Service (2017c).

Brady and Bass

5

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

(ASEC). If the reader is primarily interested in comparing income measures derived from tax
data with income measures derived from the CPS, see Brady and Pierce (2012), Bee and Mitchell
(2017), Larrimore et al. (2019), and Brady and Bass (2021).

3.1 Unit of Analysis
We use the individual as the unit of analysis9 rather than the tax return,10 the family11—that
is, the combination of all individuals reported on a tax return, inclusive of primary taxpayers,
secondary taxpayers, and dependents (along with their income)—or the household12—that is,
the combination of all families (along with their income) living in the same household. As such,
we include in our sample non-dependent filers, dependent filers, dependent nonfilers, and nondependent nonfilers.
We do not attempt to use the tax data to construct households, as is done in Larrimore et al.
(2019); nor do we classify dependents using the income reported by the filers who claim them,
as is done in Lurie and Pearce (2021). Non-dependent filers are classified by the income
reported on their returns. Non-dependent nonfilers are classified by the income reported on
information returns. Dependents are classified by their own income as reported on either a
dependent return or on information returns.
The primary reason we use the individual as our unit of observation is that the focus of our
research, using both the cross-sectional data described in this paper and panel data, is
measuring changes in the amount and composition of income over the life cycle. The
composition of an individual’s tax return, family, or household rarely remains stable across the

The individual is the unit of analysis used in Brady et al. (2017), Brady and Bass (2020a), Brady and Bass (2020b),
and Brady and Bass (2021).
9

The IRS Statistics of Income Division (SOI) uses the tax return (inclusive of both non-dependent and dependent tax
returns) as the unit of analysis in their annual Complete Report. See, for example, Internal Revenue Service, Statistics
of Income Division (2018).
10

The Joint Committee on Taxation (JCT) and the US Department of Treasury Office of Tax Analysis (OTA) use the
family as the unit of analysis in their distributional analyses (Joint Committee on Taxation 1993, Cronin 1999).
11

The Congressional Budget Office (CBO) uses the household as the unit of analysis in its distributional analyses, as
does Larrimore et al. (2019), which also includes nonfilers. CBO creates households for their microsimulation model
using a combination of tax data and CPS data. Larrimore et al. (2019) uses the mailing address on tax returns and
information returns to construct household measures of income from tax data.
12

Brady and Bass

6

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

life cycle. It is also difficult, if not impossible, to use any other of unit of analysis to study
differences in income by age.

3.2 Measure of Income
Our measure of total income is the sum of six types of income: labor (wage and salary, selfemployment earnings, unemployment compensation), Social Security (disability benefits and
retirement benefits), retirement (IRA distributions and income from pensions and annuities),
investment (taxable interest, tax-exempt interest, dividends, gains/losses),
business/farm/rents/royalties (business and farm income in excess of self-employment
earnings; income from rents, royalties, partnerships, S-corps, and trusts), and other (net
alimony [alimony received less alimony paid] and other income).
In addition to total income and its components, we also measure spendable income—that
is total income less federal income and payroll taxes.
While our total income measure is created from tax data, it differs from the tax code’s
definition of income. We include some types of income excluded from taxable income—such as
tax-exempt interest and the nontaxable portion of Social Security benefits—and exclude some
types of income included in taxable income—such as taxable state income tax refunds. In
addition, we also treat, to the extent possible, all contributions and distributions from
retirement plans the same--regardless of whether contributions were from an employer or an
employee, and regardless of their tax treatment. This means we exclude from income not only
tax-deferred employee contributions to employer plans and IRAs, but also Roth contributions
and non-Roth after-tax contributions. It also means we include in income not only taxable nonRoth distributions, but also Roth distributions and the portion of non-Roth distributions that
represents basis. Appendix Table A.2 describes how we calculate these income and tax
measures for both filers and nonfilers.
The definition of income used in this paper also differs from previous studies focused on
comparing tax data to household survey data. For example, Brady and Pierce (2012) and Brady
and Bass (2021) measure four types of income which are measured in both data sources, and
which have been found to be accurately reported on income tax returns: wage and salary

Brady and Bass

7

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

income, Social Security benefits, retirement income, and investment income excluding
gains/losses (taxable interest, tax-exempt interest, and dividends). Bee and Mitchell (2017) use
those four sources of income plus Supplemental Security Income (SSI). Larrimore et al. (2019)
use a definition like the one used in this study but exclude realized capital gains/losses reported
on Schedule D and bottom code income at zero to limit the effect of business losses.
The primary measure of income we use to analyze the incidence and amount of income by
type is per capita income, which allocates the income reported on joint tax returns equally to each
spouse. For primary and secondary taxpayers on joint returns, per capita income is the income
derived from the tax return divided by two. For primary taxpayers on non-joint tax returns, per
capita income is simply the income derived from the tax return. Similarly, for nonfilers, who are
all assumed to be non-joint, per capita income is simply the amount derived from an
individual’s information returns.
In addition to per capita income, we also report own income for labor, Social Security, and
retirement income. For joint filers, we use information returns to allocate the income reported
on joint returns to the spouse who received the income. For non-joint filers and for nonfilers,
there is no difference between own income and per capita income.
Importantly, neither per capita income nor own income adjusts for family size.13 All income
reported on a tax return is allocated to filers—the primary taxpayer in the case of non-joint
returns and the primary and secondary taxpayers in the case of joint returns. The number of
dependents claimed on a tax return has no impact on either measure, as no filer income is
allocated to dependents. Dependents only have measured income if (1) they file their own tax
return or (2) they do not file their own return, but they have income reported on an information
return. For dependent filers, the calculation of income is the same as it is for non-dependent

The discussion in the text is limited to family-size adjustments solely for brevity; the discussion also applies to
household-size adjustments. Some, but not all, distributional analysis adjust income for family or household size. JCT
does not adjust for family size before ranking families by income (Joint Committee on Taxation 1993). In contrast,
OTA adjusts for family size and CBO adjusts for household size before ranking by income (Cronin 2022,
Congressional Budget Office 2016).
13

Brady and Bass

8

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

filers. For dependent nonfilers, the calculation of income is the same as it is for non-dependent
nonfilers.
We do not adjust our income measures for family size, in part, because we do not
characterize our annual income measures as reflecting either well-being or the ability to pay
taxes. Compared with the typical adjustment for family size—dividing family income by the
square root of the number of individuals in the family14—we believe the transparency of our per
capita income measure simplifies the interpretation of changes in the amount and composition
of income by age, and that it simplifies the interpretation for both filers and dependents. We
also note that both well-being and ability to pay depend on many factors other than annual
income and the presence of children in the current year, and that the impact of children is
unlikely to be properly captured by the typical family-size adjustment in any case.15 That said,
we encourage the reader to consider how the amount and composition of an individual’s
spending may change with age—including spending related to raising children—when
interpreting changes in the amount and composition of per capita income over the lifecycle.

3.3 The Population of Interest
Our goal is to build a sample of US citizens and resident aliens who—provided their gross
income exceeded the filing thresholds—would be required to file a 2016 Form 1040, excluding
residents of US territories. This would include US citizens and resident aliens living in a state
(inclusive of the 50 states and the District of Columbia), living outside the US, or living overseas

14

Before ranking by income, both OTA and CBO divide income by the square root of the number of individuals in
the family/household in their distributional analysis (Cronin 2022, Congressional Budget Office 2016), as does
Larrimore et al. (2019).
15

The rationale both for using the family or household (rather than the individual or tax return) as the unit of
analysis, and for dividing family/household income by the square root of family/household size before ranking by
income (rather than simply ranking by family income sans the size adjustment or ranking by family income divided
by family size sans the square root), is that it produces a measure of income that better reflects well-being and/or the
ability to pay taxes. This functional form is consistent with certain assumptions about economies of scale in family
size. That is, it is consistent with the belief that two individuals living together cannot live as cheaply as one but can
live more cheaply than two individuals living on their own. It is also consistent with the belief, common in movies,
that children are “cheaper by the dozen”—that is, that the marginal cost of a second child is less than the marginal
cost of the first, the marginal cost of a third child is less than the marginal cost of the second, and so on. While we
believe it is reasonable to assume there are economies of scale in family expenses, we find it unlikely that (1) the
square root of family/household size (conveniently) captures this effect or (2) that economies of scale are proportional
to income.

Brady and Bass

9

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

as a member of the US armed forces. This would exclude nonresident aliens engaged in a trade
or business in the US or who had US-source income—who, if required, would file a
Form 1040-NR—and self-employed residents of US possessions—who, if required, would file a
Form 1040-SS/1040-PR to pay self-employment taxes. We also exclude any residents of US
territories who file a Form 1040 because bona fide residents of US territories generally do not
file an income tax return with the IRS.16
Our population of interest differs from the populations represented in official measures of
the US population and income. For example, Census releases annual estimates of the US
resident population (inclusive of the 50 states and the District of Columbia).17 Compared with
our population, the US resident population excludes US citizens and resident aliens living
abroad as well as members of the US armed forces stationed overseas. Census also produces an
annual report on the income of the US resident civilian, noninstitutionalized population.18 In
addition to those excluded from the US resident population, this would exclude certain
members of the US armed forces stationed in the US and individuals living in nursing homes,
long-term care facilities, and other institutions.19
Our population of interest also differs from previous studies that use tax data to build a
representative sample of the US population. For example, Cilke (2014), Lurie and Pearce (2021),
and Larrimore et al. (2019) all attempt to represent the US resident population. They generally
include only tax information associated with a US address (inclusive of the 50 states and the
District of Columbia). This means, compared with our population, these studies exclude
taxpayers and dependents listed on tax returns with an overseas US armed forces address or a
foreign or missing address. They also exclude non-dependent nonfilers with an overseas US

16

See note 7.

17

See, for example, U.S. Census Bureau (2021).

18

See, for example, Semega et al. (2019).

Members of the US Armed Forces are only included in the CPS sample if they live in a household with at least one
other adult who is a civilian.
19

Brady and Bass

10

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

armed forces address on their information returns, and generally have more stringent criteria
than this study for excluding non-dependent nonfilers with foreign or missing addresses.20

4. Creating the Representative Sample
Our method of sampling individuals differs from the typical method of sampling tax
returns. For example, the Individual and Sole Proprietor (INSOLE) file, which is used by the
SOI to produce its annual Individual Income Tax Returns Complete Report publication and used by
the JCT and OTA as the basis for their microsimulation models, is a sample of tax returns.21
Similarly, the panel data used in analysis and as the basis of the OTA’s panel model is also a
sample of tax returns (Nunns et al. 2008). In these data, the sample consists of tax returns and
information for all individuals on a selected tax return—primary taxpayers, secondary
taxpayers, and dependents—is included in the data. In contrast, our sample consists of
individuals, with multiple individuals from a given tax return included in our sample only if
they independently meet our sampling criterion.
We can sample individuals rather than tax returns because individuals are the focus of our
analysis. We do not use the tax data to construct families or households, so there is no need to
link dependents to the taxpayers who claim them. All income reported on tax returns is
allocated to the filers. In the case of joint filers, our per capita income measures allocate income
equally to each spouse, while our own income measures (for labor, Social Security, and
retirement income) use information returns to allocate the tax return income to the spouse who
received it. The number of dependents listed on a tax return does not affect our measure of filer
income, nor does filer income affect our measure of income for dependents listed on the return.

We only want to include non-dependent nonfilers living abroad if they are US citizens or resident aliens. As
discussed below, we include non-dependent nonfilers only if they have a US address (in one of the 50 states or the
District of Columbia) or an overseas US armed forces address on at least one information return. In contrast, Cilke
(2014) excludes individuals with only non-US addresses on their information returns (even individuals with only
overseas US armed forces addresses) and excludes individuals with both US and non-US addresses on their
information returns if taxable income reported on an information return sent to a non-US address exceeds a given
threshold.
20

21

For a description of the INSOLE sample, see Internal Revenue Service, Statistics of Income Division (2018).

