New Work or Changes in Reporting?

New Work or Changes in Reporting?

A Framework for Measuring Self-Employment Trends in

Administrative Data∗

Andrew Garin, Emilie Jackson and Dmitri Koustas

December 14, 2021

Abstract

Increasingly, researchers in the United States look to measures of self-employment from

administrative tax data, which show larger growth in self-employment than mainstream

surveys of the labor market. This divergence may reflect changes in real activity not

well-captured by traditional surveys, but may also reflect changes in workers’ propensity to report their self-employment earnings to tax authorities, potentially in response

to incentives created by the tax code. Consistent with the latter explanation, we find

the extensive-margin growth in reported self-employment in administrative tax records

is concentrated among low-income households with children with incentives to report

self-employment earnings due to tax credit phase-ins. We employ a regression discontinuity design which compares households whose children are born at the end of

the tax year—and are eligible for these incentives—with households whose children

are born just a few days later and face no change in reporting incentives. We find a

sizable increase in reported self-employment among incentivized individuals for which

the only plausible interpretation is strategic reporting. These effects have grown over

time, with increases being concentrated in regions where other proxies for knowledge

of tax credits have grown as well. Incorporating these insights, we present a framework

to adjust trends in U.S. self-employment tax filings for changes in reporting behaviors

to extract true underlying trends in the composition of the workforce. Considering

counterfactual scenarios, we find that between 28 and 59 percent of the growth and all

countercyclicality in self-employment rates can be attributed to pure reporting changes.

∗

Garin: University of Illinois Urbana-Champaign (agarin@illinois.edu). Jackson: Michigan State University (emiliej@msu.edu). Koustas: Harris School of Public Policy, University of Chicago (email: dkoustas@uchicago.edu). We are grateful

to Katharine Abraham, John Guyton, Lawrence Katz, Elira Kuka, Kara Leibel, Alicia Miller, Bruce Meyer, Jacob Mortenson,

Max Risch, Michael Weber, Andy Whitten, Eleanor Wilking, and seminar participants at the Office of Tax Analysis at the

US Treasury and National Tax Association meetings for helpful discussions. We thank Emmanuel Saez for providing us with

code used to implement the calculations in Chetty, Friedman, Saez (2013) in the restricted-access administrative data. We are

grateful to the Russell Sage Foundation and the Sloan Foundation for generous support, and to Reigne Dadey for excellent

research assistance. Jackson gratefully acknowledges support from the Peter G. Peterson Foundation for a post-doctoral fellowship at NBER during which part of this research was conducted. This research was authorized through the IRS SOI Joint

Statistical Research Program. The researchers were granted access to tax data as IRS employees through agreements under the

Intergovernmental Personnel Act.

1

Introduction

Changes in self-employment rates are often looked to as a means to assess changes in both

the structure of the labor market and the extent of entrepreneurial activity in the broader

economy.1 However, a key challenge is that self-employment is hard to measure. In fact,

the share of workers reporting self-employment earnings in administrative tax records in the

United States has been growing, in puzzling contrast with self-employment activity reported

in workforce surveys.2 This divergence in findings could be consistent with either of two main

hypotheses: First, government surveys are incompletely capturing changes in the nature of

work, and tax records are better at picking up new forms of self-employment. Alternatively,

trends documented in administrative tax data may capture other factors that do not reflect

true changes in underlying work activity, such as changes in reporting/tax filing behavior.

This paper provides a framework for measuring self-employment using administrative

data, controlling for changes in reporting/tax filing behavior that may be occurring over

time. Unlike confidential surveys, individuals have incentives to be strategic about what

they report on tax filings—and both those incentives and taxpayer responses may be prone

to change over time. This is particularly true in the case of self-employment earnings which,

unlike employment income, can be purely self-reported without any third-party verification

(Slemrod, 2016). We build on key insights from the “bunching” literature (LaLumia, 2009;

Saez, 2010; Chetty, Friedman, and Saez, 2013; Kuka, 2014; Mortenson and Whitten, 2020) to

estimate the magnitude of the self-employment trend that is explained by changes in strategic

reporting.3 Ultimately, we provide an adjusted measure of self-employment that we believe

identify true changes in real work or entrepreneurship activity over the past two decades,

and can be used to examine changes in economic activity and labor market composition.

We first examine the extent to which self-employment trends in tax data are plausibly

driven by new kinds of app-based contract work that have grown exponentially in recent

1

For instance, recent literature on business dynamism focuses on non-employer business statistics which

include sole-proprietors (Davis, Haltiwanger, Jarmin, Krizan, Miranda, Nucci, and Sandusky, 2009; Fairlie,

Miranda, and Zolas, 2019).

2

It is now well documented that over the past two decades, self-employment rates reported in the Current

Population Survey have either remained flat or have declined since 2000 (Bureau of Labor Statistics, 2018;

Katz and Krueger, 2019), while the share of tax filers reporting self-employment income on their tax returns

rose dramatically between 2000 and 2014 (Abraham, Haltiwanger, Sandusky, and Spletzer, 2013; Jackson,

Looney, and Ramnath, 2017; Abraham, Haltiwanger, Hou, Sandusky, and Spletzer, Forthcoming).

3

Prior research finds that self-employment earnings are often reported at precise levels that maximize

refundable tax credits like the Earned Income Tax Credit, whereas this type of “bunching” behavior is rarely

observed in employer-reported wage earnings or in survey measures of self-employment earnings (LaLumia,

2009; Chetty, Friedman, and Saez, 2013; Kuka, 2014; Mortenson and Whitten, 2020). Moreover, this type

of strategic reporting has become increasingly prevalent since 2000 (Chetty, Friedman, and Saez, 2013;

Mortenson and Whitten, 2020).

1

years (Harris and Krueger, 2015; Abraham, Haltiwanger, Sandusky, and Spletzer, 2018;

Farrell and Greig, 2018; Katz and Krueger, 2019) or other type of “gig” work done by independent contractors. Standard survey questions about self-employment fail to capture new

types of “gig” work that are nonetheless being reported on tax returns.4 A unique feature

of U.S. tax data is that this gig work is separately observable—contract work performed

for firms or through online platforms is required to be separately reported by firms to the

Internal Revenue Service (IRS) (Jackson, Looney, and Ramnath, 2017; Collins, Garin, Jackson, Koustas, and Payne, 2019; Lim, Miller, Risch, and Wilking, 2019). We find that while

the prevalence of Online Platform Economy (OPE) earnings has grown dramatically since

2014, this cannot explain the rise in reported self-employment rates which peaked before

2014. While increases in independent contract work more generally can account for some

of the rise in self-reported self-employment rates between 2000 and 2005, trends in contract

work remained flat since 2005 as self-employment rates continued to climb, and year-to-year

changes have essentially zero correlation with overall self-employment trends.

Second, we evaluate the extent that changes in reporting responses account for the rise

in reported self-employment rates. Refundable credits like the Earned Income Tax Credit

(EITC) and the Child Tax Credit (CTC) not only create an incentive to report income close

to certain refund-maximizing amounts—they also create incentives to report self-employment

for some individuals who might not have otherwise. In particular, tax filers with children

with non-self-employment wage earnings in the EITC phase-in range (below the first “kink”

in the EITC schedule) face negative tax rates on the marginal dollar. Individuals in this

situation would strictly benefit from reporting some self-employment earnings rather than

none—regardless of their underlying earnings. These individuals might therefore increase

their tax refunds by choosing to report informal earnings they would have otherwise not

reported to the IRS (roughly a third of Americans report having some type of informal

income (Bracha and Burke, 2018), or, in some cases, by fabricating the income entirely. Just

as the propensity for individuals to “bunch” at refund-maximizing kinks has grown over time,

it is possible that changes in tax-code-incentivized reporting behaviors have contributed to

the observed rise in the number of individuals with self-employment income.

Consistent with this story, we find that the entire increase in the propensity to report secondary self-employment income among wage employees is driven by filers in this incentivized

group. While self-employment reporting rates have risen substantially among tax filers with

4

Although technically self-employed, contract workers may nonetheless perceive themselves to be traditional employees and report as such on surveys (Abraham, Hershbein, and Houseman, 2020; Abraham,

Haltiwanger, Sandusky, and Spletzer, 2020). Further, surveys like the monthly CPS only ask about primary

work activities and therefore may fail to capture trends in secondary or temporary self-employment captured

by tax returns.

2

children and wages in the phase-in range, those increases are absent for those with firmreported wages even slightly above this “kink” point where the marginal tax rate becomes

positive. In contrast, we find no differential growth above or below the first EITC kink

point for individuals with no kids; crucially, individuals that never face negative marginal

tax rates as the subsidy rate in the EITC phase-in region for childless adults is less than

self-employment (SECA) tax rate. Notably, we find no such rise in firm-reported contract

income receipt among those incentivized individuals with low wages and children, only a

rise in self-reported self-employment. We further document within-individual increases in

self-employment reporting in the year their first child is born—at which point they begin to

face marginal tax rates. Moreover, the propensity to begin reporting self-employment after

childbirth has grown secularly since 2000. Meanwhile, there is no trend in the propensity to

begin receiving firm-reported independent contractor income around childbirth in all years

since 2000.

A key question is whether the rise in self-reported self-employment rates—concentrated

among individuals with a strong EITC incentive to report additional earned income—reflects

a true increase in non-firm-facing self-employment activity or rather a greater proclivity to

self-report to tax authorities that is not indicative of any change in underlying labor supply.

To shed light on this question, we examine a sharp discontinuity in EITC eligibility based

on the date of birth of an individual’s first child motivated by earlier work by LaLumia,

Sallee, and Turner (2014). An individual’s EITC benefit for tax year t is calculated based

on the number of children in their household during year t—a first-born child born on or

before December 31 of year t would count towards the EITC calculation for that year t. By

contrast, a child born only a few days later at the start of year t + 1 would not count towards

the EITC calculation in year t. Accordingly, the EITC creates sharp differences in reporting

incentives for tax year t for individuals having their first child days apart at the very end of

year t and those having theirs at the very beginning of year t+1. This motivates a regression

discontinuity design comparing year-t self-employment earnings for parents with first birth

right before and right after the end of tax year t.5 If, in the limit examining births right

before and after midnight on December 31, parents face complete ex-ante uncertainty about

the year of birth, then parents should make identical labor supply decisions in year t.

We find clear evidence of a discontinuity in the probability of reporting self-employment

income around the December 31 cutoff. Our estimates indicate that having a child unexpectedly qualify towards a year’s EITC calculation increases the probability of reporting

5

Feldman, Katuščák, and Kawano (2016) examine the same discontinuity to evaluate parent responses

to children ageing out of the Child Tax Credit at age 17. Their work studies changes in real labor supply

behavior the year after the age-out, whereas we focus on pure reporting behavior at the end of same tax

year that eligibility changes.

3

self-employment income by 1.5 percentage points in years since 2010, but does not impact

the probability of having freelance income reported by firms. This result does not appear to

be driven by strategic timing of births to maximize tax refunds. If individuals were retiming

births to maximize EITC credits, then we should see individuals with births in December

have firm-reported earnings closer to the one-child refund-maximizing level; this, however,

is not the case, as firm-reported wage earnings and non-employee compensation do not vary

around the threshold date. Moreover, we find that the entire self-employment effect is driven

by individuals with wages in or below the phase-in range. These reporting effects have grown

steadily over time. Between 2000 and 2014, the baseline effect size has grown nearly threefold

from 0.5 p.p. to 1.5 p.p., and the effect conditional on having wages in the phase-in range

has grown over fivefold from 1 p.p. to over 5 p.p.

What explains this rise in reporting motivated by tax credits? We find that changes in

the generosity of the federal EITC over the period we study cannot plausibly drive the observed rise. However, prior research suggests that even during periods when policy remains

fixed, knowledge of the incentives created by the EITC and related tax provisions differs

across regions and spreads gradually over time (Chetty, Friedman, and Saez, 2013). This

process of gradual learning could lead to a steady rise in reporting of previously unreported

self-employment earnings or false reporting of self-employment income. Following Chetty,

Friedman, and Saez (2013), we use sharp-bunching rates among individuals in a hold-out

sample for each three-digit ZIP code in each year as a proxy for local knowledge of EITC

incentives and estimate an interacted version of our RDD specification. We find that highknowledge regions have dramatically larger effects than low-knowledge regions, and that

cross-sectional differences in knowledge and within-region changes in knowledge are associated with similar increases in the reporting effect. The spread of knowledge associated with

the geographic diffusion of hyper-strategic bunching behavior between 2000 and 2014 can

explain a majority of the increase in reporting behavior found in our RDD design.

Finally, we use our empirical estimates to adjust the raw self-employment data for strategic reporting incentives. Our key finding is that adjusting for changes in strategic reporting

behavior dramatically lowers the raw growth in tax-based measures of self-employment. We

still find evidence of modest growth in self-employment, particularly in the early 2000s and

in recent years with the rise of gig work mediated by online platforms, that is missing from

survey data.

Our work adds to a large public economics literature that highlights that trends observed

in administrative data may be impacted by reporting incentives, and should be interpreted

with caution (Blank, Charles, and Sallee, 2009; Slemrod, 2016). Prior work has documented

that this is particularly the case with self-employment earnings reported in tax data (Chetty,

4

Friedman, Ganong, Leibel, Plumley, and Saez, 2012; Chetty, Friedman, and Saez, 2013;

LaLumia, Sallee, and Turner, 2014). Our contribution is to document that these incentives

appear to be changing over time in a way that spuriously drives the measured trends in

self-employment. This is especially important as these raw trends are receiving increased

attention amidst policy concern about the rise of gig work and other changes in the nature

of work. Our new self-employment series adjusted for reporting trends, as well as our new

series on gig work, should be valuable to other researchers in this area.

This paper is organized as follows: in the next section, we discuss the measurement of

self-employment in administrative tax data. We also discuss important institutional details

of the U.S. tax code that relate to incentives to report self-employment. Section 3 presents

important motivating facts on self-employment trends. Section 4 introduces our main research design to estimate the response of self-employment to reporting incentives. Section 5

discusses knowledge of the tax code as a driver of the change. Section 6 provides new estimates of self-employment, which control for reporting trends. We also present an application

of our new data contributions, using our insights to examine the response of self-employment

to the business cycle. Section 7 concludes.

2

Data and Institutional Background

2.1

Self-Employment in Tax Records

In the United States, firms use W-2 information returns to report employees’ wage and salary

earnings to both the Internal Revenue Service and to the recipient, who then enters those

earnings on their 1040 tax return. In contrast, self-employment proceeds are self-reported to

the IRS by workers on their 1040 returns. In the U.S. tax system, self-employment earnings

are technically active income from a “sole-proprietorship” business. Self-employed workers

must keep track of their revenues, expenses, and net profits, and report those amounts on

Schedule C of their 1040 return.6 If an individual’s profits from self-employment exceed

$433, they must file Schedule SE and pay Social Security and Medicaid (“SECA”) taxes

equivalent to the payroll taxes withheld from their wages by their employer. Prior work

typically measures self-employment rates in tax data by comparing annual self-employment

earnings on Schedules C and SE to employment earnings on W-2 forms.7 In our analysis,

we primarily focus on Schedule SE filings, which are consistently available at the individual

(rather than filing unit) level, and begin in 1996.

6

Self-employment earnings from farming are reported on Schedule F instead of Schedule C.

For example, Abraham, Haltiwanger, Sandusky, and Spletzer (2020) focus on Schedule SE filers, while

Jackson, Looney, and Ramnath (2017) focus on Schedule SE and Schedule C filers.

