NBER WORKING PAPER SERIES

THE EMPLOYMENT EFFECTS OF A GUARANTEED INCOME:

EXPERIMENTAL EVIDENCE FROM TWO U.S. STATES

Eva Vivalt
Elizabeth Rhodes
Alexander W. Bartik
David E. Broockman
Patrick Krause
Sarah Miller

Working Paper 32719

http://www.nber.org/papers/w32719

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
July 2024, Revised January 2025

We thank the non-profit organizations that implemented the program we study. We thank Leo
Dai, Ethan Sansom, Jake Cosgrove, Kevin Didi, Taryn Eadie, Malek Hassouneh, Amy Huang,
Joshua Lin, Anthony McCanny, Oliver Scott Pankratz, Idalina Sachango, Sophia Scaglioni,
Stephen Stapleton, Derek Thiele, Angela Wang-Lin, Isaac Ahuvia, Francisco Brady, Jill Adona,
Oscar Alonso, Jack Bunge, Rashad Dixon, Marc-Andrea Fiorina and Ricardo Robles for
excellent research assistance. Alex Nawar, Sam Manning, Elizabeth Proehl, Tess Cotter, Karina
Dotson, and Aristia Kinis provided invaluable support through their work at OpenResearch. We
thank Carmelo Barbaro, Janelle Blackwood, Katie Buitrago, Melinda Croes, Crystal Godina,
Kelly Hallberg, Kirsten Jacobson, Timi Koyejo, Misuzu Schexnider, and many others at the
Inclusive Economy Lab at the University of Chicago for their pivotal role in supporting the
project. This paper gratefully acknowledges funding from the NSF (#2149344) and private
donors. This study received ethics approval from Advarra and the University of Toronto's
Institutional Review Boards. The study was pre-registered on the American Economic
Association RCT Registry (AEARCTR-0006750). The views expressed herein are those of the
authors and do not necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2024 by Eva Vivalt, Elizabeth Rhodes, Alexander W. Bartik, David E. Broockman, Patrick
Krause, and Sarah Miller. All rights reserved. Short sections of text, not to exceed two
paragraphs, may be quoted without explicit permission provided that full credit, including ©
notice, is given to the source.
The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States
Eva Vivalt, Elizabeth Rhodes, Alexander W. Bartik, David E. Broockman, Patrick Krause, and
Sarah Miller
NBER Working Paper No. 32719
July 2024, Revised January 2025
JEL No. H0, J01, J08

ABSTRACT

We study the causal impacts of income on a rich array of employment outcomes, leveraging an
experiment in which 1,000 low-income individuals were randomized into receiving $1,000 per
month unconditionally for three years, with a control group of 2,000 participants receiving $50/
month. We gather detailed survey data, administrative records, and data from a mobile phone app.
The transfer caused total individual income excluding the transfers to fall by about $2,000/year
relative to the control group and a 3.9 percentage point decrease in labor market participation.
Participants reduced their work hours as a result of the transfers by 1-2 hours/week and
participants’ partners reduced their work hours by a comparable amount. Among other categories of
time use, the greatest increase generated by the transfer was in time spent on leisure. Despite asking
detailed questions about amenities, we find no impact on quality of employment, and our
confidence intervals can rule out even small improvements. We observe no significant effects on
investments in human capital, though younger participants may pursue more formal education.
Overall, our results suggest a moderate labor supply effect that does not appear offset by other
productive activities.

Eva Vivalt David E. Broockman

Department of Economics University of California, Berkeley
University of Toronto 210 Social Sciences Building #1950
150 St. George Street Berkeley, CA 94720
Toronto, ON M5S 3G7 [email protected]
and University of Oxford
[email protected] Patrick Krause
OpenResearch
Elizabeth Rhodes [email protected]
OpenResearch
[email protected] Sarah Miller
Ross School of Business
Alexander W. Bartik University of Michigan
University of Illinois at Urbana-Champaign 701 Tappan Street
1407 W. Gregory Road Ann Arbor, MI 48109
214 David Kinley Hall and NBER
Urbana, IL 61821 [email protected]
[email protected]

A randomized controlled trials registry entry is available at

https://www.socialscienceregistry.org/trials/6750/
1 Introduction
The design and success of public poverty alleviation programs depend crucially on how cash transfers

affect beneficiaries’ labor supply and other employment-related outcomes. Means-tested cash transfer

programs distort returns to work, causing beneficiaries to cut back on their work hours or earnings

in order to preserve benefits. As a result, advocates and policymakers have increasingly considered

unconditional cash transfer programs that would not generate such distortions. However, even such

unconditional programs will result in labor supply reductions if beneficiaries place a high value on

leisure or derive a high disutility from the kind of work that is available to them. Typically such reduc-

tions in work due to income effects are not thought of as distortionary and instead reflect beneficiaries’

high marginal utility of time off of work. But, even labor supply reductions due to such income effects

could increase the fiscal costs of public programs, and they could also harm participants’ long-term

labor market prospects, especially to the extent that beneficiaries are not sufficiently forward-looking.

On the other hand, cash transfers may help recipients overcome credit or liquidity constraints, allow-

ing them to search longer and potentially find higher-quality or better-fitting jobs, reduce barriers to

employment, support entrepreneurship or human capital formation, or lead to productive non-work

activities like caregiving. In this case, the benefits of cash transfers to the beneficiaries and society may

be large even if they generate some reductions in labor supply.

Given that the U.S. government spends hundreds of billions of dollars each year on programs

such as the Child Tax Credit (CTC), the Earned Income Tax Credit (EITC), Temporary Aid to Needy

Families (TANF), and many other programs, it is important to understand the effects that cash trans-

fers have on all these dimensions. Interest in these outcomes has driven extensive research on the

impacts of income on labor supply, where much of the literature reports no effect or a weak negative

effect (summarized in Krueger and Meyer 2002). Much less is known about the impact of unearned

income on other significant aspects of the labor market, such as job search, quality of employment,

entrepreneurial activities, barriers to employment and disability, human capital formation, and labor

market mobility. We also have limited understanding of how income affects other uses of a recipi-

ent’s time, or how recipients might trade off work and competing priorities such as home production,

caregiving, leisure, and self care when more resources are readily available. These outcomes, which

are difficult to measure using the administrative and survey data sets employed in existing research,

can be important in predicting the long-run impacts of cash transfers, as well as being valuable to

1
understand in their own right.

We investigate the causal effects of income on employment and these other related outcomes by

analyzing a program by two non-profit organizations that distributed $1,000 per month for three years

to 1,000 low-income individuals randomized into the treatment group. 2,000 participants were ran-

domly assigned to receive $50 per month as the control group. This is the largest unconditional cash

transfer program evaluated by a randomized controlled trial (RCT) in the U.S. to date in terms of the

amount disbursed. We merge rich survey data with administrative records and mobile phone data. By

collecting and merging a comprehensive set of outcome variables, we are able to answer questions that

have previously eluded causal estimation. For example, if people work a little less, as we might expect

from the past literature (Imbens, Rubin and Sacerdote, 2001; Cesarini et al., 2017; Golosov et al., 2023),

what do they do with their time instead? This question has important policy implications: decision-

makers may want to know whether participants engage in activities with positive spillovers, such as

education or caregiving, and understanding how participants choose to spend their time is also in-

formative of their revealed preferences. Moreover, if the transfers enable unemployed participants to

search longer for work, does that translate to any changes in the quality of their employment? Are

there effects on entrepreneurship or human capital investments?

The transfer program studied is particularly relevant to policymakers as it is targeted at lower-

income individuals, who are the target of virtually all cash transfer programs in the U.S. Individuals

between the ages of 21 and 40 whose total household income did not exceed 300% of the Federal

Poverty Level (FPL) in 2019 were eligible to participate, with the bulk of the sample targeted to fall

below 100% or 200% of the FPL. Participants reported an average household income of about $29,900

in 2019, so the transfers represented about a 40% increase in household income. The sample approxi-

mated the broader U.S. population among those who satisfied the income and age eligibility criteria,

and we ensured balance between the treatment and control group on a long list of variables. The

study’s experimental approach allows us to estimate the causal effects of the transfer with minimal

assumptions, and we pre-registered our analyses.1

Examining the effects of the cash transfers on income and labor supply using a combination of

state Unemployment Insurance (UI) records and survey data, we find total individual income exclud-

ing the transfers fell by about $2,000 per year relative to the control group, with these effects growing

over the course of the study. These decreases should be viewed in the context of increasing income
1 AEARCTR-0006750. Changes since the pre-analysis plan was registered are described in Appendix E.

2
in both the treatment and control group over the study period. The program caused a 3.9 percent-

age point reduction in the extensive margin of labor supply and a 1-2 hours/week reduction in labor

hours for participants. The estimates of the effects of cash on income and labor hours represent an

approximately 5-6% decline relative to the control group mean. This is a moderate effect: compared

to results from studies of lottery winners, these effects are arguably larger than seen in Imbens, Rubin

and Sacerdote (2001) or Cesarini et al. (2017), but smaller than those in Golosov et al. (2023).2 Inter-

estingly, partners and other adults in the household seem to change their labor supply comparably to

participants. Given the magnitude of these household effects, we cannot reject a unitary household

model in which the household pools their income and makes decisions about labor supply and con-

sumption jointly. For every one dollar received, total household income excluding the transfers fell

by around 29 cents, and total individual income fell by around 18 cents. Estimates are similar using

administrative data alone and survey data alone. We also conducted exploratory analysis of the effect

of the transfer on a small number of pre-specified subgroups. Estimated labor supply effects are weak

and even positive for some subgroups, although for the most part these subgroup estimates are not

precise enough to reject an effect equal to the one derived from the full sample.

We captured time use using a combination of survey questions adapted from the American Time

Use Survey and 24-hour time diaries delivered through a mobile phone app on a randomly-selected

weekday and and a randomly-selected weekend day each month. The time diaries and survey ques-

tions support the findings for employment. Treated participants primarily use the time gained through

working less to increase leisure, also increasing time spent on driving or other transportation and fi-

nances, though the effects are modest in magnitude. We can reject even small changes in several other

specific categories of time use that could be important for gauging the policy effects of an unearned

cash transfer, such as time spent on childcare, exercising, searching for a job, or time spent on self

improvement.

We also saw significant impacts on duration of unemployment. Over the three years of the trans-

fers, the duration of the average spell of non-employment in the control group was 7.7 months; the

treatment had the effect of increasing this by 1.1 months. Those in the treatment group were more

likely to have recently applied for work but applied to fewer positions on average.

Despite asking extremely detailed questions about workplace amenities, we find no substantive

changes in any dimension of quality of employment and can rule out even small improvements, reject-
2 Though results are not directly comparable given the differences in the transfer size, payment frequency, and samples.

3
ing changes in the index of more than 0.028 standard deviations and changes in wages of more than 73

cents. We find that those in the treatment group have more interest in entrepreneurial activities and

are willing to take more financial risks. The coefficient on whether a participant started a business

is positive, but not statistically significant. Using data from the National Student Clearinghouse on

post-secondary education, we see some suggestive evidence that younger individuals are pursuing

more education as a result of the transfers, which could potentially help to explain the labor supply

effects within this subgroup. However, there are no significant effects in the broader sample. Those

in the treatment group also self-report increased rates of disabilities that limit the work they can do,

perhaps due to getting more medical care. We see no significant reductions in barriers to employment.

Finally, we see no significant changes in marriage or divorce, and participants do not appear to use

the transfers to move to better labor markets.

The study has a number of strengths compared to existing literature. To examine the effects of a

negative income tax (NIT) on the labor supply of recipients, the U.S. government conducted four ran-

domized experiments between 1968 and 1980 (e.g., Ashenfelter and Plant 1990). While these studies

were pathbreaking and are still often referred to today, these experiments were plagued by nonran-

dom selection, errors in randomization protocols, differential attrition, nonparticipation, and system-

atic income misreporting (Hausman and Wise 1979; Greenberg and Halsey 1983). Further, these ex-

periments were begun in a very different economic and political context, so their results may not gen-

eralize to the present day, and we are able to collect much more detailed data on a much broader range

of outcome variables, including through the use of a mobile phone app. A related strand of the liter-

ature utilizes the exogenous increase in income created by the introduction of the Earned Income Tax

Credit (EITC) and subsequent expansions to examine labor market effects (Eissa and Liebman 1996;

Meyer and Rosenbaum 2001; Eissa and Hoynes 2004; Nichols and Rothstein 2016). However, there has

been debate about these estimates due to simultaneous reforms and a strong economy (Kleven 2024).

Further, this literature necessarily focuses on subgroups potentially affected by the expansions, par-

ticularly married couples and families with children, and these subgroups could respond differently

than the broader population.

Unlike unconditional cash transfers, programs like the Earned Income Tax Credit (EITC) affect

beneficiaries’ labor market incentives because the amount of the benefit is linked to the amount of

earned income. To address this limitation, several studies have examined lottery winners. However,

the lottery studies generally either had small samples (Imbens, Rubin and Sacerdote 2001) or took

4
place in policy contexts very different from the U.S. (Cesarini et al. 2017). Lottery players may also

be selected in some way, such as being generally higher-income and perhaps more risk-loving than

the individuals a public guaranteed income program might target (Golosov et al. 2023). Other recent

quasi-experimental evidence of responses to exogenous increases in income comes from studies of the

Alaska Permanent Fund (Feinberg and Kuehn 2018; Jones and Marinescu 2018), which was small in

magnitude ($1,606 USD in 2019), and casino disbursements to Native American families in the U.S.

(Akee et al. 2010).3

In contrast to the preceding literature, a key advantage of this study is the ability to combine ex-

perimental variation in a large unconditional cash transfer with uniquely rich data. Existing studies

largely rely on administrative data sets with limited information on the individuals, despite theoreti-

cal and empirical evidence that contextual factors and preferences matter (e.g., Cox and Oaxaca 1990;

Atkinson and Micklewright 1991; Krueger and Meyer 2002; DellaVigna and Paserman 2005; Boswell,

Zimmerman and Swider 2012). We collect very detailed data about participants from administra-

tive records and surveys, enabling a more nuanced understanding of their labor supply and time use

decisions situated within the context of other choices they face. The administrative data include quar-

terly earnings and employment information reported by employers to states’ unemployment insur-

ance agencies from the two states from which participants were recruited, as well as National Student

Clearinghouse data on post-secondary educational outcomes. The survey data were collected through

a combination of in-person and phone-based surveys implemented by the Survey Research Center at

the University of Michigan as well as frequent web-based surveys and a mobile phone app. We had

very high responses to these surveys, including a 97% response rate to the midline survey and a 96%

response rate to the endline survey.4

The comprehensive data collection enables us to estimate a structural model of labor and con-

sumption responses with minimal assumptions relative to the literature. For example, as we observe

consumption, savings and debt (Bartik et al., 2024), we can consider asset accumulation explicitly in

our model. Compared to the literature, participants in our study spend nearly all the money they
3 There is also an important literature on cash transfers in a developing country context. Most of this work focuses on
conditional cash transfers and children’s outcomes (reviewed, for example, in Fiszbein et al. 2009). However, some studies
leverage unconditional cash transfers and consider employment outcomes (Mostert and Castello 2020). Banerjee et al. (2017)
review seven government-run cash transfer programs and find no systematic effect on labor supply on either the intensive
or extensive margin. In a study of three-generation households in South Africa, Bertrand, Mullainathan and Miller (2003)
find a sharp decline in both the extensive and intensive margin in working-age individuals’ labor supply when an individual
in the household receives a pension. These results are important but may not generalize to the U.S., given the significant
contextual differences.
4 This does not account for mortality, so attrition due to other causes is even smaller.