Brady and Bass

11

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

We build a sample representative of the 2016 US population in three steps. In each step we
generally only include individuals with a valid Social Security Number (SSN) or Taxpayer
Identification Number (TIN) in the SSA data.
•

First, we draw a filer sample: primary and secondary taxpayers who file a tax return.

•

Second, we draw a dependent nonfiler sample: individuals who did not file a return but
were claimed as dependents by a taxpayer who did file.

•

Third, we draw a non-dependent nonfiler sample: individuals who did not file a return
and who were not claimed as a dependent by a taxpayer who did file a return, but who
receive at least one information return.

4.1 Filer Sample
To construct our filer sample, we first pull a sample of individuals appearing as primary or
secondary taxpayers on tax-year 2016 tax returns, inclusive of Form 1040, Form 1040-SS/
1040-PR, and Form 1040-NR. Primary taxpayers on non-joint returns are selected if they have an
SSN or TIN that ends in one of 500 unique 4-digit numbers and the SSN/TIN is valid. Primary
and secondary taxpayers on joint returns are selected if they have an SSN or TIN that ends in
one of 500 unique 4-digit numbers and the SSN/TIN of either the primary or secondary taxpayer
on their return is valid.22 Although return and spousal information is retained for individuals
selected from joint returns, spouses are not generally included in the sample (unless their
SSNs/TINs independently meet our criteria for inclusion). This method results in a roughly 5.0
percent sampling rate of primary and secondary taxpayers.23 The resulting sample represents
206 million primary and secondary taxpayers (Table 1).24

As explained in Cilke (2014), the IRS uses a number of methods to validate an SSN/TIN reported on a tax form,
with the primary method checking to see if the first four letters of the taxpayer’s last name match the first four letters
of the last name associated with the SSN/TIN. Married individuals who have recently changed their last name may
be coded as invalid until their Social Security Administration records are updated.
22

The last four digits of the SSN/TIN range in value from 0001 to 9999, resulting in a sampling rate slightly over
5.0 percent (=500/9,999).
23

Sample weights are calculated as the inverse probability of selection. Taxpayers receive a weight of just under 20
(=9,999/500).
24

Brady and Bass

12

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

For analysis, we focus on primary and secondary taxpayers from 2016 Form 1040 who did
not die prior to 2016, excluding residents of US possessions. This excludes individuals
representing fewer than 2,000 taxpayers on 2016 tax returns who died before January 1, 2016. It
also excludes individuals representing 748,000 non-resident aliens who filed a Form 1040-NR,
241,000 individuals who filed a Form 1040-SS/1040-PR, and 94,000 residents of US possessions
who filed a Form 1040 (Table 1).
The final filer sample represents 204 million individuals, including 150 million primary
taxpayers and 54 million secondary taxpayers (Table 1). Out of the 204 million filers, 687,000
were on returns with a foreign or missing address and 239,000 were on returns filed by US
military personnel living overseas.
Among all filers, 98 percent had at least one information return. The most common
information returns were Form 1095 (89 percent of taxpayers), Form W-2 (72 percent), and
Form 1099-G (33 percent).

4.2 Dependent Nonfiler Sample
To construct our dependent nonfiler sample, we first pull a sample of individuals listed as a
dependent on a 2016 Form 1040 or Form 1040-NR, or listed as a qualifying child on a
Form 1040-SS/1040-PR.25 The tax data include the SSN/TIN of up to four dependents/qualifying
children listed on paper returns and the SSN/TIN of all dependents listed on electronic
returns.26 From this group we select all dependents with an SSN/TIN that ends in one of the
same 500 unique 4-digit numbers used to select the filer sample. The resulting sample
represents 94 million dependents (Table 2).
For analysis, we focus on dependents listed on 2016 Form 1040 who did not die prior to
2016 and who did not file a dependent return, excluding dependents claimed by residents of US

Qualifying children are only listed on Form 1040-SS/1040-PR by bona fide residents of Puerto Rico who claim the
additional child tax credit.
25

There are two sources for dependent information. For all returns—including both paper returns and electronic
returns—the SSN/TIN of the first four dependents listed on the return is reported and the validity of the SSN/TIN is
checked. For electronic returns, the SSN/TIN of all listed dependents is reported, but the validity of the SSN/TIN is
not checked. For the first four dependents listed on all returns, we include only those with a valid SSN/TIN. For
additional dependents listed on electronic returns, we include all (because there is no validity indicator).
26

Brady and Bass

13

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

possessions. In addition to individuals representing 9,000 dependents who died before 2016,
this excludes individuals representing 25,000 dependents claimed on Form 1040-NR, 171,000
qualifying children claimed on Form 1040-SS/1040-PR, and 41,000 dependents claimed on
Form 1040 filed by residents of US possessions (Table 2). It also excludes 9.3 million individuals
claimed as a dependent who filed their own 2016 Form 1040 (and who were already included in
the taxpayer component of the sample).27
The final dependent nonfiler sample represents 84 million individuals (Table 2). Dependent
nonfilers include 260,000 claimed on returns with a foreign or missing address and 121,000
claimed on returns filed by US military personnel living overseas.
Among all dependent nonfilers, 89 percent had at least one information return. The most
common information returns were Form 1095 (88 percent of dependent nonfilers),
Form SSA-1099 (6 percent), and Form W-2 (6 percent).
Note that the dependent nonfiler sample does not include dependents claimed on paper
returns with more than four dependents if they were not among the first four listed. Provided
they did not file their own dependent return, these individuals would be included in the nondependent nonfiler sample if they had at least one tax-year 2016 information return with a valid
SSN/TIN. If they neither filed their own return nor had an information return, they would be
excluded from the final sample altogether.

4.3 Non-Dependent Nonfiler Sample
To construct our non-dependent nonfiler sample, we first pull a sample of individuals who
had at least one 2016 information return with a valid SSN/TIN and who were neither a taxpayer
who filed, nor a dependent/qualifying child claimed on, a 2016 Form 1040, Form 1040-NR, or

The estimate of dependent filers in the filer component of the sample (9.4 million) is greater than the estimate of
dependent filers excluded from the dependent nonfiler component of the sample (9.3 million). The difference in the
estimates is presumably attributable to the fact that not all dependent filers included in the filer component of the
sample were listed as dependents on a tax return. This is for two reasons. First, some dependent filers may have been
claimed as a dependent on a paper return but not included in our dependent sample because they were not among
the first four listed on the return. The paper Form 1040 only includes space for up to four dependents. Any additional
dependents are reported to the IRS separately. Second, some dependent filers may not have been claimed as
dependents, as individuals who could be claimed as a dependent must file a dependent return regardless of whether
they were claimed as a dependent or not.
27

Brady and Bass

14

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Form 1040-SS/1040-PR. From this group we select all individuals with an SSN/TIN that ends in
one of the same 500 unique 4-digit numbers used to select the filer and dependent nonfiler
samples. The resulting sample represents 50 million non-dependent nonfilers (Table 3).
For analysis, we focus on non-dependent nonfilers who did not die prior to 2016 and who
were likely to be a US citizen or resident alien and who did not reside in a US territory. The
largest group excluded from the initial sample were 4.3 million individuals who were sent 2016
information returns but died prior to 2016 (Table 3). We also exclude 3.4 million individuals for
whom all information returns were sent to addresses in US territories or to foreign or missing
addresses. Finally, we excluded 73,000 individuals who had one of three forms that indicated
they were a foreign person.28
The final non-dependent nonfiler sample represents 42 million individuals (Table 3),
including 17,000 who had at least one information return sent to an address for members of the
US armed forces living overseas. Of the 42 million non-dependent nonfilers, 31 million (73
percent) had income. The most common information returns were Form 1095 (86 percent of
non-dependent nonfilers), Form SSA-1099 (43 percent), and Form W-2 (25 percent).

4.4 Total Population Estimates
The final sample represents 331 million filers, dependent nonfilers, and non-dependent
nonfilers (Table 4). Overall, 87 percent of the population are identified using Form 1040
(inclusive of both filers and dependent nonfilers), 10 percent are non-dependent nonfilers
identified using at least one non-Form-1095 information return, and 3 percent are identified
using Form 1095 only. Three-quarters of the population have income. Most without income are
dependent nonfilers.

The three forms are Form 8288-A (Statement of Withholding on Dispositions by Foreign Persons of U.S. Real
Property Interests), Form 8805 (Foreign Partner’s Information Statement of Section 1446 Withholding Tax), and
Form 1042-S (Foreign Person’s U.S. Source Income Subject to Withholding). About 85 percent of individuals with
Form 8288-A or Form 8805 file a return, with nearly all filing Form 1040-NR. Just over one-third of individuals with
Form 8288-A file a return, with over 60 percent filing a Form 1040-NR.
28

Brady and Bass

15

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

5. Population Composition and Median Income by Age
Dependent nonfilers (individuals identified as a dependent on Form 1040 and who do not
file a dependent return) represent most of the population at younger ages, with their share
falling rapidly after age 15, remaining low through middle age, and then increasing again at
older ages (Figure 1, top panel). From over 95 percent at younger ages, the dependent nonfiler
share of the population falls to 92 percent at age 15, 50 percent at age 18, and less than 5 percent
at age 26. Their population share continues to fall with age, falling below 2 percent in the late
30s and remaining there through the early 50s. Dependent nonfilers then begin to increase as a
share of the population, increasing to 4 percent by age 70 and 7 percent by age 80.
Largely mirroring the decline in dependent nonfilers, filers (individuals who are primary
or secondary taxpayers on Form 1040) increase rapidly as a share of the population after age 15
and the share remains high for those from their mid-20s through early 60s before falling off at
older ages (Figure 1, top panel). Filers increase from 3 percent of the population at age 15 to 43
percent at age 18 to 84 percent at age 26. Although filers are typically dependent filers at
younger ages, non-dependent filers represent nearly one-quarter of filers by age 18 and nearly
all filers after age 26 (not shown in Figure 1). The filer population share peaks at 86 percent from
age 29 through age 46, remains above 84 percent through age 50, and is still 82 percent at age 60.
After age 60, the filer share begins to decline more noticeably, falling to 75 percent by age 70
and 63 percent by age 80.
Non-dependent nonfilers (individuals not identified on a tax return but who receive at least
one information return) increase as a share of the population with age (Figure 1, top panel).
Non-dependent nonfilers average a bit less than 5 percent of the population from birth through
age 15 and then increase to 11 percent of the population by age 26, with the decline in
dependent nonfilers not completely offset by the increase in filers over this age range. After age
26, their population share increases slowly to 16 percent at age 60, an average increase of just
over 0.1 percentage points per year of age. After age 60, their population share growth
accelerates to 0.9 percentage points per year of age, on average, through age 99—increasing to
21 percent by age 70 and 30 percent by age 80.

