The Effects of EITC Correspondence Audits
on Low-Income Earners*
John Guyton
IRS

Kara Leibel
IRS

Day Manoli
UT-Austin

Ankur Patel
Treasury

Mark Payne
IRS

Brenda Schafer
IRS

May 2019
Abstract
Each year, the United States Internal Revenue Service identifies taxpayers who may have
erroneously claimed Earned Income Tax credit (EITC) benefits and requests additional
documentation from these taxpayers to verify these claims. This paper exploits random variation
inherent in audit selection processes to estimate the impacts of these EITC correspondence audits
on taxpayer behaviors. Roughly 80% of EITC correspondence audits in the analysis sample have
full disallowances due to undelivered mail, nonresponse or insufficient response. Cases of
disallowances with confirmed ineligibility make up 15% of EITC correspondence audits in the
analysis sample. In years after being audited, taxpayers have decreases in the likelihoods of
claiming EITC benefits and filing tax returns so that they subsequently forego benefits from
potentially legitimate EITC claims, other refundable credit claims and withholdings. For every
$1 that is audited, roughly $0.63 to $0.73 of tax refunds is unclaimed in years after the audits.
Additionally, spillovers from audited taxpayers to other taxpayers arise because qualifying
children on audited returns are more likely to be subsequently claimed by other taxpayers after
the audits. These spillovers indicate that net overpayments may be less than gross overpayments
since ineligible qualifying children on audited returns could be potentially eligible qualifying
children on other taxpayers’ returns. Lastly, EITC correspondence audits affect real economic
activity as wage earners have changes in the likelihood of having wage employment in the years
after being audited.

*

Guyton: Internal Revenue Service, 77 K St. NE, Washington, DC 20002 (john.guyton@irs.gov). Leibel: Internal
Revenue Service, 77 K St. NE, Washington, DC 20002 (kara.e.leibel@irs.gov). Manoli: University of Texas at
Austin, 2225 Speedway Stop C3100, Austin, TX 78712 (dsmanoli@austin.utexas.edu). Patel: United States
Treasury, 1501 Pennsylvania Ave., Washington, DC 20005 (ankur.patel@treasury.gov). Payne: Internal Revenue
Service, 77 K St. NE, Washington, DC 20002 (john.m.payne@irs.gov). Schafer: Internal Revenue Service, 77 K St.
NE, Washington, DC 20002 (brenda.schafer@irs.gov). Disclaimer: This research does not represent any official
views or opinions of the Internal Revenue Service or any other government agency. Acknowledgements: Manoli
gratefully acknowledges research funding from the Laura and John Arnold Foundation, and the authors thank Lily
Batchelder, Ciyata Coleman, Patricia Gray, Lorraine Harrison, Erin King, Wojciech Kopczuk, Pat Langetieg, Lynne
Morrison, Leatta Phillips, Dan Shaviro, Joel Slemrod, Alex Turk, and numerous colleagues, seminar and conference
participants and for helpful comments and discussions.

1

 I. Introduction
In countries around the world, tax authorities rely on audits to enforce tax codes and improve tax
compliance. This paper presents an analysis of operational audits conducted by the United States
Internal Revenue Service (IRS) in the context of administering the Earned Income Tax Credit
(EITC), and we examine three central topics for tax enforcement research: deterrence of
erroneous or potentially inappropriate behaviors, spillovers to other taxpayers, and impacts on
real economic activity as opposed to just tax reporting behaviors. Audits can generally be
categorized into two groups: research audits and operational audits. Research audits are designed
to verify data integrity and help tax authorities detect areas of noncompliance and assess gaps
between revenues owed and taxes collected. Operational audits are tools to execute tax
enforcement or complying with program rules and are generally not intended for research
purposes. While a significant body of research has studied the impacts of research audits, little is
known about the impacts of operational audits on taxpayer outcomes, possibly due to a lack of
data on operational audits or insufficient institutional background on randomized variation in
operational audits. This analysis aims to overcome these obstacles and provide insights into the
impacts of operational audits on low-income earners’ behaviors.
The EITC has become the United States’ largest wage subsidy anti-poverty program, and the IRS
is charged with administering this program. Tax administration research within the IRS and in
academic contexts has demonstrated that, each year, while a significant amount of EITC benefits
subsidize working low-income households, there are also concerns about erroneous claims of
EITC benefits.1 Correspondence audits, conducted via mail, are a key enforcement tool to protect
revenue and deter improper claims of EITC benefits. Historically, there are roughly 500,000
EITC correspondence audits each year.
We estimate the causal effects of these EITC correspondence audits on low-income earners’
behavior by exploiting random variation within part of the audit selection process. We emphasize
that overall audit selection is not random or arbitrary, but there is random selection within a
1

For evidence on EITC noncompliance and erroneous payments of EITC benefits, see Holtzblatt (1991), McCubbin
(2000), Blumenthal, Erard and Ho (2005) and Leibel (2014). Related to this literature, Saez (2010), Chetty Friedman
and Saez (2013) and Mortenson and Whitten (2018) present evidence on taxpayers reporting self-employment
income to maximize EITC benefits and tax refunds.

2

 subsample of returns that are made available for audit. In particular, random variation
conditional on observables arises from the following audit selection process. First, all tax returns
are assessed for noncompliance risk. Next, returns with the greatest risk for noncompliance are
made available for audit, and there is random selection among the subsample of returns with
returns with low and intermediate risk scores. Thus, there is random selection only conditional
on having low or intermediate risk scores. By focusing on this subsample of returns with low and
intermediate risk scores for this study, we are able to estimate causal effects of EITC
correspondence audits by comparing randomly-selected audited taxpayers to taxpayers who had
similar risk scores but were randomly not selected for audit. The analysis includes EITC
correspondence audits from tax years 2008 through 2015, and the analysis sample of taxpayers
with low and intermediate risk scores consists of 432,219 audited self-employed taxpayers and
895,065 audited wage earner taxpayers. This sample is roughly one third of all EITC
correspondence audits over this period. The analysis sample also includes 473,938 scored-butnot-audited self-employed taxpayers and 1,170,290 scored-but-not-audited wage earner
taxpayers.
The analysis of audit outcomes shows that roughly 76% and 80% of the EITC correspondence
audits for the self-employed and wage earner analysis samples have EITC benefits disallowed
due to undelivered mail, nonresponse, or insufficient response (for example, this can be due to
discontinued communications and not continuing to provide requested documentation). Each of
these outcomes mechanically results in a full disallowance, so most audited individuals have
EITC benefits mechanically. While a common assumption in tax administration research (and
perhaps more broadly in public finance research) is that audits provide insight into “true”
incomes for audited taxpayers, this result indicates that this assumption may not apply in the
context of operational audits because of undelivered mail, nonresponse or insufficient response.
Moreover, the widespread undelivered mail, nonresponse and insufficient response imply that a
relatively small share of the EITC correspondences audits have confirmed ineligibility based on
information verified with the audited taxpayer. As a result, Type 2 errors, which are cases with
confirmed ineligible taxpayers claiming EITC benefits and having benefits fully disallowed, are
only about 15% of all EITC correspondences audits.

3

 While Type 2 errors on audited returns are relatively low, Type 1 errors based on potentially
EITC-eligible taxpayers not claiming their EITC benefits increase for audited taxpayers after the
EITC correspondence audits. For the self-employed and wage earner analysis samples
respectively, in the year immediately after the audits, about 30 percent to 40 percent of audited
taxpayers who may be potentially eligible for EITC benefits do not claim EITC benefits. In terms
of dollars, for every $1 of EITC benefits that audited taxpayers are potentially eligible for in the
year immediately after the audits, about $0.45 is unclaimed. These Type 1 errors fade out over
subsequent years. However, the fade out is not driven by audited taxpayers resuming EITC
claiming. Instead, the fade out is driven by nonaudited taxpayers decreasing their EITC claiming
as qualifying children age out beyond EITC qualifying child age thresholds (younger than age 19
or younger than age 24 if a full-time student). Nonetheless, to assess the efficiency costs
(deadweight losses) of the EITC correspondence audits, we consider the cumulative impact over
years after the audience. The analysis indicates that for every $1 that is audited roughly $0.63 to
$0.73 is unclaimed in years after the audits. While models and tax administration research often
assume that audited taxpayers do not leave benefits on the table, the results from the analysis
indicate that audited taxpayers may leave benefits on the table by not claiming EITC benefits and
other refundable credits they may be eligible for and by not claiming excess withholdings.
Next, the results indicate that the EITC correspondence audits have spillover impacts on other
taxpayers through qualifying children. In particular, some qualifying children on audited returns
are subsequently claimed as dependents on other taxpayers’ tax returns. For example, in the year
after being audited, the likelihood of qualifying children on the audited return being claimed by
the audited taxpayer deceases by about 0.24 and 0.25 for the self-employed and wage earner
analysis samples respectively. About 67% (=0.16/0.24) and 52% (=0.13/0.25) of these respective
decreases are due to the qualifying children not being claimed by any taxpayers, and the
remaining 33% and 48% decreases are due to the qualifying children being claimed as
dependents by other taxpayers. Models in tax administration research often assume that only
audited taxpayers are affected by audits, but these results indicate that the operational audits have
spillover impacts on other nonaudited taxpayers who subsequently claim the qualifying children
on audited returns. Additionally, the qualifying child switching between taxpayers highlights a
distinction between gross overpayments and net overpayments when trying to assess total dollars
4

 over-spent in the context of the EITC correspondence audits. Gross overpayments will include
overpayments to taxpayers incorrectly claiming qualifying children. However, if a different
taxpayer should have claimed a qualifying child, then the (under)payment to the other taxpayer
should be subtracted off of the gross overpayment to the audited taxpayer in order to assess the
total overspending on EITC claims. Qualifying child switching accounts for roughly one third to
one half of the qualifying child changes in the current analysis, so net overpayments could be
significantly smaller than gross overpayments.
The analysis also indicates that EITC correspondence audits may affect real economic activity as
opposed to just tax reporting outcomes. In particular, for audited wage earners who have wage
employment (i.e. have a W-2) in the year of selection, there are decreases in the likelihood of
having wage employment in the years just after the EITC correspondence audits, and the
decreases are larger for taxpayers with younger (ages 0-5) qualifying children. Quantitatively,
these estimated changes in wage employment imply a participation elasticity of about 0.28,
which is consistent with prior quasi-experimental estimates of participation elasticities (see
Chetty et al 2011). However, we note that there are multiple caveats to keep in mind since the
EITC correspondence audits may affect several perceptions and factors beyond just labor supply
incentives. For audited wage earners without a W-2 in the year of selection, there appear to be
gradual increases in the likelihood of having a W-2 in the years after the audits. This may reflect
gradual transitions from informal, cash-based employment to formal, W-2-documented
employment after the EITC correspondence audits.
The analysis relates to prior tax enforcement research that has examined the impacts of audits on
taxpayer behavior (see Slemrod 2016 for a survey of recent research on tax enforcement). For
example, Kleven et al (2011) present results based on randomized audits and threat-of-audit
notices in Denmark, Advani et al (2017) examine effects of randomized audits in the United
Kingdom, and perhaps most closely, DeBacker et al (2018) examine randomized IRS audits of
EITC claimants. However, each of these studies examine impacts of research audits as opposed
to operational audits. In particular, the IRS conducts research audits as part of the National
Research Program (NRP). In the institutional background below, we discuss differences between
the NRP audits and the EITC correspondence audits in detail, but to summarize, these audits
5

 differ in multiple ways. First, in terms of population size, research audits are relatively expensive
and therefore involve smaller sample sizes. For example, the IRS conducts roughly 15,000
research audits each year (DeBacker et al 2018) and over 1.5 million operational audits each year
(see the annual IRS Data Book). Second, in terms of the nature of the audits, research audits
often involve (possibly repeated) personal contact between a tax auditor and taxpayer via a
phone call or in-person meeting, and the two parties work together to assess true income and true
tax liability. In contrast, operational audits often do not involve personal contact between tax
auditors and taxpayers (correspondence audits do not include personal contact, though field
audits, which are less frequent, can involve personal contact). Furthermore, the operational audits
do not provide taxpayers with tax auditors who assist them through the examination process.
Thus, taxpayers may be confused by correspondence audits or may not learn as much as they
would from a tax auditor. These factors can lead to nonresponse (which is much lower or
negligible with research audits), and as a result, true income and true tax liability may never be
observed. Given the widespread use of operational audits to enforce tax policies and policies in
other settings, it is important for tax authorities, program administrators, and researchers to
understand the impacts of both research audits and operational audits.
The remainder of this paper is organized as follows. Section II describes the institutional
background on EITC correspondence audits and the administrative data used in the analysis.
Section III describes the empirical analysis and results. Section IV concludes.
II. Institutional Background & Data
A. EITC Correspondence Audit Process
Each year, the IRS audits selected individual federal income tax returns to verify that income,
deductions, or credits are being reported accurately. There are generally two types of operational
audits: correspondence audits, which are conducted via mail, and field or face-to-face audits that
are conducted at the taxpayer’s home, place of business, tax preparer’s office, or IRS office.
Annual statistics on the number of correspondence and field audits are publicly available in the

6

 IRS Data Book and shown in Table 1.2 As indicated by the IRS Data Book statistics for fiscal
years 2008 to 2016, there are roughly 400,000 to 500,000 correspondence audits of returns where
EITC is claimed each year, compared to roughly 30,000 to 50,000 face-to-face audits of returns
where EITC is claimed. These numbers have been declining over time due to reductions in the
IRS budget. The statistics in Table 1 also highlight that EITC correspondence and field audits
make up a large fraction of overall audits, with EITC correspondence audits being roughly 35%
to 45% of all correspondence audits and EITC field audits being roughly 10% of all field audits.
The analysis in this paper focuses on comparing returns selected for EITC correspondence audits
with similar returns that were not selected for any audits. While the exact criteria used to select
tax returns for audit are not made public by the IRS, we summarize the process for EITC
correspondence audit selection as follows. As part of standard tax return processing, all returns
claiming children for the EITC undergo a series of checks and comparison to relevant third-party
data and past tax filing history. Returns that are flagged with indicators of potential
noncompliance are assigned one or more risk scores, depending on the nature of the flagged
condition (such as the types of rules potentially broken and the number of rules potentially
broken). Returns with greatest risk for noncompliance are made available for audit, and there is
random selection among returns with intermediate- or low-risk scores (i.e. there is random
selection conditional on observables using the low and intermediate risk scores; audit selection is
not completely random or arbitrary).
Once an individual income tax return with EITC is assigned for a correspondence audit, a
notification letter is automatically generated and sent to the taxpayer. This notice, which is
typically a CP-75, informs the recipient that her tax return is being audited and requests the
taxpayer submit more information or documentation to support claimed tax benefits, as
applicable, which may include EITC, other refundable credits, and dependency exemptions.3 The
2

The 2016 IRS Data Book is available online at https://www.irs.gov/pub/irs-soi/16databk.pdf . The IRS Data Books
for fiscal years 2010 through 2015 can be found at the same link but with adjustments to the numbers to correspond
to the desired fiscal year.
3
While the CP-75 notice explains that EITC, Additional Child Tax Credit (ACTC) and Premium Tax Credit benefits
are on hold until the audit is resolved, CP-75A notices focus only on EITC benefits and do not impose a refund hold,
and CP-75D notices specify holding only a portion of EITC benefits. Appendix Figure 1 presents an example of a
CP-75 notice, and information on the notice, as well as an example, can be found on the IRS website at

7

 type of supporting documentation requested depends on the issue that the taxpayer must
substantiate, and examples of supporting documentation are provided on the notices. For
example, recipients may be asked to show that a qualifying child (QC) meets the relationship
requirement. In such a case, taxpayers may provide a birth certificate. School records may be
used to demonstrate the residency requirement. Information on business income and expenses
may be requested to verify self-employment businesses. The CP-75 notice informs the taxpayer
that she has 30 days to respond and that her refund is on hold until the audit is resolved. CP-75
notices are typically sent within four to eight weeks after returns are filed. The majority of EITC
correspondence audits are pre-refund audits: roughly 75% of correspondence audits that do not
involve self-employment income are pre-refund audits, and roughly 90% of correspondence
audits involving self-employment income are pre-refund audits.
Once an EITC correspondence audit has been initiated, there are multiple possible outcomes.
First, the audit notification may be undeliverable due to a bad or old mailing address, or the
taxpayer simply may not respond to the notice. In both of these cases, EITC is ultimately
disallowed in full. If a taxpayer responds to the initial notice, the IRS will send a notice
explaining whether more information is needed or if a decision was reached. If the EITC is
disallowed, the taxpayer can: (1) respond to the notification and actively agree with the
disallowance; (2) respond to the notification and actively disagree with the disallowance; or (3)
not to respond to the notification and passively agree with the decision. If the EITC is allowed, it
may be allowed or partially disallowed, depending on the information provided by the taxpayer.
As indicated in annual statistics reported in the IRS Data Book and shown in Table 1, each year
roughly 85% to 90% of EITC correspondence audited returns result in changes to the tax returns.
Prior reports (National Taxpayer Advocate 2007, Schneller Chilton and Bochum 2011 and
Government Accountability Office 2014) have highlighted that nonresponse and insufficient
response, potentially due to confusion, intimidation of the audit process, or undelivered mail are

https://www.irs.gov/pub/notices/cp75_english.pdf. We acknowledge that this example CP-75 focuses on the
Premium Tax Credit rather than the EITC, but notices that focus on the EITC are similar, and we provide this
example since it is the example that is published on the IRS website.