5
receive each month on increased consumption or reduced labor. The relative lack of savings is impor-

tant in calibrating our model as estimates of the marginal propensity to earn (MPE) greatly depend

on the denominator, i.e., how much of the transfers participants treat as theirs to spend that month as

opposed to saving to spend in future time periods.

Due to the detailed data collection, this study also allows us to speak to an ongoing debate in

the literature as to whether expansions of the social safety net lengthen unemployment but ultimately

result in better job matches between job seekers and employers. This literature has historically fo-

cused on changes in the generosity of employment insurance, but similar arguments could apply to

job search under the increased security of monthly cash transfers. The literature, mostly from Eu-

ropean countries, is mixed, with Centeno (2004), Caliendo, Tatsiramos and Uhlendorff (2012), and

Nekoei and Weber (2017) finding that more generous benefits enable better jobs, while another strand

of the literature finds no such effects (e.g. Card, Chetty and Weber, 2007, Lalive, 2007, van Ours and

Vodopivec, 2008). In addition to drawing from other countries with more generous social safety nets,

past papers in this literature have often had limited information on job quality, inferring job quality

from income or the duration that the post-unemployment job was held. In contrast, we have a rich

array of variables we can use to identify quality of employment and characterize the jobs participants

are applying to.

Our study also contrasts with recent work on several randomized cash transfer programs. Chelsea

Eats, in Chelsea, MA, provided $400/month for 9 months to 1,067 treated participants, with a group

of 730 residents serving as the control. The transfers ran from Nov. 2020 to Aug. 2021. They fo-

cus primarily on food consumption and financial well-being and do not find significant effects on

employment or work hours (Liebman et al., 2022). Baby’s First Years provided 400 low-income new

mothers in a "high" cash arm with $333/month for 72 months, starting in May 2018-July 2019, with

an additional 600 in a "low" cash arm receiving $20/month. These transfers were provided on a debit

card labelled "4MyBaby", and participants were spread across four U.S. cities. The evaluators did

not find any effects on maternal employment (Stillwell et al., 2024; Sauval et al., 2024). Jaroszewicz

et al. (2023) examine a U.S. program which randomized 699 individuals to receive a one-time transfer

of $2,000, 1,374 individuals to receive a one-time transfer of $500, and 3,170 individuals to receive

nothing between July 2020 and May 2021. They find small negative effects on earned income and

null effects on employment. The Compton Pledge provided transfers of $450 per month on average

over a two-year period to 695 low-income, mostly Hispanic households, with a control group of 1,402

6
households (Balakrishnan et al., 2024). They find moderate decreases in both income excluding the

transfers and consumption. Relative to the treatments investigated in these studies, the program we

study provided larger transfers over a long period of time. The duration of the program may be im-

portant given that in our study we observe different effects over time. We benefited from extremely

high survey response rates and limited differential attrition, likely due to the control group receiving

smaller transfers and extensive outreach and tracking efforts by the project team. We also importantly

leverage administrative records, which appear to show a larger effect on labor supply than the sur-

vey data alone would suggest. Finally, we collected a wider range of employment variables than any

existing study, including data from a custom mobile phone application.

Our results demonstrate that monthly cash transfers have a moderate effect on labor supply and

that this decline in formal sector production is not fully offset by substitutions towards other produc-

tive activities like human capital investments or home production. We also do not find support for

other hypothesized benefits to long-run employment, like an improved quality of job fit, though it

is possible that a subset of participants are making investments with payoffs that will take longer to

observe. For a policymaker interested in cash transfers, the main benefits would flow through the

increased choice they offer participants in how to spend their time and invest for the future, even

if relatively few use the opportunity for any one given pursuit such as obtaining a post-secondary

degree or starting a business.

In the following sections, we describe the sample and approach in more detail. After presenting

results, we explore heterogeneity and build a structural model to explain our findings and compare

our results to the existing literature.

2 The OpenResearch Unconditional income Study (ORUS)

2.1 Recruitment

The study took place in two sites: ten counties in north central Texas, including the Dallas area, where

the cash assistance program was implemented by a local 501(c)(3) non-profit organization, and nine

counties in northern Illinois, including the Chicago area, where an identical program was imple-

mented by an Illinois-based non-profit. Both sites combined participants living in urban locations

(from Dallas and Chicago, respectively), suburban locations, medium-sized urban areas, and rural

locations. The sites are depicted in Figure 1.

A total of 3,000 people were enrolled in the program. Individuals between the ages of 21 and

7
40 whose total household income did not exceed 300% of the Federal Poverty Level (FPL) in 2019

were eligible to participate. The organizations implementing the program excluded individuals from

households where at least one person receives Supplemental Security Income (SSI) or Social Security

Disability Insurance (SSDI), as well as those in publicly-subsidized housing, so that they would not

lose important benefits. Extensive effort was taken to protect eligibility for public assistance programs,

with collaboration between some of the research team, implementing partners, and state representa-

tives to pass Bill SB 1735, which protected many government-provided benefits in Illinois.5 Only

Medicaid and energy assistance were protected in Texas, but benefits are less generous and eligibility

criteria are more restrictive in Texas. A table of specific benefits and their protection status is pro-

vided in Appendix Table A1. The transfers were not conditioned on research participation and were

considered gifts from non-profit organizations and not taxable income.

The non-profit implementers recruited potential participants in three ways. Most participants

(87%) were recruited via a mailer that asked if they were interested in participating in a cash assis-

tance demonstration program and stated that they would receive “$50 or more” each month if they

were chosen to participate. Addresses within program counties were selected to receive mailers based

on information from TargetSmart, which provides address data and demographic details about res-

idents at each address. Approximately 69% of mailers were sent to individuals who appeared to be

eligible for the program on the basis of their age and income, but 31% of mailers were sent without any

targeting, to avoid systematically excluding individuals who were eligible but who would not have

appeared to be so based on the commercial data (e.g., through having missing data). The mailers were

addressed to a maximum of one person at each address, and "or Current Resident" was appended to

the address line. Interested recipients were then directed to a website that allowed them to complete

a simple intake survey to determine eligibility. Recipients were also offered randomized incentives of

$0 to $20 to complete the survey questionnaire. Upon survey completion, online gift cards were imme-

diately sent via email to increase trust. Follow-up letters were sent to those who did not respond, with

each individual randomized to receive between 0 and 4 follow-ups. A flowchart of the recruitment

process is provided in Figure 2.

A smaller number of participants were recruited by alternative methods. First, advertisements

were displayed through Facebook and Instagram to all individuals who appeared to be eligible for

the program based on their age and county. Approximately 1 percent of the sample was recruited
5 Specifically, this bill protected SNAP, TANF, child care assistance, Medicaid, and energy assistance. Further details on

the bill are provided in Appendix Figure A1.

8
through this approach. Second, advertisements were posted on "Fresh EBT", a free mobile application

that is used by over 4 million recipients of the Supplemental Nutrition Assistance Program (SNAP)

nationwide to check their balance and manage their benefits.6 Advertisements were limited to the

eligible zip codes. Approximately 12% of the sample was recruited through this app.

2.2 Randomizations

There were two randomizations, described in more detail below. The first randomized individuals to

be in the main study sample (receiving either $50/month or $1,000/month for three years) or out of

the main study sample. The second randomization occurred after all individuals in the main study

sample were enrolled. It randomized people into receiving either a high or low transfer amount.

2.2.1 Randomization to the Main Study Sample

The first randomization took the eligible applicants and randomized 3,000 individuals into being part

of the study sample. Beyond receiving further surveys, this group received a minimum of $50 per

month unconditionally. This randomization was designed so that the study sample met certain crite-

ria. There were three desiderata: 1) that the study sample included a minimum of 20% non-Hispanic

White, 20% Black, and 20% Hispanic participants; 2) that it included a minimum of 30% individuals

below 100% of the FPL, a minimum of 30% between 100% and 200% of the FPL, and no more than

25% between 200% and 300% of the FPL; and 3) that in terms of gender representation it reflected the

distribution of men and women in the eligible population according to data from the American Com-

munity Survey (ACS). To achieve the desired sample, we blocked participants on their demographic

characteristics and randomized a larger share from some blocks to the study sample.

2.2.2 Enrollment

After the first randomization, the contact information of the sample of potential participants was pro-

vided to the Survey Research Center (SRC) at the University of Michigan. Participants were first

enrolled in the cash transfer program before being invited to participate in the research. Consenting

participants then completed a comprehensive baseline survey and were asked if they wished to pro-

vide consent for the research team to analyze their administrative data. As part of the enrollment

procedures, participants also provided bank account information so that funds could be transferred

to them via direct deposit. 348 individuals did not have a bank account at enrollment, and an online
6 The application, developed by Propel, is now called Providers.

9
bank account was created for them to receive their transfers. Enrollment was conducted in person

from October 2019 to March 2020, when it switched to being conducted over the phone until all 3,000

individuals were enrolled by October 2020.

The long baseline period was intentional. First, it enabled us to obtain a large amount of baseline

data on participants, since during the baseline period we sent participants monthly surveys. Sec-

ond, we believed that attrition might be highest in the first few months of the study, and by having

a long baseline period, we could balance on attrition when conducting the second randomization to

the $50/month or $1,000/month transfers. Participants were paid $10 per survey during the baseline

period ($50 for the enrollment survey, which was much longer), and received $50/month uncondi-

tionally as a gift during this period. Participation in the program did not depend in any way on

participation in research activities.

We tested whether the population enrolling in the study is different from the broader population

by re-weighting the population in the ACS to match our FPL group and county type stratification

variables: while we cannot rule out differences in unobservables, it is reassuring that differences in

observables appear small (Table 1). The participants look comparable to the broader population on all

measures except for being slightly more likely to rent, slightly more likely to have a college degree,

and slightly more likely to be female.

2.2.3 Randomization to Treatment or Control

After enrollment, we conducted the second randomization to assign participants to either receive

$50/month ("control") or $1,000/month ("treatment") unconditionally for three years. The differences

between these two groups will be of primary interest in our analyses.

For this second randomization, all participants had an equal 1 in 3 probability of being assigned to

the treatment group.7 We implemented a blocked random assignment process to ensure balance over

key strata as well as imposing a minimum p-value for differences between the treatment and control

group on a wide range of baseline covariates. A balance table focusing on employment outcomes is

presented in Table 2. About 58% of participants were employed at baseline, with a total household

income in the year before enrollment of about $29,900. 17% of participants had a second job. 57% had

children living with them in the household, and 43% were living with a romantic partner. The average

household had 3.0 people in it, including the participant. About 20% of the sample had a bachelor’s
7 A waitlist for the $1,000 payments was also developed, however, it was not meaningfully used as we had excellent

take-up of the $1,000 payments. Appendix B provides more details.

10
degree.

During enrollment, we identified a handful of participants who knew each other. Out of an abun-

dance of caution, we grouped these individuals and anyone at the same address (such as a large

apartment building) into a "cluster", and each cluster was assigned to either treatment or control to-

gether.8 Given the random assignment of clusters to treatment or control, the standard errors in our

analyses are also clustered at this level.9 We conducted simulations to confirm that every cluster had

a 1 in 3 chance of assignment to the treatment group. Further details are provided in Appendix C.

2.3 Cash Transfers

After the second randomization, members of the treatment group were notified of the increased trans-

fer amounts. Both the treatment and control group were reminded of the transfer timeline, and the

implementing partners reminded them repeatedly about this in the final year of the program. The

cash transfers were unconditional, and participants in the treatment and control arms continued to

receive them even if they did not participate in the research.

Enrollment in ORUS was completed by October 2020. Randomization into treatment and control

took place immediately thereafter, and treatment began in November 2020 and ran through October

2023.10

3 Data Collection and Outcome Measures

We collected four types of data: (1) administrative data; (2) data from in-person/phone interviews

conducted by SRC; (3) data from web-based surveys; and (4) data collected using a custom mobile

phone application.

3.1 Administrative Data

We leveraged data on income and employment from Illinois and Texas Unemployment Insurance (UI)

agency records. Employers are required to report quarterly employment and earnings for all employ-

ees in UI-covered positions to state agencies. These data were then made available via a data use

agreement. While some jobs are excluded from the UI system–for example, independent contractors
8 In
total, this approach yielded 18 clusters which had 2 people in them and 2 which had 3 people.
9 There is one exception: the Texas UI data provider did not permit the cluster variable to be included in the environment
due to privacy concerns. In these data we cluster at the individual level, but do not expect this to meaningfully affect our
inference given that very few participants were randomized in clusters.
10 This timing means that the majority of the transfers occurred after COVID-19 vaccines became widely available. As our

analysis relies primarily on data from 2022 and 2023, with 2023 weighted particularly heavily, our results are predominantly
based on the post-COVID-19 era, particularly compared to other cash transfer pilots.

11
and those who are self-employed but not incorporated, including “gig” workers such as drivers for

ride share companies–we expect these records to be fairly comprehensive.11 87.5% of participants

consented for us to link their administrative records. Matching was done by state partners on SSN in

Illinois and name, date of birth and SSN in Texas. The match rate was very high for those participants

who provided a full SSN, but only 72% of participants in each of Illinois and Texas shared that infor-

mation. In total, among those who consented to share administrative records but not conditioning on

provision of SSN we obtained a 71% match rate in Illinois and a 73% match rate in Texas. In Illinois,

we were able to analyze survey data alongside the administrative records. This allowed us to fill in

missing earnings and employment records in the administrative data with the corresponding survey

outcomes for those who were not successfully matched.12,13

We also linked participants to administrative data from the National Student Clearinghouse (NSC)

on post-secondary educational enrollment and completion. The NSC provides excellent coverage of

post-secondary institutions in the U.S., covering 97% of institutions over the transfer period.14 These

data include information on degree attainment, enrollments, and progress in the degree, as well as

descriptive details about the fields of study pursued. We supplement these data with survey data for

those who did not consent to linkage.