Brady and Bass

16

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Despite the filer share of the population falling at older ages, the share of the population
with income remains consistently high throughout adulthood, and even increases slightly after
age 61 (Figure 1, bottom panel).29 The share of the population with income increases rapidly for
teenagers and young adults, from 12 percent at age 12 to 71 percent at age 18 and 93 percent at
age 26. The share generally remains steady for those older than 26—with 94 percent of the
population, on average, from age 27 through age 61 and 96 percent of the population, on
average, from age 62 through age 99 having income.
To inform comparisons with previous research using only data from tax returns or using
only information on nonfilers prior to the existence of Form 1095, Figure 2 shows our
population estimates broken down by how we identified those individuals.
As already noted, the bulk of the population is identified using Form 1040—inclusive of
filers and dependent nonfilers—but this group represents a declining share of the population
with age. Their population share is about 95 percent of the sample through age 16 but falls
thereafter—to 89 percent by age 24, 84 percent by age 60, and 70 percent by age 80.
The next largest component of the population is non-dependent nonfilers identified on at
least one information return other than Form 1095, who represent a small portion of population
at younger ages but who increase in importance with age. This group’s population share is less
than 1.0 percent through age 13, but increases to 2.0 percent by age 16, and to 8.8 percent by age
24. Their population share growth slows after age 24, increasing to 12 percent by age 60. After
age 60, share growth accelerates, exceeding 29 percent by age 80.
The final component of the population—which, to our knowledge, has not been included in
analysis of tax data other than in Lurie and Pearce (2021)—is non-dependent nonfilers who are
identified using Form 1095 alone. This group represents less than 5.0 percent of the population
at all ages below 100, with their highest share among children and middle-aged adults and their
lowest share among the elderly. Individuals identified only on Form 1095 represent 4.6 percent

In this study, the presence of income is defined as having nonzero income in any of our broad income categories—
labor, Social Security, retirement, investment, business/farm/rents/royalties, and other (see the definition of total income in
Table A.2).
29

Brady and Bass

17

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

of the population younger than one year of age, with their share falling slowly to 3.1 percent by
age 16, and then falling more quickly to 2.0 percent for those aged 20 through 22. After age 22,
their population share begins to increase, peaking above 3.5 percent for individuals aged 52
through 61. After age 61, the share drops sharply, to less than 1.0 percent of the population from
age 67 through age 90, before edging up at older ages.
As explained in section 3, our primary measure of income is per capita—that is, the income
reported by married couples who file joint returns is allocated equally to each spouse. Again, no
income is allocated to dependents listed on a tax return. All income reported on a tax return is
allocated to filers—the primary and secondary taxpayers in the case of joint returns and the
primary taxpayer in the case of non-joint returns. Nonfilers (inclusive of dependent nonfilers
and non-dependent nonfilers) are allocated the income reported on their information returns (if
any), with no effort made to impute marital status or to create family groups.
The composition of the population by filing-type group—joint filer, non-joint filer, and
nonfiler (inclusive of both dependent nonfilers and non-dependent nonfilers)–varies with age,
reflecting the fact that the group to which a given individual belongs changes over the life cycle
(Figure 3, top two panels).
The nonfiler share of the population falls sharply with age early in life, as more and more
have income above the filing threshold, and remains low through age 40. The nonfiler
population share then begins to increase with age, particularly after age 61. Along with the
elderly typically having lower total income, two tax code provisions reduce the share of the
population who are required to file with age. First, Social Security benefits are either fully or

Brady and Bass

18

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

partially excluded from the gross income measure on which the filing requirement is based.30
Second, the gross income filing threshold is slightly higher for individuals aged 65 or older.31
The joint-filer share of the population follows a hump-shaped pattern by age, increasing
rapidly after age 18, hitting one-quarter of the population by age 27 and half of the population
by age 37 before peaking at 56 percent at age 62. After age 62, the joint-filer share falls—at first
slowly and then at an accelerated rate. In addition to spousal death, a portion of the decline in
joint-filer share after age 62 may be attributable to a falling share of the population filing a
return. Among filers, the share filing a joint return peaks at age 69, where it reaches 71 percent.
The relationship between median per capita total income and age differs by filing-type
group (Figure 3, bottom panel).32 For all but the elderly, joint filers have the highest per capita
income. The median per capita income of joint filers increases rapidly at younger ages, peaks at
age 46, then declines through about age 80. The median income of non-joint filers increases
more slowly than that of joint filers after age 25, but continues to increase slowly through age 62
and then increases sharply between age 62 and age 67. As a result, the median income of nonjoint taxpayers is higher than that of joint taxpayers for individuals aged 67 through 98.
Nonfilers have the lowest median income at all ages, with median income increasing only
modestly through age 40 and then remains roughly constant before drifting higher after age 62.

The percentage of Social Security benefit payments included in gross income is based on a taxpayer’s modified
adjusted gross income (MAGI), which includes half of Social Security benefit payments plus other income included
in gross income. For single, head of household, and qualifying widow(er) returns: if MAGI is $25,000 or less, no
Social Security benefit payments are included in gross income; if MAGI is between $25,000 and $34,000, the lesser of
50 percent of Social Security benefit payments or 50 percent of MAGI in excess of $25,000 is included in gross income;
if MAGI is in excess of $34,000, the lesser of 85 percent of Social Security benefit payments or 85 percent of MAGI in
excess of $34,000 plus $4,500 [=50%*($34,000-$25,000)] is included in gross income. For joint returns, the MAGI
thresholds are $32,000 and $44,000, respectively. For more information on the taxation of Social Security benefits, see
Internal Revenue Service (2017b).
30

31

See note 5.

The medians presented in this study are approximate, as true medians could represent disclosure of an individual’s
tax data. To calculate approximate medians, we average the 48 th, 49th, 50th, 51st, and 52nd percentile values and then
round that average (to the nearest dollar for amounts less than $100, the nearest $10 for amounts from $100 to less
than $10,000, the nearest $100 for amounts of $10,000 or more, and two decimal places for percentages). We then only
report these approximate medians for groups with 100 or more observations. We use the same method to report
other percentile measures. Others who wish to use this measure for their own research may cite this article for
authority or simply refer to the measure as the Brady-Bass Adjusted Median (BBAM).
32

Brady and Bass

19

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Within a filing-type group, changes in median income by age reflect not only the typical
experience of individuals, but also changes in the composition of the group by age. The income
of nonfilers grows more slowly with age than that of filers because of filing requirements, which
effectively cap the amount of income an individual can have while remaining a nonfiler. At
younger ages, the median income of joint filers increases more quickly with age than that of
non-joint filers, at least in part, because single individuals with higher earning potential are
more likely to get married.33 At older ages, the tax treatment of Social Security benefits affects
the relative income of all three groups. The upward drift in nonfiler income after age 62 is
primarily because our total income measure includes Social Security benefits excluded from
gross income,34 and to a lesser extent because the gross income filing threshold increases for
individuals aged 65 or older.35 The income of joint filers declines relative to that of non-joint
filers after age 62, at least in part, because the filing requirement effectively kicks in at lower
levels of per capita total income for joint filers: although the gross income filing threshold is
roughly twice as high for married couples,36 the thresholds for determining the share of Social
Security benefits included in gross income are not.37
Examining the entire population arguably provides a clearer picture of changes in income
over the life cycle, as individuals shift between filing-type groups over the course of their
lifetimes. For the entire population, conditional on having non-zero total income,38 median per
capita income follows a hump-shaped pattern with age (Figure 3, bottom panel, black line).
Income increases rapidly early in life to $30,000 at age 30 and then continues to increase, but at a
slower rate, peaking at $41,000 at age 46. After age 46, income declines with age—to $38,000 at
age 60 and $34,000 at age 70.

See, for example, Lundberg et al. (2016), which illustrates that, among individuals aged 30 through 44 in 2010,
individuals with at least a bachelor’s degree were more likely to be married than individuals with less education.
33

34

See note 30.

35

See note 5.

36

See note 5.

37

See note 30.

38

See note 29.

Brady and Bass

20

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

6. Comparison with Census Population Estimates
This section compares our population estimates to those produced by Census to provide a
general sense of how well the tax data captures the US population. Although the two
populations that are estimated differ, there is enough overlap between the two to make the
comparison informative.
The Census produces annual estimates of the US resident population as of July 1. The
current 2016 Census estimate is postcensal—that is, it is based on the 2010 decennial census plus
more recent data on births, deaths, and immigration. The intercensal 2016 estimate—that is, an
estimate that also incorporate information from the 2020 decennial census—is scheduled to be
released in 2023.39
As already discussed, our research focuses on a different population of interest than the
Census. We are interested in individuals who would file a Form 1040 if required to file a return,
excluding residents of US territories. This population includes individuals who reside outside
the US. It also includes individuals alive at any point during the year. For example, it will
include individuals who died during the year but had a return filed on their behalf. It will also
include dependents born as late as December 31.
To make our population estimate from the tax data as comparable as possible to the Census
estimate, we remove individuals we identify as living outside the US or who were not alive on
July 1 (Table 5). This includes: 0.9 million filers and dependent nonfilers with a foreign or
missing address; and 0.4 million filers, dependent nonfilers, and non-dependent nonfilers with
an overseas US armed forces address.40 It also includes 1.3 million individuals who died prior to
July 1 and 1.9 million individuals born after June 30. The resulting Census-equivalent
population estimate is 326.5 million individuals.

39

For the release schedule, see https://www.census.gov/programs-surveys/popest/about/schedule.html.

For filers and dependents, we remove them if they have an overseas armed forces, foreign, or missing address on
their Form 1040. For non-dependent nonfilers, we remove them if at least one of their information returns has an
overseas armed forces address.
40

Brady and Bass

21

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

The final adjustment we make to our data is to tabulate the data by age as of July 1 (rather
than age at the end of the year).

6.1 Possible Explanations for Differences Between Tax Data and Census Estimates
Before we examine differences between the Census-equivalent and Census population
estimates, we discuss why the two may differ. First, both are estimates and it is likely that
neither measures the US resident population without error. Beyond measurement error,
however, there are reasons that tax data may either overestimate or underestimate the US
resident population.
One reason the tax data may overestimate the population is that, despite out best efforts,
our estimate of the resident population may include US citizens or resident aliens who are
civilians living outside the US or who are members of the US armed forces stationed overseas.
As noted in both Cilke (2014) and Lurie and Pearce (2021), eliminating Form 1040 without a US
address will not necessarily remove all non-resident filers and dependent nonfilers from the
sample, as some may choose to list a US mailing address on their tax returns. Similarly,
eliminating information returns without a US address will not necessarily remove all
non-resident non-dependent nonfilers, as some may provide a US mailing address to the
entities which generate information returns.
Another reason that the tax data may overestimate the US resident population is that, in
addition to dependents who are US citizens and resident aliens, filers may claim children who
are resident in Canada or Mexico as dependents. Both Cilke (2014) and Larrimore et al. (2019)
note that some children in the tax data are attributable to filers legitimately claiming
dependents who do not reside in the US.
The primary reason the tax data may underestimate the US resident population is that there
is “… some portion of the population that is completely untouched by the Federal income tax
system.”41 In particular, one group of individuals traditionally not captured by the tax data is
individuals solely dependent on public assistance. This is because benefit payments from such

41

Cilke (2014), p. 6.

Brady and Bass

22

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

programs as Temporary Assistance to Needy Families (TANF), Supplemental Security Income
(SSI), and Veteran Affairs (VA) are not reported to the IRS—neither on tax returns nor on
information returns.
In contrast to most previous studies using tax data, we believe we capture a large share of
the population reliant solely on public assistance because—following Lurie and Pearce (2021),
and using the data collected and processed by the authors of that study—we include
individuals identified on Form 1095, which provides information on health insurance coverage.
Although our income measure will not include benefit payments from public assistance
programs, our population counts should include most, if not all, individuals who rely solely on
public assistance because they typically are covered by government provided health insurance.

6.2 Comparison of 2016 Population Estimates
Overall and by age, the two population estimates are very close. The Census-equivalent
estimate of 326.5 million is 3.4 million, or 1.1 percent, higher than the Census estimate of
323.1 million (Table 5). The two estimates also track very closely by single year of age (Figure 4).
Nonetheless, there some age ranges over which there are notable differences between the
two population estimates, especially among children (with more in the tax data) and the elderly
(with fewer in the tax data). The largest differences in numbers (239,000 more individuals per
birth year in the tax data)—and also large as a percentage of the Census estimate (5.8 percent
more in the tax data)—are among children aged 6 through 17, with the differences peaking at
ages seven through nine (averaging 7.8 percent more in the tax data). Although much smaller in
numbers, differences as a percentage of the Census estimate are large after age 75—with 3.3
percent fewer in the tax data from age 76 through 85, 6.5 percent fewer from age 86 through 95,
and 11.6 percent fewer for those aged 96 or older.
Over other age ranges, the two estimates are much closer. By single year of age, differences
between Census-equivalent and Census estimates averaged ±1.3 percent of the Census estimate
for ages 0 through 5 and ages 18 through 75. Over these ages, about one-third of single-year-ofage population estimates were within 1.0 percent and about four-in-five were within 2.0
percent.