7

5

Self-employment activity is highly heterogeneous (Hurst and Pugsley, 2011). While some

self-employed workers directly serve consumers (for example, barbers and plumbers), others

work for firms as non-employee independent contractors (for example, freelance IT consultants). Importantly, firms are required to report all payments to self-employed contractors

in excess of $600 on 1099-MISC Box 7 (replaced by 1099-NEC in 2020). These returns are

useful to us in two respects. First, they allow us to identify which self-employed workers are

working for firms in “alternative” work arrangements.8 . This enables us to assess whether

changes in how firms classify workers drives aggregate self-employment trends. Second, 1099s

provide information on worker revenues, which is third-party reported and not sensitive to

taxpayer reporting incentives. Unlike W-2s, however, 1099s report gross revenues and not

profits after expenses. Hence, recipients can still influence their taxable self-employment

earnings through their self-reported expenses on Schedule C.

The tax data further enable us to pay special attention to a new and growing class of

independent contract work mediated by online platforms. We refer to these arrangements—

which are a subset of the broader “gig” economy”—as the “online platform economy” (OPE).

We focus specifically on labor platforms where workers directly provide services to others

(for example, ridesharing or delivery) as opposed to platforms on which individuals sell

goods or rent capital (for example, craft merchandise sites or homesharing). OPE earnings

are reported to the IRS either as independent contractor earnings on 1099-MISC/NEC, or,

in some cases, as vendor revenues on 1099-K. We identify 1099 forms issued from OPE

companies using the method in Collins, Garin, Jackson, Koustas, and Payne (2019); we

discuss several important measurement issues in the Data Appendix.

The self-employed workforce in tax data may differ from self-employed workers identified

in survey and other data sources. In tax data, we observe total annual earnings by source.

By contrast, in surveys like the Current Population Survey (CPS), workers self-identify as

employed or self-employed based on their predominant activity in a reference week. “Pointin-time” measures will generally lead to lower estimates of the prevalence of self-employment

work—in particular, they under-count secondary, informal self-employment and firm-facing

self-employment (National Academies of Sciences and Medicine, 2020; Abraham, Hershbein,

and Houseman, 2020; Abraham, Haltiwanger, Sandusky, and Spletzer, 2020). Further, even

surveys that record retrospective annual earnings like the CPS Annual Social and Economic

Supplement (ASEC) may nonetheless under-count self-employed workers if such workers

fail to report small supplemental earnings or if independent contractors incorrectly perceive

themselves as employees.

In theory, the information reported on tax returns should be more comprehensive than

8

We explore this workforce in depth in earlier work (Collins, Garin, Jackson, Koustas, and Payne, 2019)

6

information on surveys, since individuals face significant penalties if they misreport their income. In reality, enforcement is imperfect and individuals have considerable scope to strategically under-report—or over-report—income to minimize their tax burden and maximize

refundable credits. Thus, changes in strategic reporting behavior over time can potentially

drive trends in tax data, whereas confidential surveys are often immune to such concerns.

2.2

Incentives in the Tax Code to Report Self-Employment

Certain provisions in the tax code may incentivize certain individuals to report self-employment

profits. In particular, the Child Tax Credit (CTC) and Earned Income Tax Credit (EITC)

phase-in with higher earnings up to a threshold level where there is a first “kink” in the EITC

schedule as the maximum credit level is obtained. Both this threshold level and the effective

subsidy rate depend on the number of children claimed on a tax return.9 The phase-in rate

of the EITC is 7.65 percent for childless households, 34 percent for families with 1 child, 40

percent for families with two children, and 45 percent for families with three children. The

CTC further increases the effective phase-in rate by 15 percentage points.

Individuals with children face net negative marginal tax rates in the EITC and CTC

phase-in range, even after incorporating all federal taxes including payroll (FICA/SECA)

taxes.10 These subsidies are sometimes enhanced by additional credits provided by states.

By contrast, the net subsidy disappears once earnings are above the applicable phase-in

range, since additional earnings are subject to the payroll tax but no marginal subsidy.11

Moreover, households with no children always face a positive marginal tax rate on earnings,

as the payroll tax rate exceeds the lower EITC phase-in rate.12

Crucially, these negative marginal tax rates create an incentive for some individuals to

report self-employment income that may not have been reported otherwise. In particular,

individuals with EITC-eligible children under 19 and W-2 wages below the first EITC kink

are strictly better off reporting self-employment income on their 1040 return. In theory,

such individuals could report self-employment earnings to maximize their net subsidy—this

9

The CTC applies to children under 18, and the EITC applies to children under 19.

Self-employed individuals are responsible for the employer and employee portion of payroll taxes, which

together are 15.3 percent. Half of payroll taxes paid (the “employer”-share) is deductible. Thus, the effective

marginal tax rate on self-employment earnings is 0.153*(1-0.0765)=14.1%. At the same time, EITC and CTC

credits for every dollar of self-employment income are also scaled back proportionately by the deduction.

11

Of course, an important incentive to report self-employment income to tax authorities and pay payroll

taxes is to contribute to future Social Security benefits. However, these incentives are faced by all taxpayers,

not just those facing negative marginal tax rates. Moreover, the increase in Social Security benefits may be

less salient to myopic consumers, and since benefits are based on the highest 35 years of earnings, earnings

for young workers are likely to have little impact on future benefits.

12

Even with the most generous state-EITC for childless individuals in the District of Columbia, they

nonetheless face a marginal tax rate of 0 in the phase-in range.

10

7

sort of hyper-strategic behavior would result in “bunching” at the first EITC kink point

(Saez, 2010).13 In practice, though, these tax incentives may lead individuals to report

additional self-employment income without going so far as to report this “optimal” amount.

For example, many individuals have actual informal income (Bracha and Burke, 2018) that

they may choose to report in full that they may not have reported otherwise. Additionally,

strategic taxpayers may avoid reporting at the refund-maximizing level if they think it might

appear suspicious to auditors.14

Appendix Figure A.1 shows the effective tax rate on reporting a first dollar of selfemployment income at different levels of W-2 wage earnings, taking into account all federal

taxes. Households without children face a positive marginal tax rate on self-employment income across the wage distribution. In contrast, for families with children, additional income

at amounts below the credit-maximizing kink point is taxed at a negative tax rate. Households with three children with wage income below the threshold for the minimum income

to maximize the EITC ($13,870 in $2015), and above the minimum income threshold for

the CTC, face a negative tax rate that implies a credit of up to 41 cents on the dollar.15

Similarly, tax units with two children face a marginal tax rate of up to -0.37 and tax units

with one child face a negative marginal tax rate of up to -0.31. As we will show below, these

incentives play an important role in whether households report self-employment income in

practice.

2.3

Sample Construction

In our analysis, we use de-identified full-count tax records incorporating both filer-reported

returns and information returns. Most of our analysis uses data from 2000-2018. While some

microdata are available back to 1996, 2000 is when information returns are first available for

1099 independent contractors.

We determine family structure based on information in tax filings as well as links from

parents to children from the Social Security Administration. These child links from the SSA

provide comprehensive coverage for children born since the early 1980s who were issued a

13

While the first EITC kink point is the refund-maximizing level in most years, Mortenson and Whitten

(2020) note that temporary tax credits shift the level in some years.

14

We note that individuals that would have reported low levels of self-employment earnings in the credit

phase-in range, regardless of incentives, face an incentive to report additional self-employment income. However, their additional self-employment earnings would not impact the extensive margin of self-employment,

which is our focus. We also note that individuals with W-2 earnings close to the first EITC kink point face

positive marginal tax rates on any self-employment earnings that exceed the kink level, which mitigates the

incentive to report additional amounts.

15

For three children, the phase-in rates on the EITC and CTC are 0.45 and 0.15, respectively. After

deductions, the overall marginal tax rate for a tax unit with wage earnings above $2,500 and below the

standard deduction is: (1-0.0765)*(-0.45-0.15+0.153)= -0.413.

8

Social Security number, providing a complete picture of the number of children under 18

years of age by 2000. More information on how we clean and process the data is provided

in the Data Appendix.

3

Trends

3.1

Do New Forms of Gig Work Explain the Rise?

The first possibility is that the rise in the share of the workforce with self-employment income

demonstrates a fundamental change in the labor market, driven by new types of contractbased “gig” work (Katz and Krueger, 2019). We begin by reporting the time-series for

the annual share of individuals in the workforce with positive net self-employment income

reported on a Schedule SE, our baseline measure of self-employment. In our analysis, we

define the “workforce” as all individuals who receive wage (W-2) earnings, contract earnings

(1099-MISC or OPE 1099-K), and/or self-employment (Schedule 1040-SE) earnings.16

As reported in Figure 1, the portion of the workforce with self-employment rose significantly (2 percentage points) after 2000, peaking in 2014.17 However, trends in 1099-reported

contract work cannot explain the overall rise in reported self-employment on tax returns.

We find that 1099-based contract work did become more prevalent between 2000 and 2007,

potentially contributing to the self-employment trend. However, as SE rates continued to

rise between 2007 and 2014 the share with contract income remained largely constant.

To investigate the contribution of new online platforms to the overall trend, we break

out the portion of contract workers who only have 1099-reported non-employee compensation from OPE platforms.18 The vertical line in the figure represents a structural break in

our freelance/contracting series, when online platform mediated work started to represent

a meaningful share of the workforce. Strikingly, we find that where gig work grows most

dramatically (2014 onwards), administrative measures of self-employment are flat or decline

slightly. This implies that new forms of “gig” work cannot explain the rise in self-employment

in tax data. On the whole, annual changes in the SE series and the non-employee compensation series between 2000-2018 show no statistically significant correlation.

16

Following Collins, Garin, Jackson, Koustas, and Payne (2019), we only include individuals with contract

earnings when they receive a W-2 or file a tax return.

17

Consistent with prior work (Jackson, Looney, and Ramnath, 2017; Abraham, Haltiwanger, Sandusky,

and Spletzer, 2020), Appendix Figure A.2 notes the sharp contrast between these trends measured in tax

data with comparable trends measured in the benchmark CPS-ASEC survey data.

18

Note that in 2017 and 2018, we impute growth in OPE 1099-Ks using state-level data from Massachusetts

and Vermont, both of which introduced a state 1099-K with a lower $600 filing threshold. As discussed in

Collins, Garin, Jackson, Koustas, and Payne (2019), prior to 2017, although the federal filing requirement

was $20,000, in practice this was not binding.

9

Nonetheless, there is substantial growth in new online platform work after 2014. This

raises the question: why did larger numbers of platform workers not result in more workers

reporting self-employment income? The vast majority of new platform workers only earn a

small amount of money over the course of the year, and typically expense over half of their

gross revenues. As a result, many workers have net earnings below the Schedule SE filing

threshold after deducting expenses on their Schedule C. Additionally, platform workers are

less likely to file a Schedule C or SE than other types of contractors with non-employee

compensation on a 1099-MISC.19 Therefore, trends in non-employee compensation may diverge from trends in reported self-employment earnings because of changes in compliance,

or changes in the composition of types of 1099-MISC work with different propensities to be

reported on Schedules C and SE. We directly examine this possibility in the discussion of

counterfactual scenarios later in the paper.

3.2

Self-Employment Trends and Credit Eligibility

We turn to our second hypothesis and examine trends in relation to incentives. As discussed

in Section 2.2, individuals with dependents on their tax return and earned income in the

phase-in range of the EITC and CTC have an incentive to report additional income in order

to receive a larger refundable credit. In families with children and wage income below the top

of the phase-in range, these incentives could motivate individuals to report self-employment

income who otherwise would have reported none.

As shown in Figure A.3a, the growth in SE propensity is substantially higher for EITC

recipients with children, growing from about 14 percent in 2000 to over 20 percent by 2010.20

In comparison, we see only modest changes in 1099 receipt and SE growth for childless filers,

who do not face negative marginal tax rates. Figure A.3c examines growth by total earnings.

We find SE rates rise most dramatically for those with less than $50,000 in total earnings

(in constant $2015), from about 9 percentage points in 2000 to over 12 percentage points by

the 2010s, with a sharp increase during the 2007-9 recession. The time pattern for SE rates

looks different for earners with more than $50,000: SE grows between 2003-2005, and falls

sharply in the recession, before returning to its 2005 level. Figure A.3d reports additional

trends by gender, showing that most of the increase occurs among women.

In theory, only families with wage earnings in the credit phase-in range are incentivized

to report positive self-employment income. Accordingly, we next examine more closely where

in the wage distribution we see a rise in rates of self-employment reporting.

19

We examine filing and expensing behavior among contractors in further detail in a companion paper

(Collins, Garin, Jackson, Koustas, and Payne, 2019)

20

Figure A.3b shows additional trends by presence of children only.

10

Figure 2 reports the change in self-employment propensity between 2000 and 2014, by W2 wage earnings and the presence of children.21 The results for Schedule SE self-employment,

reported in Figure 2, Panel A, are striking. We find dramatic growth in self-employment

among households with children and combined wages below the first EITC kink point, indicated by the maroon and green lines on the figure. These are precisely the households that

gain from reporting some small amount of self-employment rather than none. The change

in self-employment between 2000-2014 for households without children, or for households

with higher levels of wage income is much smaller. This raises the possibility that incentives

to report self-employment income on tax returns—to either report previously unreported

earnings or to fabricate income—are a factor driving growth in reports of self-employment

earnings.

Panel B reports the change over time for 1099-reported self-employment. Importantly,

this 1099-reported work is reported by firms, and therefore not subject to discretion by tax

filers. We see much smaller changes over this period for households, regardless of whether

they have children.

We further show in Appendix B that the changes in self-employment reported on 1040-SE

occurs precisely at the time of one’s first childbirth, when they first face reporting incentives,

and continues to grow in subsequent years. Moreover, we show that this increase in selfemployment at the time of childbirth has grown steadily since 2000. However, we find no

change in firm-reported contract income around childbirth. While suggestive that households

are motivated by EITC incentives to report self-employment income, these findings could

also reflect real increases in types of self-employment activity not reported on 1099 forms.

The need to distinguish between real labor supply changes and pure reporting behavior

motivates our research design in the next section.

4

Labor Supply Versus Reporting

4.1

Regression Discontinuity Design

A key question is whether the increases in self-reported self-employment rates documented

above reflect a true increase in self-employment activity or rather a greater proclivity to selfreport that is not indicative of any change in underlying labor supply. To isolate reporting

behavior, we examine a sharp discontinuity in EITC eligibility based on the date of birth of

an individual’s first child. An individual’s EITC benefit for tax year t is calculated based on

21

The overall change in self-employment is the integral of the wage-bin propensity times the share of the

population in each wage bin.

11

the number of children in their household during year t. A first-born child born on or before

December 31 of year t would count towards the EITC calculation for that year t, creating an

incentive for low-wage parents to report additional self-employment income on tax day. By

contrast, a child born only a few days later at the start of year t + 1 would not count towards

the EITC calculation in year t and their parents would have no such incentive. Because

exact birthdates are difficult to precisely forecast far in advance, parents expecting a first

child close to the end of the year are uncertain about what year the birth will occur in—and

the corresponding EITC status for year t—until labor supply decisions for year t are largely

sunk.

This motivates a regression discontinuity design comparing year-t self-employment earnings for parents with their first birth right before and right after the end of tax year t. If,

in the limit—examining births right before and after midnight on December-31—parents

face complete ex-ante uncertainty about the year of birth, they should make identical labor

supply decisions in year t. However, ex-post after the children are born, the two sets of parents face different returns to self-reporting self-employment. Accordingly, all differences in

self-reported tax year t self-employment between parents on either side of the discontinuity

will be due entirely to differences in reporting behavior.