Finally, we leverage information on debt from individual-level linkages of these consenting par-

ticipants to credit report data from the credit reporting agency Experian and again supplement with

survey-reported debt measures for those who did not consent to external linkages.

3.2 Enumerated Survey Data

Trained enumerators from SRC conducted interviews with participants prior to the start of the cash

transfer payments (“baseline”), after approximately 18 months of transfer payments (“midline”), and

after approximately 30 months of transfer payments (“endline”). The midline ran from April 3 - Au-

gust 2, 2022, while the endline ran from March 30 - August 15, 2023. The endline surveys were planned
11 Graham et al. (2022) estimates that 95 percent of employment nationally is covered by UI. Looking at Californian tax
filers, Bernhardt et al. (2022) find 5.9% have exclusively independent contractor or self-employment earnings, while an
additional 6.2% have some independent contractor or self-employment earnings supplementing a W-2 income. Katz and
Krueger (2019) find 15.8% of US workers are in alternative work arrangements, including independent contractors, on-call
workers, temporary help agency workers, and contract workers; W-2 employees of a temp-agency or contract agency would
be represented in the UI data, as would on-call workers who receive a W-2.
12 In Texas, we were unable to bring such detailed data into the administrative data environment, though we did bring in

56 baseline covariates.
13 In general, we construct outcome variables in the administrative data so as to match the survey data, i.e., using data for

an individual based on the quarter in which they took the survey. In Texas, we use the quarter that most participants took
the survey in for post-baseline measures (Q2 in 2022 for midline and Q2 in 2023 for endline) and the latest baseline time
period (Q3 in 2020).
14 https://nscresearchcenter.org/workingwithourdata/.

12
to take place a few months prior to the end of the program so as not to capture changes in behavior

that may arise from the anticipation of no longer receiving monthly transfers. A timeline of the main

study events is included in Figure 3.

To avoid burdening respondents with overly long surveys, we partitioned some of the survey

questions we wanted to ask about and asked them in separate online surveys following the corre-

sponding SRC survey. Participants were provided with $50-$100 for answering the SRC surveys and

$15-$30 for answering each of the follow-up online surveys.15

We obtained very high response rates to the midline and endline survey. At the time of the midline

survey, approximately 1.5 years into the cash transfer period, when a participant might have been

enrolled in the study for 2 years, we obtained a response rate of 97%. At endline, a year later, we

obtained a response rate of 96%.16

3.3 Web-based Survey Data

We measured many of the outcomes using data from monthly surveys administered using the Qualtrics

web-based survey platform. These surveys included questions on time use with different lookback

periods as a complement to mobile app-based time diaries, as well as questions on job search, quality

of employment, job satisfaction, hours worked, income changes, intrahousehold employment out-

comes, housing search and mobility, and participation in formal and informal education and training,

among other outcomes. Participants were compensated $10 for every survey completed.

This frequent contact with participants enabled us to keep up-to-date on any address or contact

information changes. The questions that were asked on each survey varied by survey, but generally

each module of questions was asked multiple times per year. This gave us multiple chances to collect

information that might have been missed in any one survey. In our analyses, we collapse participant

responses within a year.

Response rates to the monthly web-based surveys were high: 98% completed at least one web-

based survey in the first year, 96% in year 2, and 94% in year 3. Appendix Figure A2 shows response

rates by survey year.

15 At baseline, we offered a $50 kept appointment bonus at the very end of the recruitment period, on top of the $50 base

incentive, and at midline and endline people were randomly assigned to receive a kept appointment bonus of $0, $25, or
$50 in addition to the base incentive. For the mobile endline, total incentives were increased to $30 in the final weeks of the
endline period.
16 For the three online surveys that followed the midline and were associated with it, we obtained response rates of 93.7%,

91.0% and 89.2%, and for the four online surveys that followed the endline, we had response rates of response rates of 95.2%,
93.2%, 91.1% and 88.6%.

13
3.4 Mobile Application Data

Participants in ORUS used a mobile phone application created for the program by Avicenna Research.

We used this mobile app for both passive and active data collection for the proposed study. Daily time

diaries are widely regarded as the gold standard of time use surveys, and the app provides a user-

friendly calendar interface that allows respondents to report all of their activities in a 24-hour period

by dragging activities into time slots. This interface also has the advantage of enabling us to collect

information on both primary and secondary activities (e.g., participants may say they were cooking

but also watching television alone at the same time). We asked respondents to complete time diaries

on a randomly-selected weekday and weekend day each month. Participants were compensated with

$5 for every time diary completed. A screenshot of the interface is provided in Figure 4.

The time diaries had a high response rate and were elicited very frequently, so we have a large

number of repeated measures in these data. The web-based surveys achieved higher response rates,

but were less frequent. Results for both modalities will be presented.

3.5 Attrition

We proactively curbed attrition and non-response by sending email and text reminders, as well as

sending postcards that appeared to be handwritten and calling non-responders by phone. At enroll-

ment, we also asked participants to provide the contact information of two other people who could be

reached in case the participant’s contact information was no longer valid, and participants were asked

to update this information at midline and endline.

We observed extremely limited differential attrition given the length of time over which we stayed

in contact with participants. At the time of the midline survey, we observed differential attrition of

only 1.7%, and at endline, only 3.2%. For the monthly online surveys, we did not observe significant

differential attrition at all in year 1 and year 2 after pooling across surveys within the year, with 4.3%

differential attrition observed in year 3. Differential attrition in the app-based time diaries was 6.0%

on average across the three years of the study.17

Despite this very low overall attrition, we take several measures to mitigate concerns that differ-

ential attrition might affect results. First, we prioritize outcomes in the administrative data, where

we do not observe differential attrition. Second, we check that respondents and non-respondents

appear similar to one another on a long list of baseline covariates (Appendix Tables A2-A6). Third,
17 4.4% in year 1, 7.5% in year 2, and 6.5% in year 3.

14
we provide Lee bounds estimates conservatively correcting for this (at the expense of less precision).

Fourth, we present a set of results restricting attention to the midline and endline surveys, to which

we had particularly high response rates. Finally, we implement a differences-in-differences approach

as a further robustness check, for those outcomes for which we have baseline values. This approach

does not require respondents in the treatment and control group to be balanced for identification, but

rather only requires these groups to have parallel trends. All robustness checks are included in the

appendix.

4 Method
Our main analyses estimate the effect of the cash transfers on employment outcomes through the

following specification:

Yi = α + βTreatedi + γXi + ε i (1)

where Y represents a given post-treatment outcome variable, i represents the individual participant,

Treated is an indicator variable denoting treatment status, and X is a matrix of Lasso-selected con-

trols.18

Given that we have outcomes data from multiple time periods, we had to pre-specify how we

would treat them. Our preferred specification pools results across time periods, leveraging the extra

power that multiple measures gives us, yielding a single aggregate measure capturing changes over

the study period, though we also show disaggregated results for completeness. We pre-specified that

we would we would place more weight on the endline outcomes (70%) than the midline outcomes

(30%), and that we would similarly place more weight on online survey responses in year 3 (50%)

than in year 2 (30%) or year 1 (20%). Placing more weight on the results from later years has the

advantage that if the transfers have effects that accrue over time, this approach would better capture

them. Another reason we preferred to place more weight on later time periods is that one of the

unique features of our study is the relatively long duration of the transfers it studies, and we are

primarily interested in changes that might occur over longer periods of time. Further, by focusing on
18 We
are unable to bring the full set of over a thousand baseline covariates into the Texas UI data environment, but we
do bring in a set of 56 baseline covariates from survey data, focusing on demographic variables, employment, income,
household composition, relationship status, and county type, and we can additionally leverage baseline covariates in the
administrative data. We run a Lasso within the administrative data environment on this more limited set of covariates
to generate the main estimates based on UI data. We are able to bring all baseline covariates into the Illinois UI data
environment.

15
this timeframe, we anticipate that our findings will have greater external validity given the COVID-19

pandemic potentially affecting the first year of the study, as it did many cash transfers around that

period. We also pre-specified that we would place more weight on the SRC survey data (70%) than

the online survey data (30%), given that these data may be higher-quality and have less non-response

bias. We further present a set of estimates that rely only on SRC data and administrative data, from

which individuals cannot attrit.

Since we have multiple outcome measures, we must correct for the fact we are conducting mul-

tiple hypothesis tests. Here, we take two approaches. First, we generate summary index measures

as a way of reducing the number of primary hypothesis tests, following Kling, Liebman and Katz

(2007). Constructing a hierarchy of outcomes, we group related measures into "families" of outcomes,

with several "components" capturing the same theoretical construct within a given "family", and spe-

cific "items" (e.g., responses to a survey question or a specific outcome variable in administrative

data) within the "component". For example, one family of outcomes we consider is the impact of the

transfers on quality of employment, but there are many dimensions to quality of employment. One

dimension that someone might care about is their day-to-day experience at work. This could include

such factors as whether they face discrimination at work, whether their boss treats them fairly, etc.

Questions asking about these factors ("primary items") could be combined with similar questions un-

der a "quality of work life" component, which in turn would be combined with other components in

the "quality of employment" family. The index measures are constructed by taking the standardized

estimates from item-level analyses and aggregating them within components using seemingly unre-

lated regression. The component-level estimates are then combined into families by averaging the

standardized effects. Prior to being combined in an index, items are reversed if necessary in order for

a positive treatment effect to represent a positive impact. We also present all item-level test results in

raw units, unadjusted, for the sake of interpretability. Sometimes a family may also contain one or

more secondary items, which are pre-specified to not be included in the index.

Our second approach to reduce the risk of "false positives" is to present false discovery rate (FDR)

adjusted q-values for our estimates in the main results. We put our estimates into tiers for the sake of

conducting multiple comparison adjustments, following Guess et al. (2023). The logic of this approach

is that some estimates may be higher-priority than others, and so long as this is pre-specified we

can also conduct secondary analyses that are clearly denoted as such without penalizing the higher-

priority tests. The family-level estimates are considered to be in the top tier, and all family-level

16
estimates are pooled when constructing the q-values. Component-level estimates occupy the next

tier, and these are pooled with the family-level and other component-level tests within the family.

Primary items are pooled with all the family-level, component-level and item-level tests within the

family. The last level of the hierarchy includes exploratory analyses, including any secondary items,

subgroup analyses, or estimates by time period. Further details are provided in Appendix D.

5 Results

5.1 Income, Labor Supply, and Time Use

The transfers led to a reduction in annual individual salaried or wage income of about $1,700 ac-

cording to our preferred estimate, which pools administrative records and survey data (Table 3).19

We asked participants survey questions about several types of income that go beyond income from

the types of employment that would be reported in UI data. Specifically, we asked participants to

state their total annual household income and total annual individual income, and we also asked

them about several categories of income, including salaried income, wage income, self-employment

income, supplemental income from gig work, passive income, and income from a variety of other

sources including government transfers. Table 3 presents results, along with an aggregated total cal-

culated income measure.

When participants are asked to report a single number capturing their total individual income,

the estimated treatment effect is $2,500; when we ask about different categories of income in survey

questions and aggregate them, the effect based on the aggregate calculated measure is $1,500. If we

expect administrative data to capture the types of income that the survey data on salaried or wage

income captured, then this aggregate calculated measure may be an under-estimate and the true ef-

fect on individual total income may be closer to $2,000.20 The largest component of the reduction in

calculated income appears to be salaried or wage income.

These estimates can be used to calculate participants’ MPE. One subtlety is that participants’ MPE

depends on two decisions: 1) the decision of how much of the transfer to allocate for consumption

today (whether consumption of leisure or other expenditures) vs. how much to allocate for consump-
19 Specifically, we use administrative records for those who consent to sharing these data and are able to be matched as

well as enumerated survey data for those who did not consent to share administrative records or, in Illinois, consented but
were not matched. As described in the data section, leveraging survey data from this last group was possible in Illinois but
not Texas.
20 Considering that the gap between individual salaried income from aggregating administrative and survey data, $1,675,

and the effect estimated from survey data, $1,150, is $525, and one might expect the calculated total income measure from
survey data, $1,503, to be understated by at least this much.

17
tion in the future; and 2) the decision of how much to consume today in leisure out of the total amount

reserved for today. The first of these decisions can be gauged by considering how much net worth par-

ticipants are accumulating. However, we have a range of estimates for effects on net worth, as these

estimates depend on whether one includes types of assets like real estate that only a few people pos-

sess and how one treats the uncertainty inherent in the estimates. To be conservative, we use a range

of different plausible numbers for effects on net worth in calculating the MPE. The point estimate of

the preferred estimate for changes in net worth from Bartik et al. (2024) is -$1,000 (i.e., treated partic-

ipants end up with $1,000 more debt than the control group over the course of the program), but an

estimate of -$2,000 is possible if excluding real estate, which is imprecisely estimated. We also include

an optimistic estimate of the impact of the transfers on net worth of $5,000, based on the confidence

intervals around the main result. Thus, Table 3 shows a range of possible values for each type of

income.

The $2,000 figure lines up well with participants’ reported reduction in work hours. Using data

on hours participants report working at their various jobs, we estimate a 1.3 hour decrease in work

hours per week (Table 4).21 In our enumerated and quarterly time use surveys, we similarly observe a

1.4 hours/week reduction in work hours (Figure A3), and in the mobile phone app-based time diaries

a 1.3 hours/week reduction in work hours plus other income-generating activities combined (Figure

5). A $2,000 reduction in annual income represents an approximate 5-6% reduction relative to the

control mean for total individual income, while a 1.3-1.4 hours/week reduction in work represents

an approximately 4-5% reduction relative to these variables’ control mean. We expect that effects

on work hours are slightly understated, similar to effects on income (the survey-based estimate of

$1,500 represented a 4-5% reduction, in line with the work hours estimate). Based in large part on

the administrative data, we estimate a 3.9 percentage point reduction in the extensive margin of labor

supply (Table 4), which means that the great majority of the effect on total hours worked is a result of

the effect on the extensive margin.22 This makes intuitive sense: most low-income jobs do not come

with a great amount of flexibility in work hours, so effects on labor hours are primarily driven by
21 As with all our preferred measures, this estimate is unconditional, i.e., inclusive of both employed and non-employed
participants, and could partially reflect both employed participants cutting back on work hours as well as non-employed
participants not beginning new employment.
22 Specifically, the estimate aggregating administrative and survey data is 3.9 percentage points. If the average person

worked about 40 hours a week, 50 weeks a year, that would translate into a reduction of 78 hours over the course of a year.
If hours worked per week declined by 1.5 due to the treatment, representing 5% of the control group mean, the entire labor
supply effect would be driven by the extensive margin. If hours worked per week declined by 1.8, representing 6% of the
control group mean, the extensive margin would represent 5/6ths of the labor supply effect. Some of the robustness checks
for hours/week from the survey questions and time diaries resulted in estimates of 1.6-1.7 hours/week (Tables A24-A25).