Brady and Bass

23

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

The Census-equivalent estimate from the tax data is higher than the Census estimate for
older children and adults in their prime working years. The tax data are higher for individuals
aged five through 20 (and markedly so, as already noted, for children aged 6 through 17). The
tax data also identify more individuals aged 28 through 61, with the Census-equivalent estimate
1.3 percent higher, on average, than the Census estimate over this age range.
The Census-equivalent estimate is lower than the Census estimate for infants, young
adults, and the elderly. There are 2.0 percent fewer newborns and 0.9 percent fewer one-year
old children in the tax data than in the Census. On average, there are 1.6 percent fewer
individuals in the tax data aged 21 through 27, with differences peaking at 2.8 percent fewer of
the Census estimate at age 25. Tax data are consistently lower than the Census estimate for
individuals older than aged 67, with (as already noted) the percentage differences increasing
with age.
The two population estimates are closest for children aged two through four and adults
aged 62 through 67. The Census-equivalent estimate averages about 10,400 more per birth year
than the Census estimate for individuals aged two through four and about 16,500 more per
birth year for individuals aged 62 through 67—both differences representing less than 0.5
percent of the Census population estimate over those ages.
We do note an anomalous pattern that occurs every five years of age from age 31 (the 1985
birth-year cohort) through age 61 (the 1955 birth-year cohort) where the Census-equivalent
estimate from the tax data declines relative to the Census estimate, reducing the differences
between the two estimates.42 This pattern occurs because there are larger swings in the size of
adjacent birth-year cohorts in the Census estimate than in the tax data. For example, differences
in the number of individuals aged 37 (the 1979 birth-year cohort) and aged 35 (the 1981 birthyear cohort) are similar within the two data sources, with the age 37 population 4.8 percent
lower than the age 35 population in the tax data and 4.5 percent lower in the Census estimate. In

When comparing Census estimates to the Census-equivalent estimates from the tax data, the age in years is
calculated as of July 1, 2016. For expositional ease, in the text we are effectively assigning the same birth year to all
individuals born in the 12-month period ending on July 1 of a given year.
42

Brady and Bass

24

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

between, however, the change in cohort size differs. In the tax data, older age groups are
progressively smaller, with 1.3 percent fewer individuals aged 36 (the 1980 birth-year cohort)
than aged 35, and 3.5 percent fewer individuals aged 37 than aged 36. In the Census estimate,
by contrast, the older age groups get larger before they get smaller, with 1.9 percent more
individuals aged 36 than aged 35, and then 6.3 percent fewer individuals aged 37 than aged 36.
We also note why differences in population estimates peak for children aged seven through
nine. Both the Census-equivalent and Census estimates show that the population aged six (the
2010 birth-year cohort) is roughly 5 percent lower than the population aged 16 (the 2000 birthyear cohort). In the tax data, cohort size drops sharply for children born after 2008, with the
population aged eight and nine (the 2008 and 2007 birth-year cohorts) actually 2 percent higher,
on average, than the population aged 16. In the Census estimate, there is a more consistent drop
in cohort size between the 2000 and 2010 birth years, with the population aged eight and nine
already 2 percent lower, on average, than the population aged 16. Further, cohort size continues
to decline for children born after 2010, with 7 percent fewer newborns (the 2016 birth-year
cohort) than children aged six in the tax data, but only 2 percent fewer in the Census estimate.

6.3 Comparison with Results of Previous Studies
This section compares our results to the results in Cilke (2014) and Larrimore et al. (2019)—
two previous studies which use tax data to derive population estimates and which compare
them to the Census estimate by single year of age. Our results are qualitatively similar to the
previous studies, in that all three studies illustrate that population estimates derived from tax
data are close to those of the Census and generally track Census estimates by age. That said, our
results differ from the previous studies, particularly for children and middle-aged individuals.
Overall, we find 1.1 percent more individuals in Census-equivalent estimate from the tax
data than the Census population estimate, whereas both Cilke (2014) and Larrimore et al. (2019)
find about 0.5 percent fewer.43

Cilke (2014) finds that, for individuals aged 16 and older, the tax data represent 99.5 percent of the Census estimate.
Larrimore et al. (2019) finds that, for the total population (across all ages), the tax data represent 99.5 percent of the
Census estimate.
43

Brady and Bass

25

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

We also find different patterns by age. Both previous studies find more children younger
than age 15 in the tax data, with the tax data noticeably higher than the Census estimate for
children as young as age one. We also find more in the tax data younger than age 15, but the
overall difference is smaller than found in the previous studies and large differences do not
emerge until children are aged five or older. Both previous studies find fewer in the tax data
starting around age 15—Cilke (2014) discusses finding fewer individuals aged 16 through 24,
Larrimore et al. (2019) discusses finding fewer individuals aged 15 to 20. In contrast, we find the
Census-equivalent estimate exceeds the Census estimate through age 20. Both previous studies
also find fewer middle-aged individuals in the tax data—Cilke (2014) discusses finding fewer
aged 50 to 64, Larrimore et al. (2019) discusses finding fewer aged 40 to 55—whereas we
generally find more individuals in tax data aged 27 through 67. Finally, we find increasingly
large percentage differences beginning after age 67 with fewer individuals in the tax data than
in the Census estimate, whereas the previous studies do not highlight similar differences.
There are two primary reasons our results differ from the previous studies.44 The first is
that we utilize tax data not available when the previous studies were done, allowing us to
identify additional individuals. The second is that we compare Census-equivalent and and
Census estimates for 2016, whereas the previous two studies compare estimates for 2011 and
2010, respectively.
To disentangle the effects of data availability from the effects of the time period analyzed,
we first redo the 2016 comparison excluding the newly available tax data, and then redo that
same comparison using 2010 data.
There are two sources of tax information that allow us to identify more individuals in 2016
than could be identified in 2010. The availability of Form 1095, which was first used in the 2014
tax year to report health insurance coverage (although it was not required until later), accounts
for most of the additional individuals identified. We are also able to identify all dependents

A third reason, which is less consequential for all but the elderly, is that the comparisons between the tax and
Census data in the previous studies are slightly different than the comparison done in this paper. This issue is
discussed in the appendix.
44

Brady and Bass

26

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

claimed on electronically filed returns rather than just the first four dependents listed, but this
has a much smaller impact on our estimate.45
To account for differences in data availability, we exclude from our Census-equivalent
estimate individuals who were (1) identified only using Form 1095 or (2) identified only as a
dependent on an electronically filed return who was not among the first four dependents listed.
Excluding these individuals reduces our estimate the most for children and middle-aged adults,
with only very small effects for the elderly (Figure 5).
Overall, we no longer find more in the tax data than in the Census estimate after adjusting
the Census-equivalent estimate down to account for data availability, but we still cannot
replicate the results of the previous studies, as we now find too few (Table 5). Excluding
individuals only identified using the newly available tax data, the 2016 Census-equivalent
estimate represents 98.0 percent of the Census estimate—slightly below the 99.5 percent found
in the earlier studies.
By age, adjusting the population estimate to account for data availability accentuates the
discrepancy between our results and the previous studies for children (Figure 5). After
removing individuals identified with the newly available tax data, there are now fewer children
younger than age 15 in the tax data than in the Census estimate—considerably fewer aged six
and younger and only a bit more aged seven through 14.
To investigate how the time period analyzed affects the results, we rerun the analysis using
2010 data. We then compare the 2010 results to the 2016 results adjusted for data availability.
When comparing the 2010 and 2016 population estimates by single year of age, we
generally see similar changes in the age profile of both the Census-equivalent (Figure 6, top
panel) and the Census (Figure 6, middle panel) estimates. In both comparisons, between 2010
and 2016 we see: more individuals in their early-20s through late-30s and in their early-50s or

Cilke (2014) only includes information on up to four dependents for both paper returns and electronically filed
returns. We believe Larrimore et al. (2019) does as well. Further, our investigation of the data indicates, that even if
information on additional dependents from electronically filed returns were used in both 2010 and 2016, the results
would not be directly comparable because the share of returns filed electronically increased considerably by 2016.
45

Brady and Bass

27

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

older; fewer older teenagers and individuals in their late-30s through early-50s; and about the
same number aged eight through 16.
The exception to this rule is the change between 2010 and 2016 in the population aged
seven or younger. For this age group, the Census estimate is about the same in both years, at
just over 4.0 million per birth year, on average. In contrast, the size of this age group declines in
Census-equivalent estimate, from 4.2 million per birth year, on average, in 2010 to 3.8 million
per birth year, on average, in 2016.
The comparison of the 2010 and 2016 results illustrates that our results differ from those of
the previous studies not only because of the availability of additional data, but also because we
analyze a different tax year. Combined with the adjustments to account for the availability of
data, switching to 2010 data allow us to generally replicate the results from the previous studies
(Figure 6, bottom panel). In particular, we replicate the findings that, in 2010, there are more
individuals younger than age 15 in the Census equivalent estimate than in the Census estimate,
and that differences arise by age two. We also find fewer individuals in the tax data aged 15
through 24 in 2010. In addition, differences between the two population estimates for those
aged 85 and older were smaller in 2010 than they were in 2016, possibly explaining why they
were not highlighted in the previous studies. The differences between the 2010 and 2016 results,
however, illustrate that the 2010 differences between the Census-equivalent and the Census
estimates by age do not generalize to other years.
Rather than being consistent across time by age, we find differences between tax data and
Census estimates are generally consistent across time by birth-year. When comparing the 2010
and 2016 population estimates by year of birth, we again see similar changes in the birth-year
profile of both the Census-equivalent (Figure 7, top panel) and Census (Figure 7, middle panel)
estimates. In both estimates between 2010 and 2016, the 1972 and later birth-year cohorts (aged
38 or younger in 2010) increase in size (implying net migration more than offsets deaths) and
the 1971 or earlier birth-year cohorts (aged 39 or older in 2010) decrease in size (implying net
migration does not fully offset deaths). As a result, the 2010 and 2016 differences between the
tax and Census population estimates, expressed as a percentage of the Census population

Brady and Bass

28

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

estimate, are more highly correlated by birth-year (Figure 7, bottom panel) than they are by age
(Figure 6, bottom panel).
Although birth-year differences in the two population estimates are correlated across time,
there are three birth-year ranges over which the 2010 and 2016 results diverge consistently.
Between 2010 and 2016, the Census-equivalent estimate from the tax data declines relative to
the Census estimate for two birth-year groups: the 1989 through 2008 birth-year cohorts (aged
two through 21 in 2010), and the 1944 and earlier cohorts (aged 66 or older in 2010). The Censusequivalent estimate increases relative to the Census estimate between 2010 and 2016 for only
one birth-year group: the 1949 through 1954 birth-year cohort (aged 56 through 61 in 2010).
Of these changes, the easiest to explain is the increase relative to the Census estimate for the
1949 through 1954 birth-year cohorts. The age of these birth-year cohorts increased from 56
through 61 in 2010 to 62 through 67 in 2016. As discussed above and illustrated in Figure 2, in
2016 the share of the population identified using only Form 1095 declines sharply from 3.7
percent for individuals aged 61 to 0.9 percent for individuals aged 67. Thus, adjusted for data
availability, the increase in the Census-equivalent estimate relative to the Census estimate for
these birth cohorts between 2010 and 2016 presumably is because we identify a higher share of
these individuals in 2016 using non-Form-1095 tax data.
The decline relative to the Census estimate for the 1944 and earlier birth-year cohorts is the
result of relatively small differences between the tax data and the Census estimate in population
changes (net migration minus deaths) between 2010 and 2016. For example, the age of the 1917
through 1944 birth-year cohorts increased from 66 through 93 in 2010 to 72 through 99 in 2016.
Over this period, their population falls by 9.85 million in the tax data and by 9.80 million in the
Census estimate—a difference of only 0.5 percent. In combination with existing 2010
differences, however, 2016 differences represent a much larger percentage of the shrinking
population for these birth-year cohorts.
We do not have a ready explanation for the decline relative to the Census estimate for the
1989 through 2008 birth cohorts, and further investigation is beyond the scope of this research.
The fact that the relative decline happened for children aged two through 21 in 2010 suggests

Brady and Bass

29

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

that the decline is associated with dependents. What is less clear is why the relative decline
would occur for children young enough to be claimed as a dependent in both 2010 and 2016.
We do note, however, that—despite the decline in the tax data relative to the Census estimate—
2016 differences between the two estimates follow a similar pattern by birth year as the 2010
differences.