The key identifying assumption is that true year-t labor supply is identical for parents on

either side of the December 31 cutoff. This assumption would be violated if parents are able to

schedule births around New Year’s Eve to maximize their tax refunds given their true year-t

labor supply.22 In our analysis, we test for strategic birth timing by examining discontinuities

in firm-reported earnings, which should exist if individuals are sorting. Additionally, we

plot the distribution of births around the end of tax years 2011-2018 (corresponding to our

benchmark sample below) in Appendix Figure A.4 and find that the distribution of births is

mostly smooth, with expected decreases around the Christmas and New Year’s holidays. To

ensure that selective avoidance of births during the New Year’s holiday does not influence

our results, we omit a three-day bandwidth “donut hole” from our baseline analysis sample

and examine robustness to the inclusion of the donut hole and alternate bandwidths.

We estimate regressions of the form:

yi = α + β1{datei ∈ December} + f (datei ) + i

22

(1)

LaLumia, Sallee, and Turner (2014) test for such behavior and, while they find some response of the timing of births to tax incentives, the effect is concentrated almost entirely among second and later births—the

effect of tax incentives on first-birth timing is negligible. However, LaLumia, Sallee, and Turner (2014) also

find that self-employed parents with births in December are much more likely to have earnings for that

year that are bunched around the first EITC kink point than parents with births in the following January,

consistent with a pure reporting response.

12

on the sample of childbirths in December and January. In this specification, datei is the

running variable—which is the child’s birth date measured as days since December 31—

and outcome yi is measured in tax year t. In our baseline analysis, we include births in a

fifteen-day bandwidth around midnight on New Year’s Eve, omitting births in the three-day

bandwidth “donut hole” around the start of the new year in case of any potential shifting

around the holiday, and let f (·) be a linear function allowing the slope to change around the

cutoff.23 We also examine alternative bandwidths and polynomial specifications as robustness

checks. The coefficient β on the indicator 1{datei ∈ December} is the RDD effect of having

a child in tax year t instead of tax year t + 1. We estimate Equation 1 on all parents with

first births in the specified window as identified in Social Security birth records. This sample

includes known parents listed in Social Security records, regardless of whether or not that

parent claims the child on their tax return.24

4.2

Baseline Results on Reporting Effects

We first estimate effects by pooling births all around the end of tax years 2011 to 2018—

focusing on recent years with elevated self-employment reporting rates. We begin by testing

for differences in third-party reported income around the discontinuity at the end of the year

in Table 1. The results in Column (1) show no effect on the propensity to have employment

income reported by firms on a W-2 form. Importantly, the results in Column (2) also show

no effect on the propensity to have self-employment income from independent contracting

work reported by firms on 1099 forms. Further, Columns (3) and (4) show that there is also

no impact on the level of W-2 reported wage and salary earnings in the tax unit.

As noted by LaLumia, Sallee, and Turner (2014), the benefits of retiming a birth do

not depend so much on the presence or amount of income per se as on the change in tax

liability that would occur as a result. Thus, as a stronger test of whether parents sort across

the cutoff date to maximize their EITC based on their firm-reported employment earnings,

we calculate a “simulated EITC” similar to Chetty, Friedman, and Saez (2013) that is the

credit each individual would earn if their birth occurred in December of tax year t based on

their and their spouse’s (if married filing jointly) W-2 wages from year t. If individuals are

sorting to maximize their EITC, then the wages of parents immediately to the left of the

discontinuity should predict higher credits than the wages of those to the right. However,

the differences in simulated credits at the discontinuity, displayed in Column (5) of Table

23

We use robust standard errors rather than clustering standard errors by date. Kolesár and Rothe (2018)

show that SEs clustered by a discrete running variable ”do not guard against model misspecification, and

[...] have poor coverage properties”.

24

We examine all parents with known births. However some births in the data are missing identifiers for

one or both parents. Missing identifiers are more common for fathers than for mothers in the birth data.

13

1 and graphically in Panel A of Figure 3, are negligible and statistically indistinguishable

from zero. Moreover, the results in Column (4) show that individuals are equally likely to

have wages in the EITC phase-in region on both sides of the cutoff. These results support

the claim that labor supply is fixed around the cutoff.

Accordingly, we can measure the difference in individuals’ incentives to report SE income around the cutoff given their wage earnings. We measure the incentive to self-report

self-employment income using the effective federal marginal tax rate (inclusive of SECA

taxes) that individuals would face on a first dollar of self-employment income beyond their

firm-reported W-2 earnings, given the year their child was actually born.25 Panel B of Figure 3 shows that 25 percent of individuals with December births—but none with January

births—face negative marginal tax rates on reported self-employed. We report point estimates separately for individuals with W-2 wages above and below the first EITC kink point

in the first two columns of Appendix Table A.1. We find that this change in reporting incentives occurs only in those with wages below the first EITC kink, among whom a December

birth virtually always makes the MTR negative (and 30 percentage points lower on net).

While these reporting incentives exist in theory, the returns on reporting can only be

actualized if an individual files a tax return and claims their child as an eligible dependent.

We examine effects on claiming behavior in Columns 3-6 of Appendix Table A.1 and within

narrower wage bins in Figure A.5. Notably, a first birth only increases their propensity

to claim a child on their return by about 60 percentage points.26 This is partly because

individuals are not able to claim children as dependents if they are claimed by another

filer (e.g. another parent on a different return or a grandparent), and partly because some

individuals already claimed children as dependents prior to the first birth identified in the

SSA data.27 This implies our first-birth RDD identifies an “intent to treat” effect rather

than the direct effect of having a claimable child. Hence, the true effect of having an eligible

child is larger by a factor of 1/0.6 = 1.7. In contrast, the propensity to report children

around a first birth is markedly lower among individuals with wages below the first EITC

kink. This low claiming rate also reflects much lower filing rates among low earners—likely

including many individuals who do not file despite being eligible for a net refund.28

25

We calculate this for all individuals regardless of filing status based solely on tax-unit-level W-2 reported

earnings and the date of birth of the child using TAXSIM

26

Note that there is an effect on 1040 filing for individuals with wages slightly above the first EITC kink

point, despite there being self employment reporting incentives or reporting effects for those individuals.

This is expected—these individuals qualify for a refundable EITC benefit given their wages so long as their

children were born during the tax year, but must file a 1040 return in order to claim that benefit.

27

This would occur, for example, if an individual had married another person with a child from a previous

marriage. This also could occur do to incorrect identification of first births due to missing SSA data on one’s

first child’s parentage.

28

A leading reason for low EITC takeup is that many eligible individuals do not file a tax return. For

14

Panel C of Figure 3 presents our main results on self employment reporting. There is

a clear break in the propensity to self-report self-employment income after December 31.

Notably, the relationship between date of birth and reported self-employment is completely

flat away from the discontinuity, further suggesting that the result is not driven by local

sorting around the cutoff date. By contrast, though the incentives are the same. There is

no change in the propensity to have 1099 non-employee compensation (reported by firms)

around the same cutoff date in Panel D.

Panel A of Table 2 reports baseline RDD estimates of the response of self-employment

reporting to reporting incentives. The estimates in Column 1 indicate that, at the discontinuity, having a qualifying child for EITC determination increases the probability of reporting

self-employment income by 1.34 percentage points (over a baseline rate of 7.19 percentage

points). Given our prior results, these estimates can be interpreted as pure reporting effects.

The result in Column (2) confirms that there is no “placebo” effect on self-employment in

the tax year prior to tax year t.29 We explore the robustness of our baseline estimates to alternative RDD specifications in Appendix Figure A.6. The estimates are not sensitive to the

inclusion of a “donut-hole” and are highly stable across a wide range of bandwidths. We also

find that our results are robust to including a quadratic polynomial in the running variable

with slope breaks around the cutoff, though overfitting the quadratic on a small number of

days results in unstable point estimates in specifications with smaller bandwidths.

Interestingly, the result in Column (5) of Table 2 implies that while some individuals

who begin reporting self-employment due to the incentive “bunch” their earnings within

$500 of the the first EITC kink, this sort of “sharp-bunching” behavior accounts for less

than one-fifth of the baseline SE reporting effect. In addition, although we found no effects

of reporting incentives on the propensity to have contract income reported on a 1099, the

result in Column (6) implies that over a quarter of the baseline self-employment reporting

effect comes from individuals who have 1099 contract income but who nonetheless would not

have reported any self employment earnings on their 1040.

The results in Panels B and C of Table 2 confirm that these reporting responses are

concentrated among individuals with a strict incentive to do so. As predicted, we find

no effect of self-employment reporting for individuals with wages above the EITC phase-in

range. By contrast, we find that the presence of incentives increases the share reporting self

instance, in tax year 2016, only single filers with wages above $10,350 are legally required to file a 1040

($20,700 for married filers). Recent work by Goldin, Homonoff, Javaid, and Schafer (2021) finds that simply

informing EITC-eligible individuals about free tax preparation services increases their propensity to file a

1040 and benefit from the EITC.

29

The magnitude of the reporting effect in Column (1) is remarkably similar to the post-2010 observed

increase in self-employment around first births in our event-study analysis in Appendix B, suggesting that

the those changes likely reflect pure reporting behavior given fixed labor supply.

15

employment earnings by 4.6 percentage points (over a base rate of 10.8 percentage points)

among individuals with W-2 wages below the first EITC kink (including individuals with

no W-2 wages). In Figure 4, we estimate effects separately for individuals within $2000

bins of year-t tax unit W-2 wages measured in constant 2015 dollars. The results in the

figure show that the reporting effect diminishes as W-2 wages approach the first EITC kink

and disappears completely above the kink, at which point individuals face strictly positive

marginal tax rates on any reported self-employment earnings.

In Columns (7) and (8), we find that individuals with wages below the first EITC kink

report $509 more in self-employment income ($647 more including their spouse) if their

child is born in December. How much do these individuals gain from reporting those selfemployment proceeds? To quantify the tax benefit from reporting self employment income,

we calculate each individual’s tax burden first using their own and their spouse’s wages and

self-employment earnings, and then based on wages alone. We then take the difference to

measure the net reduction in one’s net tax burden (inclusive of refundable credits).30 For

individuals without EITC-eligible children, this “gain” is negative, as individuals pay higher

taxes as a result of reporting additional earned income. However, the results in Column (9)

show that the realized gains from reporting self-employment increase for those with wages

below the first EITC kink and EITC-eligible children. On average, these individuals receive

net subsidies for the self-employment reported on their 1040.31 Dividing the average increase

in the propensity SE below the kink (4.6 percentage points), by the average benefit ($448.5),

implies a SE reporting response of 0.01 percentage point for every $1 in incentive.

We note that the regression discontinuity design only identifies the short-run reporting

response to one’s initial, and unexpected, exposure to negative marginal tax rates. However,

continued exposure to those same incentives over time might lead to larger shifts in reporting

behavior as individuals become aware of the incentives and how to optimally respond to them.

We can test for the presence of ongoing learning by examining reporting behavior in the

subsequent tax year t+1. In contrast to tax year t, both children born in December of tax year

t and those born in January of t+1 count the same towards EITC and CTC determination in

tax year t+1—accordingly, individuals on both sides of the year t cutoff should face identical

incentives in the following year t + 1. If individuals immediately internalize all reporting

incentives they face, then all differences between the two birth groups should disappear in

30

In this exercise, we use TAXSIM to estimate the federal tax (including FICA/SECA taxes) owed on

combined tax unit W-2 and Schedule SE earnings, or W-2 earning alone, given each individual’s martial

status and number of claimed dependents. For computational reasons, we do not incorporate additional

information on deductions or other types of income.

31

The effect for individuals with wages above the EITC phase-in range in Column (9) is negative despite

there being no self employment reporting effect in Column (1)—this simply reflects the bite of the EITC

phase-out for individuals with qualifying children.

16

year t + 1. However, if accumulated experience leads to larger responses, individuals with

December births in year t may still respond more than individuals with January births in

t + 1.32 Consistent with continued learning, we find statistically significant reporting effects

in year t + 1 in Column (4) of Table 2. The effect size is considerable, amounting to about

one-quarter of the initial year-t reporting effect. This finding suggests that the short-run

effects identified in the baseline specification understate the longer-run impact of tax-code

incentives on individuals’ self-employment reporting.

4.3

Reporting Effects Over Time

We next examine whether these behaviors have evolved over time. We estimate the main

RDD specification separately for birth cohorts in each year—that is, births in December of

each year t and in January of the corresponding year t + 1. Figure 5 contrasts the effects

among individuals with combined wages below the first EITC kink and those with wages

above. While we find negligible effects for individuals with combined tax-unit wages above

the kink in all years, we find that reporting effects have increased significantly over time

for individuals with household wages in the EITC phase-in region. This latter effect grows

approximately fivefold between 2000, when the effect size is 1 percentage point, and 2015,

when the effect size is six percentage points. Yet, for this same group, we find no effect on

having 1099-reported non-employee compensation in all years. These results indicate that

the propensity to report self-employment conditional on one’s true labor supply is far more

responsive to incentives in the tax code in recent years than two decades ago.

4.4

More Over-reporting or Less Under-reporting?

Our results imply that individuals with an incentive to report self-employment income are increasingly likely to do so irrespective of any changes in their work. One potential explanation

is that individuals who are not self-employed are more likely to fabricate self-employment

income on their tax returns to maximize refunds, artificially inflating the self-employment

rate. An alternative explanation is that individuals with actual self-employment income

that would have otherwise gone unreported to the IRS are now incentivized to report that

income on a tax return.33 Our baseline findings are therefore consistent with either a rise in

32

This comparison identifies learning as a direct result of filing experience similar to Ramnath and Tong

(2017), but not additional learning over time that is independent of past filing experience.

33

For example, Bracha and Burke (2018) report that nearly one in three workers have some type of

informal income—a rate that far exceeds the share of workers reporting self-employment income to the IRS.

Individuals with a tax incentive to report self-employment income may choose to truthfully report such

informal income that might otherwise be omitted on a tax return

17

over-reporting or a decline in under-reporting.

To shed light on this matter, we examine the the audits of a representative stratified

random sample of 1040 filers conducted in tax years 2001 and 2006–2014 as part of the IRS’s

National Research Program (NRP) Individual Income Tax Reporting Compliance Studies.

Using data from these audits, we measure how often individuals on 1040 returns inaccurately

report having self-employment earnings or inaccurately report having no self-employment

earnings. We plot the propensity of each type of audit result in Figure A.8, separately for

individuals with and without an incentive to report self-employment.34 We find, first, that

while audits consistently find that 2–2.5 percent of tax filers individuals have self-employment

income but report none, that rate is constant over time and nearly identical for individuals

with and without EITC reporting incentives (if incentives were driving behavior, we would

expect under-reporting to fall among incentivized individuals). By contrast, we find that

while almost no one incorrectly reported self-employment income in 2001, the share doing

so has risen substantially over subsequent years, but only among individuals with a tax

incentive to do so.

These findings suggest that our baseline results do, to some extent, reflect a rising propensity to report self-employment income despite having none. However, the rising rates of such

behavior in Figure A.8 can only explain a one percentage point increase in reported selfemployment among incentivized individuals, just one-fifth of the rise in reporting among

this group over the same period implied by our RDD estimates in Panel B of Figure 5.

Indeed, the NRP studies are not necessarily always able or intended to detect a decline in

under-reporting of self-employment income. In particular the studies we examine focus on

individuals who file a 1040 return, yet it is likely that many individuals with self-employment

that is not reported to the IRS do not file a 1040 at all. More generally, NRP audits may

detect many types of informal self-employment income when they go unreported—though

individuals could legitimately report such income to claim an EITC refund if they chose to

do so. Thus, while we do find evidence of a rise in over-reporting of self-employment income,

we cannot rule out simultaneous declines in under-reporting.