18
people quitting their jobs or not getting new ones as quickly rather than through participants reducing

work hours.

Participants also reported their total household income. The transfers decreased total household

income by about $4,100/year relative to the control group in survey data (Table 3); however, as this

measure was asked in a similar way to the question about their total individual income, we expect it

may also somewhat overstate the decline. If we adjust this measure using the ratio of the administra-

tive data-adjusted individual calculated income and the individual aggregate self-reported income,

we get about a $3,300 reduction in total household income.23 This represents a 6.9% decline relative to

the control group mean. We estimate that there is a decrease in labor supply for all adults in the house-

hold of approximately 2.2 hours/week (Table 4), with the component that is not due to participants

driven almost entirely by participants’ partners. This estimate represents a 4.6% decrease compared

to the control mean. The difference between this decrease and the percent implied by the estimates for

total household income could be partially due to a small, imprecisely-estimated decline in other trans-

fers received, but it could also reflect some slight substitution into employment not captured by the UI

data and some slight underestimation of the decline in total work hours. The estimates for the work

hours of others in the household were not pre-specified and so are subjected to a large penalty in the

false discovery rate corrections but are significant before those corrections are applied. Though the ef-

fect on partners’ hours worked may be slightly understated, it is of a roughly comparable magnitude

to the effect we observe on the participants’ own work hours, consistent with a unitary household

model, and we cannot reject that they are the same.

As described in section 3, participants recorded their time use on a mobile app. These time diaries

are asked on a frequent (bi-monthly) basis and elicited in 15-minute increments. Figure 5 shows the

estimated effects on time use as measured in the mobile app. Between reductions in “market work”

and “other income”, they show a reduction in about 1.3 hour/week of work, consistent with the em-

ployment module survey questions about hours worked at each job. Appendix G presents robustness

checks, including an alternative coding of overlapping activities and a LLM-based classification of text

responses for those who entered free text in an “other” category, as well as further results, including

for time spent with others (e.g., time spent with friends, children, or alone) and results from the enu-

merated and quarterly surveys. The extra time participants have from reduced work is used largely
23 Specifically, -$4,125*$2,028/$2,504=-$3,341.

19
for leisure,24 non-commuting transportation, and other activities (Figure 5).25 Survey data show a

decrease in work hours of 1.4 hours/week, which is very similar to the results from the mobile phone

app.

Figure A4 shows the effects on time use as measured in the mobile app separately by whether

participants had children living in the household at baseline. Those without children in the household

reduced their market work by more than those who did not, consistent with the earlier results for

income. Interestingly, we do not observe those with children spending more or less time on childcare

as a result of the transfers. More generally, the mobile app also asked participants who they were

conducting activities with. Figure A5 shows the effects on the amount of time spent with various

people. These effects are all small and insignificant after adjusting for the false discovery rate, though

the closest category to being significant before adjustment is time “with my boss”.

Both effects on income (Figure A13 and A15), effects on labor supply (Figure A14 and A15-A16),

and effects on time use (Figure A17-A20), suggest a time trend: the treatment effect grows over the

course of the study. This effect would be consistent with increasing separation from the labor market,

but it would also be consistent with individuals taking some time to switch into non-employment

activities such as pursuing education. We will return to consider this hypothesis when presenting

results for human capital investment, but the bottom line is that it does not appear that pursuing

higher education explains most of the reduction in labor supply that we observe.

One interesting note, however, is that it appears in quarterly survey data that treated participants

may start to "catch up" towards the end of the study (Figure A16): at least, we cannot reject null effects

for some quarters close to the end, while we can in the second year of the program.26 This could be

due to the end of the transfers drawing near, as we also observe participants taking a larger number

of actions to search for a job in the final year of the program, consistent with, e.g., Card, Chetty and

Weber (2007).
NY ∂p NY ∂h
We translate our results into labor supply elasticities according to ηe = ∂v p and ηi = ∂v h ,

where ηe and ηi are the extensive and intensive margins, respectively, NY is net-of-tax income, v is

virtual income (the transfers), p is participation and h is hours.27 We estimate ηe for participants as
24 Socialand solo leisure are not individually significant, but they represent the largest category if pooled.
25 The survey-based time use measures asked participants about different categories in which they could spend their time
and did not explicitly have a “social” or “solo” leisure category and did not distinguish between “market work” or “other
income-generating activities”. Still, the survey-based time questions showed similar decreases in hours/week worked. In
terms of increases in time spent on certain categories, the only category that stands out is “finances”; people in the treatment
group spent approximately 0.3 hours/week more on this activity (Figure A3).
26 However, this is less apparent in the administrative data (Figure A15).
27 There is a subtlety here: since the control group gets $50/month, the changes we observe in p or h are due to changes in

20
-0.15 and ηi for participants as -0.13.

5.2 Duration of Unemployment

Duration of non-employment and unemployment both go up, as one might expect if, with the trans-

fers, people feel less pressure to immediately take up a new job upon leaving one (Table 5). We

construct two types of variables to examine impacts in this domain: 1) considering the average and

longest duration of non-employment over the entire study, using an employment history timeline

that we made that captures when participants left or started any job, including second, third, or fourth

jobs, and 2) considering cross-sectional measures of how long participants were non-employed or un-

employed at the point in time at which they answered a survey. These two types of measures do

not have to coincide. The duration of the average spell of non-employment causally increased by 1.1

months relative to the control group mean of 7.7 months, with treated participants’ longest spell of

non-employment increasing by 0.8 months relative to the control group mean of 8.7 months. The cross-

sectional measures had somewhat smaller estimated effects on lower control group means: the effect

on the point-in-time measure of the duration of non-employment was 0.7 months, on an aggregate

control group mean of 6.1 months, and the corresponding effect on the duration of unemployment

at the time of being surveyed was 0.6 months on a control group mean of 2.9 months. The number

of months of non-employment in the last year increased by 0.3 months, but this item (which was not

pre-specified) is from a survey question in which participants were asked to identify which months

they were employed in the last year, and people can be unemployed for far longer than a year.28

5.3 Job Search and Selectivity

Receiving unconditional cash transfers made recipients more likely to search for a job and apply for a

job (Table 6). There is also some suggestive evidence that they took more actions to search for a job.29

However, they applied to on average about 1 fewer job in the last 3 months (compared to a control

mean of 5.5 applications in that time period) and perhaps interviewed for fewer as well.30 These

results suggest that while treated participants are more interested in finding work (perhaps in part due
unearned income of $950/month, and the elasticities are calculated accordingly.
28 If the effect of the transfer on duration of non-employment is particularly large among those who have not been em-

ployed for more than a year, that would be consistent with the duration of non-employment increasing by less when partici-
pants provide their employment status for the last 12 months than when they report how long they have not been employed
over an unrestricted period of time.
29 This is marginally significant before multiple hypothesis corrections. “Actions” here include things like looking at

job postings, directly contacting employers for a job, contacting job centers, contacting friends or relatives to find work,
contacting a professional network to find work, posting a resume online, and taking other actions to find work. These are
broken out in exploratory analysis in Table A40.
30 This item is marginally significant before the multiple hypothesis correction, but not after it.

21
to more of them not being employed), they are more selective about the jobs they apply for. The mixed

results mean that the “Active Search” component, and the overall index for this family of outcomes,

are both insignificant. Nonetheless, the results for individual items seem to tell a compelling story.

We do not see much in the way of differences in the types of jobs participants applied for (Table

7). In exploratory analysis of self-reported requirements for them to take a job, treated participants

are more likely to say that interesting or meaningful work is a requirement, but this result does not

survive the false discovery rate correction (Table A42).

5.4 Quality of Employment

As described earlier, there is debate in the literature as to whether quality of employment should go

up or down in response to a cash transfer. In order to address this question, we included a large

number of questions relating to quality of employment, divided into several components. Unlike the

other families, this family of outcomes focuses exclusively on those who are employed, as it makes

little sense to ask about quality of employment for those who are not employed. Note that since

employment changed in the treatment group relative to the control group, there is some selection into

our ability to observe these outcomes. However, since the extensive margin effects were fairly small

(3.9 percentage points), we believe these quality estimates are still largely directly interpretable.

First, we consider adequacy of employment. Many low-income individuals would like to work

more hours but are constrained by not being offered many hours of work by their employers. This

component measures the adequacy of their employment, including whether they are part-time in their

main job and would prefer to work full-time, whether they would prefer to work more hours in their

main job, and the number of jobs they hold. Second, we consider employment quality as measured by

benefits that are provided, including whether training is provided by the employer and related survey

questions, as well as whether the respondent must work irregular shifts. Third, we consider whether

the respondent reports working any informal job and, in exploratory analysis, whether they report

supplemental income from any gig economy jobs such as Uber, TaskRabbit, etc. Fourth, we elicit

participants’ hourly wage. Fifth, we consider stability of employment, including questions like how

many months the respondent has been employed in the past year, how many months the respondent

expects to remain in their main job, and whether their jobs are salaried or whether they are performing

contract or freelance work. Finally, we consider quality of work life, which aims to capture the day-to-

day experience at work, including questions such as whether the participant faces discrimination at

22
work, how satisfied they are with the compensation and non-wage aspects of their main job, whether

job demands interfere with family life, and the number of stressors in their work environment.

Despite the very detailed questions, the results do not support any changes in quality of em-

ployment, and for most items we can reject even small changes (see Table 8 for the component index

measures and Table 9 for the raw item measures).31 Overall, we can reject changes of more than 0.028

standard deviations in the family-level index. There were two main clusters of variables that showed

some significance. First, in the stability of employment component, a variable capturing how many

jobs the respondent held in the past 12 months (or, descriptively, in the past two years) was significant

before the false discovery rate adjustment, as was the exploratory outcome of how many months they

expect to remain in their main job.32 This could simply be a function of participants reducing their

labor supply, rather than being a measure of quality of employment. Second, under quality of work

life, the treatment effects appeared slightly negative for opportunities for promotion, treated partici-

pants were slightly more likely to say a scheduled shift was cancelled with less than 24 hours notice in

the last month and report a larger number of stressors in their work environment, treated participants

reported finding it slightly harder to take time off from work. Apart from results not being broadly

significant, point estimates were generally small across the board.33

5.5 Entrepreneurship

In contrast to the quality of employment measures, we see some shifts in entrepreneurship, at least

in terms of entrepreneurial orientation and intention (Table 10). The “entrepreneurial orientation”

component captures willingness to take financial risks and includes both a survey measure and risk

preferences from an incentive-compatible multiple price list experiment. The “entrepreneurial inten-

tion” component was based on questions such as whether or not the respondent has an idea for a

business and the respondent’s self-reported likelihood that they would start a business in the next

five years. Both components, as well as the overall index, are very significant.

While treated participants exhibited more entrepreneurial orientation and intentions, this did not

translate into significantly more entrepreneurial activity. The point estimate is positive, but small, and

it is possible that very few people have the inclination to become entrepreneurs in general. We pre-

specified that we would consider entrepreneurial orientation and intentions as potential precursors to
31 For example, people who had multiple jobs were asked exploratory questions about why they had multiple jobs.
32 This latter question was asked only of those participants whose main job is temporary, and not much weight should be
placed on it.
33 Table A43 provides further exploratory analyses within this family of outcomes, obtaining a more detailed breakdown

of which specific benefits are offered by participants’ employers.

23
entrepreneurial activity, and it remains possible that there is an effect that is too small to be detected

in our sample. Our confidence intervals include an increase as large as 2.8 percentage points. There

also appears to be a time trend, with this treatment effect growing over time and the point estimate

only marginally insignificant in year three (Figure A30).

5.6 Disability and Barriers to Employment

While one might expect disabilities to remain fairly constant throughout the course of the program,

this is not necessarily the case if people are able to leverage the transfers to improve their health or,

conversely, if people in the sample get more care and therefore get diagnosed more. It is also possible

that if individuals are out of the labor force more, they may be more likely to think of themselves as

disabled as a form of self-signalling (i.e., to mitigate any stigma associated with non-employment).

We find a significant increase in the likelihood that a respondent has a self-reported disability (an

increase of 4.0 percentage points on a base of 31 percentage points in the control group) and in the

likelihood they report a health problem or disability that limits the work they can do (an increase of

4.0 percentage points on a base of 28 percentage points in the control group) (Table 11). Participants

also report slightly worse disabilities or health problems that have persisted for slightly longer periods

of time. As a result of these consistent responses, the index for the family is significant. Somewhat

reassuringly, none of these measures was significant at endline (Figure A31), which might support

the hypothesis that participants received diagnoses early into the program and perhaps were able to

partially treat them or have them otherwise be less salient by the time of the survey.

We also asked participants about barriers to employment. One theoretical motivation for the

provision of cash transfers has been that it might help individuals overcome challenges preventing

them from working, such as a lack of transportation or childcare. However, we do not find significant

impacts on self-reported barriers to employment (Table 12).

5.7 Human Capital Formation

If treated participants are investing more in education, we can expect them to have better long-term

employment outcomes, all else equal. As Hoynes and Rothstein (2019) have highlighted, education is

a particularly important determinant of the long-run cost-effectiveness of cash transfers. To investi-

gate these outcomes, we leverage National Student Clearinghouse (NSC) data. 87.5% of participants

provided consent for their administrative records to be used; for those who did not, we supplement

the NSC data with survey data. The NSC includes outcomes on completion of post-secondary pro-

24
grams, total years of post-secondary education completed, and enrollment in post-secondary pro-

grams. The survey data gathers these same variables, but additionally gathers information on attain-

ment of high school degrees / GEDs and informal education.

By and large, we do not observe significantly improved education outcomes in our sample, though

there are some indicators of minor improvements. 92 percent of the control group had completed a

high school, GED or post-secondary program by endline, and the treatment group was 0.8 percent-

age points more likely to have completed such a program (Table 13). This result was not significant

when pooling across time periods but appeared to grow over time such that by endline the result was

only marginally insignificant (Figure A33). Exploratory analysis suggests that this response is more

driven by participants getting a GED than by their getting a post-secondary degree, since we observe

insignificant, negative coefficients when restricting attention to completion of post-secondary degrees

(Table A44). Total years of post-secondary education completed post-baseline and enrollment in a

post-secondary program showed positive but insignificant effects.