6.4 Discussion
To get a sense of how well the tax data represent our population of interest, we compare it
to Census estimates of the US resident population. Although our sample of interest includes
individuals living outside the US, US residents represent the bulk of our population. Therefore,
we adjust our data as best we can to conform to the Census population concept and then make
the comparison.
Consistent with previous research, the two total population estimates are similar. That said,
most previous studies find slightly fewer in the tax data than in the Census estimate, whereas
we find slightly more. This is because, consistent with Lurie and Pearce (2021), we include
individuals who are only identified on Form 1095.
Also consistent with the previous research, we find the tax data generally track the Census
estimate closely by age, but that there are differences over some age ranges. That said, we find
differences over slightly different age ranges than the previous studies.
Some of the differences between the tax data and the Census estimate could be attributable
to the tax data including non-residents even after our best efforts to remove them. For example,
compared with the Census estimate, the 2016 Census-equivalent estimate from the tax data is
higher for older children and prime working age adults. These differences would be consistent
with the explanation that some individuals work and live outside the US but, for tax purposes,
use a US mailing address for themselves and their dependents. It would also be consistent with
the explanation that some filers who are US residents validly claim dependents who are
Mexican or Canadian residents.
Our results, however, suggest some caution when suggesting explanations for the
differences between the tax data and the Census estimate. This is because, across years,

Brady and Bass

30

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

differences between the tax data and the Census estimate appear to be more consistent by birth
year than by age. For example, Larrimore et al. (2019) suggests there are fewer children older
than 16 in the tax data because they no longer qualify for the child tax credit and are, thus, less
likely to be claimed as a dependent on a return. Although that explanation is consistent with the
2010 data the study analyzes, it is not consistent with our 2016 results—which show more
children in the tax data through age 20.
To the extent that differences between the tax data and the Census estimate are caused by
the inclusion of non-residents in our Census-equivalent estimates, it is not of great concern to
our broader research agenda. Although we attempt to remove non-residents to facilitate the
comparison with the Census estimate, we include them in the data we analyze.
This exercise illustrates that the tax data appear to be representative of the population we
wish to study. The Census-equivalent population estimate from the tax data is reasonably close
to Census estimate of the US resident population. Differences between the two estimates are not
unexpected given that the tax data are not particularly well suited to measuring the US resident
population and given that both the tax data and the Census estimate likely measure the
population with error.

7. Conclusion
This paper was written to document our method of building a representative sample of the
US population from tax data, which represents the first step in a larger project measuring
changes in the amount and composition of income over the life cycle. We document our method
both for readers of our broader research, so they understand how we derive our estimates, and
for other researchers analyzing tax data, so that they can use or improve upon our methods.
To build a sample representative of the population, we supplement tax return data—which
allows us to identify filers and dependent nonfilers—with information return data—which
allows us to identify non-dependent nonfilers. Although non-dependent nonfilers are not a
large part of the overall population, their importance increases with age, making their inclusion
crucial for measuring elderly income.

Brady and Bass

31

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Inclusive of both filers and nonfilers, we find the share of the population with income
remains fairly steady after age 26 and actually increases slightly after age 61. For those with
income, the median amount follows a hump shape over the lifecycle, peaking at age 46.
Many of our findings are similar to those in the existing literature. This is not terribly
surprising given that our work builds on the work of many, especially Cilke (2014) and Lurie
and Pearce (2021). For example, as with previous studies, we find that the tax data produce a
population estimate similar to that of the Census, and that the two estimates track closely by
age. We conclude from this comparison that the tax data adequately captures our population of
interest.
Although our work builds on the existing literature, we note several innovations. First, we
sample individuals rather than tax returns. Second, we use the individual—rather than the tax
return, family, or household—as the unit of observation. All income reported on a tax return is
allocated to filers—primary taxpayers in the case of non-joint returns and primary and
secondary filers in the case of joint returns. We do not adjust filer income to account for the
number of dependents they claim. Further, dependents have income only if they file their own
tax return or have income reported on information returns. Our method allows us to examine
individual income by single year of age—which is difficult, if not impossible, to do for tax
returns or families.
Although we cannot claim credit for the innovation, we do note that—following Lurie and
Pearce (2021)—we include in our sample individuals identified only on Form 1095, which
reports health insurance coverage. There are some portions of the population that tax data
traditionally does not identify, such as individuals whose only source of income is public
assistance. We believe that the Form 1095 allows us to identify most individuals solely
dependent on public assistance because they typically are covered by government provided
health insurance.
Finally, our examination of 2010 and 2016 data show that, across years, differences between
the tax data and the Census population estimate are more correlated with birth year than they
are by age. This is important because many explanations for why the tax data would

Brady and Bass

32

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

overestimate or underestimate the true population are related to the age of individuals. Our
work suggests that explanations for differences between the two estimates should also be
consistent with differences by birth cohort that persist over time.

Brady and Bass

33

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

References
Bee, Adam C. and Joshua Mitchell. 2017. “Do Older Americans Have More Income Than We
Think?” 2017. SEHSD Working Paper No. 2017-39. Washington, DC: U.S. Census Bureau Social,
Economic, and Housing Statistics Division. Available at
https://www.census.gov/content/dam/Census/library/working-papers/2017/demo/SEHSDWP2017-39.pdf
Brady, Peter J. and Steven Bass. 2021. “Comparing the Current Population Survey to Income
Tax Data.” SOI Working Paper. Washington, DC: Internal Revenue Service. Available at
https://www.irs.gov/pub/irs-soi/21rpcomparingcpstoincometaxdata.pdf.
Brady, Peter J. and Steven Bass. 2020a. “Reconciling Form 1040 and Form 1099-R Data.” SOI
Working Paper. Washington, DC: Internal Revenue Service. Available at
https://www.irs.gov/pub/irs-soi/20rpreconciling10401099R.pdf.
Brady, Peter J. and Steven Bass. 2020b. “Decoding Retirement: A Detailed Look at Retirement
Distributions Reported on Tax Returns.” SOI Working Paper. Washington, DC: Internal Revenue
Service. Available at https://www.irs.gov/pub/irs-soi/20rpdecodingretirement.pdf.
Brady, Peter, Steven Bass, Jessica Holland, and Kevin Pierce. 2017. “Using Panel Tax Data to
Examine the Transition to Retirement.” SOI Working Paper. Washington, DC: Internal Revenue
Service. Available at https://www.irs.gov/pub/irs-soi/17rptransitionretirement.pdf.
Brady, Peter and Kevin Pierce. 2012. “The Promise and Potential Pitfalls of Using
Administrative Tax Data: A Case Study.” Unpublished manuscript, Investment Company
Institute (April).
Cilke, James. 2014. The Case of Missing Strangers: What We Know and Don’t Know About
Non-Filers” Proceedings of the 107th Annual Conference on Taxation. Washington, DC: National Tax
Association. Available at https://www.ntanet.org/wp-content/uploads/proceedings/2014/029cilke-case-missing-strangers-know-don.pdf.
Cilke, James. 1998. “A Profile of Non-Filers.” OTA Working Paper 78. Washington, DC: US
Department of the Treasury. Available at https://home.treasury.gov/system/files/131/WP-78.pdf
Congressional Budget Office. 2016. The Distribution of Household Income and Federal Taxes, 2013.
Washington, DC: Congressional Budget Office. Available at
https://www.cbo.gov/publication/51361
Cronin, Julie-Anne. 2022. “U.S Treasury Distributional Analysis Methodology.” OTA Technical
Paper 8. Washington, DC: US Department of the Treasury. Available at
https://home.treasury.gov/system/files/131/TP-8.pdf.

Brady and Bass

34

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Cronin, Julie-Anne. 1999. “U.S Treasury Distributional Analysis Methodology.” OTA Working
Paper 85. Washington, DC: US Department of the Treasury. Available at
https://home.treasury.gov/system/files/131/WP-85.pdf.
Internal Revenue Service, Statistics of Income Division. 2018. Individual Income Tax Returns 2016,
Publication 1304. Washington, DC: Internal Revenue Service. Available at
https://www.irs.gov/pub/irs-soi/16inalcr.pdf.
Internal Revenue Service. 2017a. U.S. Tax Guide for Aliens: For Use in Preparing 2016 Returns,
Publication 519. Washington, DC: Internal Revenue Service. Available at
https://www.irs.gov/pub/irs-prior/p519--2016.pdf.
Internal Revenue Service. 2017b. Tax Guide for Seniors: For Use in Preparing 2016 Returns,
Publication 554. Washington, DC: Internal Revenue Service. Available at
https://www.irs.gov/pub/irs-prior/p554--2016.pdf.
Internal Revenue Service. 2017c. Tax Guide for Individuals with Income from U.S. Possessions for Use
in Preparing 2016 Returns, Publication 570. Washington, DC: Internal Revenue Service. Available
at https://www.irs.gov/pub/irs-prior/p570--2016.pdf.
Internal Revenue Service. 2016a. Tax Guide for U.S. Citizens and Resident Aliens Abroad: For Use in
Preparing 2016 Returns, Publication 54. Washington, DC: Internal Revenue Service. Available at
https://www.irs.gov/pub/irs-prior/p54--2016.pdf.
Internal Revenue Service. 2016b. Exemptions, Standard Deduction, and Filing Information: For Use
in Preparing 2016 Returns, Publication 501. Washington, DC: Internal Revenue Service. Available
at https://www.irs.gov/pub/irs-prior/p501--2016.pdf.
Internal Revenue Service. 2016c. 1040 Instructions 2010. Washington, DC: Internal Revenue
Service. Available at https://www.irs.gov/pub/irs-prior/i1040gi--2016.pdf.
Joint Committee on Taxation. 1993. “Methodology and Issues in Measuring Changes in The
Distribution of Tax Burdens.” JCS-7-93. Available at https://www.jct.gov/publications/1993/jcs7-93/.

Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2019. “Household Incomes in Tax
data: Using Addresses to Move from Tax Unit to Household Income Distributions.” Journal
of Human Resources. Advance online publication. Available at
http://jhr.uwpress.org/content/early/2019/07/02/jhr.56.2.0718-9647R1.abstract.
Lawrence, Joshua, Michael Udell, and Tiffany Young. 2011. “The Income Tax Position of
Persons Not Filing Income Tax Returns for Tax Year 2005.” In Alan Plumley ed. Recent
Research on Tax Administration and Compliance. IRS Research Bulletin. Washington, DC:
Internal Revenue Service. Available at https://www.irs.gov/pub/irs-soi/11rescontaxposition.pdf.