8

 factors in some disallowances. We provide more details on the audits outcomes in the summary
statistics described below.
In most cases, when EITC benefits are disallowed, taxpayers are notified of the change via
Notice CP-79. This notice explains to taxpayers that to claim EITC benefits in the future, they
must include Form 8862 with the filed tax return for the year in which they first claim EITC
again.4 Form 8862 includes questions to verify the taxpayer’s eligibility for EITC benefits (and
other potentially applicable refundable tax credits). Taxpayers may also be banned from claiming
the EITC for the next two years (reckless disregard) or the next ten years (willful disregard).
In addition to the operational correspondence and field audits, the IRS also conducts research
audits through the IRS National Research Program (NRP). These NRP audits are intended to
help the IRS detect possible areas of noncompliance and assess its success and effectiveness in
collecting tax revenues. In terms of sample sizes, NRP research audits are relatively expensive,
so roughly 15,000 NRP audits are selected each year and the sample is weighted to create a
sample that is representative of the national population of tax filers. The population for
operational audits is not intended to be representative of the national population of all tax filers
since the operational audit population only selects returns that may have potential
noncompliance, and there are over 1.5 million operational audits (including all audits, not just
EITC audits).
Appendix Figures 4 and 5 provide examples of the letter and notification documents sent to
taxpayers who are selected for these research audits, we note multiple differences between these
documents and the documents for the EITC correspondence audit documents. First, the letter for
research audits explains that the return was selected at random to improve tax compliance and
better understand fairness in the tax system. The letter also explains that there will be a telephone
conversation between the tax auditor and taxpayer to explain the examination process, and the

4

Appendix Figures 2 and 3 present examples of a CP-79 notice and a Form 8862 respectively. More information
about the CP-79 notice is available on the IRS website at https://www.irs.gov/individuals/understanding-your-cp79notice.

9

 notice highlights that there may not be any errors on the tax return. Each of these elements are
different from the EITC correspondence audit CP-75 notice.
Perhaps more important than the differences in notices, the nature of the research audits is
significantly different from the operational audits. Research audits are intended to detect possible
areas of noncompliance, but operational audits are conducted because risk factors have been
detected. There are efforts made to contact and assist taxpayers through research audits and
hence there are explicit goals to minimize nonresponse and confusion, and taxpayers may learn
about rules and how to be tax compliant from tax auditors. In contrast, operational audits often
involve single contacts to taxpayers with no assistance from a tax auditor. As a result,
nonresponse and insufficient response can be important factors since taxpayers must navigate the
examination processes themselves (possibly with their tax preparers) without the assistance of a
tax auditor. As we document, nonresponse, insufficient response are common outcomes for
operational audits, and these can lead to long-term impacts of the EITC correspondence audits.
B. Analysis Data
The data used in the empirical analysis is based on the population of tax returns that claimed
EITC benefits and were scored for potential noncompliance from 2008 through 2015. The 2008
restriction is imposed because data for some mailed notices for EITC correspondence audits are
only available from 2008 onward. The 2015 restriction is imposed so that outcomes can be
observed for at least 1 year after selection for scoring, and outcome data are available through
2016.
The analysis data is constructed from this population of scored returns by imposing two sample
restrictions. First, we focus only on single or head-of-household tax returns so that the analysis
only requires tracking one individual (the primary taxpayer on the single or head-of-household
return) before and after being flagged for risk scoring. Second, we impose a common support
sample restriction. Specifically, given that the research design is based on comparing
observationally similar audited and scored-but-not-audited returns, the analysis data is
determined by creating cells based on audit selection variables for each tax year, such as the
10

 types of rules potentially broken, the number of rules broken, and risk scores. The sample is
restricted to observations in cells that have both audited and nonaudited returns. This sample
restriction ensures that there is a common support for the audit selection variables between the
audited and scored-but-not audited samples. Observations in cells with only audited returns, such
as high risk or multiple-issue (potentially breaking multiple rules) returns, are dropped since
there are no observationally similar nonaudited returns for comparison. Similarly, observations
in cells with only nonaudited returns are dropped since there are no observationally similar
audited returns for comparison. After imposing these sample restrictions, we refer to the
remaining sample as the “analysis sample,” which consists of both audited and scored but not
audited tax returns.
The analysis data and empirical analysis below are split into two analysis samples: taxpayers
who report self-employment (Schedule C) income on their selected tax returns, who are referred
to as “Self-Employed,” and taxpayers who do not have any self-employment income on their
selected tax returns, who are referred to as “Wage Earners.” This split is motivated by prior
research that has highlighted different responses to audits and threat-of-audit interventions across
taxpayers with and without third-party verified income (Slemrod, Blumenthal and Christian
2001, Kleven, Knudsen, Kreiner, Pedersen, and Saez 2011, Slemrod 2016). Furthermore, the
analysis sample of EITC correspondence audits generally consists of single-issue audits as
opposed to multiple-issue audits. For the Self-Employed the single issue is verifying the
existence of self-employment business income; for the Wage Earners, the single issue is
verifying qualifying child eligibility. Tax returns that are considered for multiple-issue audits are
generally higher risk returns that are always selected for audit. For such returns, there are no
comparable nonaudited returns, so such returns are dropped from the analysis sample based on
the common support sample restriction described above. Lastly, we note that the definition of the
self-employed and wage earner samples follows definitions from the prior literature (for
examples, see Saez 2010 and Chetty, Friedman and Saez 2013). As a result of defining wage
earners based on taxpayers without self-employment income, the wage earner sample includes
some individuals who do not have W-2 wage earnings forms, and some of these individuals may
still have income reported as “wages, salaries, and tips” on their tax returns (IRS Form 1040). In
the analysis below, we present separate results for wage earners with and without W-2s in the
11

 year of selection. The analysis samples consist of 432,219 audited taxpayers for the selfemployed analysis sample and 895,065 audited taxpayers for the wage earner analysis sample.
These audits make up roughly one third of all EITC correspondence audits over the analysis time
period (2008 through 2015). The analysis samples on returns with low and intermediate risk
scores also include 473,938 scored-but-not-audited self-employed taxpayers and 1,170,290
scored-but-not-audited wage earner taxpayers.
C. Summary Statistics and Graphical Analysis
Table 2 presents summary statistics for the analysis samples used in the empirical analysis
below.5 The summary statistics are presented separately for the Self-Employed and Wage
Earners, and for each of these groups, statistics are presented for the following subgroups:
audited tax returns, scored but not audited returns, and a 1% random sample of EITC returns. For
the 1% random sample of EITC returns, we correspondingly draw 1% random samples of single
or head-of-household Self-Employed or Wage Earner EITC returns. We focus first on comparing
the audited and scored but not audited returns to the random samples of EITC returns. We note
that the analysis samples have a higher fraction male head-of-household tax returns, and the
primary taxpayers in the analysis samples are slightly younger than those in the general EITC
population. About 50% of the taxpayers in the self-employed analysis sample have a W-2, and
this is slightly higher than the corresponding 45% figure for the comparable general EITC
population. For the analysis sample of wage earners, about 86% and 95% of the audited and
nonaudited taxpayers have W-2s, whereas about 97% of the random sample of EITC returns for
wage earners have W-2s. Furthermore, the analysis samples have slightly lower incomes and
higher refund amounts (and are more likely to be on the maximum credit portion of the EITC
benefit schedule) then the random sample of EITC returns. The analysis samples have a higher
fraction of returns with one qualifying child while the random sample of EITC returns is more
evenly distributed across the numbers of qualifying children. Tax preparation methods appear

5

We do not present summary statistics for the full EITC correspondence audit population because the IRS does not
make these statistics on this population publicly available and because we aim to avoid any possible disclosure of
audit selection criteria or risk assessment criteria based on comparisons between the lower-risk analysis samples and
the full EITC correspondence audit population.

12

 roughly similar across the analysis samples and the random samples of EITC returns with a
majority of returns involving a paid tax preparer and use of software (electronic filing).
The audit characteristics for the self-employed and wage earner analysis samples indicate that,
respectively 90% to 75% of the EITC correspondence audits are pre-refund audits. For the selfemployed analysis sample, roughly 90% of the audits focus only on verifying Schedule C
income and about 10% focus only on verifying qualifying child eligibility. For the wage earners
analysis sample, roughly 96% of the correspondence audits focus only on verifying qualifying
child eligibility. (Remaining audits could be multiple issue audits or may focus on other aspects
of income verification beyond just Schedule C income verification.)
Table 3 presents audit outcomes for the analysis samples. Focusing first on the full sample
results, the audit outcomes (which are mutually exclusive groups) highlight that almost 80% of
the audits in the analysis samples have undelivered mail, nonresponse, or full disallowance with
passive agreement. The full disallowance with passive agreement scenario arises if a taxpayer
initially responds to a correspondence audit request for supporting information but then stops
responding to additional subsequent requests for supporting information. For the self-employed
and wage earner analysis samples respectively, roughly to 13% to 15% of EITC correspondence
audits lead to a full disallowance with active agreement, and about 5% to 7% of EITC
correspondence audits have a full allowance. Partial allowances constitute less than 2% of audit
outcomes in the analysis samples.
Type 2 errors are cases in which ineligibility is confirmed, and in a strict sense, these cases only
arise with outcomes of full disallowances with active agreement since the taxpayer’s active
agreement confirms the ineligibility causing the disallowance. Thus, overall, these audit
outcomes indicate relatively low Type 2 error rates. More specifically, the relatively low Type 2
error rates appear driven by widespread undelivered mail, nonresponse and full disallowance
with passive agreement. Given that roughly 15% of EITC correspondence audits result in
confirmed Type 2 errors, within the analysis sample, in order to identify 100 cases with Type 2
errors, it is necessary to conduct about 667 EITC correspondence audits.

13

 Table 3 also presents statistics on audit outcomes across various subgroups. Across all
subgroups, undelivered mail, nonresponse and full disallowance with passive agreement
generally account for most outcomes, and when rates of undelivered mail and nonresponse are
lower, rates of both full allowances and full disallowances are higher. For both the self-employed
and wage earner analysis samples, women have slightly lower rates of undelivered mail and
nonresponse than men, and women have slightly higher rates of full allowances than men. Rates
of undelivered mail are roughly constant across age groups, and nonresponse decreases slightly
with age while full disallowance with active agreement and full allowance rates increase with
age. Across income groups, undelivered mail and nonresponse rates decrease with higher income
groups, and full allowance and full disallowance rates increase with income. (However, we note
that, for wage earners, partial allowances appear to account for an unexpectedly large share
(24%) of outcomes for taxpayers with earned income above $40,000.) Audited taxpayers with a
paid tax preparer have higher rates of full allowances and full disallowances than audited
taxpayers without paid tax preparers. Audited taxpayers with an EITC claim in the prior three
years appear less likely to have undelivered mail and nonresponse and more likely to have full
disallowances and full allowances than audited taxpayers without an EITC claim in the prior
three years. Undelivered mail and nonresponse rates decrease across groups with more
qualifying children, and rates of full allowances and full disallowances increase across groups
with more qualifying children.
In the empirical analysis below, we examine heterogeneity in the effects of the EITC
correspondence audits based on the age of the youngest qualifying child on the selected return,
whether or not a taxpayer has a W-2 in the year of selection, and the estimated propensity scores.
We discuss the motivations for each of these dimensions in more detail below, but in this section
we discuss the differences in audit outcomes across these dimensions. For both the selfemployed and wage earner analysis samples, audited taxpayers with younger qualifying children
have slightly lower rates of full disallowance due to undelivered mail, nonresponse or
insufficient response (passive disagreement) and slightly higher rates of allowances. Audited
taxpayers without a W-2 in the year of selection have higher rates of undelivered mail, and
among wage earners, this group has a lower rate of full allowance than wage earners with a W-2
in the year of selection (0.016 versus 0.078).
14

 We discuss the propensity score estimation in the empirical analysis below, but the propensity
score groups reflect groups with similar observables but different fractions of audited taxpayers.
For example, the lowest quintile consists of the twenty percent of each respective analysis
sample that has similar observables and the lowest fraction of audited individuals. Similarly, the
highest quintile consists of the twenty percent of each analysis sample that has similar
observables and the highest fraction of audited individuals. Across the quintile groups, the
fractions of audited individuals are 0.006, 0.101, 0.434, 0.852 and 0.992 for the self-employed
analysis sample and 0.006, 0.050, 0.226, 0.890 and 0.996 for the wage earner analysis sample.
Even though the fractions of audited individuals very significantly across the groups, the audit
outcomes for both analysis samples indicate that full disallowances due to undelivered mail,
nonresponse, and insufficient response account for at least 70% of the outcomes for audited
taxpayers in each group. Nonetheless, audited taxpayers in the lowest quintile do have higher
rates of partial and full allowances than audited taxpayers in the highest quintile (0.166 versus
0.061 for the self-employed analysis sample and 0.149 versus 0.078 for the wage earner analysis
sample.
Related to audit outcomes, Figure 1 presents plots of EITC claiming, tax filing and qualifying
child claiming with separate series for the random sample of EITC returns and different audit
outcome groups. Similar to the empirical analysis below, each outcome is examined both before
and after the year of selection so that differences across the groups and across the years since
selection can be visually inspected. The plots indicate mostly similar trends in the outcomes
across the groups in the years prior to selection. In the years immediately after selection, the
outcomes appear similar for the random sample of EITC returns, returns that were scored but not
audited, and audited returns that ultimately had the EITC allowed. However, there are noticeably
different trends after selection for returns that were audited and ultimately had the EITC
disallowed. Following the audits, the returns with disallowances show decreases in the likelihood
of claiming EITC benefits, decreases in the likelihood of filing tax returns, and increases in the
likelihood that the qualifying children claimed on the audited returns are subsequently claimed
by other taxpayers.

15

 The graphical patterns suggest multiple insights. First, EITC disallowances due to
correspondence audits may reduce subsequent EITC claiming, possibly through reductions in tax
filing. Second, given that there do not appear to be sharp, differential changes in outcomes for
audited taxpayers who ultimately have the EITC allowed, the effects of EITC correspondence
audits on taxpayers may be driven primarily by the disallowances of EITC benefits as opposed to
simply being selected for a correspondence audit and being sent a request for supplemental
information. Third, EITC correspondence audits may have spillover effects on other nonaudited
taxpayers through qualifying children who were previously claimed on an audited return but are
subsequently claimed by other taxpayers.
IV. Empirical Analysis
A. Research Design
Our research design exploits the random variation in audit assignment to estimate the causal
effects of the EITC correspondence audits on taxpayer outcomes. Because the random
assignment of audit status is conditional on observables, we first re-weight the analysis data
using inverse probability weighting, and we then estimate a generalized difference-in-difference
regression specification using the re-weighted data. The difference-in-difference regression
specification with the re-weighted data mimics an RCT (randomized controlled trial) in which
the differences between the randomly assigned treatment (audited) and control (nonaudited)
groups are estimated for each year before and after random assignment. For each outcome of
interest, we present graphical evidence and regression estimates for differences between the
audited and scored-but-not-audited (nonaudited) groups for each year before and after the year of
selection and random assignment of audit status. The evidence for the years prior to the year of
selection helps to confirm comparability of the groups prior to the year of selection. Even though
pre-audit selection difference may be small or statistically insignificant, we present difference-indifference estimates for the impacts of the EITC correspondence audits on outcomes of interest.
Rather than just relying on post-audit selection differences, the difference-in-difference estimates
explicitly subtract off any pre-selection differences between the audited and nonaudited groups
from the post-audit selection differences. This allows us to be more confident that the estimates
16

 reflect causal impacts of the EITC correspondence audits and not any other pre-existing
difference between these groups.
We use inverse probability weighting to ensure that observables are balanced between the
treatment and control group and eliminate bias due to selection on observables. The weights are
estimated as follows. First, we define an indicator variable 𝑨𝒊 that is equal to 1 if individual i was
selected for an EITC correspondence audit. Next, we pool the samples of audited and scored-butnot-audited individuals and estimate the propensity score via the following regression
specification
𝑨𝒊 = 𝜷𝑿𝒊 + 𝑢(
where 𝑿𝒊 denotes a rich set of covariates that we discuss in more detail below. Intuitively, the
propensity score captures the (estimated) probability that an observation with observables X is
- (𝑨𝒊 =
* 𝒊 = 𝑷𝒓
assigned to be audited. We then obtain predicted values from this regression, 𝒑
𝟏

𝟏 𝑿𝒊 ) and use these predicted values to compute weights. We use weights 𝒘
* 𝒊 = 𝟏3𝒑* for the
𝒊

𝟏

scored-but-not-audited individuals and 𝒘
* 𝒊 = 𝒑*   for the audited individuals. Intuitively, these
𝒊

weights balance observables between the audited and scored-but-not audited returns by “upweighting” audited returns that have observables similar to scored-but-not audited returns and
scored-but-not-audited returns that have observables similar to audited returns, and similarly, by
“down-weighting” audited returns that have observables similar to other audited returns and
scored-but-not-audited returns that have observables similar to other scored-but-not-audited
returns. Weights are estimated separately for the self-employed and wage earner samples.
The covariates for estimating the weights include dummies for gender, head-of-household filing
status, tax preparation method, year of birth, income percentile (measured in 50 two-percent
bins), number of qualifying children claimed on the flagged return, and indicators for filing,
claiming EITC and having a W2 in each of the last 3 calendar years. Most importantly, the
covariates also include controls based on audit selection criteria. These variables are not made
public by the IRS, so we can only summarize these covariates by mentioning that these audit
17

 selection controls include fixed effects for groups based on the types of rules broken, the number
of rules broken, and the tax year of the return. Overall, the 𝑅6 values from these regressions for
computing the weights are 0.639 and 0.763 for the self-employed and wage earner analysis
samples respectively.
Appendix Figure 6 presents the fraction audited by percentiles of the estimated propensity scores
for both the self-employed and wage earner analysis samples. We note two features from these
plots. First, for both analysis samples, observations with low and high estimated propensity
scores do have respectively low and high fractions of taxpayers that were actually assigned to be
audited. Thus, the observables used to predict audit assignment appear to correlate with the
actual outcomes as expected. Second, there is a significant portion of the estimated propensity
score distribution that has both substantial fractions of both audited and nonaudited taxpayers.
These observations that have similar observables but different audit assignment will be “upweighted”, and the observations with observables that closely predict audit assignment (i.e.
observations in the low and high ends of estimated propensity score distributions) will be “downweighted.” We also examine heterogeneity across groups with different estimated propensity
scores below.
Appendix Table 1 presents summary statistics on the re-weighted samples. For both the selfemployed and wage earner analysis samples, the re-weighted data reduce differences between
the audited and scored-but-not-audited returns relative to the differences shown in Table 2 with
the summary statistics for the un-weighted data. In particular, differences in gender are smaller
for the self-employed relative to the difference in Table 2, and for wage earners, differences in
gender, age, income and benefits measures are all smaller. We do not present formal statistical
tests of these differences because the large sample sizes lead to statistical significance even for
non-meaningful differences. Instead, in the empirical analysis below, we present graphical
evidence on the re-weighted differences between the audited and nonaudited returns for several
outcomes. This graphical evidence indicates that the differences based on the re-weighted data
are close to 0 and stable in the years prior to audit assignment so that any difference can be
subtracted off from the post-audit differences.