We pre-specified one heterogeneity analysis for this family of outcomes: a subgroup analysis

based on the age of the participant at baseline. Since education is an investment that only provides

returns over time, younger people tend to have higher rates of return to investment in education

and may be more likely to embark on post-secondary education as a result of the transfers. Indeed,

when we conduct this heterogeneity analysis, we observe that those participants who were under 30

at the time of the baseline survey appear more likely to be enrolled in a post-secondary program, and

this is even marginally significant prior to the false discovery rate correction. The formal education

component is also significant at p < 0.05 prior to these corrections.

While these results are fairly noisy, they would be consistent with individuals having different

uses for the transfers, with only some individuals using them to pursue education.

5.8 Household Stability

In order to interpret the effects on employment and income that we observe, we need to also consider

potential changes in household composition. The initial analyses of the NIT experiments in the 1970s

showed an effect on marriage dissolution and if the cash transfers in our study caused people to leave

the household, that could mediate the effects we observe on total household income.

However, the treatment did not cause any significant changes in relationships between partici-

pants and others in their household (Table 14). A relatively small share of people in our sample were

25
married at baseline, and other types of relationships are more common in our sample, so we asked

about relationships in several different ways. We observed no effect on whether the respondent is

divorced, whether the respondent has a spouse/partner, or whether the respondent is in a roman-

tic relationship. If anything, it is possible that participants had more changes in relationships with

romantic partners outside the household, but not with partners within the household.

In exploratory analysis, we consider self-reported reasons why relationships ended (Table A45),

but there are no clear trends. If anything, participants may be more likely to report they themselves

ended the relationship, rather than it being ended by their partner or by mutual agreement, but this

result does not survive the false discovery rate adjustment.

5.9 Labor Market Mobility

We observe large changes in where participants live over the course of the study, which can affect

labor market outcomes (e.g., Chetty and Hendren, 2018). On average, 43% of those in the control

move housing units since baseline, and 4.1 percentage points more people moved in the treatment

group, with the vast majority of these moves being to different neighborhoods, defined as a different

Census tract (Bartik et al., 2024). Fewer moves were to different labor markets, which we define as a

different commuting zone (Table 15). In particular, by the end of the study, 12% of control households

had moved labor markets since baseline, and 1.9 percentage points more people in the treatment group

moved labor markets, although this is not statistically significant. The treatment group reported more

active labor market search behaviors, however, and participants indicated significantly greater interest

in moving areas, such that the overall index for moving labor markets remained highly significant

with an effect size of 0.09.

Perhaps because most moves are within the same labor market, we do not see significant changes

in the quality of the labor markets participants reside in (Table 16). There are no significant dif-

ferences in the labor markets that treated and control participants live in by the end of the study,

with the exception that treated participants are perhaps more likely to live in areas with lower crime,

though this does not survive the false discovery rate adjustments. Further, all differences are relatively

small. However, just because participants are not necessarily moving to different labor markets does

not mean their moves are not economically meaningful. First, by revealed preference participants

moved; this suggests it was welfare-enhancing for them even if it did not improve their employment

prospects. Second, it is possible that even moves within a commuting zone could affect proximity to

26
certain labor markets.

5.10 Take-Up of Benefits

The welfare effects of cash transfer programs depend in part on whether recipients substitute away

from other social programs. While the cash transfer program we are studying was carefully designed

to minimize the risk that the cash transfer recipients would lose other benefits they might receive, if

the transfer has effects on work or education outcomes it may nonetheless indirectly affect eligibility

for programs such as the EITC, SNAP, Medicaid, or SSI. Further, a large literature suggests that the

take-up of benefits often depends on having sufficient time and energy to apply for benefits, and the

cash transfers might provide participants with more bandwidth. Conversely, if participants are doing

well, they may feel less need to apply for benefits.

Overall, we do not observe statistically significant effects on benefits take-up in survey data (Table

17). Benefits decrease by about $300 per year, but this is a very noisy estimate.34 To the extent that

benefits were reduced by the transfers, however, the elasticity estimates may slightly understate the

income effects of the transfers.

6 Discussion

6.1 Heterogeneity in Treatment Effects

We pre-specified several heterogeneity analyses that consider impacts by various attributes partici-

pants held at baseline. These subgroup analyses all are adjusted for multiple hypothesis corrections

as discussed in Section 4 and Appendix D. Due to these tests being pre-specified as exploratory, we

would not anticipate them to survive the multiple hypothesis corrections, however, it can still be in-

formative to consider the past estimates and broad trends observed across different measures.

The treatment effects on income appear stronger among those who were above the Federal Poverty

Level at baseline (Table A8). This is consistent with what we would theoretically expect with decreas-

ing returns to income. It is also in line with other lottery studies, in which higher-income individuals

are seen to adjust their labor supply by more than lower-income individuals (Golosov et al., 2023); we

confirm this pattern holds even at lower absolute income levels.

We observe interesting heterogeneity in treatment effects by education. Treated participants who

did not have a bachelor’s degree at baseline seemingly reduced their income and employment by
34 Benefits
in this section include non-income benefits such as SNAP and WIC which are excluded from the estimates of
government benefits under the “Income” family.

27
more than those who did (Tables A9 and A13). In fact, those with a bachelor’s degree had insignificant

increases in individual salaried/wage income, while reducing self-employment income and supple-

mental income from gig work. While these subgroup analyses are only exploratory, they align with

heterogeneity tests by age: negative labor supply effects are larger for participants in their 20s at base-

line (Table A14), and we also observe larger effects on formal education among those in this younger

age group (Table A17). This suggests a story in which younger participants may be more likely to use

the money to enroll in post-secondary education and do not work as much while they do so. However,

this is only suggestive, and it remains possible that we observe larger negative labor supply effects on

participants in their 20s without a college degree for other reasons. The quality of employment mea-

sures are broadly comparable between those who had a bachelor’s degree at baseline and those who

did not (Tables A20-A21). Again, caution should be taken in interpretation given the large number of

items tested.

We also pre-specified a close look at outcomes by sex, since the literature often finds large empir-

ical differences along this dimension (e.g., Eissa and Hoynes, 2004). In the survey data, differences in

impacts on income and employment by sex are ambiguous. Males in the treatment group appear to

have slightly larger reductions in income relative to the control group (Table A10), while females and

others appear to have slightly larger treatment effects on labor supply (Table A15). These differences

are not significant.

Finally, we pre-specified two heterogeneity analyses for the entrepreneurship family of outcomes,

looking at differences by age and education at baseline. These results are relatively noisy, but there

may be larger effects on entrepreneurial intention for those who do not have a bachelor’s degree at

baseline and who are in their 30s at baseline (Tables A18-A19).

One potential source of heterogeneity in treatment effects that we did not pre-specify that we

would consider is heterogeneity by state. We observed substantial differences in the income and labor

supply effects by state in both the UI and survey data (Tables 18, A11, and A16). While the sites

differ along several dimensions and so we cannot attribute the observed differences to any one factor,

several differences between the sites are worth highlighting. First, the Texas site had a lower cost of

living than the Illinois site. This means that the transfers could in principle go farther, and thus we

may anticipate them to have a larger effect on earned income and labor supply. Second, and related to

this, total household incomes at baseline were about $2,700 lower (p<0.001) in Texas, so the transfers

represented a relatively larger share of income (though only by about 4 percentage points). Third, the

28
Illinois site has a more generous set of existing social safety net programs. In Texas, given that there

are few public benefits, the transfers may be filling a more substantial gap in basic needs, potentially

triggering larger changes in recipients’ financial and work decisions. Fourth, employment growth was

much higher in the Texas site than in the Illinois site over the course of this study. In a high growth

environment, participants may be more likely to expect that if they left a job they would be able to

find one again quickly if needed. Finally, Texas has a lower minimum wage than Illinois. This means

that there are some jobs that pay very poorly and participants may not be interested in them if they

are in the treatment group. However, we did not see any significant changes in either participants’

reservation wage (Table 7), employed participants’ wage rate (Table 8), or the weight participants

placed on potential income in job search (Table A42), and in fact baseline wages were insignificantly

higher in our sample in Texas, so this explanation may be less likely.

Another potential source of heterogeneity in treatment effects that we did not pre-specify is whether

or not the participant had children in the household at baseline. We observe substantially larger neg-

ative effects on income for those who did not have children at baseline (Table A12). This could be

consistent with households with children having a greater need for income. Alternatively, it is pos-

sible that this relates to the earlier observation that effects tend to be larger for younger participants.

Ultimately, it should be remembered that these heterogeneity analyses are not causal and the vari-

ables considered could merely be correlated with other variables that mediate income effects, without

mediating them directly themselves.

6.2 Comparison of UI and Survey Data

In general, results from the survey data line up very well with results in the UI data. Table 18 shows

a full breakdown across the different data sources. In both sources of data, individual income and

labor supply is lower in the treatment group, the difference between the treatment and control group

increases over time, and the difference between the treatment and control group is larger in Texas. The

administrative and survey data even show similar patterns in that both data sources show a growing

gap between the treatment and control group over time (Figures A15-A16). However, the magnitude

of the treatment effect is somewhat larger in the administrative records than in the survey data.

The difference between the results in the administrative and survey data could in part be due to

treated participants switching out of jobs that are captured in UI records into less formal work. How-

ever, as we saw when considering information on the types of jobs people hold and survey questions

29
about whether they do “gig” or “temp” work, there appears to be limited substitution into informal

work or work that may not be well-captured in UI records, so this does not seem to be able to ex-

plain the difference. It is possible that the slightly different samples contribute to the gap (as a small

share of participants in Texas cannot be matched to administrative data), though those who were un-

matched in Illinois did not appear to be very different from those who were matched in survey data

(Table 18).35 Finally, the survey data may somewhat understate the declines in employment, especially

given that we observe that treated participants appear to value work more and express more negative

perceptions towards those who do not work (Broockman et al., 2024). Despite these discrepancies, the

administrative and survey data tell remarkably similar stories about the effects and the trajectory of

those effects over time, increasing our confidence in these results.

6.3 Robustness Checks

While differential attrition was very low over the study period, we nonetheless performed a number

of pre-specified robustness checks. In particular, we conducted a difference-in-differences analysis;

restricted attention to administrative data from which individuals cannot attrit or data collected at

midline or endline in the enumerated surveys, to which we expected high response rates; and es-

timated a set of results with Lee bounds. In addition, given that some variables are more likely to

contain outliers, we conducted median regression for these outcomes. We also provide a set of regres-

sions which do not include any covariates.

Overall, the results of these robustness checks appear consistent with the estimates from the main

analyses. The family-level indices which were significant in the main regressions are significant in

all the robustness checks, and no family-level index which was insignificant in the main regression

is significant in any robustness check. This is also true for all components except a few single-item

components. We show the income and labor supply estimates by item, as we do for time use, and at

the item level there is a bit more variation, but results are still broadly in line with the main estimates.

The regression on whether the respondent is employed is significant in the robustness check without

covariates and when restricting attention to data from the enumerated midline and endline surveys,

but it is not in the difference-in-differences or bounding analysis using survey data. The magnitudes

of the point estimates remain comparable.

35 We may more generally anticipate some minor differences in results given that we are unable to include the full set of

Lasso-selected controls in the Texas administrative data environment, though given the randomized nature of the treatment
we do not expect this to drive results. As described in the methods section, we included a subset of 56 baseline covari-
ates, focusing on variables capturing demographic information, employment, income, household composition, relationship
status, and county type.

30
6.4 Modeling Labor Supply Decisions

The literature on cash transfers often focuses on estimating MPEs. To estimate the MPE, one needs to

know how much individuals decide to set aside for future time periods versus how much to spend

in increased leisure or consumption this period. Out of the amount individuals decide to spend this

period, one can then observe how much they spend on leisure.

Most papers that estimate MPEs do not have data on consumption or net assets and so have to

make assumptions about how much people set aside for future time periods, i.e., via assumptions

about the discount rate. Assumptions about the intertemporal choices people make are particularly

important in studies of lottery winners, since lottery winners often receive a large lump sum payment

and it is not necessarily obvious how much people will allocate to spending each period. If people

experience diminishing marginal utility to cash, they may decide to save most of it, so a large lump

sum transfer gets reduced to a small annual spend. For example, in Golosov et al. (2023), the median

winner receives a lump sum payment of $44,000, not too far off from the $34,200 that participants in

our study receive over the duration of the program.36 The reasonable assumptions in their paper lead

to a share to be spent in the first year of 4.2%, or $1,855.37

In our setting, we observe very limited asset accumulation (on the order of $0 to $2000) and in-

creases in debt (of around $1000 to $2000) in the treatment group relative to the control over the course

of the study (Bartik et al., 2024). Participants appear to spend approximately the full amount of the

transfers each month, on average.38 Regardless of the precise extent of savings and debt, this marks an

interesting divergence from the literature on lottery winners. Since our participants are younger than

those in lottery studies, we would expect them to live for more periods of time and therefore to save

more, all else equal. However, if we were to follow the same model as in Golosov et al. (2023) and set

the same values for the discount rate and real interest rate, we would predict participants allocate less

than $100 per month to spending, whereas we observe much more than that in both our consumption

and labor data.

This lower rate of savings in our sample compared to the literature is not unreasonable and there
36 ($1,000-$50)*36,
accounting for the control group payments.
37 Based
off a k-year old household with remaining lifetime of T − k years and discount rate d allocating share λ of a
 t    T − k +1  −1
+r
lump-sum transfer to the first t̄ years: λ(r, d) = Σt̄t=1 11+ d
1 − 1
.
d 1+ d 1+ d
38 There is a portion of the transfers that is not captured in our labor, consumption, and assets data, and it is possible
that some of this unobserved amount flows to assets. However, the literature is clear on the tendency for consumption
to be particularly underreported, so it is more likely that the bulk of this is underreported consumption (Meyer, Mok and
Sullivan, 2015).

31
are many potential explanations for it. First, our sample is lower-income on average, which implies

that more of the transfers could go to meeting immediate basic needs. The program we study also

had a monthly disbursement schedule, and perhaps this might encourage participants to think of

the transfers as money to spend that month, as a kind of mental accounting. At the same time, it is

possible that in taking on new financial activities, participants also faced new unexpected shocks. For

example, a participant that purchases a car may find that they have to do unexpected maintenance

on it, or fix a broken window. Alternatively, perhaps participants in our study have different time

preferences than those in the lottery studies. If participants fear that others will ask them for a portion

of the transfers, they may have an extra incentive to spend it quickly.