5

Mechanisms Driving Changes Over Time

Our analysis indicates a substantial rise in tax-code-motivated reporting of self-employment

income over the past two decades. One potential explanation for the growth in reporting

effects over time is increased generosity of tax credits that might amplify the incentives

34

We classify individuals on the basis of their firm-reported W-2 income and the number of eligible children

determined by the audit.

18

to report. To test this explanation, we examine how the impact on post-wage MTRs has

evolved over time by estimating the specifications in Column (5) of Table A.1 for each year

2000–2018. The results, plotted in A.7, show that the only major change in self-employment

reporting incentives for families with one child occurred in 2009 when the Child Tax Credit

was expanded, which lowered the federal MTR for low-wage families from about -25% to

about -30%. This result holds even when incorporating state EITC expansions in our analysis

in Panel B. While this expansion may have impacted reporting behavior, we observe that the

vast majority of the increase in reporting effects apparent in Figure 5 occurs from 2000 to

2007, prior to the announcement of the 2009 reform. Accordingly, it appears that changes in

policy cannot account for the observed increase in self-employment reporting effects observed

over this period.

Even during periods when policy remains fixed, knowledge of the incentives created by

the EITC and related tax provisions can differ across regions and spread gradually over

time (Chetty, Friedman, and Saez, 2013). This process of gradual spread of awareness of

incentives in the tax code could lead to a steady rise in reporting of previously unreported

self-employment earnings or false reporting of self-employment income. Learning dynamics

have been shown to be particularly important in the case of the EITC. Consistent with the

gradual learning story, earlier work has documented that hyper-strategic “sharp-bunching”

behavior—reporting exactly the amount of self-employment income that qualifies you for the

maximum tax benefit, has become more common over time (Chetty, Friedman, and Saez,

2013; Mortenson and Whitten, 2020). Chetty, Friedman, and Saez (2013) observe that this

spread is partly geographic in nature: The more common “sharp-bunching” is in a locale,

the more likely new mothers are to claim more generous benefits after having their first child.

While the prevalence of hyper-strategic “sharp bunching” is a useful measure of knowledge,

knowledge of the tax code can lead to increases in self-employment reporting that are not

associated with bunching behavior. Indeed, our results in Table 2 imply that sharp bunching

behavior accounts for only a small fraction of reporting responses to tax incentives.

To directly assess whether changes in local knowledge might explain the evolution of the

RDD estimates in Figure 5 over time, we estimate an interacted version of the main RDD

specification in Equation 1 that allows the effect to vary with local knowledge, proxied by the

share of sharp bunchers in each ZIP3 in each year.35 To avoid any mechanical dependence,

35

We replicate the sharp-bunching measure from Chetty, Friedman, and Saez (2013) and extend it through

2018. This measure is defined as the share of 1040 filers with children in a 3-digit ZIP code and year with

AGI under $50,000 in adjusted 2010 dollars who report self-employment earnings and total earned income

within $1,000 of the first EITC kink for that year. The resulting aggregate series is presented in Appendix

Figure A.9. ZIP codes for individuals in the RDD sample are taken from W-2s and 1099-MISCs in the

current year, or most recent of the prior two years if missing.

19

we tabulate the sharp-bunching share in each ZIP3 in each year following Chetty, Friedman,

and Saez (2013) only among individuals without new births in the December at the end of

the tax year or in the following January. We then fully interact the estimating equation:

∆yizmt =α + βBunchSharezmt × 1{datei ∈ December} + δBunchSharezmt

+ γ1{datei ∈ December} + f (datei , BunchSharezmt )

+ ζzm + φzt + izmt

(2)

Here, t denotes the RDD cohort (the tax year corresponding to the year-end DecemberJanuary pair) and m denotes the calendar month. We interact all terms, including the

function of the running variable, with the bunching share for each ZIP3 and tax year. Since

nationwide average bunching rates increase systematically over time, we include ZIP-bycalendar-month and cohort-by-calendar month fixed effects in some specifications to isolate

only within-ZIP variation in bunching rates. The fixed effects absorb the RDD main effect (η)

in each ZIP and each cohort, but not the interaction effect (β). We estimate this specification

on all observations in our sample with non-missing ZIP codes, pooling all birth cohorts in the

December and subsequent January of years 2000 through 2018 to capture the full evolution

of knowledge since 2000.

The results, presented in Table 3, suggest that local knowledge plausibly plays an important role determining the propensity to report self-employment income in response to

incentives. When sharp-bunching is more common among parents with older children in the

same region, new parents are significantly more likely to report self-employment income in

response to changing incentives around their first birth. The results in Column (1) indicate

that reporting effects in regions with low knowledge are only a small fraction of the baseline

estimates in Table 2, but reporting effects grow significantly with greater local knowledge

of tax incentives. The coefficients on the interaction terms are nearly the same in magnitude when including ZIP-month and cohort-month fixed effects. This indicates that the

interaction term estimates reflect differential increases in knowledge within regions rather

than aggregate time trends. The results in Columns (3) and (4) show that as knowledge

spreads, new parents exposed to incentives are more likely to bunch at the first EITC kink,

but increases in that type of bunching behavior accounts for less than one-third of the overall

increase in self-employment behavior. Reassuringly, there is no RDD interaction effect for

individuals with wages above the first EITC kink in Panel C, or any interaction effects on

the propensity to have of 1099-reported payments or to sort around the cutoff based on the

EITC value of wages in Columns (5)-(8).

The magnitude of the interaction term estimates in Table 2 are large enough to explain

20

most of the increase in RDD effects over time in Figure 5. The estimates in Columns (1)

and (2) imply that an increase in knowledge corresponding to a 3 percentage point rise in

sharp bunching—approximately the national increase observed in Appendix Figure A.9—

should increase the RDD effect among incentivized individuals by 2.5–3 percentage points.

Accordingly, local knowledge spread can account for the majority of the roughly 4 percentage

point rise in the RDD estimates from 2000 to 2018 in Panel B of Figure 5. However, this

exercise only reflects localized knowledge transmission; any other knowledge spread that is

not mediated by ZIP codes (e.g. broad-based dissemination via the internet or broader social

networks) would not be captured by our analysis. Thus, gradual transmission of knowledge

through both localized and non-localized channels plausibly explains the observed increase

in reporting behavior over time.

Focusing on first childbirths may understate the full impact of increasing local knowledge

of tax incentives if individuals absorb local knowledge gradually over time after their initial

eligibility, as we found above. In that case, the spread in local knowledge might have a

greater impact on self-employment reporting for the average eligible individual with children

more than captured by short-run effects after their first birth. For comparison, Appendix

Table A.2 estimates panel regressions of ZIP-3 self-employment reporting rates on the ZIP3 bunching measure used in the interacted RDD specification.36 For comparability to the

estimates in Column (2) of Table 3, we include ZIP and year fixed effects and weight by

population. Whereas a 1 percentage point increase in bunching increased self-employment

reporting rates among incentivized individuals by less than 1 percentage point in our interacted RDD analysis, we find in Appendix Table A.2 that the same increase in bunching is

associated with 3.6 percentage point higher self-employment reporting rates among all incentivized individuals with children. While the ZIP-level regressions cannot isolate reporting

effects from labor supply effects as the RDD estimates do, we nonetheless find no effect on

the propensity to have 1099-reported contract income and only negligible impacts on selfemployment reporting by individuals not facing reporting incentives (with high wages or no

children), consistent with pure reporting effects. This suggests that spreading local knowledge could increase overall self-employment reporting rates among incentivized individuals

over time 3–4 times as much as implied by the growth of our RDD estimates in Figure 5.

36

In this analysis, ZIP-3 self-employment rates are defined as a share of all individuals in the workforce

(among whom ZIP codes are always observed), not as a share of all individuals with children, as we do

not observe ZIP codes for all individuals with children. For comparison, 93% of the RDD sample with

non-missing ZIP codes in Table 3 are in the workforce by our definition

21

6

Quantification/Counterfactuals

6.1

Accounting for Changes in Reporting Behavior

The empirical analysis above finds that individuals with incentives to report self-employment

have become increasingly likely to do so. Further, our RDD analysis finds that the entire

increase in the propensity to report self-employment after first births is accounted for by

a pure reporting effect with no change in underlying labor supply, and this reporting effect has grown in magnitude over time. That is, incentivized individuals are increasingly

likely to report self-employment income on their tax return that they would not have in the

past—these could be informal sources of income that would have otherwise been unreported

or even fabricated earnings. In this section we quantify the effect that reporting incentives

have on measured growth in self-employment and present self-employment trends under an

assortment of counterfactual scenarios.

First, we apply our estimates from Section 4 to adjust the observed trend in Schedule SE.

The regression discontinuity estimates can be interpreted as the short-run effect of having a

single child eligible towards their EITC calculation, after labor supply decisions have been

sunk. We use the annual effect estimates presented in Figure 5 to estimate the share of

all individuals with children under 19 and tax unit W-2 wages below the first EITC kink

who would not have reported self-employment had they not faced negative marginal tax

rates.37 We then consider how much self-employment reporting rates would have grown by

using the annual effect estimates presented in Figure 5 to account for those only reporting

self-employment due to incentives. Specifically, our adjustment is given as follows:

SEt − βtRDD,∆AnySE,BT K · P OPtk,BT K

SEt∗ RDD

=

W Ft∗

W Ft − βtRDD,∆SEOnly,BT K · P OPtk,BT K

(3)

where βtRDD,∆AnySE,BT K and βtRDD,∆SEOnly,BT K are our RDD coefficients for reporting any

self-employment and only self-employment, respectively, among those below the kink, in

year t. Our RDD coefficients are at the population level and refer to the increase in selfemployment among all new parents, not just those in the workforce. To convert back to a

measure as a share of the workforce, we multiply these coefficients by the population with

children under 19 (the main eligibility criteria for the EITC) and who have wages below the

EITC kink, P OPtk,BT K , and also adjust the denominator to remove individuals only in our

workforce measure due to the incentive.

37

To reduce noise, we impose the effect on individuals with W-2 wages above the kink is zero in all our

years; this is consistent with our estimates in in Figure 5.

22

To implement this adjustment, we use population-level estimates of individuals with

children under 19 in their household from the Social Security birth records. We then apply

our annual regression coefficients to estimate the implied number of individuals who would

not have reported self-employment if they had not been incentivized to do so. The result of

this exercise is presented as “Scenario 1” in Figure 6a. This adjustment reduces the share

of the workforce reporting self-employment by about half of one percentage point in 2014,

about one quarter of the total increase since 2000.

The evidence presented above indicates that the RDD adjustment identifies a strict lower

bound on the share of individuals with qualifying dependents who report self-employment

solely due to reporting incentives. There are several reasons we think these estimates provide

a lower bound. First, the RDD effect is a short-run effect identified off of individuals who

have not previously been eligible for refundable credits and learn only shortly before filing

their taxes that their child was born prior to the end of the tax year. Our results in Column

4 of Table 2 indicate that individuals become increasingly aware of the incentives provided

by the tax code in the longer term, and adjust their behavior accordingly. Consistent with

that hypothesis, our findings in Appendix Table A.2 suggest that the effects of spreading

knowledge of the tax code on the overall incentivized workforce may be significantly larger

than the effects on new parents shortly after their first births. Second, the generosity of the

credit increases as individuals have subsequent children, which could lead to higher rates of

credit-motivated self-employment filing among those with additional children.

To account for how changes in awareness of incentives may have impacted reporting

behavior more broadly, we next consider how self-employment reporting rates would have

evolved if trends among individuals with incentives to report additional self-employment

income paralleled trends among individuals without such an incentive. More precisely, as a

second adjustment, we replace SE rates for low-wage individuals with children—who have an

incentive to report SE—with the rate among comparable individuals without children who

have no such incentive. We implement these comparisons within age, gender, and narrow

wage bins.38

38

Formally, our approach is to replace the actual SE rate for incentivized individuals in each year with

children with those for unincentivized individuals without children in the same cell. We compute the counterfactual self employment rate as follows:





X

X

X

X

∗

∗



SEt∗ =

ωg,t

ζw,g,k,t

SEw,g,0,t +

ζw,g,k,t

SEw,g,k,t 

(4)

g

k

w≤w∗ (k)

w>w∗ (k)

where we denote g a gender-age cell (e.g. women in a certain age band), k ∈ {0, 1, 2+} the number of children,

w the wage bin for a tax unit’s total wages,39 and ωg,t is g’s share of the overall workforce in year t. Individuals

are in the incentivized region of the tax schedule if they have children and wages below the first EITC kink

point for k, w∗ (k) (i.e. k = 1 and w < w∗ (k = 1) or k = 2+ and w < w∗ (k = 2+)). We further refine our

23

This second adjusted series is presented as “Scenario 2” in Panel A of Figure 6a. Under

this counterfactual scenario, the share of the workforce observed with self-employment earnings would have been nearly a full percentage point lower in 2014—cutting the increase in

those years by roughly half. The adjustment to the self-employed share eliminates most of

the change between 2005 and 2012, during which time observed self-employment filings rose

substantially. This difference reflects not only increasing rates of self-employment reporting

among individuals with qualifying children and wages below the first EITC kink, but also an

increase in the share of individuals with low wages—and hence incentives to report additional

self-employment earnings—in the wake of the Great Recession.

To evaluate whether individuals without children provide a valid counterfactual for individuals with children within narrow wage bins (i.e. differing only in reporting behavior

but not in underlying labor supply), we perform the same adjustments to the share of the

workforce with third-party-reported independent contractor earnings on 1099 forms in Figure 6b as a placebo exercise. We find the same adjustments have almost no impact on the

1099 series. This latter finding implies that firm-reported independent contracting activity

among low-wage individuals with kids evolved in parallel to rates for other individuals since

2000; this supports the validity of the adjustment exercise.

This second approach allows for a straightforward adjustment at finer levels of geography,

such as state. We simply redefine the cell to include state, where we determine state based

on the recipient address on information returns. The results of our state-level adjustment

are reported in Appendix Figure A.12. Panel (a) shows the raw percentage point growth

in schedule SE by state. The deeper red indicates more growth. We see dramatic growth

in the South, as well as California, New York, Nevada, and Michigan. Panel (b) shows

the size of our Scenario (2) adjustment by state. Many of the states that showed the most

growth in Schedule SE are also places that appear to have the most incentivized growth.

This figure also mirrors the geographic spread of knowledge in Chetty, Friedman, and Saez

(2013). Finally, Panel C shows what our adjusted series under Scenario 2 looks like, using

the same color breaks as the first map. Growth is dramatically scaled back in much of the

definition to exclude individuals with total earnings (wages+SE) above the phase-out region. Individuals

with no wages require special consideration since in the absence of the incentive, some of these individuals

would no longer be part of the tax workforce at all. In order to account for this, the weights must be adjusted.

To calculate the counterfactual number in the workforce in the absence of the incentive, we use populationlevel estimates of individuals with children under 19 in their household from the Current Population Survey

(CPS) Annual Social and Economic Supplement (ASEC). Denote P OP0,g,k,t the population of g, k with

0 total wages in year t. We calculate a counterfactual number of such individuals with k kids who would

SE∗

SE

otherwise be in the workforce in the absence of incentives as W F0,g,k,t

= W Fw,g,0,t

/P OP0,g,0,t ·P OP0,g,k,t , i.e.

we apply the self-employment-to-population ratio for those without children, and multiply by the population

with k children. We report the implied change in the workforce in Appendix Figure A.11. We recalculate

∗

the weights of our wage bins using this counterfactual level, and denote these adjusted weights ζw,g,k,t

.