18

 Using the re-weighted data, we use a difference-in-differences strategy to exploit the random
variation in audit assignment and estimate the causal effects of the correspondence audits on
taxpayer outcomes. First, we define event time as the years since the year of random assignment
of audit status. Specifically, for individual i in year t, event time 𝒆𝒊𝒕 is defined as 𝒆𝒊𝒕 = 𝒂𝒊 − 𝒕
where 𝒂𝒊 denotes the year that individual i’s tax return is flagged and randomly assigned for an
EITC correspondence audit or not. Next, the impacts of EITC correspondence audits on an
outcome y are estimated via the following regression specification:
𝟒

𝒚𝒊𝒕 = <

𝟒

𝒌?3𝟕

𝜷𝒌 𝟏(𝒆𝒊𝒕 = 𝒌) + <

𝒌?3𝟕

𝜹𝒌 𝑨𝒊 𝟏(𝒆𝒊𝒕 = 𝒌) + 𝜺𝒊𝒕 .

The coefficients 𝜷𝒌 reflect the means of the outcome variable at each event time for the scoredbut-not-audited group, and the coefficients 𝜹𝒌  reflect the differences in the means for the audited
group relative to the nonaudited group for each event time. The standard errors for the
coefficients are clustered based on tax year, the year of random assignment and the indicator for
being audited or not. We plot the estimated 𝜷𝒌 and 𝜹𝒌 coefficients from the regressions.
Additionally, we estimate difference-in-differences estimates of the impacts of the
correspondence audits on outcome y at event time 𝒌 = +𝟏, +𝟐, … by subtracting off the average
pre-selection difference from the post-selection difference at event time k:
𝒅𝒌 = 𝜹𝒌 − 𝟎. 𝟑𝟑𝟑(𝜹3𝟐 + 𝜹3𝟑 + 𝜹3𝟒 ).
We examine a variety of outcomes for primary taxpayers on audited and scored-but-not-audited
returns, including claiming EITC benefits, reporting self-employment income, filing a tax return
(as either a primary or secondary taxpayer), and tax refund amounts. Additionally, we estimate a
similar regression specification based on tracking qualifying children claimed on audited and
nonaudited tax returns across tax years before and after being selected for risk assessment. In
particular, the regression specification is the same as the regression specification described
above, but instead of using the subscript i to refer to an individual taxpayer, the subscript i refers
to a qualifying child claimed on an audited or scored-but-not-audited return. By tracking the
qualifying children, we are able to examine the extent to which qualifying children on audited
19

 returns are likely to be claimed as qualifying children by other taxpayers after the audits, as well
as the characteristics of the (primary) taxpayers claiming audited qualifying children before and
after the EITC correspondence audits.
B. Results
1. Impacts on EITC Claiming & Tax Outcomes
Figure 2 presents the estimated impacts of the EITC correspondence audits on EITC claiming,
tax filing, and tax refunds for the self-employed and wage earner analysis samples. For the selfemployed taxpayers, there are some differences between audited and nonaudited taxpayers in the
pre-audit assignment trends for EITC claiming and tax filing. For the wage earner taxpayers, the
pre-audit trends for these outcomes appear more similar for audited and nonaudited taxpayers.
For both the self-employed and wage earner samples, the plots highlight that in the year just after
audit assignment, there are significant decreases in EITC claiming and tax filing (as either a
primary or secondary taxpayer) and tax refunds for the audited group relative to the nonaudited
group.
Based on the difference-in-difference estimates in Table 4, the declines in the likelihood of filing
are smaller than the declines in the likelihood of EITC claiming. This indicates that, in addition
to reducing EITC claiming through decreases in filing, the EITC correspondence audits also
appear to cause individuals to subsequently not claim EITC benefits even when they file tax
returns. Over subsequent years after the EITC correspondence audits, the impacts on EITC
claiming, tax filing and tax refunds fade out. This fade out could be due to qualifying children
aging beyond the EITC qualifying child age thresholds (less than age 19 or less than age 24 for
full-time students) so that EITC claiming, tax filing and tax refunds for the nonaudited group
ultimately converge to the corresponding rates and values for the audited group.
The changes in EITC claiming following the EITC correspondence audits indicate that, for the
low- and intermediate risk returns in the analysis samples, audited taxpayers appear to have Type
1 errors (cases of eligible taxpayers not claiming EITC benefits) in years after the audits. Using
20

 the scored-but-not-audited taxpayers as a counterfactual for what EITC claiming would have
been for the audited taxpayers had they not been assigned to the EITC correspondence audits, we
compute a Type 1 error rate associated with not claiming potentially legitimate EITC benefits for
each year after the audits by expressing the change in EITC claiming in each year after the audits
as a fraction of baseline EITC claiming for the scored-but-not-audited taxpayers in each
corresponding year after the audits. Intuitively, this fraction measures the likelihood of
incomplete take-up of EITC benefits for potentially EITC-eligible taxpayers. These Type 1 error
rates are presented in Table 4 for both the self-employed and wage earner analysis samples. The
decreases in EITC claiming after the audits are significant relative to the baseline mean EITC
claiming rates for the scored-but-not-audited groups: in the year just after the audits, the Type 1
error rates are between 0.33 and 0.43 for the self-employed and wage earner analysis samples
respectively. Intuitively, among audited taxpayers who may be eligible to claim EITC benefits in
the year after audit, about 33% and 43% of the self-employed and wage earner taxpayers
respectively do not claim their EITC benefits. Over subsequent years, these effects on Type 1
errors fade out as the impacts on EITC claiming fade out.
We note that, in addition to leaving potential EITC benefits on the table, audited taxpayers may
also leave benefits from other refundable tax credits (such as the Additional Child Tax Credit) or
their federal income tax withholdings on the table after the EITC correspondence audits either by
not claiming refundable credits on their tax returns when they file or by not filing tax returns at
all. For example, when low-income earners do not file tax returns, they may leave potential tax
refunds based on federal income tax withholdings on the table if they would have been in the 0%
tax bracket and had no federal income tax liability. (If some audited individuals anticipate that
they will no longer file tax returns after being audited, they may reduce their federal income tax
withholdings so that they receive this income through wage earnings payments and not through a
tax refund after filing. In separate analyses not shown, we do not observe any evidence of
decreases in the likelihood of having withholdings. Instead, the likelihood of having
withholdings and not filing increases after the EITC correspondence audits due to the decreases
in the likelihood of filing.)

21

 Similar to computing Type 1 errors based on EITC claiming, we also compute Type 1 errors
based on dollar amounts by expressing the change in average tax refunds received after the
audits (which accounts for changes in not claiming refundable credits and withholdings) as a
fraction of the average tax refund received for the nonaudited group. These estimates are also
shown in Table 4. Since tax refunds and EITC benefits conditional on tax filing and claiming are
similar between the audited and nonaudited taxpayers in the analysis samples, these Type 1
errors based on dollar amounts are similar in magnitude to the Type 1 errors based on EITC
claiming. Overall, in the year just after the audits, audited taxpayers appear to receive roughly
$0.55 of every dollar of tax refunds that they would have been eligible for in the absence of the
EITC correspondence audits.
We have examined heterogeneity in the impacts of EITC correspondence audits along various
dimensions. Examining heterogeneity based on gender and the number of qualifying children is
motivated by the prior literature on labor supply responses to EITC benefits particularly among
single mothers. Examining heterogeneity based on access to a paid tax preparer is motivated by
the intuition that paid tax preparers may mitigate any misperceptions and help with any
corrections for taxpayers. Appendix Figure 7 presents the effects of the EITC correspondence
audits on the likelihood of claiming EITC benefits splits by gender of the selected taxpayer, the
number of qualifying children claimed on the selected return, and whether or not the selected
taxpayer had a paid tax preparer. Overall, we do not find clear evidence of heterogeneity in the
responses to the EITC correspondence audits based on these dimensions.
We also examine heterogeneity in the impacts of the EITC correspondence audits across ages of
the youngest qualifying child claimed on the selected return. Intuitively, taxpayers with older
qualifying children may not be leaving as much money on the table as taxpayers with younger
qualifying children because older qualifying children may age out beyond the EITC qualifying
age thresholds (less than age 19 or less than age 24 if a full-time student). Table 5 presents the
effects of the EITC correspondence audits on EITC claiming and tax refunds received split by
the age of the youngest qualifying child claimed on the selected tax return. For both the selfemployed and wage earner analysis samples, these results indicate larger and more persistent
decreases in EITC claiming and tax refunds received for taxpayers with younger qualifying
22

 children. For the self-employed, the cumulative decrease in tax refunds over seven years after the
audit selection is roughly $5500 for taxpayers with younger (ages 0-5) qualifying children and
roughly $1100 for taxpayers with older (ages 13+) qualifying children. For the wage earners, the
cumulative decrease in tax refunds over seven years after the audit selection is roughly $6100 for
taxpayers with younger (ages 0-5) qualifying children and roughly $3800 for taxpayers with
older (ages 13+) qualifying children.
2. Spillovers through Qualifying Children and Net Overpayments
Figure 3 illustrates the impacts of EITC correspondence audits on outcomes related to tracking
the qualifying children claimed on audited and nonaudited returns.6 The first outcome we
examine for the qualifying children is an indicator equal to one if the qualifying child is claimed
as a dependent by the primary taxpayer on the selected return in any years before or after the
year of selection. These results, shown in plots A and B of Figure 3 for the self-employed and
wage earner groups respectively, demonstrate that, just after the EITC correspondence audits,
there is a sharp decrease in the likelihood that qualifying children on audited tax returns are
claimed as dependents on subsequent tax returns by the selected taxpayers. This is consistent
with the sharp decreases in the probabilities of claiming EITC benefits and filing tax returns for
audited taxpayers in the years just after the audits.
We also examine changes in the likelihood of the qualifying children on selected returns being
claimed as a dependent on any tax return (including those filed by other taxpayers). The results
are show in plots C and D of Figure 3 for the self-employed and wage earners respectively.
These plots show a decrease in the likelihood of being claimed as a dependent on any tax return.
Turning to the quantitative results in Table 4, the difference-in-difference estimates indicate that,
for the qualifying children in both the self-employed and wage earner groups, the decreases in
the likelihood of being claimed as a dependent on any tax return is smaller (in absolute value)
6

For the analysis of the sample of qualifying children, we include dummies for the ages (in years) of qualifying
children when calculating the weights for the qualifying children. (These dummies are in addition to the variables
included when calculating the weights for primary taxpayers.) This explicitly ensures that the age distribution is
similar across the qualifying children in the audited and nonaudited groups. Thus, any differential patterns in the
claiming of the qualifying children are not due to differences in the age distribution of the qualifying children across
the audited and nonaudited groups.

23

 than the decrease in the likelihood of being claimed as a dependent by the selected taxpayer. This
indicates that while many of the qualifying children claimed on audited tax returns are not
subsequently claimed on any tax returns after the EITC correspondence audits, many of the
qualifying children also switch to being claimed as dependents by other taxpayers. Thus, the
EITC correspondence audits appear to have spillovers to other taxpayers.
For the qualifying children in the self-employed group, in the year after being audited, the
likelihood of being claimed by the selected taxpayer decreases by 0.24, and the likelihood of
being claimed as a dependent on any tax return decreases by 0.16. Thus, the likelihood of being
claimed as a dependent by another taxpayer increases by roughly 0.08. For the qualifying
children in the wage earner group, in the year after being audited, the likelihood of being claimed
by the selected taxpayer decreases by 0.25, and the likelihood of being claimed as a dependent
on any tax return decreases by 0.13. Thus, the likelihood of being claimed as a dependent by
another taxpayer increases by roughly 0.12. Over subsequent years after the EITC
correspondence audits, the changes in the likelihood of being claimed as a dependent mostly fade
out as qualifying children age beyond the age thresholds for being qualifying children (less than
age 19 or less than age 24 if a full-time student). Overall, these results indicate that after the
EITC correspondence audits, a relatively small share of all qualifying children claimed on
audited returns are subsequently induced to not be claimed on any returns after the audits
(roughly 0.16 and 0.13 of qualifying children on audited self-employed and wage earner returns
respectively). This may indicate that taxpayers may generally be aware of tax benefits associated
with claiming dependents and that current enforcement procedures may be effective at verifying
the existence of qualifying children. However, the switching of qualifying children on audited
returns to being claimed by other taxpayers may indicate that data or documentation for
verification of EITC-qualifying relationships between taxpayers and qualifying children may be
more difficult for current enforcement procedures or taxpayers to obtain.
Next, we examine EITC amounts associated with the qualifying children on audited and scoredbut-not-audited returns. While EITC claiming may decrease for audited taxpayers after the
audits, EITC amounts associated with qualifying children on audited returns may not decrease
significantly since some of the qualifying children on audited returns are subsequently claimed
24

 as dependents by other taxpayers. Plots E and F of Figure 3 show the changes in EITC amounts
associated with the qualifying children, and the corresponding difference-in-difference estimates
are presented in Table 4. Overall, the changes in EITC benefits associated with the qualifying
children on audited returns are relatively small.
We have examined heterogeneity in the effects of the EITC correspondence audits along
multiple dimensions. Similar to the analysis of heterogeneity in the effects on EITC claiming and
tax outcomes, we not find much evidence of heterogeneity in the effects of the EITC
correspondence audits on qualifying child outcomes along the dimensions of gender of the
selected taxpayer, the number of qualifying children claimed on the selected tax return or the use
of a paid tax preparer for the selected tax return. However, across ages of the qualifying children
we find evidence that older qualifying children are more likely to not be claimed after the EITC
correspondence audits, and younger qualifying children are more likely to switch to being
claimed as dependents on other taxpayers returns. Table 5 present the effects of the EITC
correspondence audits on the likelihood of being claimed by the selected taxpayer and by any
taxpayer split by the age of the qualifying children in the year of selection. For both the selfemployed and wage earner groups, there are larger decreases in the likelihood of being claimed
by the audited taxpayer for the younger qualifying children than the older qualifying children,
and there are larger decreases in the likelihood of being claimed as a dependent by any taxpayer
for the older qualifying children than the younger qualifying children. We focus on the first year
just after the audits, but these patterns continue for the other years after the audits as well. For the
self-employed, the decrease in being claimed by any taxpayer accounts for about 83%
(=.156/.189) of the decrease in being claimed by the selected taxpayer for older (ages 13+)
qualifying children, and about 49% (=.129/.263) for younger (ages 0-5) qualifying children.
Thus, the switching to being claimed as dependents by other taxpayers accounts for the
remaining 17% for older qualifying children and about 51% for younger qualifying children. For
the wage earners, the decrease in being claimed by any taxpayer accounts for about 71%
(=.156/.221) of the decrease in being claimed by the selected taxpayer for older (ages 13+)
qualifying children, and about 41% (=.109/.269) for younger (ages 0-5) qualifying children.
Thus, the switching to being claimed as dependents by other taxpayers accounts for the
remaining 29% for older qualifying children and about 59% for younger qualifying children.
25

 These spillovers to other taxpayers claiming qualifying children on audited tax returns and the
lack of sharp changes in EITC amounts associated with the qualifying children highlight the
distinction between gross overpayments of EITC benefits and net overpayment of EITC benefits.
Aggregate gross overpayments of EITC benefits will include any overpayments of EITC benefits
arising from taxpayers erroneously claiming qualifying children. However, if some of the
erroneously claimed qualifying children should have been claimed as qualifying children by
other taxpayers who then would have received EITC benefits, then these underpayments of EITC
benefits for these other taxpayers could be net out from the aggregate gross overpayments to
determine how many dollars were actually overspent in aggregate. The results based on the
current analysis samples indicate that roughly one third to one half of the changes in claiming
qualifying children after audits can be accounted for by the qualifying children being claimed by
other taxpayers, so aggregate net overpayments could be two-thirds or half as large as aggregate
gross overpayments.
3. Impacts on Employment and Earnings
In this section we analyze the impacts of the EITC correspondence audits on real economic
activity: specifically, wage employment and wage earnings. 7 Wage employment is measured
based on having a Form W-2 reported by an employer to the IRS, and wage earnings are
measured as the amounts reported on the W-2s. Before turning to any results, we discuss
possible theoretical channels and mechanisms through which EITC correspondence audits may
affect the likelihood of having a W-2 for wage employment.
First, a significant body of prior research on labor supply effects of EITC benefits has
highlighted how the EITC provides incentives for individuals to participate in the labor force (i.e.
extensive margin labor supply incentives) so that they have positive earned income and qualify

7

We have also examined changes in the likelihood of having 1099-MISC (contractor employment) income. This
analysis did not indicate any statistically significant or economically meaningful changes in the likelihood of having
contractor employment income. Roughly ten to twenty percent of the taxpayers in the self-employed analysis sample
have 1099-MISC income in any tax year, and roughly three to seven percent of taxpayers in the wage earner
analysis sample have 1099-MISC income in any tax year.