In order to more closely compare our paper to the literature, we use a standard model to estimate

the MPE. We consider the following objective function:

T
max ∑ δt−1 [u(ct − γc ) − Bv(lt )]
ct ,lt ,At+1 t=1

s.t.

At+1 = (1 + r )[ At + wt lt − ct + It + Tt ]

ct ≥ 0

A T +1 = 0

where c is consumption, γ is a subsistence level of consumption, B is a scaling factor, v is the marginal

disutility of labor or the marginal utility of leisure, l is labor supply, δ is the discount rate, A are

assets, r is the real interest rate, w are real wages, I is non-labor income including passive income,

government transfers, and gifts from family and friends, and T are the unconditional cash transfers.

Consumption is assumed to be positive, and assets in the final period are assumed to be spent down,

as is typical in the literature.

We assume individuals have a CRRA utility function for consumption:

c1t −σ
u(ct ) =
1−σ

and the marginal disutility of labor is governed by:

32
 η 1−η η
v ( lt ) = l
1−η t

where η is the Frisch elasticity of labor supply.

We estimate this model using moments from data on employment, consumption, and net assets.39

We focus on estimating the unitary household model, as it appears to have the most support in our

estimates. We then use the fitted parameter values to estimate the MPE.40

Table 19 compares the MPE as estimated by the model with the MPE estimates calculated from

our reduced form estimates under different assumptions about changes in net assets over the study

period. The aggregate estimate for net worth excluding real estate assets and real estate debts in

Bartik et al. (2024) is about -$2000; we also provide estimates that assume no net change in assets and

a relatively high value of accumulation of $5000 over the study period, representing approximately

the upper end of the confidence interval and a value that might be a reasonable upper bound if there

is under-reporting of net asset accumulation. Modest changes in net assets do not affect the MPE

estimates much. For example, for a decline in total household income of $3,341 per year, the pooled

estimates range from -0.28 - -0.34.

The MPE estimates from the fitted model are roughly in line with the calculated MPE estimates

but slightly lower, perhaps in part due to the fact that income decreases between midline and endline,

and the model weights all time periods equally while our pooled results place more weight on results

towards the end of the study. On the whole, these estimates are larger than in some of the lottery

studies (Imbens, Rubin and Sacerdote, 2001; Cesarini et al., 2017) but smaller than in Golosov et al.

(2023). However, the model overly smooths across time periods, indicating that one would need some

kind of friction in order to get the time trends we observe. There could be several sources of frictions.

For example, our results would be consistent with participants finding it harder than expected to find

jobs over time.41 Alternatively, individuals may take different amounts of time to decide what to
39 Specifically, we use the quarterly employment estimates from UI records, the quarterly consumption data from surveys,

and the data on net assets from the enumerated midline and endline. We believe that assets and debt are fairly reliably
estimated in our data (debt is from Experian for those who consent to share administrative records), while consumption
is somewhat underreported, similar to how studies leveraging the Consumer Expenditure Survey show it underreports
consumption. In order to account for this, we generate a scaling factor using the employment and net asset estimates. We
leverage the fact that by midline and endline, the total transfers up to that point minus the observed decreases in income
minus the net assets accumulated should equal consumption up to that point).
40 We assume r=0.02 yearly. Figure A8 shows the trajectory of labor earnings, consumption and net assets over time per

the best-fitting parameter values: σ=1, η=0.3, δ=0.96.

41 We observe some suggestive evidence that participants who leave a job are overly optimistic about their chances of

finding an acceptable job quickly. When participants leave a job and are asked how likely it is they think they will find a
new acceptable job within 6 months, approximately 59% of people consider this “Very likely”, however, only 34% actually

33
do with the transfers. If, over time, more treated participants decide to pursue education or start an

informal business, this could also contribute to the patterns we observe. Since there are many potential

explanations that fit our data, we do not take a stance on which potential explanation is correct but

note that estimates based on different snapshots of time could come to quite different conclusions

about the MPE and this heterogeneity can be obscured by the standard model.

6.5 Comparison to Forecasts from NBER Affiliates

We can also compare our estimates, more generally, to the current received wisdom about cash trans-

fers by surveying experts in economics as to what they think we will find. As described in DellaVigna,

Pope and Vivalt (2019), expert forecasts can be a valuable tool for judging the novelty of research find-

ings. We elicited forecasts from a subset of researchers affiliated with the National Bureau of Economic

Research (NBER). These researchers were affiliated with at least one of several NBER Programs.42 The

survey was designed such that each person was encouraged to answer a small set of questions relating

to their main field of expertise, but they were allowed to take other survey modules if they wished.

In total, we sent 795 researchers an email with an individualized link to take the forecasting survey,

and 136 (17.1%) completed it, of whom 43 completed the employment module, primarily affiliates of

Labor Studies, Public Economics, and Economics of Health. While this response rate is relatively low,

it is commensurate with what one might normally expect for researchers at this level of seniority.43

Researchers were not compensated, and the survey was unincentivized.44

We supplemented the sample by eliciting forecasts from users of the Social Science Prediction Plat-

form (SSPP), including its Superforecaster Panel.45 The Superforecaster Panel is a panel of researchers

interested in forecasting who take nearly every survey posted on the platform. Panellists are paid a

flat fee every quarter for their services and receive other benefits. For the version of the survey posted

on the platform, participants were offered accuracy-based incentives.

Table 20 presents results. Interestingly, NBER Labor Studies affiliates and SSPP users perform

fairly comparably, with the exception of the question about individual salaried income, where SSPP
do take up a new job in this timeframe. This optimism is not significantly different between the treatment and control group,
but it is possible that if the treatment group feels more able to act on this optimism due to having more of a safety net, it
could lead to the kind of path we observe. It is also possible that the labor supply effects in the earliest time periods could
have been affected by the tight labor market or other benefits available during the COVID-19 pandemic.
42 Children, Development Economics, Development of the American Economy, Health Care and Health Economics (now

merged into Economics of Health), Labor Studies, Political Economy, and Public Economics.
43 Ferguson et al. (2023) suggest a 10-24% rate is typical.
44 Given the researchers’ level of seniority, this is appropriate as those taking the survey would tend to be taking it out

of personal interest and not be swayed by small cash incentives. See Ferguson et al. (2023), who randomize $75 and $100
incentives to faculty.
45 https://www.socialscienceprediction.org/.

34
users predicted substantially more positive effects. NBER program affiliates and SSPP users were

asked overlapping but non-identical sets of questions, as we wanted to maximize the attention paid by

NBER domain experts to particular topics, but for the Superforecaster Panel we wanted respondents

to answer as many questions - independent of field - as possible.

We observe that the NBER affiliates had fairly accurate assessments of the effects of the transfers

on the intensive and extensive margin of labor supply and on individual salaried income as measured

in the administrative data, as judged by their mean and median responses. These forecasts somewhat

understated the observed effects, but are within their confidence intervals. However, there was great

heterogeneity in beliefs. Figure A9 shows the distribution of responses. While the group as a whole

may be reasonably accurate in their responses about labor supply, any one given individual is likely

to be off by a large margin.

NBER affiliates also slightly underestimated the effects on the average duration of unemployment,

though the median and mean lie within the confidence interval of the observed treatment effect. They

predicted increases in the hourly wage, whereas the estimated effects on hourly wage were -$0.23

at endline. The mean and median NBER affiliate’s forecast are outside of the confidence interval

associated with this point estimate, as is the mean but not the median forecast from NBER affiliates

in Labor Studies. NBER affiliates also believed that participants would search for work less, whereas

we observed participants searching for work 7.1 percentage points more towards the end of the study,

and all mean and median forecasts are far outside the confidence intervals associated with this result.

It is possible that forecasters were not thinking about how, if participants reduce labor supply as a

result of the transfers, they may also seek employment more, particularly as the end of the transfers

approaches. It is also true that the point estimate on the number of jobs applied to is negative, i.e.,

they were searching less intensively. Finally, NBER affiliates expected enrollment in a formal post-

secondary program to increase slightly (2.5-4.4 percentage points), while our point estimate for the

final year of the program was 0.0. Again, the confidence interval on the point estimate excludes the

mean and median of any subgroup’s forecasts.46

Overall, this analysis suggests that economists have more of a sense for effects on labor supply

than they do for other important employment outcomes such as hourly wages, human capital invest-

ments, and job search, underscoring the benefits of the diverse array of outcome variables considered
46 Six NBER affiliates also answered questions about time use, however, this is too small a sample to draw inferences from.

Many SSPP forecasters answered these questions, however, and they tended to overestimate the amount of time spent on
social and solitary leisure.

35
in this study.

7 Conclusion
After decades of shifting welfare assistance from direct cash payments to in-kind benefits, cash trans-

fers have increasingly been proposed as a way to alleviate poverty and provide beneficiaries the flex-

ibility to purchase what they need. At the same time, some policymakers have raised concerns that

such transfers may lead beneficiaries to pull back from the labor market, which may in turn increase

the need for and reliance on future transfers and dampen beneficiaries long-term job prospects, while

raising the fiscal cost of the transfers themselves. Alternatively, if cash transfers help beneficiaries

search for higher quality or better fitting jobs, start new businesses, or invest in their future earnings

via human capital, a reduction in labor supply may ultimately be productive.

Our results provide support for both sides of this debate. On the one hand, we do find that the

transfer we study generated significant reductions in individual and household market earnings. The

reductions in individual labor supply we observe are smaller than what has been documented in some

settings (e.g., Golosov et al., 2023), but larger than what has been observed in others (e.g., Imbens, Ru-

bin and Sacerdote, 2001; Cesarini et al., 2017). The spillovers onto other household members–who also

reduced their labor supply in response to the transfer–implies the total amount of work withdrawn

from the market is fairly substantial. Further, we do not find evidence of the type of job quality or

human capital improvements that advocates have hoped might accompany the provision of greater

resources, and our confidence intervals allow us to rule out even very small effects of the transfer on

these outcomes. On the other hand, we find that participants showed more interest in entrepreneurial

activities and willingness to take risks due to the transfers, which could improve future earnings and

lead to additional economic benefits over time. And, exploratory analysis of subgroups suggests that

not all responses to the transfer were identical: older participants experienced very little change in

either labor supply or human capital, while younger participants reduced time spent working but

appeared to pursue more education. Finally, the fact that some of the transfer was used to reduce

work shows the high value that participants place on leisure at the margin or, equivalently, the high

disutility they have for the kind of work that is available to them.

While the duration, magnitude of the transfers, and comprehensive nature of our data collection

is unprecedented for a study of this size, future work would improve our understanding of the long-

term impacts of income on employment. In particular, follow-ups could consider to what extent labor

36
market effects persist after the end of the transfer period and shed light on effects on participants’

children, which may be particularly important in policy decisions. Additional work would be needed

to understand the potential general equilibrium effects that might arise should such a program be

scaled up.

Our analysis demonstrates that even a fully unconditional cash transfer results in moderate la-

bor supply reductions for recipients. Virtually all existing large-scale cash transfer programs in the

U.S. are means-tested, which provides additional disincentives to work. Rather than being driven by

such program features, participants in our study reduced their labor supply because they placed a

high value, at the margin, on additional leisure. While decreased labor market participation is gen-

erally characterized negatively, policymakers should take into account the fact that recipients have

demonstrated–by their own choices–that time away from work is something they prize highly.

37
 Table 1: Study Sample Characteristics Compared to Eligible Population
Eligible Population Comparison (ACS) Study Sample
Full US Population Study Counties Eligible Screener Respondents Enrolled Active
Survey Group
Unweighted Reweighted to Match Reweighted to Match Unweighted Reweighted to Match Unweighted
Enrolled Sample FPL and Enrolled Sample FPL Enrolled Sample FPL
County Type Distribution County Type Distribution County Type Distribution
(1) (2) (3) (4) (5) (6)
Panel A. Key active group stratification variables
Income <100% of FPL 0.25 0.34 0.34 0.30 0.34 0.33
Income 100-200% of FPL 0.36 0.41 0.41 0.33 0.41 0.40
Income 200% + of FPL 0.38 0.24 0.24 0.37 0.24 0.24

Rural County 0.26 0.13 0.13 0.13 0.13 0.13

Suburban County 0.32 0.18 0.18 0.22 0.18 0.18
Medium-Sized Urban County 0.16 0.16 0.16 0.15 0.16 0.16
Large Urban County 0.24 0.53 0.53 0.51 0.53 0.53

38
Panel B. Demographic Characteristics
Any Children 0.59 0.59 0.62 0.57 0.59 0.57
HH Size 3.36 3.25 3.34 3.14 3.20 2.98
Age <30 0.52 0.54 0.53 0.54 0.54 0.54
White (non-hispanic) 0.59 0.46 0.41 0.48 0.46 0.47
Black (non-hispanic) 0.17 0.25 0.29 0.25 0.26 0.30
Hispanic 0.17 0.22 0.25 0.22 0.22 0.22
Female 0.57 0.59 0.61 0.68 0.69 0.67
HH Income 36,199 30,521 31,204 32,327 29,245 29,942
College Degree or more 0.17 0.16 0.16 0.28 0.25 0.20
Renter 0.56 0.68 0.66 0.82 0.84 0.79

N 919395 904792 35086 14573 14573 3000

Notes: This table compares characteristics of our sample with characteristics of the full US population and the population of the study
counties, reweighted to match the enrolled sample’s FPL and county type distribution. Our sample is very similar along most dimensions,
though our participants are a little more likely to be renters. It should be noted that columns (4) and (5) use data from the online screener
while column (6) uses baseline survey data, so the numbers may differ slightly.
 Table 2: Descriptive Statistics: Baseline Covariate Balance

Treatment Control p-value

Demographic
Age 30.169 30.035 0.542
Male 0.328 0.319 0.627
Female 0.669 0.678 0.628
Non-binary/other 0.003 0.003 0.999
Non-Hispanic Black 0.295 0.305 0.554
Non-Hispanic Asian 0.036 0.038 0.790
Non-Hispanic White 0.473 0.463 0.597
Non-Hispanic Native American 0.020 0.025 0.428
Hispanic 0.220 0.214 0.694
Household Size 2.943 2.996 0.435
Number of Other Adults in the Household 0.684 0.716 0.347
Any Children 0.568 0.571 0.897
Has Disability 0.338 0.311 0.130
Bachelor’s Degree 0.202 0.205 0.866
Employed 0.578 0.586 0.675
Income and Employment
Total Household Income (1000s) 29.991 29.917 0.922
Total Individual Income (1000s) 21.355 21.217 0.861
Work Hours/Week 21.207 21.780 0.487
Has a Second Job 0.168 0.173 0.712
Months Employed in the Past Year 7.215 7.268 0.778
Number of Jobs in the Past 1 Year 1.403 1.439 0.457
Number of Jobs in the Past 3 Years 2.684 2.620 0.485
Searching for Work 0.495 0.510 0.429
Started or Helped to Start a Business 0.315 0.296 0.268
Housing
Lived Temporarily with Family or Friends 0.262 0.281 0.286
Stayed in Non-Permanent Housing 0.086 0.084 0.811
Housing Search Actions in Last 3 Months 0.255 0.242 0.447
Number of Times Moved in the Past 5 Years 1.328 1.358 0.468
Relationships
Is in a Romantic Relationship 0.627 0.621 0.749
Lives with a Romantic Partner 0.441 0.431 0.586
Married 0.221 0.222 0.951
Divorced 0.077 0.081 0.706

Notes: This table shows the baseline levels of a number of different variables relating to the em-
ployment outcomes considered in this paper. The treatment and control groups look comparable
for all items.