24

US. Maryland and New York still have fairly dramatic growth, but otherwise our adjusted

series shows much more muted growth across the country.

In Appendix Table A.3, we report what the implied changes to the measured level of

the self-employment rate would be if individuals were entirely unaware of tax incentives

to report self-employment in all years. These level adjustments report a specific scenario

where incentivized people would have otherwise reported their self-employment earnings in

the same way unincentivized individuals do in the absence of the incentives.40

6.2

Additional Quantification Exercises

In addition to accounting for incentives in our adjustments, we also consider two additional

factors that may be quantitatively important for explaining observed SE trends: changes

in the propensity for under-reporting and changes in the demographic composition of the

workforce.

First, we assess the concern raised in Section 3 that trends in SE income reporting may

not line up with trends in the propensity to have non-employee income on 1099 forms due

to changes in accurate reporting of non-employee compensation on Schedule C or changes

in the composition of 1099 work. Changes in compliance along these lines might directly

contribute to the observed rise in SE reporting in excess of changes in the prevalence of 1099reported contract work. Note, however, that our adjustments above should already take

into account changes in the propensity to report a 1099 due to the incentives we consider.

Our SE adjustment incorporating these trends is described in Appendix C and presented

in Column (4) of Table A.3 and as “Scenario 3” in Figure C.2a. We find our measure

of SE underreporting is declining through 2009, thereby implying increasing rates of SE

filing among 1099 recipients in these years; since then, underreporting has been increasing,

reaching its highest levels in recent years amidst the rise in OPE work. In the end, however,

we find new online platform only makes a modest contribution to SE growth in recent years,

given the small dollar amounts and high rates of expensing, which we take at face value for

this exercise.

As documented in Collins, Garin, Jackson, Koustas, and Payne (2019), older workers—particularly retirement-age workers over 65—are dramatically more likely to have selfemployment income and non-employee compensation reported on 1099 forms conditional on

having any labor income, in all years since 2000. Accordingly the aging of the workforce

40

However, we note that this does not necessarily correspond to a “true” self-employment rate. As discussed above in Section 4.4, although it could be the case that incentivized individuals are fabricating selfemployment income, it could alternatively be the case that individuals facing positive tax rates incentives

under-report their self-employment income and, once incentivized, households reduce their underreporting.

25

could result in higher overall self-employment rates in our baseline measure. To examine

the role that the changing demographic composition of the workforce might contribute to

the remaining observed rise in self-employment reporting since 2000, we calculate what the

results of the Scenario 2 adjustment would be if the demographic composition of the workforce remained fixed at 2000 levels. Further details are described in Appendix C and this

adjustment is presented as Column (5) in Table A.3 and “Scenario 4” in Figure C.2b. Adjusting for both demographic shifts and for differential reporting trends among incentivized

individuals accounts for a large majority of the observed rise in self-employment reporting

since 2000. This adjustment accounts for nearly 75% of the observed increase since 2000.

Less than one half of a percentage point increase since 2000 remains unexplained.

6.3

Application: Reporting Over the Business Cycle

To illustrate the importance of adjusting for reporting behavior, we close by examining the

relationship between self-employment reported on tax returns and the state of the aggregate

economy. One question studied in the self-employment literature is whether people turn

to self-employment to smooth economic shocks. While unemployed and underemployed

workers may seek additional opportunities, at the same time, returns to self-employment

may be lower during aggregate recessions, reducing the value of this outside option. This

question has mainly been studied using survey data such as the CPS, with recent attention

placed on alternative work arrangements and independent contracting.41

Tax data has potential benefits for studying this question. It is well known that measurement error in survey data will lead to imprecise estimates; survey data may also be

biased if it is not picking up changes in freelance/contract work. However, the reporting

issues highlighted above may become more severe during economic downturns, as the the

number of individuals facing incentives to report self-employment to increase tax refunds

is also countercyclical. As shown in Appendix Figure A.10, as the wage distribution shifts

down, more families have wages fall into the credit phase-in range.

Our goal is to separate out reporting-effect from real changes in self-employment in

response to recessions using our adjusted series. To examine the response of self-employment

to the business cycle, we follow a standard approach in this literature and exploit variation

in local labor market conditions. We estimate the following specification:

ys,t = βU Rs,t + αs + αt + es,t

41

(5)

Parker (2009) provides a review of the literature, and Fairlie (2013) provides a more recent update

examining the Great Recession. Farber (1999) and Katz and Krueger (2017) focus on alternative work as

captured in the Contingent Worker Supplement (CWS) to the CPS.

26

where U Rs,t is the average unemployment rate in state s in year t. We include state and year

fixed effects, so our main coefficient of interest is identified off of panel variation across our

time period of study, 2000-2018. The outcome ys,t is a measure of the state-level prevalence

of self-employment based on either on raw self-employment reporting on tax returns or on the

series adjusting for reporting trends in Scenario 2 above. For comparison, we also construct

analogous outcomes using survey responses on the CPS-ASEC.

The results, reported in Table 4a, show that the share of workers with self-employment

income is significantly countercyclical in raw tax data: Column (1) shows that the baseline

self-employment share of the workforce derived from tax data increases in response increases

in unemployment. To put the coefficient into context, the unemployment rate increased by

about 5 percentage points during the Great Recession, so our estimated coefficient implies

an increase in self-employment as a share of the workforce of (0.144*5 =) 0.72 p.p., or about

(0.72/10.9=) 6.6%. In contrast, we see no response of self-employment when using our

adjusted series in Column (2), 1099s (Column 3) or using self-employment from the CPS

(Column 4).

As an alternative approach, we examine how the prevalence of each type of work in the

broader adult population varies over the business cycle in Table 4b. The interpretation of

our coefficients becomes the log point change in the employment-to-population ratio (approximately a percent change) for a percentage point increase in the unemployment rate.42

(The denominator of the dependent variable is the same across these specifications, only

the numerator is changing.) We find that while the number of adults with wage income

reported on W-2s (Column 1) or contractor compensation reported on 1099s (Column 5)

is significantly pro-cyclical, the share of adults reporting self-employment income (Column

3) is constant around the business cycle. However, after adjusting for changes in reporting

behavior (Column 4)—which increase self-employment reporting during recessions—we estimate that self-employment actually moves in close proportion to work reported on W-2s

and 1099s. Further, the coefficients for the adjusted series are similar to effects on both employment rates and self-employment measured in the CPS-ASEC. One importance difference

compared with using the CPS is that the tax outcomes are more precisely estimated.43

These exercises point to the importance of accounting for time-varying reporting behaviors when using self-reported self-employment measures in administrative data. Further, we

believe that they demonstrate the value of our new adjusted SE series and 1099 series. By

carefully considering the incentives created by the tax code, it is possible to recover useful

42

We measure the adult population in each year and state using the CPS-ASEC.

The point estimate on self-employment in the CPS-ASEC is imprecisely estimated, consistent with

classical measurement error in a dependent variable.

43

27

estimates of underlying work behavior from self-reported information on tax filings.

7

Conclusions

Taken together, we highlight that new types of online platform work, or “gig” work, cannot

explain increasing trends in self-employment in administrative data. Instead, we find that a

substantial portion of the growth is driven by strategic reporting behavior. More precisely,

we find an increase among low income individuals with kids—a group that faces negative

marginal tax rates if they report self-employment income. Our regression discontinuity

analysis comparing households with births in December versus January indicates that this

increase is driven by changes in pure reporting behavior rather than an underlying labor

response. When we consider counterfactual scenarios in which reporting behavior remained

constant at 2000 level, we find that between 25 and 55 percent of the average increase in

self-employment rates since 2000 can be attributed to pure reporting changes. Once we

additionally adjust for the aging of the workforce and other trends in reporting, we account

for nearly all the observed change in self-employment in U.S. tax data.

By employing an RDD methodology comparing households with births in December

versus January, of the following tax year, we find a significant increase in self-reported but

not firm-reported self-employment among December births relative to January births. This

well identified RDD specification leads us to believe that this is driven by changes simply in

reporting behavior rather than an underlying labor response. We apply our estimates of the

reporting response to provide an adjusted self-employment trend.

Our framework may be important going forward to interpret changes in self-employment

as tax policy continues to evolve. For instance, in the latest relief bill the American Rescue

Plan Act of 2021, the phase-in rate for childless adults will now be 15.3 percent for 2021,

up from 7.65 percent, completely covering the payroll tax; thus, childless adults now face

reporting incentives in states and territories that top-up the federal EITC credit. Furthermore, our adjustments have also highlighted the impact that demographic shifts within the

population over time (e.g. declining birth rates, aging population, etc.) have had on the

self-employment share of the workforce. Similarly, business cycles play an important role in

fluctuations in the self-employment share by increasing or decreasing the size of the group

incentivized by negative marginal tax rates.

Overall, while we find that tax data on self-employment suffer from reporting incentives,

we believe that such data can still be highly valuable for measuring labor-market trends so

long as reporting incentives are kept in mind. We show that these incentives are measurable

and can be accounted for with our framework. Moreover, our work highlights that even while

28

self-reported earnings are sensitive to reporting incentives, third-party income reported by

firms provides a valuable benchmark. Despite not driving the diverging self-employment

trends between survey and tax data, increases in gig work and secondary forms of contract

work are being well captured in firm-reported tax data. To this end, our new self-employment

series adjusted for reporting trends, as well as our new series on gig work, should be valuable

to other researchers in this area.

References

Abraham, K., J. Haltiwanger, K. Sandusky, and J. Spletzer (2013): “Exploring Differences in Employment between Household and Establishment Data,” Journal of Labor Economics,

31(2), S129–S172.

(2018): “Driving the Gig Economy,” .

(2020): Measuring the Gig Economy: Current Knowledge and Open Issues,University of

Chicago Press.

Abraham, K. G., J. C. Haltiwanger, C. Hou, K. Sandusky, and J. R. Spletzer (Forthcoming): “Reconciling Survey and Administrative Measures of Self-Employment,” Journal of

Labor Economics.

Abraham, K. G., B. J. Hershbein, and S. N. Houseman (2020): “Most Self-Employed Workers Are Independent Contractors,” Policy Brief, W.E. Upjohn Institute for Employment Research.

Blank, R. M., K. K. Charles, and J. M. Sallee (2009): “A Cautionary Tale about the Use

of Administrative Data: Evidence from Age of Marriage Laws,” American Economic Journal:

Applied Economics, 1(2), 128–49.

Bracha, A., and M. Burke (2018): “How Big is the Gig?,” Working Paper.

Bureau of Labor Statistics (2018): “Contingent and Alternative Employment Arrangements

- May 2017,” USDL-18-0942.

Chetty, R., J. N. Friedman, P. Ganong, K. E. Leibel, A. H. Plumley, and E. Saez

(2012): “Taxpayer Response to the EITC: Evidence from IRS National Research Program,”

Tables available at: www.rajchetty.com/chettyfiles/eitc nrp tabs.pdf.

Chetty, R., J. N. Friedman, and E. Saez (2013): “Using Differences in Knowledge across

Neighborhoods to Uncover the Impacts of the EITC on Earnings,” American Economic Review,

103(7), 2683–2721.

Collins, B., A. Garin, E. Jackson, D. Koustas, and M. Payne (2019): “Is Gig Work

Replacing Traditional Employment? Evidence from Two Decades of Tax Returns,” SOI Working

Paper.

Davis, S. J., J. Haltiwanger, R. S. Jarmin, C. Krizan, J. Miranda, A. Nucci, and

K. Sandusky (2009): Measuring the Dynamics of Young and Small Businesses: Integrating the

Employer and Nonemployer Universespp. 329–366. University of Chicago Press.

Fairlie, R. W. (2013): “Entrepreneurship, Economic Conditions, and the Great Recession,”

Journal of Economics & Management Strategy, 22(2), 207–231.

29

Fairlie, R. W., J. Miranda, and N. Zolas (2019): “Measuring Job Creation, Growth, and

Survival among the Universe of Start-ups in the United States Using a Combined Start-up Panel

Data Set,” ILR Review, 72(5), 1262–1277.

Farber, H. S. (1999): “Alternative and Part-Time Employment Arrangements as a Response to

Job Loss,” Journal of Labor Economics, 17(4), S142–S169.

Farrell, D., and F. Greig (2018): “The Online Platform Economy in 2018: Drivers, Workers,

Sellers, and Lessors,” JPMorgan Chase Institute.

Feldman, N. E., P. Katuščák, and L. Kawano (2016): “Taxpayer Confusion: Evidence from

the Child Tax Credit,” American Economic Review, 106(3), 807–35.

Goldin, J., T. Homonoff, R. Javaid, and B. Schafer (2021): “Tax Filing and Take-Up: Experimental Evidence on Tax Preparation Outreach and EITC Participation,” Discussion paper,

National Bureau of Economic Research.

Harris, S. D., and A. B. Krueger (2015): “A Proposal for Modernizing Labor Laws for

Twenty-First-Century Work: The “Independent Worker”,” The Hamilton Project Discussion

Paper 2015-10.

Hurst, E., and B. Pugsley (2011): “What Do Small Businesses Do?,” Brookings Papers on

Economic Activity, 42(2 (Fall)), 73–142.

Jackson, E., A. Looney, and S. Ramnath (2017): “The Rise of Alternative Work Arrangements: Evidence and Implications for Tax Filing and Benefit Coverage,” Office of Tax Analysis

Working Paper 114.

Katz, L., and A. Krueger (2019): “The Rise and Nature of Alternative Work Arrangements in

the United States, 1995-2015,” ILR Review, 72(2), 382–416.

Katz, L. F., and A. B. Krueger (2017): “The Role of Unemployment in the Rise in Alternative

Work Arrangements,” American Economic Review, 107(5), 388–92.

Kolesár, M., and C. Rothe (2018): “Inference in Regression Discontinuity Designs with a

Discrete Running Variable,” American Economic Review, 108(8), 2277–2304.

Kuka, E. (2014): “EITC and the Self-employed: Real or Reporting Effects?,” Public Finance

Review, 42(6), 691–719.

LaLumia, S. (2009): “The Earned Income Tax Credit and Reported Self-Employment Income,”

National Tax Journal, 62(2), 191–217.

LaLumia, S., J. M. Sallee, and N. Turner (2014): “New Evidence on Taxes and the Timing

of Birth,” American Economic Journal: Economic Policy, 7(2), 258–293.

Lim, K., A. Miller, M. Risch, and E. Wilking (2019): “Independent Contractors in the U.S.:

New Trends from 15 years of Administrative Tax Data,” SOI Working Paper.

Mortenson, J. A., and A. Whitten (2020): “Bunching to Maximize Tax Credits: Evidence

from Kinks in the US Tax Schedule,” American Economic Journal: Economic Policy, 12(3),

402–32.

National Academies of Sciences, E., and Medicine (2020): Measuring Alternative Work

Arrangements for Research and Policy. The National Academies Press, Washington, DC.

Parker, S. C. (2009): “Entrepreneurship, Self-employment and the Labour Market,” The Oxford

Handbook of Entrepreneurship.

30

Ramnath, S. P., and P. K. Tong (2017): “The Persistent Reduction in Poverty from Filing a

Tax Return,” American Economic Journal: Economic Policy, 9(4), 367–94.

Saez, E. (2010): “Do Taxpayers Bunch at Kink Points?,” American Economic Journal: Economic

Policy, 2(3), 180–212.

Slemrod, J. (2016): “Caveats to the Research Use of Tax-Return Administrative Data,” National

Tax Journal, 69(4), 1003–1020.

Figures

Figure 1: Share of Workforce with Self-Employment and 1099 Information Returns

Share of Overall Workforce

.12

.11

.1

.09

1985

1990

1995

SE

2000

1099

2005

2010

2015

1099 - Excl. OPE

Notes: Figure shows the share of the overall tax workforce by tax year with any SE income as

filed on Schedule SE (black line) and individuals who receive a 1099 Information Return (maroon

line).