Brady and Bass

35

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Lundberg, Shelly, Robert A. Pollak, and Jenna Stearns. 2016. "Family Inequality: Diverging
Patterns in Marriage, Cohabitation, and Childbearing." Journal of Economic Perspectives, 30 (2):
79-102. Available at https://www.aeaweb.org/articles?id=10.1257/jep.30.2.79.
Lurie, Ithai Z. and James Pearce. 2021. “Health Insurance Coverage in Tax and Survey Data.”
American Journal of Health Economics, 7(2): 164-184. Available at
https://www.journals.uchicago.edu/toc/ajhe/2021/7/2.
Mortenson, Jacob A., James Cilke, Michael Udell and Jonathan Zytnick. 2009. “Attaching the
Left Tail: A New Profile of Income for Persons Who Do Not Appear on Federal Income Tax
Returns.” Proceedings of the 102nd Annual Conference on Taxation. Washington, DC: National Tax
Association. Available at https://ntanet.org/wp-content/uploads/proceedings/2009/011mortenson-attaching-left-tail-2009-nta-proceedings.pdf
Nunns, James R., Deena Ackerman, James Cilke, Julie-Anne Cronin, Janet H. Janet Holtzblatt,
Gillian Hunter Emily Lin, and Janet McCubbin. 2008. “Treasury’s Panel Model for Tax
Analysis.” OTA Technical Papers 3. Available at https://home.treasury.gov/system/files/131/TP3.pdf.
Saez, Emmanuel. 2016. “Statistics of Income Tabulations: High Incomes, Gender, Age, Earnings
Split, and Non-filers” SOI Working Paper. Washington, DC: Internal Revenue Service. Available
at https://www.irs.gov/pub/irs-soi/16rpsaeztabulations.pdf.
Sailer, Peter and Michael Weber. 1998. “The IRS Population Count: An Update.“ Washington,
DC: Statistics of Income Division, Internal Revenue Service. Available at:
https://www.irs.gov/pub/irs-soi/indpopct.pdf.
Semega, Jessica, Melissa Kollar, John Creamer, and Abinash Mohanty. 2019. “Income and
Poverty in the United States: 2018.” Current Population Reports, P60-266. Washington, DC: U.S.
Census Bureau. Available at
https://www.census.gov/content/dam/Census/library/publications/2019/demo/p60-266.pdf.
U.S. Census Bureau. 2021. Annual Estimates of the Resident Population by Single Year of Age
and Sex. Available at: https://www.census.gov/programs-surveys/popest/technicaldocumentation/research/evaluation-estimates/2020-evaluation-estimates/2010s-nationaldetail.html.

Brady and Bass

36

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞϭ
&ŝůĞƌ^ŚĂƌĞĞĐůŝŶĞƐǁŝƚŚŐĞ͕ďƵƚ^ŚĂƌĞǁŝƚŚ/ŶĐŽŵĞZĞŵĂŝŶƐ^ƚĞĂĚLJ

WŽƉƵůĂƚŝŽŶďLJƐĂŵƉůĞĐŽŵƉŽŶĞŶƚĂŶĚƉƌĞƐĞŶĐĞŽĨŝŶĐŽŵĞ͕ƚĂdžͲLJĞĂƌϮϬϭϲ

ϱ

ϭϬϬ

EŽŶͲĚĞƉĞŶĚĞŶƚŶŽŶĨŝůĞƌƐ

ϴϬ
WĞƌĐĞŶƚĂŐĞŽĨŝŶĚŝǀŝĚƵĂůƐ

DŝůůŝŽŶƐŽĨŝŶĚŝǀŝĚƵĂůƐ

ϰ

ĞƉĞŶĚĞŶƚ
ŶŽŶĨŝůĞƌƐ
ϯ

&ŝůĞƌƐ
Ϯ

ϭ

EŽŶͲĚĞƉĞŶĚĞŶƚŶŽŶĨŝůĞƌƐ
ĞƉĞŶĚĞŶƚ
ŶŽŶĨŝůĞƌƐ

ϲϬ

&ŝůĞƌƐ
ϰϬ

ϮϬ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱ

ϭϬϬ

ϰ

ϴϬ

ϯ

tŝƚŚŽƵƚ
ŝŶĐŽŵĞ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

WĞƌĐĞŶƚĂŐĞŽĨŝŶĚŝǀŝĚƵĂůƐ

DŝůůŝŽŶƐŽĨŝŶĚŝǀŝĚƵĂůƐ

ŐĞ

tŝƚŚŝŶĐŽŵĞ

Ϯ

ϭ

tŝƚŚŽƵƚ
ŝŶĐŽŵĞ

ϲϬ

tŝƚŚŝŶĐŽŵĞ

ϰϬ

ϮϬ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

Ϭ

ŐĞ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ŐĞ

^ŽƵƌĐĞ͗ƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ

Brady and Bass

37

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞϮ
zŽƵŶŐĞƌ/ŶĚŝǀŝĚƵĂůƐDŽƌĞ>ŝŬĞůLJƚŽĞ/ĚĞŶƚŝĨŝĞĚǁŝƚŚ&ŽƌŵϭϬϵϱKŶůLJ

WŽƉƵůĂƚŝŽŶďLJŵĞƚŚŽĚŽĨŝĚĞŶƚŝĨŝĐĂƚŝŽŶ͕ƚĂdžͲLJĞĂƌϮϬϭϲ
EŽŶͲĚĞƉĞŶĚĞŶƚŶŽŶĨŝůĞƌƐŽŶůLJŝĚĞŶƚŝĨŝĞĚŽŶ&ŽƌŵϭϬϵϱ

ϱ

DŝůůŝŽŶƐŽĨŝŶĚŝǀŝĚƵĂůƐ

ϰ

ϯ

EŽŶͲĚĞƉĞŶĚĞŶƚŶŽŶĨŝůĞƌƐ
ŝĚĞŶƚŝĨŝĞĚǁŝƚŚŶŽŶͲ&ŽƌŵͲϭϬϵϱ
ŝŶĨŽƌŵĂƚŝŽŶƌĞƚƵƌŶ

/ĚĞŶƚŝĨŝĞĚƵƐŝŶŐ&ŽƌŵϭϬϰϬ
;&ŝůĞƌƐĂŶĚĚĞƉĞŶĚĞŶƚƐͿ
Ϯ

ϭ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

EŽŶͲĚĞƉĞŶĚĞŶƚŶŽŶĨŝůĞƌƐŽŶůLJŝĚĞŶƚŝĨŝĞĚŽŶ&ŽƌŵϭϬϵϱ
ϭϬϬ

EŽŶͲĚĞƉĞŶĚĞŶƚŶŽŶĨŝůĞƌƐ
ŝĚĞŶƚŝĨŝĞĚǁŝƚŚŶŽŶͲ&ŽƌŵͲϭϬϵϱ
ŝŶĨŽƌŵĂƚŝŽŶƌĞƚƵƌŶ

WĞƌĐĞŶƚĂŐĞŽĨŝŶĚŝǀŝĚƵĂůƐ

ϴϬ

ϲϬ

/ĚĞŶƚŝĨŝĞĚƵƐŝŶŐ&ŽƌŵϭϬϰϬ
;&ŝůĞƌƐĂŶĚĚĞƉĞŶĚĞŶƚƐͿ
ϰϬ

ϮϬ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

^ŽƵƌĐĞ͗ƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ

Brady and Bass

38

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞϯ
ŽŵƉŽƐŝƚŝŽŶŽĨ&ŝůŝŶŐͲdLJƉĞ'ƌŽƵƉƐŚĂŶŐĞƐŽǀĞƌƚŚĞ>ŝĨĞLJĐůĞ

WŽƉƵůĂƚŝŽŶĂŶĚŝŶĐŽŵĞďLJĨŝůŝŶŐƚLJƉĞ͕ƚĂdžͲLJĞĂƌϮϬϭϲ
DŝůůŝŽŶƐŽĨŝŶĚŝǀŝĚƵĂůƐ
ϱ

EŽŶĨŝůĞƌ

ϰ

ϯ

EŽŶͲũŽŝŶƚĨŝůĞƌ
Ϯ

:ŽŝŶƚĨŝůĞƌ

ϭ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

WĞƌĐĞŶƚĂŐĞŽĨŝŶĚŝǀŝĚƵĂůƐ
ϭϬϬ

ϴϬ

EŽŶĨŝůĞƌ
ϲϬ

EŽŶͲũŽŝŶƚĨŝůĞƌ

ϰϬ

ϮϬ

:ŽŝŶƚĨŝůĞƌ
Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ
ŐĞ

DĞĚŝĂŶƉĞƌĐĂƉŝƚĂƚŽƚĂůŝŶĐŽŵĞ;ƚŚŽƵƐĂŶĚƐŽĨĚŽůůĂƌƐͿΎ
ϱϬ

ůů
:ŽŝŶƚĨŝůĞƌ
EŽŶͲũŽŝŶƚĨŝůĞƌ
EŽŶĨŝůĞƌ

ϰϬ

ϯϬ

ϮϬ

ϭϬ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ
ŐĞ

ΎDĞĚŝĂŶƉĞƌĐĂƉŝƚĂƚŽƚĂůŝŶĐŽŵĞŝƐĐĂůĐƵůĂƚĞĚĨŽƌŝŶĚŝǀŝĚƵĂůƐĂůŝǀĞĂƚLJĞĂƌͲĞŶĚϮϬϭϲ͘
^ŽƵƌĐĞ͗ƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ

Brady and Bass

39

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞϰ
ĞŶƐƵƐͲƋƵŝǀĂůĞŶƚWŽƉƵůĂƚŝŽŶƐƚŝŵĂƚĞĨƌŽŵdĂdžĂƚĂůŽƐĞůLJdƌĂĐŬƐĞŶƐƵƐƐƚŝŵĂƚĞ

WŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞĂƐŽĨ:ƵůLJϭ͕ϮϬϭϲ

ϱ͕ϬϬϬ͕ϬϬϬ

ĞŶƐƵƐƉŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞ
ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶ

ϰ͕ϬϬϬ͕ϬϬϬ

ϯ͕ϬϬϬ͕ϬϬϬ

Ϯ͕ϬϬϬ͕ϬϬϬ

ϭ͕ϬϬϬ͕ϬϬϬ

Ϭ
Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ϵϬ

ϭϬϬн

ŐĞ

ŝĨĨĞƌĞŶĐĞ;ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚ

ĞŶƐƵƐͿĂƐĂƉĞƌĐĞŶƚĂŐĞŽĨĞŶƐƵƐƉŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞ

ϭϬ

Ϭ

ͲϭϬ

ͲϮϬ
Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ŐĞ

^ŽƵƌĐĞƐ͗h͘^͘ĞŶƐƵƐƵƌĞĂƵϮϬϮϬĂŶĚĂƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ

Brady and Bass

40

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞϱ
ĚĚŝƚŝŽŶŽĨ&ŽƌŵϭϬϵϱĂŶĚdžƚƌĂĞƉĞŶĚĞŶƚƐ,ĂƐ>ĂƌŐĞƐƚĨĨĞĐƚŽŶEŽŶͲůĚĞƌůLJƐƚŝŵĂƚĞƐ

ŝĨĨĞƌĞŶĐĞďĞƚǁĞĞŶĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚĂŶĚĞŶƐƵƐĞƐƚŝŵĂƚĞƐĂƐĂƉĞƌĐĞŶƚĂŐĞŽĨĞŶƐƵƐ͕ϮϬϭϲ

ϭϬ

ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶ

ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶ͕ĞdžĐůƵĚŝŶŐ&ŽƌŵϭϬϵϱaŽŶůLJĂŶĚĂĚĚŝƚŝŽŶĂůĚĞƉĞŶĚĞŶƚƐb