39
Table 3: Impact of Guaranteed Income on Annual Earned and Other Unearned Income (in
$1,000s)

Control Mean Treatment Effect MPE Elasticity N

Total household income (self-reported) 48.2 (33.9) -4.1***††† -0.34 - 0.42 -0.30 2898
(1.0)
[0.00]
Total individual income 36.6 (27.0) -1.5* -0.12 - -0.15 -0.15 2881
(0.9)
[0.13]
Total individual income (self-reported) 33.5 (25.1) -2.5**† -0.21 - -0.26 -0.26 2855
(1.0)
[0.09]
Individual salaried/wage income 24.3 (25.0) -1.7** -0.14 - -0.17 -0.27 2614
(pooled UI and survey data) (0.9)
[0.13]
Individual salaried/wage income 26.0 (26.2) -1.1 -0.10 - -0.12 -0.18 2920
(0.8)
[0.30]
Self-employment income 5.9 (13.7) -0.1 -0.01 - -0.01 -0.08 2902
(0.5)
[0.42]
Income from gig work 0.4 (1.3) -0.1 0 - -0.01 -1.65 2925
(0.0)
[0.19]
Passive income 0.0 (0.2) 0.0 0.00 - 0.00 0.70 2923
(0.0)
[0.19]
Other income 4.7 (6.1) -0.1 -0.01 - -0.01 -0.04 2935
(0.2)
[0.38]
Government transfers 3.6 (4.9) -0.2 -0.01 - -0.02 -0.10 2962
(0.1)
[0.43]

Notes: This table shows the impacts of an unconditional cash transfer on other income outcomes
for participants and their households, excluding the transfers, in $1,000s. As an exception, the
income family of outcomes was pre-specified to not have its components aggregated in the same
way as other families. Standard errors are provided in parentheses, and the FDR-adjusted q-value
in square brackets below it. Items that are italicized are secondary outcomes for the sake of the FDR
corrections, and unitalicized rows here refer to single-item components. The MPE range associated
with each estimate is calculated assuming net asset accumulation of -$2000 to $5000 over the course
of the study. The preferred MPE estimate for the total household income adjusts for the fact it may
be misreported by adjusting it according to the ratio of the total calculated individual income and
the aggregate self-reported individual income measure, as described in the text. Estimates are
provided in terms of raw units ($). All measures are survey-based except for the pooled UI and
survey data estimate. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

40
 Table 4: Impact of Guaranteed Income on Employment

Control Mean Treatment Effect Elasticity N

Labor Supply Elasticity Index -0.06**†† .
(0.03)
[0.04]
Labor Supply Elasticity Component -0.06**†† .
(0.03)
[0.04]
Whether the respondent is employed 0.72 (0.39) -0.04***†† -0.15 2638
(pooled UI and survey data) (0.01)
[0.01]
Whether the respondent is employed 0.74 (0.39) -0.02* -0.07 2962
(0.01)
[0.67]
Hours worked per week 30.28 (19.83) -1.28**†† -0.13 2940
(0.64)
[0.03]
Number of other household members which 0.47 (0.61) -0.02 -0.11 2943
are employed (0.02)
[1.00]
Total number of hours participant and 40.69 (24.84) -2.16*** -0.16 2945
spouse/partner works per week (0.78)
[0.31]
Total number of hours all household members 48.22 (29.64) -2.21** -0.13 2945
(including the participant) work per week (0.92)
[0.37]
Total number of hours participant’s parents in 3.22 (12.07) -0.13 -0.09 2941
household work per week (0.35)
[1.00]
Total number of hours participant’s adult 1.23 (6.75) 0.30 1.68 2945
children in household work per week (0.29)
[0.99]

Notes: This table shows the impacts of an unconditional cash transfer on the labor supply of par-
ticipants. The top-level index, "Labor Supply", in bold, declines by about 0.06 standard deviations.
There is a single component, with two primary items under it. The q-values on the component
and the top-level family index measures are different even as the point estimate is the same as they
adjust for different sets of estimates in the FDR corrections (see Appendix D for details). Items
that are italicized were exploratory items for the sake of the FDR corrections (post-pre-analysis
plan, i.e., the lowest FDR tier). Standard errors are provided in parentheses, and the FDR-adjusted
q-value in square brackets below it. Except for the family- and component-level index values, es-
timates are provided in terms of raw units. All measures are survey-based except for the pooled
UI and survey data estimate. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value
thresholds.

41
 Table 5: Impact of Guaranteed Income on Duration of Unemployment

Control Mean Treatment Effect N

Duration of Unemployment Index -0.10***††
(0.04)
[0.026]
Single-item Component: Average length of continuous 7.68 (11.16) 1.09**†† 2930
spells of non-employment in months, over the study (0.46)
duration [0.018]
Length of longest continuous spell of non-employment in months, 8.73 (11.73) 0.85***†† 2930
over the study duration (0.32)
[0.031]
Duration of unemployment in months at time of survey 2.87 (8.05) 0.60**†† 2940
(0.29)
[0.036]
Duration of non-employment in months at time of survey 6.07 (12.21) 0.72**†† 2938
(0.36)
[0.036]
Number of months of non-employment in the last year 3.38 (4.41) 0.26**† 2934
(0.13)
[0.071]

Notes: This table shows the impacts of an unconditional cash transfer on the duration of non-
employment and unemployment of participants. The top-level index, "Duration of Unemploy-
ment", in bold, declines by about 0.10 standard deviations. As there is a single primary item in the
component (average length of continuous spells of non-employment), it is "promoted" to act as a
component as per appendix D, but it is still presented in raw units. Several items that are italicized
represent secondary outcomes for the sake of the FDR corrections. Standard errors are provided in
parentheses, and the FDR-adjusted q-value in square brackets below it. Except for the family-level
index value, estimates are provided in terms of raw units. * p < 0.1, ** p < 0.05, *** p < 0.01; †
refers to comparable q-value thresholds.

42
 Table 6: Impact of Guaranteed Income on Employment Preferences and Job Search

Control Mean Treatment Effect N

Employment Preferences and Job Search Index 0.02
(0.02)
[0.377]
Active Search Component 0.03
(0.02)
[0.735]
Whether participant searched for a job 0.60 (0.38) 0.06***††† 2943
(0.01)
[0.001]
Whether the respondent is seeking a new, additional, or any job 0.39 (0.41) 0.03* 2939
(0.01)
[0.235]
Number of different actions taken to search for a job 1.69 (1.72) 0.09* 2942
(0.05)
[0.235]
Whether the participant applied for a job 0.49 (0.39) 0.04***†† 2942
(0.01)
[0.015]
Number of job applications sent 5.45 (11.83) -0.89**† 2942
(0.35)
[0.079]
Whether the participant interviewed for a job 0.36 (0.36) 0.01 2942
(0.01)
[0.777]
Number of jobs interviewed for 0.73 (1.72) -0.09* 2942
(0.05)
[0.235]
Preferences for Employment Component 0.01
(0.02)
[0.735]
How many work hours the respondent wants (less, same, more) 2.18 (0.52) 0.02 2927
(0.02)
[0.282]
Whether a respondent is employed or, if unemployed, would prefer to be 0.90 (0.26) -0.01 2942
working (0.01)
[0.463]

Notes: This table shows the impacts of an unconditional cash transfer on employment preferences
and job search. The top-level index increases insignificantly by about 0.02 standard deviations.
There are two components in this family of outcomes: Active Search and Preferences for Employ-
ment, both presented in standard deviations in order to aggregate primary items beneath them.
Several items that are italicized represent secondary outcomes for the sake of the FDR corrections.
Standard errors are provided in parentheses, and the FDR-adjusted q-value in square brackets be-
low it. Except for the family- and component-level index values, estimates are provided in terms
of raw units. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

43
 Table 7: Impact of Guaranteed Income on Selectivity of Job Search

Control Mean Treatment Effect N

Selectivity of Job Search Index -0.00
(0.02)
[0.752]
Perceived likelihood of finding an acceptable job in 6 months (1 - 4 3.38 (0.81) -0.15*** 889
scale) (0.05)
[0.146]
Participant’s reservation wage, reported in minimum hourly 18.30 (8.73) -0.17 1068
remuneration (0.49)
[1.000]
Selectivity Component -0.00
(0.02)
[1.000]
Natural log of average income of jobs which the respondent 10.67 (0.34) -0.00 2071
applied to (0.01)
[1.000]
Whether the respondent is willing to take any job offered 0.16 (0.36) -0.01 1050
(0.02)
[1.000]
Number of sacrifices participants would be willing to make 2.18 (1.06) 0.05 2496
to secure a job (0.04)
[1.000]
If searching for a job, how long respondent is willing to 7.15 (8.64) 0.08 2476
search in months (0.34)
[1.000]

Notes: This table shows the impacts of an unconditional cash transfer on selectivity of job search.
The top-level index decreases insignificantly by less than 0.01 standard deviations. There is one
component with primary items in it (Selectivity) and two components pre-specified as containing
only secondary items regarding participants’ expectations and their reservation wage (which do
not contribute to the index). Standard errors are provided in parentheses, and the FDR-adjusted
q-value in square brackets below it. Except for the family- and component-level index values,
estimates are provided in terms of raw units. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to
comparable q-value thresholds.

44
Table 8: Impact of Guaranteed Income on Quality of Employment: Summary of Top-Level
Components

Control Mean Treatment Effect N

Quality of Employment Index -0.01
(0.01)
[0.449]
Adequacy of Employment Component 0.01
(0.03)
[1.000]
Employment Quality Component -0.01
(0.02)
[1.000]
Single-item Component: Whether the respondent reports 0.24 (0.37) -0.00 2404
working any informal job (0.01)
[1.000]
Single-item Component: Average hourly income from all 17.26 (9.72) -0.18 2408
jobs, weighted by hours worked at each job (0.37)
[1.000]
Stability of Employment Component -0.02
(0.02)
[1.000]
Quality of Work Life Component -0.02
(0.02)
[1.000]

Notes: This table shows the impacts of an unconditional cash transfer on quality of employment.
The top-level index decreases insignificantly by about 0.01 standard deviations. This table shows
summary measures of each component in the family; two are single-primary-item components
and are reported in raw units, while the others are reported in terms of standard deviations as
they aggregate a number of primary items. Standard errors are provided in parentheses, and the
FDR-adjusted q-value in square brackets below it. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to
comparable q-value thresholds.

45
Table 9: Impact of Guaranteed Income on Quality of Employment: Item-Level Analyses
Control Mean Treatment Effect N
Adequacy of Employment
The respondent is employed part-time in their main job and would prefer 0.24 (0.39) -0.00 2336
to work full-time (0.02)
[1.000]
The respondent would prefer to work more hours in their current main 0.21 (0.36) 0.01 2409
job (0.02)
[1.000]
The number of jobs held by the respondent apart from their main job 0.38 (0.70) -0.03 2407
(0.03)
[1.000]
Employment Quality
Whether training is offered by the respondent’s main employer 0.53 (0.45) 0.01 2399
(0.02)
[1.000]
Whether training is offered during work hours by the respondent’s main 0.49 (0.45) 0.01 2398
employer (0.02)
[1.000]
Whether formal training is offered by the respondent’s main employer 0.13 (0.29) -0.00 2397
(0.01)
[1.000]
Number of non-wage benefits at respondent’s job(s), weighted by hours 3.62 (2.90) -0.12 2408
worked at each job (0.11)
[1.000]
Whether the respondent must work an irregular shift at each job, 0.19 (0.34) 0.01 2405
weighted by hours worked at each job (0.01)
[1.000]
Number of non-wage benefits at respondent’s job(s), alternate measure 4.53 (2.97) -0.17 2342
(0.11)
[1.000]
Informality of Employment
Whether the respondent reports any gig economy jobs such as Uber, TaskRabbit, 0.09 (0.25) -0.00 2403
or online surveys (0.01)
[1.000]
Stability of Employment
How many months the respondent has been employed in the past year 10.69 (2.66) -0.03 2396
(0.10)
[1.000]
How long the respondent has spent at their current main job and other 24.88 (34.85) 1.70 2403
jobs (months), weighted by hours worked at each job (1.15)
[1.000]
How many jobs the respondent has held in the past 12 months 1.76 (1.60) -0.12** 2390
(0.05)
[1.000]
How many jobs the respondent has held in the past two years 2.33 (3.67) -0.17* 2389
(0.09)
[1.000]
Whether the respondent’s main job is a temp job 0.10 (0.26) 0.01 2404
(0.01)
[1.000]
Whether each of the respondent’s jobs is salaried, weighted by hours 0.23 (0.39) -0.01 2403
worked at each job (0.01)
[1.000]
Whether the respondent is performing contract or freelance work at each 0.25 (0.38) 0.00 2402
job, weighted by hours worked at each job (0.01)
[1.000]
How many months the respondent expects to remain in their main job 8.97 (6.56) -1.30* 341
(0.70)
[1.000]