After the entry of OPE, we additionally distinguish the receipt of 1099 Information Returns including and excluding those received from OPE firms (dashed maroon line). For years prior

to 2000, SE workers and the overall workforce are drawn from annual SSA records on workers:

https://www.ssa.gov/policy/docs/statcomps/supplement/2020/4b.pdf.

31

Figure 2: Growth Over 2000-2014 in Self-Employment, by Wage Income

2000-2014 Change, Share within Wage Bin

(a) Any SE Filing

.2

.15

.1

.05

0

-.05

0

10000

20000

30000

40000

50000

Wages (Individual + Spouse) ($2015)

No Children

1 Child

60000

2+ Children

2000-2014 Change, Share within Wage Bin

(b) Any 1099

.2

.15

.1

.05

0

-.05

0

10000

20000

30000

40000

50000

Wages (Individual + Spouse) ($2015)

No Children

1 Child

60000

2+ Children

Notes: Figure shows the change in propensity to file SE (Panel A) or receive a 1099 Information Return

(Panel B), for tax units with wage earnings. Wage earnings, in $500 bins, are reported on the x-axis. Change

is calculated between 2000 and 2014. Wages are determined based on W2 information returns. Number of

children and spouse as reported on tax return. The area to the left of the vertical lines report the earnings

where reporting an additional dollar of self-employment income would face a negative marginal tax rate due

to the EITC, for households with 1 (maroon line) and 2 (green line) children.

32

Figure 3: Regression Discontinuity Design

(a) Simulated EITC Based on W2

0

800

Simulated Amount

820

840

Has Negative MTR on 1st Dollar of SE

.05

.1

.15

.2

.25

860

(b) Negative MTR After W2 Wages

-30

-25

-20

-15

-10

-5

0

5

10

15

Day Relative to December 31

Mean Within Birthdate

Estimation Sample Only

20

25

30

-30

Regression Fit

-25

-20

-15

-10

-5

0

5

10

15

Day Relative to December 31

Mean Within Birthdate

20

25

30

Regression Fit

Any: Change from t-1 to t

.005

.01

.015

0

0

Any: Change from t-1 to t

.005

.01

.015

.02

(d) Change: Any 1099 NEC

.02

(c) Change: Reports SE Earnings

Estimation Sample Only

-30

-25

-20

-15

-10

-5

0

5

10

15

Day Relative to December 31

Mean Within Birthdate

Estimation Sample Only

20

25

30

-30

Regression Fit

-25

-20

-15

-10

-5

0

5

10

15

Day Relative to December 31

Mean Within Birthdate

Estimation Sample Only

20

25

30

Regression Fit

Notes: Figure graphically presents results from the baseline regression discontinuity design specification in

Equation (1), pooling births in each December 2011-2018 and each subsequent January. Panel (a) examines

the change in “simulated EITC” that is the credit each individual would earn if their birth occurred in

December of tax year t based on their and their spouse’s (if married filing jointly) W-2 wages from year t.

Panel (b) examines whether one would face a negative federal marginal tax rate (including SECA taxes) on

a first dollar of self-employment earnings beyond one’s W-2 reported wage/salary earnings and those of any

spouse in year t, given the year their child was actually born. Panel (c) examines the change in whether

the one reports any Schedule SE earnings in tax year t relative to the prior year t − 1. Panel (d) examines

the change in having non-employee income reported on a 1099-MISC in tax year t relative to the prior year

t − 1. Each dot is the average outcome for parents with births on the corresponding date, pooled across

years. The solid dots are the calendar dates used in the main estimation window. The line segments are the

regression fits allowing for the estimated discontinuity between December 31 and January 1, with 95 percent

confidence bands from robust standard errors displayed.

33

Figure 4: RDD Estimates by Tax Unit W-2 Wage Earnings

RDD Effect: Starts Reporting After Childbirth

.08

.06

.04

.02

0

-.02

0

2

4

6

8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50

W2 Wages (Individual + Spouse), Thousands of 2015 Dollars

Notes: Figure presents results from the regression discontinuity design specification in Equation (1) pooling

births in each December 2011-2018 and each subsequent January, estimated separately for individuals within

$2000 bins of year-t tax unit (self plus spouse) W2 wages, measured in constant 2015 Dollars. The dashed

maroon line is the amount where the first EITC kink occurs for families with one child based on the 2015

schedule.

.08

.06

.04

.02

0

-.02

RDD Effect: Starts Reporting After Childbirth

Figure 5: RDD Estimates by Year

2000

2002

2004

2006

2008

2010

2012

2014

2016

2018

Any SE: Wage Income Below First EITC Kink (Incentive to Report)

Any SE: Wage Income Above First EITC Kink (No Incentive)

Any 1099 NEC: Wage Income Below First EITC Kink (Incentive)

Notes: Figure reports our baseline RDD estimates from estimating Specification 1 in the text within individual cohorts. We report separate estimates for individuals with combined wages below the first EITC kink

and those with wages above the first EITC kink point. Years correspond to the tax year t, at the end of

which the births occur in the corresponding December or January.

34

Figure 6: Adjusted SE Shares Under Counterfactual Assumptions

(b) Adjusted Share of Workforce with Non-employee Compensation

Percentage Points (Relative to 2000)

35

Percentage Points (Relative to 2000)

(a) Adjusted Share of Workforce Filing Self-Employment

.02

.015

.01

.005

.025

.02

.015

.01

.005

0

2000

0

2000

2003

2006

2009

2012

2015

Baseline

Scenario 1: RDD

Scenario 2: Replace with SE Rates of Unincentivized

2018

2003

2006

2009

2012

2015

2018

Baseline

... Excluding OPE Work

Scenario (1): RDD

Scenario (2): Replace with 1099 Rates of Unincentivized

... Excluding OPE Work

Notes: Figure presents the percentage point change in the share of the workforce with self employment income or 1099-reported contract payments

relative to 2000, along with counterfactual increases in each series under alternative scenarios. Panel (a) applies these adjustments to self-employment,

and Panel (b) to non-employee compensation. “Scenario (1)” adjusts self-employment downward using our annual RDD estimates reported in Figure

5, under the counterfactual assumption that the RDD effects remained constant in all years. “Scenario (2)” reports how overall rates would have

evolved if trends for individuals who have incentives to report self employment followed those rates among comparable individuals without such

incentives. Dashed lines in Panel (b) exclude the OPE.

Tables

Table 1: RDD Estimates: Third-Party Reported Earnings

∆ Any

Wages

∆ Any

1099-NEC

Tax Unit

W-2 Wages

Tax Unit Wages

< 1st Kink

Sim. 1-Child EIC

W-2 Only

(1)

(2)

(3)

(4)

(5)

0.003

0.00003

46.1

-0.0009

-0.8

(0.002)

(0.001)

(297.5)

(0.002)

(5.8)

1382740

1382740

1382740

1382740

1382740

0.8

0.08

55244.5

0.3

834.5

December Births

Coeff

N

DV Mean Level, Jan Births

Notes: Table displays estimates from the baseline regression discontinuity design specification in Equation

(1) on third-party reported earnings. The sample is all individuals with births in the last fifteen days

of December of each tax year t in 2011-2018 or the first fifteen days of January immediately following

tax year t, omitting births within three days of the start of the new year. Outcomes are from year t or

are changes from year t relative to the prior year t − 1, as specified. Simulated 1-Child EIC Levels are

calculated as the EIC amount one would receive if their first child had been born in year t (irrespective

of when the true birth occurred) given only one’s W-2 reported wage/salary earnings and those of any

spouse reported on a 1040. We report mean year t levels of each dependent variable for individuals with

first births in January of t + 1. Robust standard errors are displayed in parentheses.

36

Table 2: RDD Estimates: Main Effects on Self-Employment Reporting

Any SE Earnings

∆ Any

SE

Tax Year t-1

Tax Year t

(1)

(2)

(3)

Tax Year t+1

∆ Any SE &

Sharp Buncher

∆ Has 1099 NEC

& Reports SE

∆ Individual

SE Earnings

∆ Tax Unit

SE Earnings

Tax Benefi

Reportin

(4)

(5)

(6)

(7)

(8)

(9)

Panel A. All Parents

Coeff

0.0134**

-0.00101

0.0124**

0.00395**

0.00254**

0.00388**

130.8**

153.6**

90.62

(0.00118)

(0.00120)

(0.00127)

(0.00135)

(0.000326)

(0.000950)

(11.57)

(19.53)

(9.37

N

1382740

1382740

1382740

1382740

1382740

1382740

1382740

1382740

13827

DV Mean Level, Jan Births

0.0719

0.0683

0.0719

0.0878

0.00203

0.0406

855.4

1770.5

-417.

Panel B. With Wages < 1st EITC Kink

Coeff

-0.00139

0.0446**

0.0146**

0.00939**

0.0128**

509.0**

646.5**

448.5

(0.00270)

(0.00315)

(0.00335)

(0.00121)

(0.00197)

(30.10)

(43.90)

(15.4

N

349240

349240

349240

349240

349240

349240

349240

349240

34924

DV Mean Level, Jan Births

0.108

0.0900

0.108

0.143

0.00804

0.0503

1403.1

2319.4

-404.

0.00240

-0.000849

0.00155

0.000433

0.000229

0.000858

3.108

-12.85

-30.45

(0.00128)

(0.00131)

(0.00131)

(0.00140)

(0.000152)

(0.00108)

(11.59)

(21.44)

(11.3

N

1033500

1033500

1033500

1033500

1033500

1033500

1033500

1033500

10335

DV Mean Level, Jan Births

0.0597

0.0610

0.0597

0.0692

0

0.0374

670.6

1585.3

-422.

37

0.0460**

(0.00272)

Panel C. With Wages ≥1st EITC Kink

Coeff

Notes: Table displays estimates from the baseline regression discontinuity design specification in Equation (1) on third-party reported earnings. The sample is all individuals with

in the last fifteen days of December of each tax year t in 2011-2018 or the first fifteen days of January immediately following tax year t, omitting births within three days of the s

the new year. “Wages < 1st kink” subsample includes all individuals in tax units (self plus spouse if filing a 1040 jointly) with year t wages in the EITC phase-in region for hou

with one child in that year (irrespective of whether their birth actually occurred in December or January); the complementary subsample includes all other individuals. Outcom

from year t or are changes from year t relative to the prior year t − 1, as specified. ”Sharp bunchers” are individuals with earning income within $500 of the level where the first

kink occurs. ”Net Tax Benefit from Reporting SE” is the increase in net taxes one would pay if both they and their spouse (if present on a 1040) did not report their self-emplo

earnings to the IRS. We report mean levels of each dependent variable for individuals with first births in January of t + 1 in each subsample; levels are from year t except in colu

and 4, which report year t − 1 and t + 1 means, respecively. Robust standard errors are displayed in parentheses.

Table 3: RDD Estimates: Interactions with Knowledge Spread

∆ Any 1040-SE

Income

(1)

∆ Any 1040-SE

& Sharp Buncher

(2)

(3)

(4)

∆ Any 1099-NEC

(5)

(6)

Simulated W2-Only

One Child EIC

(7)

(8)

Panel A. All Parents

December Birth

0.00250*

-

-0.000716*

-

0.000982

-

-0.162

-

(0.00126)

-

(0.000323)

-

(0.00138)

-

(6.197)

-

December Birth

× ZIP Bunching

0.337**

0.303**

0.0998**

0.0831**

0.0120

-0.0250

-8.682

44.61

(0.0453)

(0.0597)

(0.0146)

(0.0183)

(0.0477)

(0.0633)

(198.0)

(266.4)

N

3094399

3094387

3094399

3094387

3094399

3094387

3094399

3094387

0.0134**

-

-0.000445

-

0.00335

-

8.664

-

(0.00336)

-

(0.00130)

-

(0.00340)

-

(11.71)

-

December Birth

× ZIP Bunching

0.999**

0.868**

0.311**

0.254**

0.0788

-0.0616

-221.5

344.4

(0.123)

(0.143)

(0.0526)

(0.0665)

(0.112)

(0.144)

(339.5)

(467.4)

N

665093

665074

665093

665074

665093

665074

665093

665074

0.00151

-

-0.000294

-

0.000666

-

-1.938

-

(0.00132)

-

(0.000151)

-

(0.00149)

-

(7.200)

-

Panel B. With Wages < 1st EITC Kink

December Birth

Panel C. With Wages ≥1st EITC Kink

December Birth

December Birth

× ZIP Bunching

0.0557

0.0506

0.0156*

0.0135

-0.0199

-0.0191

65.15

-108.6

(0.0457)

(0.0611)

(0.00654)

(0.00800)

(0.0514)

(0.0694)

(238.5)

(320.1)

N

2429306

2429295

2429306

2429295

2429306

2429295

2429306

2429295

Zip × Month FE

X

X

X

X

Cohort × Month FE

X

X

X

X

Notes: Table displays estimates from the regression discontinuity specification in Equation (2) in the text including interactions

with local sharp-bunching rates at the 3-Digit ZIP code level for each year. Bunching rates are calculated omitting individuals

in the RDD sample following Chetty, Friedman, and Saez (2013). The sample is all individuals with births in the December of

each tax year t in 2001-2018 or the January immediately following tax year t with a valid ZIP code reported on an information

return or tax filing in year t. The bunching variable is interacted with the discontinuity and the running-variable slope terms.

Columns 2, 4, 6, and 8 include year t (cohort) by birth calendar month and ZIP3 by birth calendar month fixed effects so that

all bunching variation in the interaction term comes from within-ZIP changes over time; these FEs absorb the RDD main

effect. Robust standard errors are displayed in parentheses. See notes to 2 for additional details.

38

Table 4: Response of Self-Employment to the Business Cycle Under Various Measures

(a) Share of Workforce

Tax: Raw SE

Incentives-Adjusted

1099

ASEC: SE

(1)

(2)

(3)

(4)

0.144

0.0253

0.0119

-0.0151

(0.0304)

(0.0176)

(0.0245)

(0.0514)

Unemployment Rate

N

969

969

969

969

R2

0.894

0.936

0.893

0.764

Dep. Mean

0.109

0.098

0.096

0.076

State FE

X

X

X

X

Year FE

X

X

X

X

(b) Log employment-to-adult population

Unemployment Rate

N

R

2

Tax: Wage

ASEC: Wage

Tax: SE

Incentives-Adj

1099

ASEC: SE

(1)

(2)

(3)

(4)

(5)

(6)

-1.259

-0.749

0.278

-0.929

-0.950

-1.156

(0.156)

(0.151)

(0.246)

(0.188)

(0.251)

(0.714)

969

969

969

969

969

969

0.919

0.904

0.924

0.958

0.914

0.805

State FE

X

X

X

X

X

X

Year FE

X

X

X

X

X

X

Notes: Table reports the results from running specification (5) in the text for the dependent variable and

source indicated by table subtitles and column headers. Standard errors clustered on state are displayed

in parentheses.

39

Appendix (For Online Publication)

A

Appendix Figures

Federal Marginal Tax Rate (Including Payroll Taxes)

Figure A.1: Effective Federal Marginal Tax Rate For Reporting Additional Dollar of SelfEmployment Income

60

40

20

0

-20

-40

-60

0

10,000

20,000

30,000

40,000

Before Tax Wage Earnings

One child

No Children

Two children

50,000

60,000

Three children

Notes: Figure shows the effective marginal tax rate for reporting an additional dollar of self-employment

income, for a given level of before-tax wage earnings. Calculation takes into account the full tax schedule in

tax year 2015 with no other credits/deductions except the EITC, CTC and standard deductions, and assumes

Schedule SE payroll taxes are paid on the self-employment income and taxpayers deduct the employer-share of

the payroll tax on self-employment income. Calculation assumes married filing jointly, however the marginal

tax rates below the first kink point are identical for married and single parents who claim children. The

area to the left of the vertical lines indicate the first kink-point of the EITC schedule, for households with 1

(maroon line) and 2 or more (green line) children.