26

 for EITC benefits. If some audited taxpayers (possibly erroneously) perceive that they are no
longer eligible for the EITC, labor force participation (the likelihood of having a W-2) may
decrease after the EITC correspondence audits because of the perceived reduction in extensive
margin labor supply incentives.
While this first channel is based on perceptions and losses of EITC incentives, a potential second
channel through which the EITC correspondence audits may affects labor force participation is
through the losses of EITC benefits and reduced tax refunds. The losses of benefits may leave
audited taxpayers less able to finance to finance costs associated with employment (such as
transportation and childcare costs). Moreover, since individuals with younger children are more
likely to have childcare costs, these decreases in labor force participation may be larger for them.
The impacts of the EITC correspondence audits on the likelihood of having a W-2 for wage
employment may also vary based on whether or not taxpayers have a W-2 at the time of audit
selection. For example, in the year of selection, some wage earners may not have W-2s in the
year of selection but may still have earned income from cash-based employment. After the EITC
correspondence audits, these taxpayers may seek to obtain formal, W-2-documented employment
instead of their informal, cash-based employment.
Based on these possible mechanisms, we present the impacts of the EITC correspondence audits
on the likelihood of having a W-2 first for the full self-employed and wage earner analysis
samples, and then we examine the impacts for taxpayers with and without a W-2 in the year of
selection and based on the age of the youngest qualifying child. Figure 4 presents plots for the
likelihoods of having wage employment for these groups, and Table 6 presents the corresponding
difference-in-difference estimates. Plot A in Figure 4 for the self-employed analysis sample
illustrates that there may be slight increases in the likelihood of having wage employment in the
years after the audits, though these results are generally not statistically significant. Plot B of
Figure 4 for the wage earner analysis sample shows gradual increases in the likelihood of having
wage employment, and the change in the likelihood of having wage employment is statistically
significant by seven years after the EITC correspondence audits. Plots C and D of Figure 4 for
wage earners with and without W-2s in the year of selection respectively indicate that the
27

 increases in wage employment appear to be driven by increases for wage earners who do not
have W-2s in the year of selection. For wage earners who do have a W-2 in the year of selection,
the graphical evidence indicates that decreases in the likelihood of having a W-2 for wage
employment in the years after the audits relative to the years before. Furthermore, plot E
indicates slightly larger decreases in wage employment for wage earners with a W-2 in the year
of selection and with younger qualifying children, and plot F indicates slightly larger increases in
wage employment for wage earners who do not have a W-2 in the year of selection and had
younger qualifying children.
The estimates in Table 6 indicate that wage earners with a W-2 in the year of selection and
younger (ages 0-5) qualifying children have a 0.03 decrease in the likelihood of wage
employment in the year after the audits. To put the magnitude of this change in wage
employment in perspective, we compute the implied extensive margin (labor force participation)
elasticity. The numerator of the elasticity expresses the change in wage employment as a fraction
J.JK

of the baseline mean (J.LK = 0.032). The denominator of the elasticity is the change in the
average marginal net-of-tax rate. Since the marginal tax rates in the EITC phase-in portion of the
benefit schedule are 0.34, 0.40 and 0.45 for taxpayers with one, two and three or more qualifying
children respectively, we use the phase-in (subsidy) rate of 0.40 as a rough average marginal netof-tax rate for audited taxpayers. Next, since the Type 1 error rate is EITC claiming is about 40%
for wage earners, we assume that about 40% of audited wage earners perceive a loss of labor
force participation incentives from losing EITC benefits. The denominator of the elasticity is
J.SJ

then (0.40) ∗ RT.SJ U = (0.40 ∗ .286) = 0.114. Based on these assumptions, the implied
J.JK6

participation elasticity is RJ.TTS U = 0.280. We note that assuming a higher fraction of audited
taxpayers perceiving losses in EITC labor force participation incentives implies a lower
elasticity. In the extreme case that all audited taxpayers perceive losses of EITC participation
J.JK6

incentives just after the audits, the implied participation elasticity is RJ.6YZ U = 0.112. Chetty et al
(2011) survey evidence on extensive margin (labor force participation) elasticities and highlight
that quasi-experimental evidence indicates elasticities of roughly 0.25 across a variety of
settings. Thus, these estimates are consistent with this prior evidence. However, we note that
there are multiple caveats to keep in mind. First, this estimate is based on transitions from being
28

 employed to not having wage employment when losing EITC benefits. In contrast, prior EITCbased estimates of labor supply elasticities are based on transitions into employment when
gaining larger EITC benefits. Second, the observed changes in labor force participation
following the EITC correspondence audits may be driven by (mis)perceptions, qualifying child
changes or other factors affected by the EITC correspondence audits and not just labor supply
incentives.
We examine changes in the distributions of W-2 wages over subsequent years after being
selected for the audited or nonaudited groups. For this analysis, we follow a distribution
regression strategy by creating indicators for having W-2 wage earnings in $5000 wage bins
centered around $0, $5000, $10000, … and ³ $40000 and then estimating the above event time
regression specifications separately for each indicator. These estimates are presented in
Appendix Tables 2 and 3. Overall, the estimates for the $0 wage earnings bin are consistent with
the extensive margin, wage employment results described above. For the other wage earnings
bins, the results are frequently small and statistically insignificant, and there are no clear patterns
of changes in the distributions of wage earnings emerge.
4. Heterogeneity Based on Estimated Propensity Score
We examine heterogeneity across across groups with different estimated propensity scores (i.e.
different estimated probabilities of audit) to examine whether the results are robust to focusing
explicitly on observations with similar observables but different audit assignment and to examine
heterogeneity across groups with different fractions of observations that were assigned to be
audited. As described above, when computing the inverse probability weights, we estimate the
propensity score, or probability of being assigned to audit based on covariates that include the
types of rules broken, numbers of rules broken and other audit selection variables. While it is not
possible to present differences across groups with different types of roles broken or other specific
audit selection variables because the IRS does not publicly disclose these variables, we are able
to examine differences across groups with different estimated probabilities of being assigned to
be audited.

29

 Appendix Figure 6 presents the fraction of individuals that are audited by percentiles of the
estimated propensity scores for both the self-employed and wage earner samples. The plots
highlight that, while there are audited and nonaudited individuals in each percentile bin due to
the common support sample restriction, the lowest percentiles and highest percentiles have
relatively low overlapping audited and nonaudited individuals, while the middle percentiles have
higher overlapping audited and nonaudited populations. Based on this overlap in the middle of
the percentile distribution, we divide each analysis sample into quintiles (20 percentile bins)
based on the estimated propensity scores and then focus on observations in the 2nd quintile (20th
percentile up to 40th percentile), the 3rd quintile (40th percentile up to the 60th percentile) and
the 4th quintile (60th percentile up to the 80th percentile). For the self- employed and wage
earner samples respectively, the estimated probabilities of audit across these groups are roughly
0.10, 0.43 and 0.85 and 0.05, 0.23 and 0.89. Thus, observations in the 4th quintile have distinctly
higher estimated probabilities of being audited than observations in the other two lower quintiles.
Figure 5 presents results across these different quintiles for the self- employed, wage earners
with a W-2 in the year of selection, and wage earners without a W-2 in the year of selection. For
each of these three samples, the figure includes plots of two outcomes, EITC claiming and
having a W-2 for wage employment, across the three quintile groups based on the estimated
probabilities of being audited. The plots for EITC claiming highlight that, for each sample, the
groups with lower estimated probabilities of being audited (the 2nd and 3rd quintile) have
sharper decreases in EITC claiming in subsequent years after the EITC correspondence audits
than the quintile with the higher estimated probability of audit (the 4th quintile). Consistent with
these results on EITC claiming across the quintiles, the plots for having a W-2 for wage
employment show that the labor force participation patterns discussed above are most
pronounced for the quintiles with the lower estimated probabilities of being audited (the 2nd and
3rd quintiles). Specifically, for wage earners with a W-2in the year of selection, the decrease in
the likelihood of having a W-2 for wage employment just after the EITC correspondence audits
are most pronounced for the 2nd and 3rd quintiles, and for wage earners without a W-2 in the
year of selection, the gradual increase in the likelihood of having a W-2 for wage employment is
more pronounced for the 2nd end 3rd quintiles. Intuitively, the EITC correspondence audits may
be most surprising or unexpected for taxpayers in the lower quintiles, and taxpayers most
30

 surprised by the audits may be most likely to discontinue claiming EITC benefits after the EITC
correspondence audits and most likely to have the labor force participation changes described
above.
Corresponding to the graphical evidence in Figure 5, Table 7 presents the difference-indifference estimates across the samples and quintile groups. The decrease in EITC claiming are
largest for the lower (2nd) quintile, and it is persistent for the self- employed and wage earners
with a W-2 in the year of selection. In terms of dollars, the cumulative decreases in tax refunds
received after the EITC correspondence audits are $7635, $8610 and $4926 for the lower (2nd)
quintiles of the self- employed, wage earners with a W-2 in the year of selection and wage
earners without a W-2 in the year of selection respectively. Turning to the estimates for having a
W-2, in the year just after the EITC correspondence audits, the decrease in the likelihood of
having a W-2 is 0.08 for the lower quintile of wage earners with a W-2 in the year of selection,
and the decrease in the likelihood of having a W-2 appears to persist.
Overall, these results provide insights into possible heterogeneity and mechanisms behind the
main impacts described above, and these results indicate that the main results are robust to
dropping outliers with low or high estimated probabilities of audit (though this may not be
surprising given that these observations would get relatively low weighting based on the inverse
probability weighting).
5. Self-Employed EITC Maximizers
Prior analysis has documented widespread EITC maximizing or bunching behavior among EITC
recipients with self-employment income (see Saez 2010, Chetty Friedman and Saez 2013 and
Mortenson and Whitten 2018). Specifically, this behavior refers to EITC recipients with selfemployment income reporting exactly or very close to EITC Kink 1, which is the minimum
earned income necessary to receive maximum EITC benefits. Motivated by this prior research,
we examine differences in audit outcomes and subsequent behaviors across different levels of
earned income relative to EITC Kink 1. These results are presented in Figure 6. Plot A presents
the distributions of earnings relative to EITC Kink 1 for the self-employed analysis sample and a
31

 random sample of EITC recipients with self-employment income. The plot highlights that,
consistent with the random sample of EITC recipients with self- employment income, there is
widespread EITC maximizing behavior in the self-employed analysis sample, and the analysis
sample consists of a higher fraction of taxpayers reporting earned income at or just around EITC
Kink 1 relative to the random sample.
Plot B in Figure 6 illustrates that audit outcomes do not appear to very much across different
levels of earned income relative to EITC Kink 1. The result that the full disallowance rate does
not vary substantially around EITC Kink 1 may be striking given the clear spike in the
distribution of returns at EITC Kink 1. However, this may be due to EITC correspondence audits
verify only the existence of a self-employment business and not verifying specifically whether
self-employment income or expenses are over- or under-reported.
Plots C through F of Figure 6 presents plots of EITC claiming across different levels of earnings
relative to EITC Kink 1 and across different event times before and after the year of selection.
These plots are constructed by categorizing taxpayers into bins of earned income relative to
EITC Kink 1 in the year of selection, and then within each bin, we calculate the fraction of
taxpayers in each bin who claim the EITC at different years before and after the year of
selection. These plots illustrate that audited taxpayers with earned income close to EITC Kink 1
in the year of selection appear to have similar patterns as audited taxpayers with earned incomes
further away from EITC Kink 1 in the year of selection. Thus, audited EITC maximizers appear
to respond to the EITC correspondence audits similar to the way non-EITC maximizers respond.
In each year after selection, the decrease in EITC claiming for audited taxpayers relative to
nonaudited taxpayers is similar across different levels of earned income relative to EITC Kink 1.
If EITC maximizers had larger (smaller) decreases in EITC claiming rates after the
correspondence audits than non-maximizers, we would have expected more of a V-shaped
(hump-shaped) pattern in the differences across earned income relative to EITC Kink 1. Based
on these results, the factors behind EITC-maximizing or bunching decisions may be independent
from the factors behind responses to the EITC correspondence audits. For example, among two
EITC claimants with self-employment income, one may be more likely to report earned income

32

 at EITC Kink 1 than the other, but when audited, these taxpayers appear equally likely to not
respond (or not provide a sufficient response) and not claim EITC benefits subsequently.
6. Efficiency Costs of EITC Correspondence Audits
We consider two strategies to assess the efficiency costs (deadweight losses) associated with the
cumulative effects of the EITC correspondence audits. The first approach is from a longitudinal
perspective and the second approach is from a cross-sectional perspective. For the longitudinal
perspective, we start by assuming that there are N returns that are audited. In each year k = 1, 2,
... after the audits, the change in the number of subsequent EITC claims is 𝑁 ∗ 𝛿] where 𝛿] is the
estimated change in the probability of claiming EITC benefits k years after selection. The total
change in EITC claims across multiple years after the audits is then ∑_]?T 𝑁 ∗ 𝛿] , so the ratio of
the total change in EITC claims to the number of audited claims is given by
∆=

_
∑_]?T 𝑁 ∗ 𝛿]
= < 𝛿]
𝑁
]?T

The cumulative difference-in-difference estimates therefore reflect the cumulative impacts of an
audit on subsequent EITC claims and tax filing. Similarly, the sum of the difference-indifference impacts on tax refunds can be divided by the amount of dollars audited in the year of
selection to estimate the total (cumulative) change in dollars of tax refunds per dollar audited.
Based on the yearly impacts shown in Table 4, Table 8 presents the estimated cumulative
difference-in-difference impacts with standard errors. The cumulative impacts for the selfemployed analysis sample imply that, for every 100 EITC correspondence audits of this sample,
over subsequent years there are roughly 33 fewer EITC claims, 13 fewer filed tax returns, and
cumulative tax refunds decrease by roughly $3200 per audited individual (beyond the amount
disallowed on the selected return). In terms of audited dollars, for every $1 dollar of tax refunds
that is subject to an EITC correspondence audit, total future tax refunds are lower by roughly
$0.72. For the wage earners, the cumulative impacts imply that for every 100 EITC
correspondence audits of this sample, over subsequent years there are roughly 68 fewer EITC
claims and 14 fewer filed tax returns, and cumulative tax refunds decrease by roughly $3800 per
33

 audited individual. For every $1 of tax refund that is subject to an EITC correspondence audit for
the wage earner sample, total future tax refunds are lower by roughly $0.63. Since 15% of audits
result in full disallowances with confirm ineligibility, these estimates could be multiplied by 6.67
(=1/0.15) to put the impacts in terms of dollars of EITC benefits disallowed with confirmed
ineligibility.
Taking a cross-sectional perspective, we examine the cumulative impacts of the EITC
correspondence audits in terms of changes in the annual EITC participation (take-up) rate in each
year. We suppose that there are N EITC-eligible individuals in a given year. Within this
population, we assume that there is a fraction of individuals who are k=1, 2, … years since they
were audited. We denote this fraction by 𝑎] so the fraction of individuals who have never been
audited is given by 1 − ∑_]?T 𝑎] . For the individuals who have never been audited, we assume
the baseline EITC participation rate is 𝜃, and for individuals who have been audited, this
baseline participation rate is reduced due to the audits to 𝜃(1 − 𝑑] ) where 𝑑] is the estimated
percentage reduction in the probability of claiming EITC benefits (i.e. the difference-indifference estimate at event time k divided by the fraction of the nonaudited group claiming the
EITC at event time k; these estimates are presented in Appendix Table 1). The overall EITC
participation rate is the given by
_
𝑁(1 − ∑_]?T 𝑎] )𝜃 + (∑_]?T 𝑁𝑎] 𝜃(1 − 𝑑] )
∆𝑝𝑎𝑟𝑡 =  
= 𝜃[1 − < 𝑎] 𝑑] ]
𝑁
]?T

Thus, [1 − ∑_]?T 𝑎] 𝑑] ] reflects the percent change in the EITC participation rate (i.e. in the
absence if the audits, the baseline participation rate would have been 𝜃, but since some EITC
eligible individuals have been audited, the participation rate is reduced). For a back-of-theenvelope calculation, we assume that the EITC population is constant each year at 25 million
returns, that there are 500,000 audits each year so that 𝑎] = 𝑎 = 0.02, and that the estimated
impacts of the audits apply to all audited individuals. This last assumption would be an upper
bound on the number of taxpayers that the current estimates could apply, though it is not clear
whether or not taxpayers with higher risk returns would respond similarly to taxpayers with
lower risk returns. Using the difference-in-difference estimates as a fraction of baseline EITC
34

 claiming for the nonaudited group, we calculate [1 − ∑_]?T 𝑎] 𝑑] ] =[1 − (0.02) ∑_]?T 𝑑] ] =
0.988 for the self-employed and 0.971 for wage earners. Thus, because of some individuals
having previously experienced the EITC correspondence audits, the EITC participation rate may
be roughly one to or three percent lower in each year.
V. Conclusions
While prior studies have often focused on randomized research audits, this project exploits
random variation inherent in audit processes to estimate how operational audits affect taxpayer
behaviors. Research audits typically involve tax auditors making direct contact with audited
taxpayers and assisting them through the examination process, whereas operational EITC
correspondence audits do not involve such direct contact or assistance. The empirical analysis
documents that roughly 80% of EITC correspondence audits in the analysis sample have
outcomes of undelivered mail, nonresponse and full disallowance with passive agreement. As a
result, true incomes are often never observed in these audits (even though this is often a common
assumption in tax enforcement models of audits), and Type 2 error corrections (cases of
disallowances with confirmed ineligibility) make up only 15% of EITC correspondence audits in
the analysis sample.
The analysis provides insights for three central topics in tax enforcement: deterrence, spillovers
and impacts on real economic activity. Regarding deterrence, there are significant decreases in
EITC claiming and tax filing following the audits, but some audited taxpayers may leave benefits
on the table by foregoing potentially legitimate EITC claims or not claiming tax refunds based
on excess withholding. Regarding spillovers, qualifying children on audited tax returns are often
claimed by other taxpayers after the audits, so the EITC correspondence audits appear to cause
spillovers to these taxpayers. Regarding changes in real economic activity, audited taxpayers
have changes in the likelihood of having wage employment in the years after the EITC
correspondence audits, and the changes appear larger for taxpayers with younger (ages 0-5)
qualifying children than older (ages 13+) qualifying children.

35

 The impacts on many outcomes appear to fade out over subsequent years. This fade out can be
drive by qualifying children aging beyond EITC qualifying child age thresholds thereby causing
the EITC claiming rate for the nonaudited group to gradually converge to the lower EITC
claiming rate of the audited group. Future research may consider the impacts of soft-touch postaudit assignment outreach to audited and nonaudited taxpayers. For example, clarifications on
rules may be sent to taxpayers filing intermediate-risk returns but who are not randomly selected
for audit. Similar clarifications of rules and reminders to file could be sent to audited taxpayers
in the years after audit. Overall, further research can help improve the design and efficiency of
operational audits by aiming to reduce undelivered mail and increase appropriate responses and
by aiming to decrease potential mistakes by taxpayers in years after the EITC correspondence
audits.