Ϭ

ͲϭϬ

ͲϮϬ
Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

a&ŽƌŵϭϬϵϱŝŶĐůƵĚĞƐŝŶĨŽƌŵĂƚŝŽŶĂďŽƵƚŚĞĂůƚŚŝŶƐƵƌĂŶĐĞĐŽǀĞƌĂŐĞƚŚƌŽƵŐŚƚŚĞ,ĞĂůƚŚ/ŶƐƵƌĂŶĐĞDĂƌŬĞƚƉůĂĐĞ
;&ŽƌŵϭϬϵϱͲͿ͕ŚĞĂůƚŚŝŶƐƵƌĂŶĐĞĐŽǀĞƌĂŐĞƚŚƌŽƵŐŚŽƚŚĞƌƉƌŝǀĂƚĞŽƌŐŽǀĞƌŶŵĞŶƚƐŽƵƌĐĞƐ;&ŽƌŵϭϬϵϱͲͿ͕ĂŶĚ
ĂŶŽĨĨĞƌŽĨŚĞĂůƚŚŝŶƐƵƌĂŶĐĞĐŽǀĞƌĂŐĞďLJĐĞƌƚĂŝŶĞŵƉůŽLJĞƌƐ;&ŽƌŵϭϬϵϱͲͿ͘
bĚĚŝƚŝŽŶĂůĚĞƉĞŶĚĞŶƚƐĂƌĞƚŚŽƐĞŶŽƚůŝƐƚĞĚĂƐŽŶĞŽĨƚŚĞĨŝƌƐƚĨŽƵƌĚĞƉĞŶĚĞŶƚƐŽŶ&ŽƌŵϭϬϰϬ͘

EŽƚĞ͗WŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞƐĂƌĞĂƐŽĨ:ƵůLJϭ͕ϮϬϭϲ͘
^ŽƵƌĐĞƐ͗h͘^͘ĞŶƐƵƐƵƌĞĂƵϮϬϮϬĂŶĚĂƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ

Brady and Bass

41

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞϲ
ŝĨĨĞƌĞŶĐĞƐďLJŐĞƌĞ/ŶĐŽŶƐŝƐƚĞŶƚĐƌŽƐƐzĞĂƌƐ

WŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞĂƐŽĨ:ƵůLJϭ
ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶ

a

ϱ͕ϬϬϬ͕ϬϬϬ

ϮϬϭϬ
ϮϬϭϲ
ϰ͕ϬϬϬ͕ϬϬϬ

ϯ͕ϬϬϬ͕ϬϬϬ

Ϯ͕ϬϬϬ͕ϬϬϬ

ϭ͕ϬϬϬ͕ϬϬϬ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

ĞŶƐƵƐƉŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞ
ϱ͕ϬϬϬ͕ϬϬϬ

ϮϬϭϬ
ϮϬϭϲ
ϰ͕ϬϬϬ͕ϬϬϬ

ϯ͕ϬϬϬ͕ϬϬϬ

Ϯ͕ϬϬϬ͕ϬϬϬ

ϭ͕ϬϬϬ͕ϬϬϬ

Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

ŝĨĨĞƌĞŶĐĞ;ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚ

ĞŶƐƵƐͿĂƐĂƉĞƌĐĞŶƚĂŐĞŽĨĞŶƐƵƐƉŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞ

ϭϬ

ϮϬϭϬ
ϮϬϭϲ
Ϭ

ϭϬ

ϮϬ
Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

a/ŶƚŚŝƐĨŝŐƵƌĞ͕ƚŚĞĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶĞdžĐůƵĚĞƐŝŶĚŝǀŝĚƵĂůƐŝĚĞŶƚŝĨŝĞĚŽŶůLJƵƐŝŶŐ&ŽƌŵϭϬϵϱ;ŝŶĐůƵƐŝǀĞ
ŽĨ&ŽƌŵϭϬϵϱͲ͕&ŽƌŵϭϬϵϱͲ͕ĂŶĚ&ŽƌŵϭϬϵϱͲͿŽƌŽŶůLJĂƐƚŚĞĨŝĨƚŚŽƌŚŝŐŚĞƌĚĞƉĞŶĚĞŶƚŽŶ&ŽƌŵϭϬϰϬ͘
^ŽƵƌĐĞƐ͗h͘^͘ĞŶƐƵƐƵƌĞĂƵϮϬϮϬĂŶĚĂƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ
Brady and Bass

42

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞϳ
WĂƚƚĞƌŶ/ƐDŽƌĞŽŶƐŝƐƚĞŶƚďLJzĞĂƌŽĨŝƌƚŚ

WŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞĂƐŽĨ:ƵůLJϭ
ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶ

a

ϱ͕ϬϬϬ͕ϬϬϬ

ϮϬϭϬ
ϮϬϭϲ
ϰ͕ϬϬϬ͕ϬϬϬ

ϯ͕ϬϬϬ͕ϬϬϬ

Ϯ͕ϬϬϬ͕ϬϬϬ

ϭ͕ϬϬϬ͕ϬϬϬ

ϮϬϭϲ

ϮϬϭϭ

ϮϬϬϲ

ϮϬϬϭ

ϭϵϵϲ

ϭϵϵϭ

ϭϵϴϲ

ϭϵϴϭ

ϭϵϳϲ

ϭϵϳϭ

ϭϵϲϲ

ϭϵϲϭ

ϭϵϱϲ

ϭϵϱϭ

ϭϵϰϲ

ϭϵϰϭ

ϭϵϯϲ

ϭϵϯϭ

ϭϵϮϲ

ϭϵϮϭ

ϭϵϭϲ

ϭϵϭϭ

zĞĂƌŽĨďŝƌƚŚ

ĞŶƐƵƐƉŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞ
ϱ͕ϬϬϬ͕ϬϬϬ

ϮϬϭϬ
ϮϬϭϲ
ϰ͕ϬϬϬ͕ϬϬϬ

ϯ͕ϬϬϬ͕ϬϬϬ

Ϯ͕ϬϬϬ͕ϬϬϬ

ϭ͕ϬϬϬ͕ϬϬϬ

ϮϬϭϲ

ϮϬϭϭ

ϮϬϬϲ

ϮϬϬϭ

ϭϵϵϲ

ϭϵϵϭ

ϭϵϴϲ

ϭϵϴϭ

ϭϵϳϲ

ϭϵϳϭ

ϭϵϲϲ

ϭϵϲϭ

ϭϵϱϲ

ϭϵϱϭ

ϭϵϰϲ

ϭϵϰϭ

ϭϵϯϲ

ϭϵϯϭ

ϭϵϮϲ

ϭϵϮϭ

ϭϵϭϲ

ϭϵϭϭ

zĞĂƌŽĨďŝƌƚŚ

ŝĨĨĞƌĞŶĐĞ;ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚ

ĞŶƐƵƐͿĂƐĂƉĞƌĐĞŶƚĂŐĞŽĨĞŶƐƵƐƉŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞ

ϭϬ

ϮϬϭϬ
ϮϬϭϲ
Ϭ

ϭϬ

ϮϬ
ϮϬϭϲ

ϮϬϭϭ

ϮϬϬϲ

ϮϬϬϭ

ϭϵϵϲ

ϭϵϵϭ

ϭϵϴϲ

ϭϵϴϭ

ϭϵϳϲ

ϭϵϳϭ

ϭϵϲϲ

ϭϵϲϭ

ϭϵϱϲ

ϭϵϱϭ

ϭϵϰϲ

ϭϵϰϭ

ϭϵϯϲ

ϭϵϯϭ

ϭϵϮϲ

ϭϵϮϭ

ϭϵϭϲ

ϭϵϭϭ

zĞĂƌŽĨďŝƌƚŚ

a/ŶƚŚŝƐĨŝŐƵƌĞ͕ƚŚĞĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶĞdžĐůƵĚĞƐŝŶĚŝǀŝĚƵĂůƐŝĚĞŶƚŝĨŝĞĚŽŶůLJƵƐŝŶŐ&ŽƌŵϭϬϵϱ;ŝŶĐůƵƐŝǀĞ
ŽĨ&ŽƌŵϭϬϵϱͲ͕&ŽƌŵϭϬϵϱͲ͕ĂŶĚ&ŽƌŵϭϬϵϱͲͿŽƌŽŶůLJĂƐƚŚĞĨŝĨƚŚŽƌŚŝŐŚĞƌĚĞƉĞŶĚĞŶƚŽŶ&ŽƌŵϭϬϰϬ͘
^ŽƵƌĐĞƐ͗h͘^͘ĞŶƐƵƐƵƌĞĂƵϮϬϮϬĂŶĚĂƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ
Brady and Bass

43

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Table 1

Derivation of the Taxpayer Component of Sample
Population estimate of primary and secondary taxpayers filing any 2016 tax return (thousands)
Primary Secondary
Total
filers
filers taxpayers
151,268
54,309
205,577
Initial sample (Taxpayers who filed 2016 Form 1040,a
Form 1040-NR, or Form 1040-SS/1040-PR)
─ Died before January 1, 2016
1
1
2
─ Filed Form 1040-NR
748
0
748
─ Filed Form 1040-SS/1040-PR
178
63
241
a
─ Filed Form 1040 with a US territory address
68
26
94
= Final taxpayer sample
150,273
54,219
204,492
Memo: Composition of final taxpayer sample
Taxpayers on non-dependent returns
Taxpayers on dependent returns

195,090
9,402

Taxpayers with a state address
Taxpayers with an overseas US armed forces address
Taxpayers with a foreign or missing address
Memo: Most common types of information returns among taxpayers
Any Information return
Form 1095b
Form W-2
Form 1099-G
Form 5498
Form 1099-INT
Form 1098
Form 1099-R
Form SSA-1099
Form 1099-DIV
Form 1099-MISC
a

203,566
239
687
200,802
182,447
147,828
67,417
60,275
54,298
51,972
48,724
39,403
36,234
24,509

The term "Form 1040" is inclusive of Form 1040, Form 1040-A, or Form 1040-EZ.

b

Form 1095 includes information about health insurance coverage through the Health Insurance Marketplace (Form
1095-A), health insurance coverage through other private or government sources (Form 1095-B), and an offer of health
insurance coverage by certain employers (Form 1095-C).
Source: Author's tabulation of IRS data

Brady and Bass

44

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Table 2

Derivation of the Dependent Nonfiler Component of Sample
Population estimate of dependents a claimed on any 2016 tax return (thousands)
Initial sample (Dependentsa claimed on 2016 Form 1040,b
Form 1040-NR, or Form 1040-SS/1040-PR)
─ Died before January 1, 2016
─ Claimed on Form 1040-NR
─ Claimed on Form 1040-SS/1040-PR
b
─ Claimed on Form 1040 with a US territory address
b
─ Filed a Form 1040 dependent return
= Final dependent nonfiler sample
Memo: Composition of final dependent nonfiler sample
With income
Without income
Form 1095c only
At least one form other than Form 1095c
No information return
Dependent nonfilers with a state address
Dependent nonfilers with an overseas US armed forces address
Dependent nonfilers with a foreign or missing address

Dependents
93,766
9
25
171
41
9,285
84,235
13,337
70,898
59,389
2,504
9,005
83,854
121
260

Memo: Most common types of information returns among dependent nonfilers
Any information return
75,230
Form 1095c
74,075
Form SSA-1099
4,860
Form W-2
4,765
Form 1098-T
3,621
Form 1099-INT
2,770
Form 1099-DIV
2,384
Form 1099-MISC
600
Form 1099-B
526
Form 1099-G
518
Form 1099-R
450
a

Data include up to four dependents claimed on paper returns and all depdendents claimed on electronic
returns.
b

The term "Form 1040" is inclusive of Form 1040, Form 1040-A, or Form 1040-EZ.

c

Form 1095 includes information about health insurance coverage through the Health Insurance Marketplace
(Form 1095-A), health insurance coverage through other private or government sources (Form 1095-B), and an
offer of health insurance coverage by certain employers (Form 1095-C).
Source: Author's tabulation of IRS data