40

Figure A.2: Share of Workforce with Self-Employment, in Tax Returns and CPS ASEC

.12

Schedule SE

Share of Overall Workforce

.11

.1

.09

.08

ASEC CPS

.07

1985

1990

1995

2000

2005

2010

2015

Notes: Figure shows the share of the overall tax workforce by tax year with any SE income as filed

on Schedule SE (black line) and individuals who report self-employment in the CPS ASEC supplement

(maroon line). The workforce is defined as all individuals with either W2 or Schedule SE income. For

years prior to 2000, SE workers and the overall workforce are drawn from annual SSA records on workers:

https://www.ssa.gov/policy/docs/statcomps/supplement/2020/4b.pdf.

41

Figure A.3: Share of Workforce with Self-Employment and 1099 Information Returns

(a) By EITC and Kids

EITC Recipients with Kids

All Others

Share of Workforce, Subset with no EITC or no Kids

Share of Workforce, Subset with EITC and Kids

.2

.15

.1

2000

2005

2010

.2

.15

.1

2015

SE

2000

1099

2005

2010

2015

1099 - Excl. OPE

(b) By Presence of Children

With Kids

Without Kids

.16

Share of Workforce, Subset without Kids

Share of Workforce, Subset with Kids

.16

.14

.12

.1

.08

.14

.12

.1

.08

2000

2005

2010

SE

2015

2000

1099

42

2005

2010

1099 - Excl. OPE

2015

(c) By Total Earnings

Total Earnings ≥$50,000

Total Earnings <$50,000

.13

Share of Workforce, Subset with Earnings ≥ $50,000

Share of Workforce, Subset with Earnings <$50,000

.13

.12

.11

.1

.09

2000

2005

2010

.12

.11

.1

.09

2015

SE

2000

1099

2005

2010

2015

1099 - Excl. OPE

(d) By Gender

Male

Female

.14

Share of Workforce, Subset Female

Share of Workforce, Subset Male

.14

.12

.1

.08

.06

.12

.1

.08

.06

2000

2005

2010

SE

2015

2000

1099

2005

2010

2015

1099 - Excl. OPE

Notes: Figure shows the share of the overall tax workforce by tax year with any SE income as filed on

Schedule SE (black line) and individuals who receive a 1099 Information Return (maroon line). After the

entry of OPE, we additionally distinguish the receipt of 1099 Information Returns including and excluding

those received from OPE firms (dashed maroon line). In Panel (a), the workforce definition is split on EITC

recipients with kids claimed on their 1040. In panel (b), we split by presence of kids on their 1040, In panel

(c), total earnings refers to the sum of wage and self-employment income by a primary tax filer and their

spouse as reported on a 1040. In Panel (d), we split by gender.

43

0

.005

Density

.01

.015

.02

Figure A.4: Distribution of First Births Around End of Tax Years 2011-2018

-30

-15

0

Date Relative to Dec 31

15

30

Notes: Histogram reports distribution of all first births in December of each tax year 2011-2018 or the

following January in our SSA sample (corresponding to the sample in our baseline analysis). The solid red

line denotes the end of tax year t and the dashed grey lines correspond to the Federal holidays on Christmas

day (December 25) and New Year’s day (January 1).

44

Figure A.5: RDD Filing Effects by Tax Unit W2 Wage Earnings

(a) Change: Files 1040

.8

1

.15

.6

Filing Share

.05

.4

RDD Effect

.1

.2

0

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

22000

24000

26000

28000

30000

32000

34000

36000

38000

40000

42000

44000

46000

48000

50000

-.05

W2 Wages (Individual + Spouse)

RDD Estimate

Baseline Filing Share, January Births

(b) Change: Any Children Claimed on 1040 (c) Change: Any Deps Claimed on Sch. EIC

.6

.7

.6

RDD Effect

RDD Effect

.4

.5

.4

.2

.3

.2

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

22000

24000

26000

28000

30000

32000

34000

36000

38000

40000

42000

44000

46000

48000

50000

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

22000

24000

26000

28000

30000

32000

34000

36000

38000

40000

42000

44000

46000

48000

50000

0

W2 Wages (Individual + Spouse)

W2 Wages (Individual + Spouse)

Notes: Figure presents results from the baseline regression discontinuity design specification in Equation (1)

pooling births in each December 2011-2018 and each subsequent January, estimated separately for individuals

within $2000 bins of year-t tax unit (self plus spouse) W2 wages, measured in constant 2015 Dollars. The

dashed maroon line is the earnings amount where the first EITC kink occurs for families with one child

based on the 2015 schedule.

45

Figure A.6: Robustness of Main RDD Effects

.08

Starts Reporting SE After Childbirth

-.02

0

.02

.04

.06

Baseline Specification

10 Day BW

15 Day BW

20 Day BW

30 Day BW

Linear, Donut is:

None

2 Day

3 Day

5 Day

Quadratic, Donut is:

None

2 Day

3 Day

5 Day

Notes: Figure displays main regression discontinuity effects on the change in whether the one reports any

Schedule SE earnings in tax year t relative to the prior year t − 1 from Column 1 in Table 2 under alternative

specifications. Donut hole widths are bandwidths omitted from the regression sample. Quadratic specifications allow slopes to differ across the threshold. The horizontal black line corresponds to the size of the

benchmark estimate in Table 2.

46

Figure A.7: RDD Effects on MTRs After Wages by Year

-30

-20

-10

0

(a) Federal Taxes Only (Inlcuding SECA)

2000

2005

2010

2015

2020

Wage Income Above First EITC Kink (No Incentive)

Wage Income Below First EITC Kink (Incentive to Report)

-40

-30

-20

-10

0

(b) Federal + State Taxes

2000

2005

2010

2015

2020

Wage Income Above First EITC Kink (No Incentive)

Wage Income Below (Incentive to Report)

Notes: Figure reports our baseline RDD estimates from estimating Equation 1 in the text within individual

cohorts. Years correspond to the tax year t, at the end of which the births occur in the corresponding

December or January. Outcomes are marginal tax rates on a first dollar of self employment earnings,

conditional on own and spouse’s W2 wage earnings, calculated using TAXSIM.

47

0

.005

.01

.015

.02

.025

Figure A.8: Changes in Self-Employment Status after NRP Audits

2000

2005

2010

2015

tax_yr

SE Added After Audit:

SE Removed After Audit:

W2 Below Kink

W2 Below Kink

W2 Above Kink

W2 Above Kink

Notes: Figure displays results of audits of a representative stratified random sample of 1040 filers conducted

in tax years 2001 and 2006–2014 as part of the IRS’s National Research Program (NRP) Individual Income Tax Reporting Compliance Studies. Using sampling weights for representativeness, the figure plots

the share of individuals with 1040 returns who are found to have incorrectly not reported self-employment

income on Schedule SE when they should have, and the share of individuals found to have reported positive

self-employment income on Schedule SE when they actually should have reported none. Each propensity is

calculated separately for individuals with and for individuals without an incentive to report self-employment.

Individuals are classified based on their firm-reported W2 income and the number of eligible children determined by the audit.

48

.01

CFS-style Sharp Bunching Share

.02

.03

.04

Figure A.9: CFS-style Sharp Bunching Share Among Eligible Taxpayers with Children,

1996-2017

1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

Notes: Figure plots the average of the share of tax payers who are sharp bunchers, following the methodology

of Chetty, Friedman, and Saez (2013).

49

.08

Share of Workforce Incentivized

.09

.1

.11

.12

Figure A.10: Share of Workforce with Incentive to Report SE

2000

2005

2010

2015

2020

Notes: The share incentivized in each year represents the number of individuals with children and wages

below the corresponding EITC kink point as a share of the tax workforce.

50

Figure A.11: Counterfactual Change in Workforce Due to Incentive

.012

Share of Workforce

.01

.008

.006

.004

2000

2005

2010

2015

2020

Notes: Figure shows the implied share of the

Pworkforce who are in the workforce only due to the incentive.

∗

We define this as (W F0,g,k,t − W F0,g,k,t

)/( w W Fw,g,k,t ). We calculate a counterfactual number of such

∗

=

individuals with k kids who would otherwise be in the workforce in the absence of incentives as W F0,g,k,t

W Fw,g,0,t /P OP0,g,0,t · P OP0,g,k,t , i.e. we apply the self-employment-to-population ratio for those without

children, and multiply by the population with k children.

51

Figure A.12: Raw SE Filing and Application of Our Adjustment By State, 2000-2014

(a) Raw SE Filing Growth

(b) With Incentives Adjustment (S2)

52

(c) Growth in Adjusted SE Series

Notes: Panel (a) reports raw percentage-point growth in SE filing by state, 2000-2014. The colors refer to quintiles of growth over time, with darker

colors indicating higher levels of growth. Panel (b) reports the size of our Scenario (2) adjustment, see text for more details. Panel (c) reports our

adjusted-SE series by state, using the same scale as Panel (a)

Appendix Tables

Table A.1: RDD Estimates: “First-Stage” Effects on Filing Status and Reporting Incentives

MTR After

Wages

Has Neg MTR

After Wages

∆ Any

1040

∆ Any

Children

∆ Any EITC

Dependents

(1)

(2)

(3)

(4)

(5)

-6.534**

0.251**

0.0221**

0.598**

0.225**

(0.0826)

(0.00138)

(0.00153)

(0.00180)

(0.00162)

N

1382740

1382740

1382740

1382740

1382740

DV Mean Level, Jan Births

22.72

0

0.860

0.131

0.0758

Panel A. All Parents

Coeff

Panel B. With Wages < 1st EITC Kink

Coeff

-30.69**

0.991**

0.0614**

0.299**

0.252**

(0.0565)

(0.000634)

(0.00472)

(0.00366)

(0.00354)

N

349240

349240

349240

349240

349240

DV Mean Level, Jan Births

8.370

0

0.528

0.105

0.0912

1.619**

0.000393**

0.00875**

0.699**

0.215**

(0.0352)

(0.0000662)

(0.00129)

(0.00192)

(0.00181)

N

1033500

1033500

1033500

1033500

1033500

DV Mean Level, Jan Births

27.56

0

0.973

0.140

0.0707

Panel C. With Wages ≥1st EITC Kink

Coeff

Notes: Table displays estimates from the baseline regression discontinuity design specification in Equation (1) on

third-party reported earnings. The sample is all individuals with births in the last fifteen days of December of each

tax year t in 2011-2018 or the first fifteen days of January immediately following tax year t, omitting births within

three days of the start of the new year. “Wages < 1st kink” subsample includes all individuals in tax units (self plus

spouse if filing a 1040 jointly) with year t wages in the EITC phase-in region for households with one child in that

year (irrespective of whether their birth actually occurred in December or January); the complementary subsample

includes all other individuals. Outcomes are from year t or are changes from year t relative to the prior year t − 1, as

specified. Marginal tax rates (MTRs) after wages are calculated as the federal marginal tax rate on the first dollar

of self-employment earnings (including SECA taxes) beyond one’s W-2 reported wage/salary earnings and those of

any spouse reported on a 1040, given the year their child was actually born. We report mean year t levels of each

dependent variable for individuals with first births in January of t + 1 in each subsample. Robust standard errors

are displayed in parentheses.

53

Table A.2: Panel Relationship Between ZIP Bunching and SE Reporting

Individuals With Children

& Wages Below Kink

Individuals With Children

& Wages Above Kink

Individuals Without

Children

(1)

(2)

(3)

Outcome: Workforce Share with SE

ZIP Bunching Share

3.636**

0.123**

0.113**

(0.290)

(0.0267)

(0.0325)

15709

15744

15782

-0.00210

0.0674**

0.0797**

(0.148)

(0.0179)

(0.0228)

N

15709

15744

15782

Zip FE

X

X

X

Year FE

X

X

X

N

Outcome: Workforce Share with 1099 NEC

ZIP Bunching Share

Notes: Panels display estimates of panel regressions of self-employment rates and non-employee compensation reported on

1099-MISC within each specified workforce segments on the year-by-ZIP3 bunching measures calculated as in Chetty, Friedman, and Saez (2013). Sample is all individuals in the tax workforce 2000–2018, collapsed to the ZIP-year-subgroup level.

Regressions are weighted by the workforce population in each cell. Standard errors are clustered by year and ZIP3.

54

Table A.3: Counterfactual Self-Employment Rates

Baseline

Scenario 1

RDD Adjusted

Scenario 2

Incentivized =

Unincentivized

Scenario 3

Scenario 2 &

1099-to-SE Adj

Scenario 4

Scenario 2 &

Fixed Demog.

2000

0.0974

0.0969

0.0901

0.0901

0.1095

2001

0.0980

0.0968

0.0898

0.0896

0.1084

2002

0.0989

0.0969

0.0897

0.0891

0.1088

2003

0.1030

0.1014

0.0928

0.0917

0.1122

2004

0.1062

0.1033

0.0957

0.0944

0.1150

2005

0.1095

0.1088

0.0985

0.0970

0.1166

2006

0.1102

0.1082

0.0989

0.0973

0.1174

2007

0.1112

0.1074

0.0990

0.0969

0.1165

2008

0.1095

0.1065

0.0962

0.0936

0.1136

2009

0.1128

0.1091

0.0971

0.0937

0.1134

2010

0.1150

0.1100

0.0973

0.0935

0.1141

2011

0.1172

0.1129

0.0985

0.0941

0.1161

2012

0.1175

0.1129

0.0994

0.0947

0.1176

2013

0.1170

0.1121

0.0991

0.0942

0.1171

2014

0.1182

0.1132

0.1005

0.0955

0.1192

2015

0.1175

0.1128

0.1003

0.0953

0.1195

2016

0.1174

0.1123

0.1008

0.0957

0.1216

2017

0.1160

0.1127

0.1000

0.0946

0.1206

2018

0.1151

0.1103

0.0995

0.0941

0.1214

Notes: Table reports baseline (unadjusted) share of workforce with self-employment

earnings alongside counterfactual series adjusted for shifts in reporting behavior and

demographic change “Scenario (1)” examines how self employment would have evolved

in the absence of any reporting incentives captured in our RDD estimates; specifically,

it reports the counterfactual replacing our RDD estimates reported in Figure 5 with

zero in all years. The adjustment in “Scenario (2)” replaces SE rates for individuals

with incentives to report SE with the rates among comparable individuals without this

incentive in each year. “Scenario (3)” applies an adjustment accounting for changes

in the propensity of unincentivized individuals with 1099 contract income to report

their self-employment proceeds on a 1040 over time. “Scenario (4)” accounts for demographic shifts over this time frame by holding the gender and age composition of

the workforce constant at its 2000 levels. See text for further details.

55

Table A.4: Level Adjustments: Share of Workforce with 1099-Reported Non-Employee Compensation

Baseline

Scenario 1

RDD Adjusted

Scenario 2

Incentivized =

Unincentivized

Scenario 4

Scenario 2 &

Fixed Demog.