36

 References
Advani, A., Elming, W., & Shaw, J. (2017). The dynamic effects of tax audits (No. W17/24).
Institute for Fiscal Studies.
Bastian, J., & Michelmore, K. (2018). The Long-Term Impact of the Earned Income Tax Credit
on Children’s Education and Employment Outcomes. Journal of Labor Economics, forthcoming.
Blumenthal, M., Erard, B., & Ho, C. C. (2005). Participation and compliance with the earned
income tax credit. National Tax Journal, 189-213.
Chetty, R., Friedman, J. N., & Saez, E. (2013). Using Differences in Knowledge across
Neighborhoods to Uncover the Impacts of the EITC on Earnings. American Economic Review,
103(7), 2683-2721.
Chetty, R., Guren, A., Manoli, D., & Weber, A. (2011). Are micro and macro labor supply
elasticities consistent? A review of evidence on the intensive and extensive margins. American
Economic Review, 101(3), 471-75.
Dahl, G. B., & Lochner, L. (2012). The impact of family income on child achievement: Evidence
from the earned income tax credit. American Economic Review, 102(5), 1927-56.
Dahl, Gordon B., and Lance Lochner. 2017. "The Impact of Family Income on Child
Achievement: Evidence from the Earned Income Tax Credit: Reply." American Economic
Review, 107(2): 629-31.
DeBacker, J., Heim, B. T., Tran, A., & Yuskavage, A. (2018). The Effects of IRS Audits on
EITC Claimants. National Tax Journal, 71(3), 451-484.
Eissa, N., & Liebman, J. B. (1996). Labor supply response to the earned income tax credit. The
Quarterly Journal of Economics, 111(2), 605-637.
Evans, W. N., & Garthwaite, C. L. (2014). Giving mom a break: The impact of higher EITC
payments on maternal health. American Economic Journal: Economic Policy, 6(2), 258-90.
Government Accountability Office, (2014). “IRS Correspondence Audits: Better Management
Could Improve Tax Compliance and Reduce Taxpayer Burden.” Report to the Committee on
Finance, U.S. Senate. GAO-14-479.
Holtzblatt, J. (1991). Administering Refundable Tax Credits: Lessons from the EITC
Experience. In Proceedings of the Annual Conference on Taxation Held under the Auspices of
the National Tax Association-Tax Institute of America (Vol. 84, pp. 180-186). National Tax
Association.
Hoynes, H., Miller, D., & Simon, D. (2015). Income, the earned income tax credit, and infant
health. American Economic Journal: Economic Policy, 7(1), 172-211.
37

 Hoynes, H. W., & Patel, A. J. (2018). Effective policy for reducing poverty and inequality? The
Earned Income Tax Credit and the distribution of income. Journal of Human Resources, 53(4),
859-890.
Kleven, H., Knudsen, M., Kreiner, C., Pedersen, S., & Saez, E. (2011). Unwilling or Unable to
Cheat? Evidence from a Tax Audit Experiment in Denmark. Econometrica, 79(3), 651-692.
Leibel, K. (2014). “Tax Compliance and Sources of Error for the Earned Income Tax Credit
Claimed on 2006-2008 Returns.” Internal Revenue Service, Technical Paper.
Lundstrom, Samuel. 2017. "The Impact of Family Income on Child Achievement: Evidence
from the Earned Income Tax Credit: Comment." American Economic Review, 107(2): 623-28.
Manoli, D., & Turner, N. (2018). Cash-on-Hand & College Enrollment: Evidence from
Population Tax Data and the Earned Income Tax Credit. American Economic Journal: Economic
Policy, forthcoming.
Maxfield, M. (2013). The effects of the Earned Income Tax Credit on child achievement and
long-term educational attainment. Michigan State University Job Market Paper. November.
McCubbin, J. (2000). EITC noncompliance: The determinants of the misreporting of children.
National Tax Journal, 1135-1164.
Meyer, B. D. (2010). The Effects of the Earned Income Tax Credit and Recent Reforms. Tax
policy and the economy, 24(1), 153-180.
Meyer, B. D., & Rosenbaum, D. T. (2001). Welfare, the earned income tax credit, and the labor
supply of single mothers. The quarterly journal of economics, 116(3), 1063-1114.
Mortenson, J., & Whitten, A. (2017). “Bunching to Maximize Tax Credits: Evidence from Kinks
in the US Tax Schedule.” American Economic Journal: Economic Policy, forthcoming.
National Taxpayer Advocate (2007). “2007 Annual Report to Congress” Volume 2.
Saez, E. (2010). Do taxpayers bunch at kink points?. American Economic Journal: Economic
Policy, 2(3), 180-212.
Schneller, J. P., Chilton, A. S., & Boehm, J. L. (2011). The Earned Income Tax Credit, LowIncome Workers, and the Legal Aid Community. Colum. J. Tax L., 3, 176.
Slemrod, J. (2016). “Tax compliance and enforcement: New research and its policy
implications.” University of Michigan, Ross School of Business working paper 1302.

38

 Slemrod, J., Blumenthal, M., & Christian, C. (2001). Taxpayer response to an increased
probability of audit: evidence from a controlled experiment in Minnesota. Journal of public
economics, 79(3), 455-483.
Strully, K. W., Rehkopf, D. H., & Xuan, Z. (2010). Effects of prenatal poverty on infant health:
state earned income tax credits and birth weight. American Sociological Review, 75(4), 534-562.

39

 Figure 1.
Background Patterns for Analysis Sample
B. EITC Claiming, Wage Earners

0

0

.2

.2

Fraction
.4
.6

Fraction
.4
.6

.8

.8

1

1

A. EITC Claiming, Self-Employed

-8

-7

-6

-5

-4

-3

-2 -1
0
1
2
Years since Selection

3

4

5

6

7

-8

-7

-6

-5

-4

-3

EITC Returns
Scored, Not Audited
Scored, Audited, Disallowed
Scored, Audited, Allowed

-2 -1
0
1
2
Years since Selection

3

4

5

6

7

4

5

6

7

EITC Returns
Scored, Not Audited
Scored, Audited, Disallowed
Scored, Audited, Allowed

D. Tax Filing, Wage Earners

0

0

.2

.2

Fraction
.4
.6

Fraction
.4
.6

.8

.8

1

1

C. Tax Filing, Self-Employed

-8

-7

-6

-5

-4

-3

-2 -1
0
1
2
Years since Selection

3

4

5

6

7

-8

-7

-6

-5

-4

-3

EITC Returns
Scored, Not Audited
Scored, Audited, Disallowed
Scored, Audited, Allowed

-2 -1
0
1
2
Years since Selection

3

EITC Returns
Scored, Not Audited
Scored, Audited, Disallowed
Scored, Audited, Allowed

.8
Fraction
.4
.6
.2
0

0

.2

Fraction
.4
.6

.8

1

F. Qualifying Children Claimed by Selected
Taxpayer, Wage Earners

1

E. Qualifying Children Claimed by Selected
Taxpayer, Self-Employed

-8

-7

-6

-5

-4

-3

-2 -1
0
1
2
Years since Selection

3

EITC Returns
Scored, Not Audited
Scored, Audited, Disallowed
Scored, Audited, Allowed

4

5

6

7

-8

-7

-6

-5

-4

-3

-2 -1
0
1
2
Years since Selection

3

4

5

6

7

EITC Returns
Scored, Not Audited
Scored, Audited, Disallowed
Scored, Audited, Allowed

Notes: Each plot is constructed by computing fractions of the specified outcome for each sample by years since selection. The year of selection refers to the year a return
is selected for risk scoring and random assignment to audit or non-audit status. The EITC Return sample is a 1% random sample of EITC returns for tax years 2008 through
2015, and the year of selection refers to the year the return is randomly drawn. Data used in creating these plots is unweighted.

 Figure 2.
Effects of EITC Correspondence Audits on Tax Outcomes
B. EITC Claiming, Wage Earners

1

2

3

4

5

6

Audited
Difference (Audited - Not Audited)

.6
.4
.2
-8 -7 -6 -5 -4 -3 -2 -1

0

1

2

3

4

5

6

7

Years since Random Assignment

Not Audited

Audited
Difference (Audited - Not Audited)

Not Audited

D. Tax Filing, Wage Earners

.8
Mean

0

3

4

5

6

7

-8 -7 -6 -5 -4 -3 -2 -1

Not Audited

4

5

6

Audited
Difference (Audited - Not Audited)

7

Difference

-2500 -2000 -1500 -1000 -500

3

4

5

6

7

Not Audited

5000

0

5000
3000
2000
1000
2

Years since Random Assignment

Mean

4000

500
0
-500
-1000

1

3

F. Tax Refund, Wage Earners

-1500

0

2

Audited
Difference (Audited - Not Audited)

E. Tax Refund, Self-Employed

-8 -7 -6 -5 -4 -3 -2 -1

1

4000

Audited
Difference (Audited - Not Audited)

0

Years since Random Assignment

Mean

2

3000

1

2000

0

Years since Random Assignment

1000

-8 -7 -6 -5 -4 -3 -2 -1

.4

-.2

.4

-.2

-.15

-.1

.6

.6

Difference

Mean

-.05
-.1

Difference

.8

.1

0

.05

1

.2

C. Tax Filing, Self-Employed

Difference

Mean

.8

1

.1
0
-.1
-.2
-.3

Difference

.4
7

Years since Random Assignment

1

0

.2

.6
-8 -7 -6 -5 -4 -3 -2 -1

Mean

-.1
-.3

-.2

Difference

0

.8

1

.1

A. EITC Claiming, Self-Employed

-8 -7 -6 -5 -4 -3 -2 -1

0

1

2

3

4

5

6

7

Years since Random Assignment

Not Audited

Audited
Difference (Audited - Not Audited)

Not Audited

Notes: Each plot illustrates estimated regression coefficients from regressing the outcome variable specified in the plot title on event time dummies, an indicator for being
an audited individual, interactions between the event time dummies and the audited indicator. The difference estimates and standard error bands refer to the estimated
coefficients and standard errors on the event time dummies interacted with the audited indicator. Means of the specified outcome variables are computed for each event
time for the non-audited group, and means for the audited group are computed as the means for the non-audited group plus the estimated difference for the
corresponding event time. Data used in the regressions is re-weighted using inverse probability weights. Standard errors are clustered based on tax year and audit
assignment group (audited or non-audited) and the year of selection.

 Figure 3.
Effects of EITC Correspondence Audits on Qualifying Child Outcomes
B. Claimed by Selected Taxpayer,
Wage Earners
1
.8
.6
.4

Mean

-.1

Difference

-8 -7 -6 -5 -4 -3 -2 -1

0

1

2

3

4

5

6

0

0

-.3

-.3

.2

.2

-.2

-.2

.4

Mean

Difference

.6

-.1

0

.8

0

1

.1

A. Claimed by Selected Taxpayer,
Self-Employed

7

-8 -7 -6 -5 -4 -3 -2 -1

Years since Random Assignment

Audited
Difference (Audited - Not Audited)

0

1

2

3

4

5

6

7

Years since Random Assignment

Not Audited

Audited
Difference (Audited - Not Audited)

Not Audited

D. Claimed on Any Tax Return, Wage Earners
1
.8
.6

Mean

0

Difference

.6

-.1

Mean

-.05
-.1

3

4

5

6

7

Not Audited

3

Audited
Difference (Audited - Not Audited)

3

4

5

6

7

4

5

6

7

Not Audited

4000

1000
0
-500

Difference

3000

500

5000

2

Years since Random Assignment

Mean

3000
2000
1000
1

2

F. EITC Amount Associated with Qualifying Child,
Wage Earners

4000

500
0
-500

Difference

-1000
-1500

0

1

Audited
Difference (Audited - Not Audited)

E. EITC Amount Associated with Qualifying Child,
Self-Employed

-8 -7 -6 -5 -4 -3 -2 -1

0

Years since Random Assignment

-1500 -1000

Audited
Difference (Audited - Not Audited)

.4
-8 -7 -6 -5 -4 -3 -2 -1

2000

2

Mean

1

1000

0

Years since Random Assignment

0

-8 -7 -6 -5 -4 -3 -2 -1

-.2

-.2

.4

-.15

Difference

.8

0

.1

.05

1

C. Claimed on Any Tax Return, Self-Employed

-8 -7 -6 -5 -4 -3 -2 -1

0

1

2

3

4

5

6

7

Years since Random Assignment

Not Audited

Audited
Difference (Audited - Not Audited)

Not Audited

Notes: Each plot illustrates estimated regression coefficients from regressing the outcome variable specified in the plot title on event time dummies, an indicator for being
an audited individual, interactions between the event time dummies and the audited indicator. The difference estimates and standard error bands refer to the estimated
coefficients and standard errors on the event time dummies interacted with the audited indicator. Means of the specified outcome variables are computed for each event
time for the non-audited group, and means for the audited group are computed as the means for the non-audited group plus the estimated difference for the
corresponding event time. Data used in the regressions is re-weighted using inverse probability weights. Standard errors are clustered based on tax year and audit
assignment group (audited or non-audited) and the year of selection.

 Figure 4.
Effects of EITC Correspondence Audits on Wage Employment
B. Wage Earners

0

1

2

3

4

5

6

.9
.7

Mean

.1

.6

7

-8 -7 -6 -5 -4 -3 -2 -1

Years since Random Assignment

Not Audited

2

3

Audited
Difference (Audited - Not Audited)

4

5

6

7

D. Wage Earners without W-2 in Year of Selection

-8 -7 -6 -5 -4 -3 -2 -1

0

1

2

3

4

5

6

7

.3
0

-.05

.6

-.02

0

.1

0

.2

Difference

.1

.9
.8

Mean

.7

.02

.4

1

.08
.06
.04

Not Audited

.15

C. Wage Earners with W-2 in Year of Selection

Difference

1

.05

Audited
Difference (Audited - Not Audited)

0

Years since Random Assignment

-8 -7 -6 -5 -4 -3 -2 -1

Years since Random Assignment

Audited
Difference (Audited - Not Audited)

0

1

2

3

4

5

6

7

Years since Random Assignment

Not Audited

Audited
Difference (Audited - Not Audited)

Not Audited

F. Wage Earners without W-2 in Year of Selection,
by Age of Qualifying Child
.15
.05
-.05

-.01

0

0

.01

Difference

.1

.02

.03

E. Wage Earners with W-2 in Year of Selection,
by Age of Qualifying Child

Difference

Mean

-8 -7 -6 -5 -4 -3 -2 -1

.5

-.1

.46

-.1

.48

-.05

0

Difference

.52

Mean

.5

0

Difference

.05

.8

.54

.2

.1

.56

A. Self-Employed

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

6

Years since Random Assignment

QC Age 0-5

QC Age 6-12

7

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

6

7

Years since Random Assignment

QC Age 13+

QC Age 0-5

QC Age 6-12

QC Age 13+

Notes: Each plot illustrates estimated regression coefficients from regressing the outcome variable specified in the plot title on event time dummies, an indicator for being
an audited individual, interactions between the event time dummies and the audited indicator. The difference estimates and standard error bands refer to the estimated
coefficients and standard errors on the event time dummies interacted with the audited indicator. Means of the specified outcome variables are computed for each event
time for the non-audited group, and means for the audited group are computed as the means for the non-audited group plus the estimated difference for the
corresponding event time. Data used in the regressions is re-weighted using inverse probability weights. Standard errors are clustered based on tax year and audit
assignment group (audited or non-audited) and the year of selection.

 Figure 5.
Heterogeneity based on Propensity Score
B. Has Wage Employment,
Self-Employed

0

Difference

-.05

-.1

-.1

-.3

-.2

Difference

0

.05

.1

.1

A. EITC Claiming,
Self-Employed

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

6

7

-8

-7

-6

-5

Years since Random Assignment

2nd Quintile

-3

-2

-1

0

1

2

3

4

5

6

7

Years since Random Assignment

3rd Quintile

4th Quintile

2nd Quintile

3rd Quintile

4th Quintile

D. Has Wage Employment,
Wage Earners with W-2 in Year of Selection

.05
-.05

-.3

0

-.2

-.1

Difference

0

.1

.1

.15

C. EITC Claiming,
Wage Earners with W-2 in Year of Selection

Difference

-4

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

6

7

-8

-7

-6

-5

Years since Random Assignment

2nd Quintile

3rd Quintile

-4

-3

-2

-1

0

1

2

3

4

5

6

7

Years since Random Assignment

4th Quintile

2nd Quintile

3rd Quintile

4th Quintile

F. Has Wage Employment,
Wage Earners without W-2 in Year of Selection,

-.4

-.05

-.3

0

.05

Difference

-.1
-.2

Difference

0

.1

.1

E. EITC Claiming,
Wage Earners without W-2 in Year of Selection

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

Years since Random Assignment

2nd Quintile

3rd Quintile

6

7

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

6

7

Years since Random Assignment

4th Quintile

2nd Quintile

3rd Quintile

4th Quintile

Notes: Each plot illustrates estimated regression coefficients from regressing the outcome variable specified in the plot title on event time dummies, an indicator for being
an audited individual, interactions between the event time dummies and the audited indicator. The difference estimates refer to the estimated coefficients on the event
time dummies interacted with the audited indicator. Data used in the regressions is re-weighted using inverse probability weights.