Brady and Bass

45

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Table 3

Derivation of the Non-Dependent Nonfiler Component of Sample
Number of individuals who are not identified on any 2016 tax return but who have at least one
2016 information return (thousands)
Other
nonfilers
a
50,077
Initial sample (Non-dependent nonfilers with at least
one 2016 information return)
─ Died before January 1, 2016
4,285
─ All information returns sent to US territory addresses
1,782
b
─ All information returns sent to foreign, missing, or US territory addresses
1,569
c
─ Presence of information return indicating foreign person
73
= Final non-dependent nonfiler sample
42,368
Memo: Composition of final non-dependent nonfiler sample
With income
Without income
Form 1095d only
At least one form other than Form 1095d
Non-dependent nonfilers without an overseas US armed forces address
Non-dependent nonfilers with an overseas US armed forces address

30,879
11,489
9,251
2,238
42,351
17

Memo: The most common types of information returns among non-dependent nonfilers
Any information return
42,368
c
Form 1095
36,500
Form SSA-1099
18,374
Form W-2
10,786
Form 1099-R
6,457
Form 1099-INT
4,427
Form 1099-MISC
3,101
Form 5498
2,994
Form 1098
2,981
Form 1099-G
2,944
Form 1099-DIV
1,991
a

Non-dependent nonfilers exclude all individuals identified on Form 1040 (inclusive of Form 1040-A and Form 1040-EZ),
Form 1040-NR, or Form 1040-SS/1040-PR, including primary and secondary taxpayers and individuals claimed as a
dependent.
b

Individuals in this group have at least one information return sent to a foreign or missing address.

c

The forms used to identify foreign persons are: Form 1042-S (Foreign Person’s U.S. Source Income Subject to
Withholding), Form 8288-A (Statement of Withholding on Dispositions by Foreign Persons of U.S. Real Property
Interests), and Form 8805 (Foreign Partner's Information Statement of Section 1446 Withholding Tax).
d

Form 1095 includes information about health insurance coverage through the Health Insurance Marketplace (Form
1095-A), health insurance coverage through other private or government sources (Form 1095-B), and an offer of health
insurance coverage by certain employers (Form 1095-C).
Source: Author's tabulation of IRS data

Brady and Bass

46

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Table 4

Composition of the Total Population

Number of unique individuals identified on 2016 Form 1040 a or 2016 information return, excluding
individuals with a US territory address and non-dependent nonfilers with a foreign or missing address
Number of Share of total
individuals
population
(thousands) (percentage)
Total population estimate
331,095
100
Filers
204,492
62
Non-dependent filer
195,090
59
Dependent filer
9,402
3
Dependent nonfilers
84,235
25
With income
13,337
4
Without income
70,898
21
Non-dependent nonfilers
42,368
13
With income
30,879
9
Without income
11,489
3
Memo: Share of population by various characteristics (percentage)
Identified on Form 1040 (filers plus dependent nonfilers)
Non-dependent nonfilers with at least one non-Form-1095b information return
Non-dependent nonfilers with Form 1095b only

a

87
10
3

Taxpayers on joint return
Taxpayers on non-joint return
Nonfilers (dependent plus non-dependent)

33
29
38

Taxpayers plus nonfilers with income
Dependent nonfilers without income
Non-dependent nonfilers without income

75
21
3

The term "Form 1040" is inclusive of Form 1040, Form 1040-A, or Form 1040-EZ.

b

Form 1095 includes information about health insurance coverage through the Health Insurance Marketplace (Form 1095-A),
health insurance coverage through other private or government sources (Form 1095-B), and an offer of health insurance
coverage by certain employers (Form 1095-C).
Source: Author's tabulation of IRS data

Brady and Bass

47

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Table 5

Derivation and Composition of the Census-Equivalent Population
a

Number of unique individuals identified on 2016 Form 1040 or 2016 information return, excluding individuals
with a US territory address and non-dependent nonfilers with a foreign or missing address
Number of
individuals
(thousands)
Total population estimate from tax data
331,095
─ Taxpayers and dependent nonfilers with a foreign or missing address
947
─ Taxpayers and dependent nonfilers with an overseas armed forces address
360
─ Non-dependent nonfilers with an overseas US armed forces addresss
17
─ Individuals who died before July 1, 2016
1,345
─ Individuals born on or after July 1, 2016
1,918
= Census-equivalent population estimate from tax data
326,508
b
c
Identified only using Form 1095 or as an additional dependent
9,873
Other
316,635
Memo:
Census population estimate as of July 1, 2016

323,072

Census-equivalent population as a percentage of Census estimate
Full Census-equivalent population
Excluding individuals identified only using Form 1095b or as an additional dependentc
a

101.1
98.0

The term "Form 1040" is inclusive of Form 1040, Form 1040-A, or Form 1040-EZ.

b

Form 1095 includes information about health insurance coverage through the Health Insurance Marketplace (Form 1095-A), health
insurance coverage through other private or government sources (Form 1095-B), and an offer of health insurance coverage by certain
employers (Form 1095-C).
c

Additional dependents are those not listed as one of the first four dependents on Form 1040.
Sources: U.S. Census Bureau 2021 and author's tabulation of IRS data

Brady and Bass

48

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Appendix
Comparison of Tax Data and Census Estimates in the Previous Literature
Beyond data availability and the year analyzed, the results of this study differ from those of
Cilke (2014) and Larrimore et al. (2019) because the comparisons between the tax data and
Census population estimates differ.
Without adjustment, population estimates from the tax data are not directly comparable to
Census estimates. Annual Census population estimates provide a snapshot of the population
alive on July 1. Tax data provide information on activities that could have occurred at any time
during the year.
In this study, we create a Census-equivalent population estimate from the tax data and then
compare that to the Census estimate. This is done by: (1) removing individuals who died before
July 1 or who were born after June 30; and (2) tabulating the data by age as of July 1.
Cilke (2014) essentially creates a tax-equivalent population estimate from the Census data
and then compares that to the tax data. This is done by (1) averaging, by single-year of age, the
July 2011 and July 2012 Census population estimates to derive a population estimate as of
December 31, 2011; and (2) adding back into the population an estimate of individuals who died
during the year.
It is possible that the method used by Cilke (2014) could produce different results than our
method. In theory, either approach produces valid comparisons. That said, the exact method
used to add individuals who died during the year back into the Census estimates could affect
the results, and we have not attempted to replicate that method here.
Unlike this study and Cilke (2014), Larrimore et al. (2019) does not appear to adjust either
the tax data or the Census estimates. That is, it appears the study compares 2010 tax-year data—
which represents individuals alive at any point during the year tabulated by age on December
31—to the decennial 2010 Census estimates.
To illustrate how failing to adjust the data would impact the results, we make two
adjustments to the tax data in addition to adjusting for data availability. First, we include
individuals alive at any point in 2016. Second, we retabulate the data by age as of December 31

Brady and Bass

49

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

(while still comparing to Census estimates of the population on July 1 tabulated by age as of
July 1).
With these additional adjustments, the 2016 population estimate from tax data represent
99.0 percent of the Census estimate (up from 98.0 percent), which is closer to replicating the
finding in Larrimore et al. (2019) that the tax data represent 99.5 percent of the Census
population estimate.
By age, these additional changes affect the elderly comparisons the most (Figure A.1).
Shifting the age groups by six months (from age as of July 1 to age as of December 31) generally
has a small impact, except for individuals aged 70 or a bit younger, who were born as the baby
boom accelerated in the late 1940s. The impact of including individuals who died before July 1
has a larger impact—at least among the elderly, where the tax data estimates switch from fewer
individuals relative to the Census estimates to more individuals.

Brady and Bass

50

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

&ŝŐƵƌĞ͘ϭ
ŽŵƉĂƌŝƐŽŶƚŽĞŶƐƵƐZĞƋƵŝƌĞƐĚũƵƐƚŵĞŶƚƐƚŽDĂƚĐŚWŽƉƵůĂƚŝŽŶ/ŶĐůƵĚĞĚŝŶĞŶƐƵƐƐƚŝŵĂƚĞ

ŽŵƉĂƌŝƐŽŶŽĨĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶĞƐƚŝŵĂƚĞƐ͕ϮϬϭϲ

ϯϬ

ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶ͕ĂŐĞĂƐŽĨ:ƵůLJϭ͕ĂůŝǀĞŽŶ:ƵůLJϭ
ĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶ͕ĂŐĞĂƚLJĞĂƌͲĞŶĚ͕ĂůŝǀĞĂƚĂŶLJƉŽŝŶƚĚƵƌŝŶŐƚŚĞLJĞĂƌ
ϮϬ

ϭϬ

Ϭ

ͲϭϬ

ͲϮϬ
Ϭ

ϭϬ

ϮϬ

ϯϬ

ϰϬ

ϱϬ

ϲϬ

ϳϬ

ϴϬ

ϵϬ

ϭϬϬн

ŐĞ

a/ŶƚŚŝƐĨŝŐƵƌĞ͕ƚŚĞĞŶƐƵƐͲĞƋƵŝǀĂůĞŶƚƉŽƉƵůĂƚŝŽŶĞdžĐůƵĚĞƐŝŶĚŝǀŝĚƵĂůƐŝĚĞŶƚŝĨŝĞĚŽŶůLJƵƐŝŶŐ&ŽƌŵϭϬϵϱ;ŝŶĐůƵƐŝǀĞ
ŽĨ&ŽƌŵϭϬϵϱͲ͕&ŽƌŵϭϬϵϱͲ͕ĂŶĚ&ŽƌŵϭϬϵϱͲͿŽƌŽŶůLJĂƐƚŚĞĨŝĨƚŚŽƌŚŝŐŚĞƌĚĞƉĞŶĚĞŶƚŽŶ&ŽƌŵϭϬϰϬ͘
^ŽƵƌĐĞƐ͗h͘^͘ĞŶƐƵƐƵƌĞĂƵϮϬϮϬĂŶĚĂƵƚŚŽƌΖƐƚĂďƵůĂƚŝŽŶŽĨ/Z^ĚĂƚĂ

Brady and Bass

51

April 24, 2023

Imagine All the People: Using Tax Data to Build a Representative Sample of the US Population

Table A.1

List of Information Returns Used to Identify Nonfilers
Form number

Form name

Forms used to identify nonfilers and measure nonfiler income
Income reporting forms
Form W-2
Form W2-G
Form SSA-1099
Form 1065
Form 1099-DIV
Form 1099-G
Form 1099-INT
Form 1099-MISC
Form 1099-R
Form 1120-S

Wage and Tax Statement
Certain Gambling Winnings
Social Security Benefit Statement
Partner’s Share of Income, Deductions, Credits, etc.
Dividends and Distributions
Certain Government Payments
Interest Income
Miscellaneous Income
Distributions From Pensions, Annuities, Retirement or Profit-Sharing Plans,
IRAs, Insurance Contracts, etc.
Shareholder’s Share of Income, Deductions, Credits, etc.

Qualified account contribution reporting forms
Form 5498

IRA Contribution Information

Forms used to identify nonfilers but not used to measure nonfiler income
Income reporting forms
Form 1041
Form 1042-Sa
Form 1099-C
Form 1099-OID
Form 1099-PATR
Form 8805a

Beneficiary’s Share of Income, Deductions, Credits, etc.
Foreign Person’s U.S. Source Income Subject to Withholding
Cancellation of Debt
Original Issue Discount
Taxable Distributions Received From Cooperatives
Foreign Partner's Information Statement of Section 1446 Withholding Tax

Expenses and charitable contribution reporting forms
Form 1098
Form 1098-T
Form 1098-E
Form 1098-C

Mortgage Interest Statement
Tuition Statement
Student Loan Interest Statement
Contributions of Motor Vehicles, Boats, and Airplanes

Transaction and financial account reporting forms
Form 1099-A
Form 1099-B
Form 1099-CAP
Form 1099-K
Form 1099-S
Form 8288-Aa
Form 8300
FinCEN Form 103
FinCEN Form 104

[Text truncated at 120,000 characters. The full text is on the page linked above.]

---

Source: Frix Law Library, https://www.frixlaw.com/law-library/documents/agency%3Airs%3A6dc5004f6676fdb9. Public record. Not legal advice.