2000

0.0906

0.0903

0.0875

0.0876

2001

0.0887

0.0889

0.0860

0.0859

2002

0.0928

0.0923

0.0899

0.0896

2003

0.0947

0.0947

0.0911

0.0905

2004

0.0965

0.0963

0.0926

0.0918

2005

0.0962

0.0962

0.0922

0.0914

2006

0.0979

0.0985

0.0939

0.0929

2007

0.0983

0.0984

0.0942

0.0929

2008

0.0970

0.0969

0.0927

0.0910

2009

0.0938

0.0939

0.0893

0.0870

2010

0.0951

0.0948

0.0906

0.0880

2011

0.0969

0.0972

0.0923

0.0895

2012

0.0975

0.0975

0.0927

0.0897

[0.0974]

[0.0975]

[0.0927]

[0.0896]

2013

2014

2015

2016

2017

2018

0.0975

0.0979

0.0926

0.0895

[0.0973]

[0.0979]

[0.0924]

[0.0893]

0.0991

0.0991

0.0941

0.0911

[0.0980]

[0.0991]

[0.0930]

[0.0899]

0.1011

0.1016

0.0960

0.0931

[0.0973]

[0.1016]

[0.0922]

[0.0891]

0.1044

0.1048

0.0992

0.0964

[0.0968]

[0.1048]

[0.0917]

[0.0885]

0.1063

0.1065

0.1011

0.0974

[0.0995]

[0.1065]

[0.0945]

[0.0904]

0.1070

0.1075

0.1017

0.0979

[0.1002]

[0.1075]

[0.0952]

[0.0909]

Notes: Table reports baseline (unadjusted) share of workforce with 1099-non-reported non-employee compensation alongside counterfactual series adjusted for reporting incentives and demographics. Shares in square

brackets exclude OPE work. “Scenario (1)” adjusts self-employment downward according using our annual

RD estimates reported in Figure 5a. “Scenario (2)” replaces SE rates for individuals who have incentives to

report SE with the rates among comparable individuals without this incentive. “Scenario (3)” is not relevant for adjustments to 1099-reported non-employee compensation and thus is not included. “Scenario (4)”

accounts for demographic shifts over this time frame by holding the composition of the workforce constant

at its 2000 levels.

56

B

Self-Employment Reporting and Changing EITC Incentives: Event Study Around Childbirth

As discussed in the main text, only households with children face a negative marginal tax rate

for reporting self-employment income. To further test the hypothesis that self-employment

growth is tied to EITC incentives, we follow Chetty, Friedman, and Saez (2013) and examine

how self-employment reporting changes around a person’s first childbirth, when they become

eligible for a generous credit. We expand upon Chetty, Friedman, and Saez (2013) in two

main ways. First, to investigate the extent behavior is changing over time, we examine

the change across different time periods. Second, we separate the rise in self-employment

around childbirth into 1099-reported self-employment and self-reported work. An increase

in 1099-reported work may suggest changing worker needs around childbirth draw workers

into self-employment for the first time. We begin with a simple exercise, examining the raw

change in self-employment at childbirth, before formalizing our analysis in an event-study

framework.

We start by examining the simple raw change in self-employment in the year of childbirth.

We take childbirths for all parents reported in the SSA database whether or not the child

is claimed as a dependent on tax filings by that parent. Figure B.1a reports the change in

self-employment filing in the year of childbirth from the year before, for every cohort of first

births from 1997-2018. The figure shows that the extent to which individuals begin reporting

self-reported self-employment exactly when it becomes advantageous to do so has increased

over this period by 0.9 percentage points, from a level of 0.9 percentage points in 1997 to 1.8

percentage points by 2014. 1099-reported work—which individuals have no discretion over

reporting—differs in two key ways. First, on average, there is no increase in 1099-reported

work in the year of childbirth. Second, there is no underlying trend in the rate of doing

1099-reported work in the year of childbirth. Appendix Figure B.2 further breaks down

the trends by gender of the parent. We find that all of this increase comes from mothers:

the change in self-employment in the year of childbirth among mothers has gone from 0.4

percentage points in 1997 to 2 percentage points by 2014. In contrast, 1099-reported work

decreases in the year of childbirth for mothers; the decrease is actually slightly greater in

magnitude today than in the past.

We next proceed to formalize this analysis and examine additional periods after childbirth

using an event-study specification that will control for aging and business cycle effects. Our

57

event study specification is standard and given as follows:

yit =

X

p

p

βkp I{FirstChildbirthi = t + k} + γa(i)×g(i)

+ γt×g(i)

+ epit

(6)

k∈K

where i indexes parent, t indexes year, a(i) gives the age of i. FirstChildbirthi is i’s year

of first birth. g(i) is the parent’s gender, thus allowing for time and age effects to differ by

parental gender.44 We examine two key outcomes: having any contract/freelance work, and

being an S.E. taxpayer with no contract/freelance work. We run separate regressions for

different 3-year rolling windows, p ∈ {2003 − 2005, 2004 − 2006, ..., 2012 − 2014}. We exclude

an indicator for the period one year prior to first birth, so that the event-time coefficients

are all relative to period -1, and examine an event window of 4 years pre and post event

(k ∈ {−4, ...4}\ − 1). Standard errors are clustered at the individual level.

Figure B.1b plots the full set of event study coefficients we estimate for two cohorts of

births: 2003-2005 births and 2011-2013 births. As in the raw means, we find that 1099reported self-employment is flat around childbirth for both cohorts. But self-reported selfemployment is a different story—the propensity to self-report self-employment income increases sharply in the year of birth and by about 0.75 percentage points in subsequent years.

Moreover, the magnitude of this time 0 response has grown over time: while self-reported

self-employment rates grew by 0.5 percentage point after childbirth in 2003–2005, the corresponding increase was around 1.25 percentage points in 2011–2013. This contrasts with

firm-reported contract work, which did not become more common after childbirth in either

time period. Appendix Figure B.3 reports estimates separately by gender of the parent. As

we found earlier, these changes over time are largely driven by mothers.

44

Accordingly, the event-study coefficients are the average of coefficients run separately for men and women,

which we report in Appendix Figure B.3.

58

Figure B.1: Change in Self-Employment Around First Childbirth, 1997-2018 Births

-.01

Change from Previous Year

0

.01

.02

(a) Changes by year of First Birth: 1997-2018 Births

1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017

Year of Childbirth

SE

1099

-.01

0

Percent Participating

.01

.02

.03

.04

(b) Childbirth Event Study Estimates, 2003-2005 Births Versus 2011-2013 Births

-4

-3

-2

2003-2005 Births

2011-2013 Births

-1

0

1

Years to First Childbirth

SE, with 1099

SE, with 1099

2

3

4

SE, No 1099

SE, No 1099

Notes: Panel A shows the average change in propensity to file SE (solid line) or receive a 1099 Information

Return (dashed line), in the year of first childbirth reported on the x-axis. Panel B plots event study

coefficients for separate regressions run on the indicated time-period and for the indicated outcome. See text

for more details.

59

Figure B.2: Change in Self-Employment Around First Childbirth, 1997-2018 Births, By Gender of Parent

1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017

Year of Childbirth

SE

See notes for Figure B.1a.

1099

-.005

60

-.01

Change from Previous Year

0

.01

Change from Previous Year

0

.005

.01

.015

.02

(b) New Fathers

.02

(a) New Mothers

1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017

Year of Childbirth

SE

1099

Figure B.3: Childbirth Event Study Estimates, Additional Estimates

Percent Participating

.01

.02

.03

0

-.01

-.01

0

Percent Participating

.01

.02

.03

.04

(b) Fathers Only

.04

(a) Mothers Only

-4

-3

-2

2

SE, with 1099

SE, with 1099

3

4

-4

SE, No 1099

SE, No 1099

-3

-2

-1

0

1

Years to First Childbirth

2003-2005 Births

2011-2013 Births

SE, with 1099

SE, with 1099

0

Percent Participating

.01

.02

.03

.04

(c) All Parents, December Births Only

-.01

61

2003-2005 Births

2011-2013 Births

-1

0

1

Years to First Childbirth

-4

-3

-2

2003-2005 Births

2011-2013 Births

Notes: See notes for Figure B.1b.

-1

0

1

Years to First Childbirth

SE, with 1099

SE, with 1099

2

3

SE, No 1099

SE, No 1099

4

2

3

SE, No 1099

SE, No 1099

4

C

Additional Counterfactual Exercises

This Appendix describes the details of two additional counterfactual exercises.

First, to estimate changes in 1099-reporting over time, we examine the filing behavior of

1099 recipients. We run the following regression on the population of 1099 recipients who

file Schedule C:

t,p

t,p

1{F ilesSE}t,p

i = γs×r + i

(7)

t

where s denotes state and r denotes a bin for the dollar value of 1099 receipts. Hence, γs×r

report the share of individuals in a bin of non-employee compensation in a state, who file

SE in year t. We run this regression separately for OPE and non-OPE 1099s (denoted p) in

each year, and only among our unincentivized population, so that the phenomenon that we

capture will be independent from our earlier adjustments. In a second step, we calculate outof-sample predicted values in each year for individuals who do not file Schedule C. We sum

these values to get the predicted share of non-C filers who should file Schedule SE. Appendix

Figure C.1 reports the implied share of the workforce with non-employee compensation who

are underreporting SE.

To account for demographic changes over time, we hold constant the weights of our

demographic groups, ωg,t at their 2000 levels, i.e.



SEt∗ =

X

g

ωg,2000

X



X



k

∗

ζw,g,k,t

SEw,g,0,t +

w≤w∗ (k)

X

w>w∗ (k)

62

∗

ζw,g,k,t

SEw,g,k,t 

Figure C.1: Estimated Share of Workforce with 1099-Non-employee Compensation Underreporting SE

Share of Workforce

.022

.02

.018

.016

2000

2005

2010

2015

2020

Notes: Share of workforce with self-employment contract payments reported on 1099 forms but not reported

on their 1040-SE, estimated using on individuals with no self-employment reporting incentives following the

methodology described in Section 6.2 (which accounts for expensing behavior).

63

Figure C.2: Adjusted SE Shares Under Additional Counterfactual Assumptions

(b) Controlling for Demographics

.02

Percentage Points (Relative to 2000)

64

Percentage Points (Relative to 2000)

(a) Constant 1099-Filing Behavior

.015

.01

.005

0

2000

2003

2006

2009

2012

2015

Baseline

Scenario 2: Replace with SE Rates of Unincentivized

Scenario 3: S2 + 1099-SE Reporting Adjustment

2018

.02

.015

.01

.005

0

2000

2003

2006

2009

2012

2015

2018

Baseline

Scenario 2: Replace with SE Rates of Unincentivized

Scenario 4: S2 + Holding Demographics at 2000 Levels

Notes: Figure presents the baseline and implied trend in SE share of the overall tax workforce under counterfactual scenarios. In panel (a), “Scenario

(3)” applies an adjustment accounting for changes in the propensity of unincentivized individuals with 1099 contract income to report their selfemployment proceeds on a 1040 over time. In panel (b), “Scenario (4)” accounts for demographic shifts over this time frame by holding the

composition of the workforce constant at its 2000 levels.

D

Data Appendix

This appendix describes the technical details of our data construction where we combine

data from a variety of different tax forms.

The core of our analysis draws on de-identified, or “masked”, W2, 1099-MISC, and

1099-K information returns along with 1040 individual tax returns and associated schedules

(e.g. Schedule SE). We begin with the population of individuals who appear as primary or

secondary filers on a 1040 in each year. We create a record of all de-identified individuals,

using masked Taxpayer Identification Numbers (TINs) appearing on these forms, attributed

to either the primary filer or the attached spouse.

For all years, we merge in self-employment information for individuals and their spouses

from Schedule SE. On Schedule SE (a schedule of Form 1040), individuals report all selfemployment income subject to SECA taxation, so long as the total exceeds $400. This includes active income from wholly-owned businesses on Schedule C, income from partnerships

on Schedule K1, and farm income on Schedule F. Importantly, SECA taxes are assessed on

individuals, not income tax filing units, so Schedule SE is always identified at the individual

level.

We next turn to cleaning and processing the information returns. For Form W-2, we

pull all W-2s with TINs that have been validated by the IRS. We eliminate duplicate or

amended returns, and we drop a small number of invalid TINs (approximately 50,000 in

2016) and TINs considered “unmatchable” (approximately 5.2 million). Both of these are

small compared to the overall number of W-2s, which exceeded 240 million in 2016. We use

the recipient TINs to match W-2s to our main file of individuals. Since a large number of

individuals with low W-2 earnings are not required to file 1040 returns, we add all cases with

valid W-2s but no 1040 to our population file.

We then merge on information from Form 1099-MISC. We pull everyone with non-zero

non-employee compensation reported in Box 7. To identify the online platform economy, we

use the list of roughly 50 large labor platforms from Collins, Garin, Jackson, Koustas, and

Payne (2019) that are mentioned in public databases than can be identified in the tax data

(along with the corresponding EIN) using the unmasked firm name. Using the corresponding

masked EIN, we then identify all 1099-MISCs in our cleaned file coming from these platforms

and classify them as OPE income.

Reporting rules for intermediaries have changed over time in important ways that affect

our measurement of the OPE. In 2011, a new law went into effect requiring companies that

processed credit cards, electronic payments, or other transactions to report each recipient’s

65

payments on a new information return, “Form 1099-K.”45 Starting in 2012, several online

intermediaries in the OPE began issuing the new Form 1099-K instead of 1099-MISC for nonemployee compensation. The income paid to gig workers on OPE labor platforms is, for all

practical purposes, non-employee compensation. However, 1099-Ks are also issued for income

from sales that is not non-employee compensation. We therefore also identify and track

the 1099-Ks issued by the approximately 50 important online “gig” platforms where selfemployed individuals offer labor services to firms or individual clients mentioned above. We

then measure the total payments individuals receive from these companies that are reported

on either a 1099-K or a 1099-MISC with non-employee compensation. We also explore

alternative approaches to identifying OPE work, as some companies cannot be identified

by this method.46 For example, we use mentions of platform names in taxpayer-reported

descriptions of business activity (line A) on Schedule C to identify additional instances of

OPE work.

A potentially important limitation to studying the 1099-K is that companies in the labor

OPE classifying themselves as third party networks are only required to file this form if the

total amount of such transactions exceeds $20,000 and the aggregate number of such transactions exceeds 200. In practice, this does not appear to impact our analysis through 2016,

as we find most of the major platforms have issued 1099-Ks to all platform participants,

regardless of the earnings level, in at least some years. However, beginning in 2017, more

platforms begin to abide by the reporting thresholds, and so our measure of gig work is underestimated after 2016. In our analysis, we use Box 1 gross receipts to measure payments.

We clean these forms using the same methodology described for the 1099-MISCs. We attribute 1099-K OPE payments to individuals, and add this to OPE income. We consider this

income to be a part of the “1099 economy” and include it in measures of “1099 recipients”

or “1099 income.” So that our definition is more comparable over time, we only classify

someone as an OPE worker if they receive a 1099-MISC or have 1099-K earnings of $600 or

more; (Collins, Garin, Jackson, Koustas, and Payne, 2019) provides tabulations that include

full counts of 1099-K workers, regardless of amount earned.

Worker characteristics Marital status and claimed dependents are defined for 1040 filers

only. Marriage is determined from listing a spouse on a 1040. Dependents are determined

from listing dependents (other than the spouse) on the 1040 and from a database of parentchild links maintained by the Social Security Administration. For measures of household

earnings, wages and 1099 earnings are merged in for the spouse. Additional characteristics

45

This measure was included in The Housing and Economic Recovery Act of 2008, but did not take effect

until the 2011 tax year.

46

For some platforms that pay through the payment processor Paypal, the 1099 will be issued by Paypal,

and cannot be separately tied to a company in the OPE.

66

are merged in from other sources. Birth dates and gender are pulled from the DM-1 file,

populated by the Social Security Administration.

67