 Figure 6.
Effects of EITC Correspondence Audits for EITC Maximizers
B. Audit Outcomes by Earnings Relative to EITC Kink 1

.02

Fraction
.04
.06

.08

Fraction
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1

.1

A. Distributions of Earnings Relative to EITC Kink 1

-3000

-2000

-1000

0

1000

2000

3000

0

Earned Income relative to EITC Kink 1 (in Year of Audit Assignment)

-3000

-2000

-1000

0

1000

2000

3000

Full Disallowance
Undelivered Mail
Nonresponse
EITC Allowed

Earned Income relative to EITC Kink 1 (in Year of Audit Assignment)

EITC Returns
Audited, Disallowed

Scored, Not Audited
Audited, Allowed

D. EITC Claiming 1 Year Before Selection

0

0

.2

.2

Fraction
.4
.6

Fraction
.4
.6

.8

.8

1

1

C. EITC Claiming 4 Years Before Selection

-3000

-2000

-1000

0

1000

2000

3000

-3000

Earned Income relative to EITC Kink 1 (in Year of Audit Assignment)

Not Audited

-2000

-1000

0

1000

2000

3000

Earned Income relative to EITC Kink 1 (in Year of Audit Assignment)

Audited

Not Audited

Audited

F. EITC Claiming 4 Years After Selection
1
.8
Fraction
.4
.6
.2
0

0

.2

Fraction
.4
.6

.8

1

E. EITC Claiming 1 Year After Selection

-3000

-2000

-1000

0

1000

2000

Earned Income relative to EITC Kink 1 (in Year of Audit Assignment)

Not Audited

Audited

3000

-3000

-2000

-1000

0

1000

2000

3000

Earned Income relative to EITC Kink 1 (in Year of Audit Assignment)

Not Audited

Audited

Notes: Plot A is constructed by creating $100 bins of earned income relative to EITC Kink 1 and computing the fraction of each sample within each bin.
Values for EITC Kink 1 are determined based on filing status, number of qualifying children and tax year. Plot B is also constructed by creating $100
bins of earned income relative to EITC Kink 1 and then computing the fraction within each bin that has the specified audit outcome. Data used in these
plots are unweighted. Plots C through F present the fractions of individuals claiming EITC by earnings relative to EITC Kink 1, which is defined as the
lowest earned income level necessary to qualify for maximum EITC benefits. Data used in these plots are re-weighted using inverse probability
weights.

 Table 1: IRS Audit Frequencies & Outcomes
EITC Correspondence Audits
EITC Field Audits
Returns Examined as
Returns Examined as
Year
Percentage of Returns
Percentage of Returns
Returns Examined Percentage of All Individual
Returns Examined
Percentage of All
Examined with No Change
Examined with No Change
Correspondence Audits
Individual Field Audits
2008
420,879
0.379
0.074
41,378
0.096
0.100
2009
450,524
0.399
0.072
33,301
0.074
0.105
2010
551,836
0.434
0.083
33,366
0.072
0.100
2011
536,174
0.447
0.105
38,198
0.073
0.101
2012
513,156
0.444
0.083
45,375
0.090
0.086
2013
492,251
0.451
0.091
46,311
0.099
0.076
2014
437,430
0.445
0.102
43,559
0.109
0.066
2015
439,862
0.441
0.092
38,170
0.101
0.101
2016
391,490
0.475
0.072
36,717
0.107
0.094
Notes: Statistics are taken from the IRS Databook for the corresponding years. The table reports data from Table 9a: Examination Coverage. The statistics reported in the table
are based on total business and nonbusiness returns with Earned Income Credit benefits. Statistics are based on returns examined by fiscal year.

 Table 2: Summary Statistics
Audited
N = 432,219
Mean
Std. Dev.
0.657
0.475
34.431
12.758
0.757
0.429
14197.920 5249.188
4554.177 8779.902
0.477
0.499

Self-Employed
Scored but Not Audited
N = 473,938
Mean
Std. Dev.
0.526
0.499
33.781
12.138
0.773
0.419
14273.400
4886.133
5033.120
9523.917
0.502
0.500

Variable
Fraction Male
Age
Fraction with Filing Status = HOH
Total Income
Wages on Form 1040
Has Wage Income on Form 1040
Only Wage Income on Form 1040
Has Form W-2
0.537
0.499
0.563
Schedule C Income
9243.554 7659.331
8781.387
Adjusted Gross Income
13449.670 5223.848
13541.340
Balance Due (refund if negative)
-4511.418 1839.887
-4701.204
Earned Income
13192.800 3877.336
13211.820
Fraction with 1 Qualifying Child
0.467
0.499
0.537
Fraction with 2 Qualifying Children
0.465
0.499
0.256
Fraction with 3+ Qualifying Children
0.067
0.250
0.206
EITC Amount
4018.636 1134.660
4030.132
Fraction on Phase-In
0.273
0.445
0.273
Fraction on Maximum Credit
0.630
0.483
0.614
Fraction on Phase-Out
0.097
0.297
0.113
Fraction Filing with Paid Preparer
0.632
0.482
0.640
Fraction Filing with Software
0.989
0.104
0.991
Fraction Filing with VITA or IRS Centers
0.002
0.048
0.003
Fraction Self Prepared Paper Returns
0.009
0.096
0.008
Fraction Incarcerated
0.010
0.101
0.006
Fraction Claimed EITC in prior 3 years
0.596
0.491
0.701
Fraction Filed Return in prior 3 years
0.743
0.437
0.826
Fraction with W-2 in prior 3 years
0.684
0.465
0.726
Fraction Filed Return and Reported Self-Emp Income in prior 3 years 0.942
0.234
0.946
Fraction Incarcerated at any time in prior 3 years
0.016
0.124
0.014
Notes: Statistics are based on tax returns in 2008 through 2015. Dollar values are CPI-adjusted to 2016.

0.496
8302.554
4885.415
1838.557
4140.964
0.499
0.436
0.404
1245.107
0.446
0.487
0.317
0.480
0.097
0.051
0.087
0.079
0.458
0.379
0.446
0.226
0.118

1% Random Sample of EITC Returns
N = 330,116
Mean
Std. Dev.
0.389
0.487
39.281
11.479
0.580
0.494
12141.940
465933.100
5747.565
11845.260
0.454
0.498
0.474
6191.168
11998.150
-2751.619
12248.630
0.331
0.252
0.076
2433.629
0.480
0.326
0.194
0.692
0.932
0.015
0.034
0.003
0.816
0.942
0.658
0.944
0.005

0.499
465750.900
79464.500
3269.059
7786.941
0.471
0.434
0.264
1937.279
0.500
0.469
0.396
0.462
0.252
0.121
0.180
0.057
0.387
0.233
0.474
0.229
0.071

Audited
N = 895,065
Mean
Std. Dev.
0.669
0.471
35.250
13.385
0.812
0.390
16792.110 5557.037
16438.040 5482.780

Wage Earners
Scored but Not Audited
N = 1,170,290
Mean
Std. Dev.
0.640
0.480
34.546
12.907
0.821
0.384
17999.510
7501.304
17479.020
7475.454

1% Random Sample of EITC Returns
N = 1,203,713
Mean
Std. Dev.
0.328
0.469
36.516
11.125
0.639
0.480
17674.780
11756.560
16994.870
10678.680

0.873
0.856

0.333
0.351

0.840
0.948

0.367
0.221

0.720
0.967

0.449
0.178

16747.420
-6300.041
16410.690
0.488
0.449
0.062
3751.446
0.162
0.487
0.351
0.605
0.989
0.007
0.010
0.008
0.628
0.843
0.871
0.093
0.013

5564.786
1930.875
5447.369
0.500
0.497
0.240
1007.645
0.368
0.500
0.477
0.489
0.104
0.080
0.098
0.090
0.483
0.364
0.335
0.290
0.115

17952.870
-5679.055
17431.410
0.679
0.241
0.077
3212.959
0.189
0.355
0.456
0.569
0.993
0.010
0.006
0.003
0.686
0.901
0.933
0.091
0.013

7491.854
2051.504
7419.491
0.467
0.428
0.267
1203.738
0.391
0.479
0.498
0.495
0.084
0.097
0.078
0.059
0.464
0.299
0.251
0.288
0.113

17562.530
-4033.304
16879.830
0.396
0.254
0.074
2051.956
0.387
0.174
0.439
0.567
0.956
0.034
0.033
0.005
0.790
0.947
0.952
0.102
0.007

11770.210
3087.181
10700.600
0.489
0.435
0.261
1650.786
0.487
0.379
0.496
0.495
0.204
0.181
0.178
0.069
0.407
0.224
0.213
0.303
0.083

 Table 3: Audit Outcomes
Self-Employed

Wage Earners

Undelivered
Mail

Nonresponse

Full Disallowance with Full Disallowance with
Passive Disagreement
Active Agreement

Partial
Allowance

Full
Allowance

Full Sample

0.111

0.472

0.219

0.128

0.011

0.054

0.129

0.433

0.201

0.150

0.013

0.069

Age < 31, Men
Ages 31-40, Men
Ages 41-50, Men
Ages 51+, Men

0.123
0.113
0.117
0.122

0.541
0.484
0.453
0.428

0.203
0.230
0.229
0.222

0.094
0.122
0.142
0.161

0.007
0.011
0.012
0.014

0.027
0.036
0.041
0.046

0.138
0.134
0.138
0.148

0.499
0.460
0.426
0.372

0.190
0.212
0.218
0.217

0.118
0.137
0.157
0.189

0.009
0.015
0.015
0.016

0.043
0.037
0.040
0.052

Age < 31, Women
Ages 31-40, Women
Ages 41-50, Women
Ages 51+, Women

0.107
0.098
0.083
0.073

0.485
0.403
0.364
0.309

0.216
0.230
0.236
0.239

0.121
0.139
0.174
0.216

0.008
0.015
0.017
0.017

0.058
0.110
0.121
0.141

0.133
0.127
0.091
0.081

0.458
0.396
0.331
0.247

0.179
0.196
0.220
0.218

0.146
0.156
0.184
0.233

0.010
0.018
0.018
0.021

0.070
0.103
0.150
0.193

Earned income < $10k
Earned income $10k-$20k
Earned income $20k-$30k
Earned income $30k-$40k
Earned income $40k+

0.125
0.110
0.060
0.046
0.000

0.511
0.466
0.376
0.375
0.241

0.181
0.228
0.267
0.257
0.517

0.131
0.123
0.197
0.226
0.138

0.007
0.011
0.020
0.012
0.034

0.041
0.057
0.072
0.076
0.069

0.151
0.139
0.091
0.068
0.016

0.492
0.437
0.397
0.375
0.336

0.158
0.196
0.236
0.237
0.208

0.135
0.140
0.186
0.214
0.096

0.012
0.012
0.018
0.023
0.240

0.048
0.073
0.065
0.077
0.088

No paid preparer
Has paid preparer

0.142
0.093

0.499
0.456

0.188
0.237

0.112
0.138

0.012
0.010

0.042
0.062

0.158
0.110

0.439
0.428

0.190
0.208

0.136
0.159

0.016
0.011

0.056
0.077

No EITC claim in prior 3 years
Has EITC claim in prior 3 years

0.148
0.086

0.525
0.436

0.188
0.240

0.099
0.148

0.006
0.014

0.029
0.072

0.183
0.097

0.469
0.411

0.178
0.215

0.114
0.171

0.009
0.016

0.043
0.084

1 QC
2 QCs
3+ QCs

0.121
0.107
0.063

0.494
0.473
0.309

0.188
0.238
0.306

0.132
0.119
0.160

0.009
0.012
0.014

0.050
0.046
0.139

0.162
0.101
0.072

0.458
0.421
0.311

0.161
0.234
0.278

0.132
0.166
0.183

0.014
0.011
0.023

0.068
0.062
0.124

QC Age 0-5
QC Age 6-12
QC Age 13+

0.087
0.101
0.122

0.442
0.456
0.493

0.256
0.236
0.190

0.130
0.129
0.131

0.010
0.013
0.010

0.070
0.060
0.050

0.102
0.116
0.149

0.412
0.420
0.446

0.227
0.215
0.176

0.160
0.160
0.141

0.011
0.014
0.014

0.082
0.069
0.070

No W-2 in Year of Selection
Has W-2 in Year of Selection

0.136
0.089

0.491
0.455

0.200
0.236

0.107
0.147

0.009
0.012

0.054
0.055

0.326
0.096

0.490
0.423

0.114
0.216

0.044
0.168

0.008
0.014

0.016
0.078

Propensity Score Quintile 1 (Lowest)
Propensity Score Quintile 2
Propensity Score Quintile 3
Propensity Score Quintile 4
Propensity Score Quintile 5 (Highest)

0.095
0.093
0.112
0.124
0.102

0.369
0.421
0.454
0.493
0.467

0.254
0.235
0.215
0.196
0.240

0.114
0.136
0.134
0.127
0.126

0.075
0.031
0.009
0.008
0.011

0.091
0.079
0.071
0.048
0.050

0.098
0.108
0.147
0.122
0.132

0.395
0.421
0.427
0.442
0.427

0.208
0.191
0.189
0.201
0.205

0.147
0.159
0.138
0.149
0.153

0.054
0.036
0.016
0.012
0.012

0.096
0.082
0.078
0.069
0.066

Notes: Characteristics for heterogeneity are based on characteristics in the year of audit selection.

Partial
Allowance

Full
Undelivered
Full Disallowance with Full Disallowance with
Nonresponse
Allowance
Mail
Passive Disagreement
Active Agreement

 1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

Table 4: Impacts of EITC Correspondence Audits, Difference-in-Difference Estimates
A. Self-Employed
Type 1 Error,
Type 1 Error, Qualifying Child Claimed Qualifying Child Claimed
EITC Claiming Filing Tax Return
Tax Refund
EITC Claiming
Tax Refund
by Selected Taxpayer
by Any Taxpayer
-0.201
-0.144
-1287.518
0.327
-0.436
-0.236
-0.155
(0.018)
(0.022)
(176.156)
(0.022)
(0.046)
(0.026)
(0.017)
-0.098
-0.061
-728.849
0.199
-0.309
-0.205
-0.135
(0.023)
(0.029)
(200.528)
(0.04)
(0.071)
(0.025)
(0.022)
-0.042
-0.016
-547.136
0.097
-0.251
-0.155
-0.106
(0.023)
(0.026)
(212.439)
(0.05)
(0.082)
(0.027)
(0.028)
-0.004
0.018
-211.126
0.009
-0.116
-0.125
-0.089
(0.029)
(0.028)
(175.696)
(0.073)
(0.089)
(0.034)
(0.034)
-0.006
0.014
-250.083
0.016
-0.137
-0.075
-0.070
(0.030)
(0.030)
(200.271)
(0.076)
(0.099)
(0.038)
(0.033)
0.004
0.022
-188.717
-0.010
-0.106
-0.079
-0.085
(0.033)
(0.038)
(190.325)
(0.087)
(0.098)
(0.040)
(0.036)
0.022
0.035
-28.598
-0.064
-0.018
-0.060
-0.059
(0.035)
(0.032)
(260.809)
(0.108)
(0.159)
(0.048)
(0.043)

EITC Claiming Filing Tax Return
1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

-0.248
(0.032)
-0.186
(0.031)
-0.123
(0.029)
-0.071
(0.030)
-0.050
(0.026)
-0.008
(0.036)
0.003
(0.028)

-0.121
(0.034)
-0.099
(0.034)
-0.062
(0.039)
-0.011
(0.039)
0.010
(0.038)
0.059
(0.053)
0.078
(0.042)

Tax Refund
-1519.185
(202.489)
-972.350
(244.702)
-758.749
(175.255)
-375.781
(168.696)
-142.354
(235.925)
-43.645
(183.243)
52.545
(179.078)

Type 1 Error,
EITC Claiming
0.422
(0.036)
0.385
(0.045)
0.300
(0.056)
0.196
(0.071)
0.153
(0.073)
0.027
(0.126)
-0.013
(0.111)

B. Wage Earners
Type 1 Error, Qualifying Child Claimed
Tax Refund
by Selected Taxpayer
-0.454
-0.251
(0.048)
(0.047)
-0.359
-0.202
(0.083)
(0.045)
-0.339
-0.149
(0.069)
(0.046)
-0.202
-0.111
(0.082)
(0.050)
-0.089
-0.048
(0.138)
(0.027)
-0.030
-0.034
(0.122)
(0.027)
0.038
-0.007
(0.132)
(0.024)

Qualifying Child Claimed
by Any Taxpayer
-0.133
(0.021)
-0.101
(0.023)
-0.070
(0.023)
-0.059
(0.026)
-0.024
(0.029)
-0.018
(0.033)
0.017
(0.032)

EITC Associated with
Qualifying Child
-892.541
(127.869)
-563.525
(126.538)
-352.885
(122.149)
-265.851
(131.580)
-200.377
(160.462)
-262.983
(165.948)
-224.115
(158.042)

EITC Associated with
Qualifying Child
-479.442
(212.972)
-224.720
(170.241)
-40.475
(155.599)
55.072
(150.685)
214.593
(146.281)
265.842
(164.410)
485.964
(138.743)

Notes: Estimates are based on regression coefficients from regressing the outcome variable specified in the column heading on event time dummies, an indicator for being an audited
individual, interactions between the event time dummies and audited indicator. Data used in the regressions is re-weighted using inverse probability weights. Standard errors are
clustered based on tax year and audit assignment group (audited or non-audited) and the year of selection. The Type 1 Error Rate at each event time is comupted by dividing the
difference-in-difference estimate for the change in EITC claiming or tax refunds at the event time by the corresponding mean of the non-audited group at that event time.

 Dependent Variable = EITC Claiming
1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

QC Age 0-5 QC Age 6-12 QC Age 13+
-0.229
-0.211
-0.206
(0.018)
(0.018)
(0.018)
-0.138
-0.117
-0.079
(0.023)
(0.025)
(0.023)
-0.080
-0.057
-0.026
(0.024)
(0.025)
(0.018)
-0.043
-0.018
0.015
(0.028)
(0.030)
(0.022)
-0.043
-0.014
0.013
(0.028)
(0.033)
(0.023)
-0.033
0.004
0.012
(0.026)
(0.042)
(0.028)
-0.016
0.025
0.019
(0.029)
(0.045)
(0.015)

Dependent Variable = EITC Claiming
1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

QC Age 0-5 QC Age 6-12 QC Age 13+
-0.304
-0.301
-0.241
(0.023)
(0.019)
(0.044)
-0.229
-0.225
-0.170
(0.025)
(0.022)
(0.036)
-0.161
-0.157
-0.097
(0.022)
(0.020)
(0.035)
-0.102
-0.097
-0.058
(0.020)
(0.020)
(0.028)
-0.080
-0.064
-0.036
(0.019)
(0.022)
(0.023)
-0.055
-0.044
-0.007
(0.029)
(0.029)
(0.030)
-0.042
-0.031
0.014
(0.024)
(0.023)
(0.024)

Table 5: Impacts of EITC Correspondence Audits, Heterogeneity by Qualifying Child Age
A. Self-Employed
Dependent Variable = Qualifying Child
Dependent Variable = Tax Refunds
Claimed by Selected Taxpayer
QC Age 0-5
QC Age 6-12 QC Age 13+
QC Age 0-5
QC Age 6-12 QC Age 13+
-1707.090
-1368.767
-982.332
-0.263
-0.213
-0.189
(159.592)
(163.753)
(210.586)
(0.022)
(0.027)
(0.021)
-1114.634
-816.486
-411.332
-0.190
-0.138
-0.086
(189.615)
(192.57)
(221.746)
(0.023)
(0.027)
(0.019)
-714.882
-1223.954
-112.162
-0.139
-0.082
-0.033
(164.421)
(748.04)
(170.641)
(0.019)
(0.023)
(0.014)
-517.180
-265.773
99.577
-0.114
-0.058
-0.003
(181.749)
(146.885)
(166.118)
(0.018)
(0.022)
(0.012)
-584.917
-237.289
50.710
-0.088
-0.032
0.014
(204.808)
(205.61)
(158.235)
(0.021)
(0.025)
(0.011)
-523.441
-93.341
90.854
-0.099
-0.033
0.020
(164.171)
(222.958)
(172.306)
(0.018)
(0.026)
(0.011)
-347.463
10.042
163.563
-0.104
-0.034
0.022
(223.906)
(293.703)
(163.916)
(0.015)
(0.029)
(0.01)

Dependent Variable = Qualifying Child
Claimed by Any Taxpayer
QC Age 0-5
QC Age 6-12 QC Age 13+
-0.129
-0.113
-0.156
(0.013)
(0.012)
(0.016)
-0.083
-0.068
-0.074
(0.015)
(0.015)
(0.018)
-0.059
-0.034
-0.034
(0.014)
(0.013)
(0.014)
-0.049
-0.025
-0.013
(0.013)
(0.013)
(0.013)
-0.036
-0.020
-0.008
(0.015)
(0.012)
(0.013)
-0.045
-0.021
-0.004
(0.015)
(0.015)
(0.012)
-0.038
-0.012
-0.002
(0.018)
(0.017)
(0.011)

B. Wage Earners
Dependent Variable = Qualifying Child
Dependent Variable = Tax Refunds
Claimed by Selected Taxpayer
QC Age 0-5
QC Age 6-12 QC Age 13+
QC Age 0-5
QC Age 6-12 QC Age 13+
-1940.469
-1854.447
-1441.870
-0.269
-0.276
-0.221
(213.184)
(175.597)
(194.875)
(0.03)
(0.03)
(0.053)
-1467.025
-702.072
-1124.302
-0.192
-0.196
-0.128
(205.729)
(643.076)
(154.103)
(0.03)
(0.032)
(0.04)
-1010.878
-919.608
-633.115
-0.134
-0.129
-0.062
(187.533)
(150.981)
(190.132)
(0.032)
(0.033)
(0.034)
-604.786
-529.195
-344.429
-0.096
-0.083
-0.022
(135.346)
(126.993)
(172.833)
(0.032)
(0.036)
(0.028)
-491.712
272.054
-232.111
-0.050
-0.026
0.007
(127.761)
(733.004)
(162.344)
(0.017)
(0.021)
(0.022)
-348.238
-264.743
-78.611
-0.041
-0.012
0.014
(161.048)
(147.914)
(167.044)
(0.018)
(0.021)
(0.021)
-260.365
-188.655
87.716
-0.021
0.012
0.019
(152.361)
(126.352)
(139.126)
(0.009)
(0.017)
(0.021)

Dependent Variable = Qualifying Child
Claimed by Any Taxpayer
QC Age 0-5
QC Age 6-12 QC Age 13+
-0.109
-0.127
-0.156
(0.013)
(0.015)
(0.035)
-0.067
-0.080
-0.083
(0.015)
(0.017)
(0.032)
-0.043
-0.049
-0.039
(0.014)
(0.018)
(0.028)
-0.033
-0.037
-0.024
(0.017)
(0.022)
(0.023)
-0.022
-0.012
0.000
(0.01)
(0.013)
(0.020)
-0.015
-0.005
0.000
(0.012)
(0.016)
(0.016)
-0.007
0.012
0.013
(0.012)
(0.014)
(0.012)

Notes: Estimates are based on regression coefficients from regressing the outcome variable specified in the column heading on event time dummies, an indicator for being an audited individual, interactions
between the event time dummies and audited indicator. Data used in the regressions is re-weighted using inverse probability weights. Standard errors are clustered based on tax year and audit assignment group
(audited or non-audited) and the year of selection.

 1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

Self-Employed

Wage Earners

0.005
(0.026)
0.014
(0.025)
0.020
(0.023)
0.030
(0.021)
0.014
(0.020)
0.021
(0.024)
0.039
(0.039)

-0.022
(0.043)
-0.022
(0.042)
-0.011
(0.043)
0.025
(0.042)
0.034
(0.045)
0.081
(0.039)
0.093
(0.046)

QC Age 0-5
-0.033
(0.013)
-0.034
(0.014)
-0.029
(0.014)
-0.026
(0.014)
-0.028
(0.013)
-0.024
(0.016)
-0.018
(0.023)

Table 6: Impacts of EITC Correspondence Audits on Wage Employment
Dependent Variable = Has W-2 for Wage Employment
Wage Earners with W-2 in Year of Selection
Wage Earners without W-2 in Year of Selection
QC Age 6-12
QC Age 13+
QC Age 0-5
QC Age 6-12
QC Age 13+
-0.026
-0.026
-0.004
0.023
0.033
(0.015)
(0.019)
(0.039)
(0.034)
(0.027)
-0.024
-0.021
0.008
0.007
0.047
(0.016)
(0.021)
(0.036)
(0.032)
(0.027)
-0.023
-0.013
0.024
0.027
0.078
(0.016)
(0.021)
(0.042)
(0.037)
(0.033)
-0.010
-0.008
0.095
0.068
0.069
(0.016)
(0.018)
(0.048)
(0.038)
(0.034)
-0.010
-0.010
0.059
0.065
0.081
(0.015)
(0.017)
(0.045)
(0.044)
(0.037)
-0.008
-0.006
0.088
0.078
0.073
(0.016)
(0.017)
(0.039)
(0.037)
(0.023)
-0.009
-0.006
0.112
0.073
0.066
(0.017)
(0.017)
(0.037)
(0.040)
(0.027)

Notes: Estimates are based on regression coefficients from regressing an indicator for having a W-2 on event time dummies, an indicator for being an audited individual, interactions between the event time dummies and audited
indicator. Data used in the regressions is re-weighted using inverse probability weights. Standard errors are clustered based on tax year and audit assignment group (audited or non-audited) and the year of selection.

 Table 7: Impacts of EITC Correspondence Audits, Heterogeneity by Quintile of Probability of Audit
A. Dependent Variable = EITC Claiming
Self-Employed
1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

Lower Quintile Middle Quintile Higher Quintile
-0.345
-0.214
-0.018
(0.039)
(0.055)
(0.036)
-0.235
-0.085
0.078
(0.041)
(0.049)
(0.026)
-0.154
-0.033
0.117
(0.056)
(0.050)
(0.023)
-0.101
-0.006
0.137
(0.033)
(0.032)
(0.018)
-0.077
0.009
0.139
(0.035)
(0.030)
(0.028)
-0.055
0.024
0.155
(0.027)
(0.028)
(0.034)
-0.044
0.023
0.167
(0.024)
(0.045)
(0.046)

Wage Earners with W-2 in Year of Selection
Lower Quintile Middle Quintile
-0.430
-0.229
(0.03)
(0.02)
-0.322
-0.134
(0.028)
(0.021)
-0.226
-0.083
(0.028)
(0.024)
-0.173
-0.055
(0.032)
(0.018)
-0.142
-0.035
(0.024)
(0.021)
-0.117
-0.014
(0.02)
(0.023)
-0.103
-0.004
(0.019)
(0.012)

Higher Quintile
-0.048
(0.018)
0.012
(0.017)
0.055
(0.018)
0.092
(0.02)
0.113
(0.03)
0.111
(0.021)
0.115
(0.026)

Wage Earners with No W-2 in Year of Selection
Lower Quintile Middle Quintile Higher Quintile
-0.378
-0.182
-0.028
(0.031)
(0.052)
(0.053)
-0.206
-0.082
-0.020
(0.046)
(0.043)
(0.041)
-0.077
-0.036
-0.007
(0.064)
(0.048)
(0.047)
0.007
-0.007
0.008
(0.098)
(0.022)
(0.029)
0.089
0.011
0.025
(0.04)
(0.023)
(0.031)
0.060
0.044
0.043
(0.049)
(0.017)
(0.018)
0.042
0.033
0.029
(0.023)
(0.017)
(0.018)

B. Dependent Variable = Tax Refunds
Self-Employed
1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

Lower Quintile Middle Quintile Higher Quintile
-2078.250
-1055.569
-271.885
(267.156)
(337.670)
(200.745)
-1457.988
-397.967
220.903
(265.192)
(291.062)
(201.661)
-1787.536
-78.053
432.985
(565.722)
(282.142)
(199.546)
-729.290
60.566
568.883
(219.342)
(228.128)
(145.896)
-693.230
138.968
560.500
(178.254)
(183.457)
(182.606)
-503.400
223.543
656.744
(230.987)
(180.221)
(187.126)
-385.423
214.556
729.021
(122.179)
(235.579)
(207.659)

Wage Earners with W-2 in Year of Selection
Lower Quintile Middle Quintile
-2645.626
-1320.622
(202.042)
(126.879)
-1947.350
93.186
(171.23)
(952.424)
-1379.648
-538.043
(148.137)
(112.886)
-1055.505
-366.973
(170.434)
(96.204)
-117.033
-281.059
(881.468)
(101.235)
-766.809
-134.585
(131.246)
(83.192)
-697.710
-123.079
(118.558)
(70.597)

Higher Quintile
-381.181
(115.973)
-8.901
(115.467)
150.580
(115.915)
365.937
(89.678)
433.741
(93.788)
418.403
(112.699)
382.620
(118.706)

Wage Earners with No W-2 in Year of Selection
Lower Quintile Middle Quintile Higher Quintile
-2611.880
-1100.805
-221.832
(253.829)
(299.288)
(426.813)
-1556.895
-593.298
-543.941
(252.328)
(263.34)
(543.03)
-810.628
-325.890
-26.492
(263.524)
(196.729)
(279.87)
-278.625
-168.159
224.873
(448.209)
(107.043)
(162.336)
124.214
-46.258
267.517
(180.947)
(121.263)
(153.873)
64.340
174.737
258.987
(270.63)
(82.89)
(125.383)
143.264
159.863
349.520
(146.652)
(69.718)
(109.042)

C. Dependent Variable = Has Wage Employment
Self-Employed
1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

Lower Quintile Middle Quintile Higher Quintile
-0.042
-0.022
0.050
(0.030)
(0.029)
(0.024)
-0.041
-0.002
0.064
(0.029)
(0.029)
(0.023)
-0.033
0.004
0.067
(0.025)
(0.027)
(0.024)
-0.033
0.002
0.064
(0.027)
(0.029)
(0.025)
-0.031
0.015
0.048
(0.024)
(0.023)
(0.018)
-0.019
0.018
0.051
(0.025)
(0.025)
(0.021)
-0.028
0.025
0.075
(0.020)
(0.020)
(0.022)

Wage Earners with W-2 in Year of Selection
Lower Quintile Middle Quintile
-0.085
-0.031
(0.02)
(0.023)
-0.073
-0.013
(0.02)
(0.023)
-0.063
-0.004
(0.021)
(0.023)
-0.055
0.003
(0.021)
(0.024)
-0.057
0.004
(0.021)
(0.024)
-0.041
0.016
(0.021)
(0.025)
-0.035
0.021
(0.02)
(0.021)

Higher Quintile
-0.008
(0.019)
0.025
(0.021)
0.034
(0.023)
0.036
(0.023)
0.043
(0.018)
0.060
(0.017)
0.033
(0.017)

Wage Earners with No W-2 in Year of Selection
Lower Quintile Middle Quintile Higher Quintile
-0.054
-0.012
0.065
(0.038)
(0.032)
(0.057)
-0.027
-0.007
0.063
(0.042)
(0.034)
(0.06)
0.024
-0.004
0.064
(0.055)
(0.034)
(0.061)
0.053
0.029
0.096
(0.058)
(0.029)
(0.052)
0.111
0.031
0.087
(0.031)
(0.029)
(0.053)
0.071
0.055
0.102
(0.034)
(0.041)
(0.05)
0.079
0.050
0.090
(0.03)
(0.026)
(0.048)

Notes: Estimates are based on regression coefficients from regressing the outcome variable specified in the column heading on event time dummies, an indicator for being an audited
individual, interactions between the event time dummies and audited indicator. Data used in the regressions is re-weighted using inverse probability weights. Standard errors are
clustered based on tax year and audit assignment group (audited or non-audited) and the year of selection.

 Table 8: Cumulative Impacts of EITC Correspondence Audits
A. Self-Employed

B. Wage Earners

EITC Claiming

-0.325
(0.088)

-0.683
(0.112)

Tax Filing

-0.131
(0.103)

-0.146
(0.141)

Tax Refunds (impact per audited return)

-3242.03
(779.71)

-3759.52
(825.82)

Tax Refunds (impact per audited dollar)

-0.725
(0.174)

-0.628
(0.138)

Notes: Estimates are based on sums of the difference-in-difference estimates from 1
to 7 years after selection. Data used in the regressions is re-weighted using inverse
probability weights. Standard errors are clustered based on tax year and audit
assignment group (audited or non-audited) and year of selection.

 For Online Publication
Appendix Figure 1. Example of CP-75 Notice

Department of Treasury
Internal Revenue Service
5333 Getwell Road Stop 822

IRS Mem phis, TN 37501-0111

8018999546718

ERIC 0. 
123 HARRIS ST
HARVARD TX 12345

 

 

 

 

 

 

 

Notice CP75

Tax year 2016

Notice date October 15. 2017
Social Security number 999-99-9999

Your Caller ID 99999

To contact us Phone 1-366-897-0161
Page 1 of 3

We?re auditing your 2016 Form 1040
Supporting documentation requested

We need you to send us
information to support items
you claimed on your tax return.

We are holding the Earned
Income Credit (EIC), and/or the
Additional Child Tax Credit
(ACTC) portion(s) of your
refund pending the results of
the audit. If you claimed the
Premium Tax Credit (PTC), we
may also hold all or a part of
your refund due to a
discrepancy with your PTC.

Be sure to respond within 30
days from the date of this
notice or we'll disallow the
items being audited, and you
may owe additional tax.

 

What you need to do immediately

0 Review the list of itemswe're auditing and provide copies
of documentation to verify what you claimed on your tax
return. See the enclosed forms for complete instructions
for what you need to send.

0 Complete the Response form at the end of this notice, and
mail or fax it to us along with any documentation within 30
days from the date of this notice.

0 If you can't get your documentation ready in time, call us at
1-866-897-0161 to discuss your options.

 

Itemsthat require supporting
documentation

To qualify for:

The list below summarizesthe itemsthat require supporting
documentation. For complete instructions on what to send, see
the enclosed forms.

You shOuld:

 

Premium Tax Credit
Form 1040

Review the enclosed Form 14950, Premium Tax Credit
Verification

0 Sibmit documentation to verify what you claimed on your
return.

 

Appendix Figure 2. Example of CP-79 Notice

 

 

 

 

 

 

 

Department of the Treasury :06? (23:17:
. at year
Internal Revenue Servrce Notice date January 26. 2017
PO Box 149342 5 - be 
IRS Austin, TX 78714-9342 on: security nurn 
To contact us Phone: 
Your caller ID
ERIC D. JOHNSON Page 1 of 1

123 HARRIS ST
HARVARD, TX 12345

We denied one or more of the credits claimed on your tax return

 

We recently denied the following credits What you need to do
you claimed on your 2016 income tax

retum: 0 You don?t need to take any action at this time.
Earned Income Tax Credit (EIC) if you claim these credits in the future, make sure you meet all
0 American Opportunity Tax Credit the qualifying rules to get every credit for which you're eligible.
(AOTC) 0 Keep a copy of this notice for your records.
0 Child Tax Credit or Additional Child
Tax Credit (CTC or ACTC)

As a result, the next time you claim the
credits listed above, you must complete
and attach Form 8862, Information To
Claim Earned Income Credit, Child Tax
Credit. Additional Child Tax Credit or
American Opportunity Tax Credit After
Disallowance, to your tax return.

 

Claiming the credits on future returns In the future, if you claim the credits you must submit Form 8862
with your tax return. You will not receive the credits until we receive
your Form 8862.

After we receive your Form 8862, we?ll review your tax return. We
may send you an audit letter asking for additional information to
con?rm you?re eligible for the credits.

If we audit your return and deny the credits, we could impose a two-
year ban on your claiming the credits if we find you recklessly or
intentionally disregarded the rules. We could impose a ten-year ban
if we ?nd you fraudulently claimed the credits.

 

Additional information 0 Visit 
0 For tax forms or publications, visit or call
1-800-TAX-FORM (1-800-829-3676).
0 The following publications may be helpful:
- Publication 596, Earned Income Credit (EIC),
- Publication 972. Child Tax Credit
- Publication 970, Tax Bene?ts for Education

Appendix Figure 3. Example of Form 8862

 

8862 IMonna?on To Chin lnoomo Omit 
?my? Anal Dbalowanco 
:Im' punt:

mum
?mam mun-??

 

 

 

Ddonyoubo?r I 


I 




 

 

Muhammad?
lull??Fiac-qu 
a ?m

manta-:2. 
moo-rum



out? 0 can? 6 can?




macaw-m.

I 

nmmdan(woqvmm


a 


I ?tumour-am

Watercolor:


0 


 

 

 

 

 

 

 

magnum: 

7 
- - Dam UM

I'Van. 

I Nuns


?punt-mind?: I?D-mud?i. 
Nl?m


cum-momma: 

ammonia-t
Nina
must-promos:


mama unwanwma

 

 

 

 

 

 

 

Appendix Figure 4.
Example of IRS Letter 2205-B for Research (NRP) Audits

Department of the Treasury our
lntemal Revenue Service 

IRS 



Dear [enter Name]:

selectedatrandomfora complianceresearch
examination. We must examine randomly-selected tax returns to better tmderstand tax compliance and improve
We'll 
examination

The results of this and other compliance research examinations will improve our e?'orts to help taxpayers
andrednceburden
on taxpayers. Please read the enclosed Notice 1332, Why Your Ream! is Being Examined.

?'hat you need to do
Please call me on or before [insert date]. You may contact me from [insert time] to [insert time] at the
telephone number provided above.

What we will discuss

During our telephone conversation, we will discuss:
0 Items 

0 The examination process.
0 Any concerns or questions you may have.
~Thedate,timeandagenda forour?rstmeeting.

Someone may represent yon

You may have someone represent you during any part of this examination Ifyou decide you want
representation, the representative you authorize will need a completed Form(s) 2848, Pawn qf Attorney and
Declaration quapresanariw, before we can discuss any of your tax matters.

If you choose to have someone represent you, please provide a completed Form 2848 by our ?rst appointment.
You can mail or fax the form to me or have your representative provide it at the ?rst appointment, if you won't
be present You can obtain Form 2848 from our o?'ice, from our web site1 or by calling

(800) 829-3676.

Letter 2205-8 (Rev. 3-2017)
(2:09 38759?!

Appendix Figure 4 (continued).
Example of IRS Letter 2205-8 for Research (NRP) Audits

Ifyou ?led ajoint rem you and your spouse may attend the examination. Ifyou and."or your amuse choose
not to attend with your representative, you must provide completed Form(s) 2848. You should provide a
separate Form 2848 for each spouse ifyou ?ledjointly even ifyou use the same representative.

Your rights as a taxpayer

We have enclosed Publication 1, Your Rights as a Taxpayer and Notice 609, PrivacyAct Notice. The
Declaration of Taxpayer Rights found in Publication 1 discusses general rules and procedures we follow in
examinations. It explains what happens before, during, and a?er an examination, and provides additional
sources of information

A video presentation, "Your Guide to an IRS Audit," is available at The video
explains the examination process and will assist you in preparing for your audit.

Thank you for your cooperation and I look forward to hearing from you by [insert date].

Sincerely,
[Nmel
[Title]

Enclosures:

Publication 1

Publication 4134, Low Income Taxpayer Clinic List

Notice 609

Notice 1332

Letter 2205-8 (Rev. 3-2017)
(2:09 35759!

Appendix Figure 5.
Example of IRS Notice 1332 for Research (NRP) Audits

Why Your Return is
Being Examined

 

Your return was selected at random for a
research exam'naion. We usualy
select returns ior general exam'nations because

incorrect- We also raidomly select returns for
oompimoe research exam'mal'ons it order 
gather datalorusethroumouttheServiceb
inprave our tax system. We reoomize that
taxpayers who consistently meet al at their tax
obioations bear their iair share of the overal tax
burden. Our mission, however, iicludes exanin-
'ng enouui tax teams to ensure that the federal

errors on the exanined returns wil be corrected.



manner
abomhowtaxpayers meet meirtaxresponsibii-
ties.Tlis iiformation help usdeterm'newhat
changesto IRS forms, ptuications,andtaxlaws


are enlorwd, mdtoprogams desimedto help


system.

Theremaynotbemyenors in your return;




pendtiesand hterestdueas remiredby law.


everyone pays their fair share of taxes in accor-
dmcewiththelawsenacted byCongress.We
appreciate your cooperation with the examination
of your retum.

 

Wont-My 
Cmmamsv

mngav

 

Appendix Figure 6.
Effects of EITC Correspondence Audits, Unweighted
B. Fraction Audited by Propensity Score,
Wage Earners

0

0

.2

.2

Fraction Audited
.4
.6

Fraction Audited
.4
.6

.8

.8

1

1

A. Fraction Audited by Propensity Score,
Self-Employed

0

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Propensity Score Percentile

0

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Propensity Score Percentile

Notes: Each plot is constructed by computing five percentile bins based on the estimated probabilities of audit, and within each bin, each point is the fraction audited.
Horizontal lines show the overall fraction audited for each sample.

 Appendix Figure 7.
Heterogeneity in Effects on EITC Claiming
B. Gender, Wage Earners

0
-.3

-.3

-.2

-.1

Difference

-.1
-.2

Difference

0

.1

.1

A. Gender, Self-Employed

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

6

7

-8

-7

-6

-5

Years since Random Assignment

-4

-3

-2

-1

0

1

2

3

4

5

6

7

Years since Random Assignment

Female
Male

Female
Male

D. Number of Qualifying Children, Wage Earners

0

Difference

-.3

-.3

-.2

-.1

-.1
-.2

Difference

0

.1

.2

.1

C. Number of Qualifying Children, Self-Employed

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

6

7

-8

-7

-6

-5

Years since Random Assignment

-4

-3

2+ Qualifying Children
1 Qualifying Child

-1

0

1

2

3

4

5

6

7

6

7

2+ Qualifying Children
1 Qualifying Child

F. Paid Tax Preparation, Wage Earners

-.1
-.3

-.3

-.2

-.2

-.1

Difference

0

0

.1

.1

E. Paid Tax Preparation, Self-Employed

Difference

-2

Years since Random Assignment

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

Years since Random Assignment

No Paid Preparer
Has Paid Preparer

3

4

5

6

7

-8

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

Years since Random Assignment

No Paid Preparer
Has Paid Preparer

Notes: Each plot illustrates estimated regression coefficients from regressing the outcome variable specified in the plot title on event time dummies, an indicator for being
an audited individual, interactions between the event time dummies and the audited indicator. The difference estimates and standard error bands refer to the estimated
coefficients and standard errors on the event time dummies interacted with the audited indicator. Means of the specified outcome variables are computed for each event
time for the non-audited group, and means for the audited group are computed as the means for the non-audited group plus the estimated difference for the
corresponding event time. Data used in the regressions is not weighted. Standard errors are clustered based on tax year and audit assignment group (audited or nonaudited) and the year of selection.

 Appendix Table 1: Summary Statistics for Re-weighted Data
Self-Employed
Audited
Scored but Not Audited
N = 432,219
N = 473,938
Mean
Std. Dev.
Mean
Std. Dev.
0.613
0.487
0.568
0.495
35.148
12.695
35.148
12.578
0.773
0.419
0.745
0.436
14076.100 5177.822
13917.310
8229.068
4783.658 9343.996
4890.495
9860.987
0.477
0.499
0.463
0.499

Wage Earners
Audited
Scored but Not Audited
N = 895,065
N = 1,170,290
Mean
Std. Dev.
Mean
Std. Dev.
0.651
0.477
0.643
0.479
35.329
13.130
35.002
13.083
0.822
0.382
0.705
0.456
17287.170 6515.430
18821.320 11271.430
16830.410 6455.795
18146.020 10633.870

Variable
Fraction Male
Age
Fraction with Filing Status = HOH
Total Income
Wages on Form 1040
Has Wage Income on Form 1040
Only Wage Income on Form 1040
0.857
Has Form W-2
0.527
0.499
0.524
0.499
0.862
Schedule C Income
8848.828 8114.515
8579.310
9670.999
Adjusted Gross Income
13336.290 5065.490
13176.180
8072.444
17240.170
Balance Due (refund if negative)
-4471.062 1861.569
-4365.905
2023.737
-5985.815
Earned Income
13045.170 3988.967
12741.010
4723.704
16799.050
Fraction with 1 Qualifying Child
0.539
0.498
0.560
0.496
0.603
Fraction with 2 Qualifying Children
0.353
0.478
0.272
0.445
0.336
Fraction with 3+ Qualifying Children
0.106
0.308
0.139
0.346
0.061
EITC Amount
3939.854 1160.742
3781.431
1319.883
3464.698
Fraction on Phase-In
0.267
0.443
0.286
0.452
0.169
Fraction on Maximum Credit
0.631
0.483
0.586
0.492
0.442
Fraction on Phase-Out
0.102
0.303
0.128
0.334
0.389
Fraction Filing with Paid Preparer
0.621
0.485
0.610
0.488
0.588
Fraction Filing with Software
0.987
0.112
0.971
0.169
0.990
Fraction Filing with VITA or IRS Centers
0.002
0.047
0.003
0.054
0.007
Fraction Self Prepared Paper Returns
0.011
0.103
0.021
0.145
0.009
Fraction Incarcerated
0.010
0.099
0.010
0.099
0.007
Fraction Claimed EITC in prior 3 years
0.639
0.480
0.674
0.469
0.671
Fraction Filed Return in prior 3 years
0.775
0.417
0.812
0.390
0.869
Fraction with W-2 in prior 3 years
0.691
0.462
0.686
0.464
0.887
Fraction Filed Return and Reported Self-Emp Income in prior 3 years 0.941
0.236
0.937
0.243
0.100
Fraction Incarcerated at any time in prior 3 years
0.018
0.132
0.016
0.127
0.016
Notes: Statistics are based on tax returns in 2008 through 2015. Dollar values are CPI-adjusted to 2016. Statistics are based on re-weighted data.

0.350
0.345

0.838
0.845

0.369
0.361

6512.813
1956.106
6417.904
0.489
0.472
0.239
1108.871
0.374
0.497
0.488
0.492
0.099
0.085
0.093
0.084
0.470
0.338
0.317
0.301
0.124

18705.180
-5404.987
16322.530
0.607
0.196
0.054
2748.213
0.161
0.344
0.495
0.516
0.977
0.009
0.018
0.011
0.617
0.860
0.860
0.115
0.018

11186.770
6896.725
8755.102
0.488
0.397
0.226
1499.244
0.367
0.475
0.500
0.500
0.151
0.096
0.132
0.105
0.486
0.347
0.347
0.319
0.133

 Appendix Table 2: Impacts of EITC Correspondence Audits on Distributions of Wage Earnings
Difference-in-Difference Estimates by Wage Bin (columns) and Event Time (Rows)
A. Self-Employed
$0
$5,000 $10,000 $15,000 $20,000 $25,000 $30,000 $35,000
1 Year After Audit -0.004 0.007
0.002
0.001
-0.001
0.001
-0.001
-0.001
(0.023) (0.007) (0.004) (0.004) (0.002) (0.003) (0.003) (0.001)
2 Years After Audit -0.015 0.010
0.008
0.003
0.000
0.000
-0.001
0.000
(0.023) (0.006) (0.005) (0.003) (0.003) (0.003) (0.002) (0.002)
3 Years After Audit -0.019 0.006
0.008
0.006
0.002
0.001
0.001
0.000
(0.022) (0.006) (0.004) (0.003) (0.004) (0.003) (0.003) (0.002)
4 Years After Audit -0.032 0.007
0.007
0.006
0.005
0.002
0.003
0.003
(0.018) (0.006) (0.004) (0.003) (0.002) (0.002) (0.002) (0.001)
5 Years After Audit -0.020 0.003
0.004
0.003
0.002
0.003
0.003
0.001
(0.017) (0.006) (0.004) (0.002) (0.002) (0.002) (0.002) (0.001)
6 Years After Audit -0.028 0.010
0.002
0.003
0.002
0.002
0.003
0.004
(0.022) (0.009) (0.004) (0.003) (0.004) (0.001) (0.001) (0.002)
7 Years After Audit -0.038 0.010
0.010
0.004
0.001
0.006
0.007
0.001
(0.033) (0.007) (0.008) (0.004) (0.006) (0.004) (0.003) (0.002)

1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

0
0.032
(0.046)
0.025
(0.045)
0.012
(0.046)
-0.025
(0.046)
-0.037
(0.049)
-0.083
(0.044)
-0.104
(0.049)

$40,000
-0.003
(0.003)
-0.005
(0.004)
-0.005
(0.005)
-0.002
(0.005)
0.001
(0.005)
0.001
(0.005)
-0.001
(0.006)

B. Wage Earners
$5,000 $10,000 $15,000 $20,000 $25,000 $30,000 $35,000 $40,000
-0.001 -0.002
0.000
-0.005
-0.003
-0.006
-0.006
-0.011
(0.006) (0.009) (0.013) (0.016) (0.009) (0.006) (0.004) (0.004)
0.003
0.001
-0.003
-0.002
-0.002
-0.003
-0.006
-0.013
(0.005) (0.007) (0.012) (0.013) (0.009) (0.006) (0.004) (0.006)
0.003
0.000
-0.002
0.003
0.002
-0.002
-0.002
-0.015
(0.006) (0.007) (0.010) (0.012) (0.010) (0.006) (0.004) (0.008)
0.000
0.004
0.002
0.013
0.010
0.009
0.000
-0.013
(0.005) (0.006) (0.010) (0.010) (0.007) (0.007) (0.005) (0.010)
0.002
0.003
0.000
0.013
0.011
0.011
0.002
-0.006
(0.004) (0.006) (0.011) (0.012) (0.008) (0.006) (0.006) (0.010)
0.002
0.002
0.008
0.023
0.020
0.022
0.006
0.000
(0.005) (0.006) (0.007) (0.008) (0.007) (0.008) (0.007) (0.015)
-0.002
0.002
0.004
0.028
0.026
0.026
0.012
0.008
(0.005) (0.004) (0.005) (0.007) (0.012) (0.009) (0.009) (0.014)

Notes: Each coilumn represents a separate regression. Estimates are based on regression coefficients from
regressing an indicator variable for having wages in the wage bin specified in the column heading on event time
dummies, an indicator for being an audited individual, interactions between the event time dummies and audited
indicator. Wage bins are computed as $5000 earnings bins which are centered around the values given in the
headings. Data used in the regressions are re-weighted using inverse probability weighting. Standard errors are
clustered based on tax year and audit assignment group (audited or non-audited) and the year of selection.

 Appendix Table 3: Impacts of EITC Correspondence Audits on Distributions of Wage Earnings
Difference-in-Difference Estimates by Wage Bin (columns) and Event Time (Rows)
A. Wage Earners with W-2 in Year of Selection
$0
$5,000 $10,000 $15,000 $20,000 $25,000 $30,000 $35,000
1 Year After Audit 0.043 -0.003 -0.003
0.000
-0.005
-0.004
-0.007
-0.007
(0.020) (0.005) (0.005) (0.009) (0.013) (0.006) (0.008) (0.005)
2 Years After Audit 0.039 0.001
-0.001
-0.004
-0.003
-0.004
-0.005
-0.008
(0.022) (0.005) (0.004) (0.008) (0.010) (0.006) (0.006) (0.005)
3 Years After Audit 0.033 0.002
-0.002
-0.006
0.001
0.000
-0.004
-0.004
(0.022) (0.005) (0.004) (0.007) (0.009) (0.007) (0.005) (0.004)
4 Years After Audit 0.025 -0.005
0.001
-0.005
0.008
0.004
0.003
-0.006
(0.021) (0.004) (0.004) (0.007) (0.008) (0.004) (0.005) (0.005)
5 Years After Audit 0.022 0.000
-0.001
-0.006
0.005
0.004
0.003
-0.004
(0.021) (0.004) (0.004) (0.008) (0.009) (0.004) (0.005) (0.005)
6 Years After Audit 0.014 -0.001 -0.003
0.001
0.013
0.007
0.009
-0.006
(0.025) (0.005) (0.004) (0.006) (0.007) (0.005) (0.007) (0.006)
7 Years After Audit 0.016 -0.005 -0.003
-0.004
0.014
0.010
0.010
-0.002
(0.028) (0.004) (0.004) (0.003) (0.005) (0.010) (0.008) (0.008)

1 Year After Audit
2 Years After Audit
3 Years After Audit
4 Years After Audit
5 Years After Audit
6 Years After Audit
7 Years After Audit

0
-0.001
(0.025)
-0.019
(0.024)
-0.033
(0.029)
-0.050
(0.034)
-0.049
(0.038)
-0.048
(0.027)
-0.074
(0.024)

$40,000
-0.013
(0.006)
-0.016
(0.007)
-0.020
(0.008)
-0.026
(0.010)
-0.022
(0.011)
-0.033
(0.014)
-0.035
(0.012)

B. Wage Earners with No W-2 in Year of Selection
$5,000 $10,000 $15,000 $20,000 $25,000 $30,000 $35,000 $40,000
0.009
0.007
-0.002
-0.008
-0.002
-0.001
-0.001
-0.002
(0.009) (0.005) (0.005) (0.004) (0.003) (0.002) (0.001) (0.003)
0.015
0.008
-0.001
-0.001
0.002
0.000
-0.003
0.000
(0.007) (0.005) (0.005) (0.003) (0.003) (0.002) (0.002) (0.004)
0.010
0.010
0.008
0.004
0.003
0.000
-0.001
-0.001
(0.009) (0.006) (0.005) (0.004) (0.004) (0.002) (0.002) (0.005)
0.019
0.007
0.011
0.003
0.004
0.003
0.003
0.001
(0.010) (0.006) (0.007) (0.005) (0.004) (0.004) (0.001) (0.005)
0.014
0.010
0.002
0.008
0.003
0.007
0.002
0.004
(0.009) (0.007) (0.007) (0.006) (0.005) (0.002) (0.003) (0.006)
0.014
0.002
0.000
0.005
0.008
0.009
0.002
0.008
(0.007) (0.006) (0.006) (0.003) (0.004) (0.003) (0.002) (0.005)
0.009
0.003
-0.002
0.020
0.011
0.015
0.002
0.015
(0.008) (0.004) (0.003) (0.004) (0.005) (0.003) (0.001) (0.002)

Notes: Each coilumn represents a separate regression. Estimates are based on regression coefficients from
regressing an indicator variable for having wages in the wage bin specified in the column heading on event time
dummies, an indicator for being an audited individual, interactions between the event time dummies and audited
indicator. Wage bins are computed as $5000 earnings bins which are centered around the values given in the
headings. Data used in the regressions are re-weighted using inverse probability weighting. Standard errors are
clustered based on tax year and audit assignment group (audited or non-audited) and the year of selection.