46
Quality of Work Life
Advance notice of schedule provided at the respondent’s main job (1-4 2.52 (1.24) -0.03 2361
scale) (0.05)
[1.000]
The work activities are not boring at the respondent’s main job (1-5 scale) 3.11 (1.05) -0.01 2249
(0.04)
[1.000]
Satisfaction with compensation at the respondent’s main job (1-5 scale) 3.51 (1.06) -0.02 2405
(0.04)
[1.000]
Whether the respondent faces age discrimination at work 0.06 (0.21) 0.00 2249
(0.01)
[1.000]
Whether the respondent faces sex discrimination at work 0.08 (0.25) 0.00 2248
(0.01)
[1.000]
Whether the respondent faces racial or ethnic discrimination at work 0.08 (0.25) 0.01 2248
(0.01)
[1.000]
Whether the respondent experienced fair treatment by their supervisor 4.05 (0.91) 0.04 2252
(1-5 scale) (0.04)
[1.000]
Whether job demands do not interfere with family life (1-4 scale) 2.91 (0.87) 0.02 2405
(0.03)
[1.000]
Whether the job is a good fit with the respondent’s experience and skills 4.19 (0.92) -0.05 2403
(1-5 scale) (0.04)
[1.000]
Flexibility of schedule at the respondent’s main job (1-4 scale) 1.91 (0.91) 0.01 2346
(0.04)
[1.000]
Overall satisfaction with the respondent’s main job (1-5 scale) 3.96 (0.96) 0.03 2404
(0.04)
[1.000]
Whether the respondent has decision-making input in their job (1-4 scale) 2.67 (0.98) -0.03 2404
(0.04)
[1.000]
Satisfaction with non-wage aspects of respondent’s main job (1-5 scale) 3.69 (1.12) 0.02 2402
(0.04)
[1.000]
Whether the respondent does not plan to leave their job in the next year 2.27 (0.72) -0.04 2403
(1-3 scale) (0.03)
[1.000]
Opportunities for promotion at the respondent’s main job (1-5 scale) 3.41 (1.27) -0.10* 2398
(0.05)
[1.000]
Safety and health conditions at the respondent’s main job (1-5 scale) 4.22 (0.79) 0.02 2253
(0.03)
[1.000]
Whether a scheduled shift was canceled with less than 24 hours notice in 0.09 (0.26) 0.02* 2485
the last month (0.01)
[1.000]
Number of stressors in their work environment at respondent’s main job 1.25 (1.24) 0.09* 2243
(0.05)
[1.000]
How easy is it to take time off from the respondent’s main job? (1-4 scale) 3.18 (0.81) -0.06* 2405
(0.03)
[1.000]
Notes: This table shows the impacts of an unconditional cash transfer on items within quality of employment. Under
various component headers, the table presents results for primary and secondary items in raw units. Items that are
italicized are secondary outcomes in the FDR corrections. Standard errors are provided in parentheses, and FDR-adjusted
q-values in square brackets below it. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

47
 Table 10: Impact of Guaranteed Income on Entrepreneurship

Control Mean Treatment Effect N

Entrepreneurship Index 0.05***††
(0.02)
[0.012]
Entrepreneurial Orientation Component 0.07***†††
(0.02)
[0.009]
The respondent’s self-reported willingness to take financial 4.52 (2.09) 0.08† 2866
risks (1-10 scale) (0.06)
[0.094]
Midpoint of the constant relative risk aversion (CRRA) 1.82 (1.55) -0.16***†† 2910
range implied by a participant’s coin flip gamble (0.06)
[0.026]
Entrepreneurial Intention Component 0.06**††
(0.02)
[0.013]
Whether or not the respondent has an idea for a business 0.58 (0.42) 0.03**†† 2909
(0.01)
[0.026]
The respondent’s likelihood rating that they will start a 4.95 (3.05) 0.15*†† 2909
business in the next 5 years (1-10 scale) (0.08)
[0.045]
The respondent’s interest in starting a business (1-10 scale) 6.21 (2.96) 0.12† 2910
(0.09)
[0.094]
Entrepreneurial Activity Component 0.01
(0.02)
[0.162]
If a family member who started a business lives in the 0.06 (0.21) -0.01**†† 2907
respondent’s household (0.01)
[0.044]
If the respondent knows someone who started or helped 0.60 (0.41) 0.03***†† 2907
start a business (0.01)
[0.026]
If the respondent ever started or helped start a business 0.30 (0.40) 0.00 2908
(0.01)
[0.281]

Notes: This table shows the impacts of an unconditional cash transfer on entrepreneurship. The
top-level index increases significantly by about 0.05 standard deviations. There are three compo-
nents with estimates in standard deviations (Entrepreneurial Orientation, Entrepreneurial Inten-
tion, and Entrepreneurial Activity), two of which are positive and significant. Each component
contains more than one primary item under it. The item representing the midpoint of the CRRA
range implied by a participant’s gamble in an incentive-compatible multiple price list experiment
is flipped before combining in the index, since low values represent comfort with risks. Standard
errors are provided in parentheses, and the FDR-adjusted q-value in square brackets below it. Ex-
cept for the family- and component-level index values, estimates are provided in terms of raw
units. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.
48
 Table 11: Impact of Guaranteed Income on Disability

Control Mean Treatment Effect N

Disability Index -0.09***††
(0.03)
[0.011]
Disability Component -0.09***†††
(0.03)
[0.002]
Whether the participant has a disability 0.31 (0.42) 0.04***††† 2874
(0.01)
[0.003]
Whether the respondent has a health problem or disability 0.28 (0.41) 0.04***††† 2872
that limits the work they can do (0.01)
[0.003]
How much the respondent’s worst disability or health 1.11 (1.71) 0.15***††† 2872
problem limits the of work they can do (1-7 scale) (0.05)
[0.003]
How long the respondent’s health problem or disability has 0.73 (1.06) 0.09***††† 2873
affected the work they can do (more than 1 year (0.03)
continuously or intermitently, less than 1 year) [0.003]

Notes: This table shows the impacts of an unconditional cash transfer on disability. The top-
level index decreases significantly by about 0.09 standard deviations, representing an increase in
disability. The q-values on the component and the top-level family index measures are different
even as the point estimate is the same as they adjust for different sets of estimates in the FDR
corrections (see Appendix D for details). There are several primary items under the component.
Standard errors are provided in parentheses, and the FDR-adjusted q-value in square brackets
below it. Except for the family- and component-level index values, estimates are provided in terms
of raw units. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

49
 Table 12: Impact of Guaranteed Income on Barriers to Employment

Control Mean Treatment Effect N

Barriers to Employment Index -0.03
(0.02)
[0.377]
Barriers to Employment Component -0.03
(0.02)
[0.425]
Whether the respondent missed work due to lack of 0.02 (0.13) 0.00 2941
childcare in the last month (0.01)
[1.000]
Whether the respondent missed work due to illness in the 0.20 (0.34) 0.00 2940
last month (0.01)
[1.000]
Whether the respondent missed work due to lack of 0.03 (0.13) 0.00 2940
trasportation in the last month (0.00)
[1.000]

Notes: This table shows the impacts of an unconditional cash transfer on barriers to employment.
The top-level index decreases insignificantly by about 0.03 standard deviations, representing an
insignificant increase in barriers. The q-values on the component and the top-level family index
measures are different even as the point estimate is the same as they adjust for different sets of
estimates in the FDR corrections (see Appendix D for details). There are several primary items
under the component. Standard errors are provided in parentheses, and the FDR-adjusted q-value
in square brackets below it. Except for the family- and component-level index values, estimates
are provided in terms of raw units. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable
q-value thresholds.

50
 Table 13: Impact of Guaranteed Income on Human Capital

Control Mean Treatment Effect N

Human Capital Index 0.22 (0.32) 0.02
(0.01)
[0.219]
Formal Education Component 0.02
(0.02)
[0.404]
Completed a high school degree, GED or post-secondary 0.92 (0.26) 0.01 2986
program (0.01)
[0.607]
Total years of post-secondary education completed 0.13 (0.33) 0.01 2623
post-baseline (0.01)
[0.611]
Enrolled in post-secondary program 0.15 (0.29) 0.01 2998
(0.01)
[0.607]
Average hours of school per week (full-time, part-time, withdrawn, 3.80 (8.70) 0.29 2615
etc.) in post-secondary program (0.31)
[0.936]
Participation in informal education 0.10 (0.21) 0.01 2987
(0.01)
[0.936]
Extent of participation in informal education (full-time, part-time, 0.07 (0.18) -0.00 2987
not enrolled) (0.01)
[0.947]
Whether the participant plans to receive job training 0.03 (0.14) 0.01** 2940
(0.01)
[0.405]

Notes: This table shows the impacts of an unconditional cash transfer on human capital. The top-
level index increases insignificantly by about 0.02 standard deviations. Apart from the component
"Formal Education", there is a component "Informal Education" comprised of only secondary items
that do not contribute to the index (so the component-level result is not printed). Items that are
italicized are secondary outcomes for the sake of the FDR corrections. Standard errors are provided
in parentheses, and the FDR-adjusted q-value in square brackets below it. Except for the family-
and component-level index values, estimates are provided in terms of raw units. * p < 0.1, **
p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

51
 Table 14: Impact of Guaranteed Income on Relationship Status

Control Mean All Surveys N

Relationship Status Index -0.013
(0.017)
[0.947]
Relationship Stability Component -0.037
(0.025)
[1.000]
How long the respondent has been in their relationship 2.496 (2.483) 0.013 2900
(0.064)
[0.971]
Number of times the respondent said they started or ended 0.428 (0.723) 0.057** 2903
a relationship in the last year (0.027)
[0.864]
Relationship Status Component 0.011
(0.016)
[1.000]
Whether the respondent is divorced 0.096 (0.289) -0.008 2943
(0.006)
[0.864]
Whether the respondent has a spouse 0.284 (0.441) -0.012 2945
(0.009)
[0.864]
Whether the respondent is in a romantic relationship 0.584 (0.435) 0.014 2989
(0.012)
[0.864]

Notes: This table shows the impacts of an unconditional cash transfer on relationship status. The
top-level index decreases insignificantly by about 0.01 standard deviations. Both the Relationship
Stability and Relationship Status component are null. There are a number of primary items under
each component. One item under Relationship Stability is signficant before adjusting for FDR: this
item looks at relationships a participant might have, regardless of whether that individual lives
in the household. Standard errors are provided in parentheses, and the FDR-adjusted q-value in
square brackets below it. Except for the family- and component-level index values, estimates are
provided in terms of raw units. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value
thresholds.

52
 Table 15: Impact of Guaranteed Income on Moving Labor Markets

Control Mean Treatment Effect N

Move Labor Market Index 0.09***†††
(0.03)
[0.003]
Single-item Component: Moved Labor Markets Since Baseline 0.12 (0.30) 0.02 2993
(0.01)
[0.121]
Search New Labor Market Component 0.11***†††
(0.03)
[0.003]
Any active area-search behaviors 0.10 (0.22) 0.03***††† 2851
(0.01)
[0.005]
Interested in moving areas 0.23 (0.36) 0.04***††† 2851
(0.01)
[0.005]
Number of active labor market-search behaviors 0.27 (0.67) 0.08***††† 2851
(0.03)
[0.005]

Notes: This table shows the impacts of an unconditional cash transfer on moving labor markets.
The top-level index for moving labor markets increases by about 0.09 standard deviations. A single
primary item component and a component with several primary items are listed. Standard errors
are provided in parentheses, and the FDR-adjusted q-value in square brackets below it. Except for
the family-level index value, estimates are provided in terms of raw units. * p < 0.1, ** p < 0.05,
*** p < 0.01; † refers to comparable q-value thresholds.

53
 Table 16: Impact of Guaranteed Income on Quality of Labor Market

Control Mean Treatment Effect N

Labor Market Quality Index -0.01
(0.01)
[0.116]
Labor Quality Component 0.00
(0.01)
[0.507]
Mean wage by education in 2022 (in dollars per hour) 29.69 (12.38) -0.02 2988
(0.05)
[0.907]
Employment to population ratio for ages 25 to 64 for respondent’s 0.76 (0.07) 0.00 2995
education group (0.00)
[0.733]
BLS projected job-growth for respondent’s education group 11.78 (8.94) -0.01 2961
(percent change) (0.13)
[0.907]
Median annual income for respondent’s education group (in 41630.43 (13472.24) 136.88 2995
dollars) (108.00)
[0.733]
Recent population growth for respondent’s education group 5.21 (11.36) -0.11 2995
(percent change) (0.14)
[0.822]
Mean wage growth 2019-2022 by education (percent change) 12.21 (4.17) -0.06 2988
(0.05)
[0.733]
Labor Market Amenities Component -0.02
(0.01)
[0.507]
Mean percentile household income rank for children whose parents 0.40 (0.01) 0.00 2993
were in the 25th percentile of income (0.00)
[0.907]
Natural amenities index (ranges from -5 to 9) -0.38 (1.51) 0.03 2992
(0.03)
[0.822]
Pollution index (mean PM2.5, RSEI, and AQI z-score) 0.99 (0.33) 0.00 2993
(0.01)
[0.822]
Consumption amenities index (PCA log scale, ranges from -2 to 2) 0.25 (0.27) 0.01 2993
(0.01)
[0.733]
Annual violent crime rate (crimes per 100,000 residents) 393.31 (110.91) -8.18* 2967
(4.60)
[0.733]
Annual property crime rate (crimes per 100,000 residents) 2329.32 (408.11) -44.15** 2967
(17.54)
[0.236]
Annual per-pupil school spending (in dollars) 13695.98 (3433.77) -10.87 2993
(134.00)
[0.907]

Notes: This table shows the impacts of an unconditional cash transfer on quality of labor market.
The top-level index decreases insignificantly by 0.01 standard deviations. Two components (Labor
Quality and Labor Market Amenities) are both null. All the primary items under them are also null,
with the exception that the labor markets participants move to perhaps have less crime, though this
does not survive the FDR adjustment. Standard errors are provided in parentheses, and the FDR-
adjusted q-value in square brackets below it. Except for the family- and component-level index
values, estimates are provided in terms of raw units. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to
comparable q-value thresholds. 54
 Table 17: Impact of Guaranteed Income on Benefits

Control Mean Treatment Effect N

Benefits Index -0.02
(0.02)
[0.242]
Take-Up Benefits Component -0.02
(0.02)
[0.639]
Total amount of government benefits received 5623.44 (6918.62) -299.13 2911
during the previous year (189.00)
[0.517]
Number of government benefits programs 1.85 (1.56) 0.00 2912
received during the previous year (0.04)
[1.000]
Number of government benefits programs received 1.75 (1.50) -0.01 2911
during the previous year (excluding educational (0.06)
assistance) [1.000]

Notes: This table shows the impacts of an unconditional cash transfer on benefits. The top-level
index decreases insignificantly by about 0.02 standard deviations. There is a single component,
with two primary items under it. The q-values on the component and the top-level family index
measures are different even as the point estimate is the same as they adjust for different sets of
estimates in the FDR corrections (see Appendix D for details). Standard errors are provided in
parentheses, and the FDR-adjusted q-value in square brackets below it. Except for the family- and
component-level index values, estimates are provided in terms of raw units. * p < 0.1, ** p < 0.05,
*** p < 0.01; † refers to comparable q-value thresholds.

55
 Table 18: Comparison of Administrative and Survey Data

Illinois Texas Aggregate

Midline Endline Pooled Midline Endline Pooled Midline Endline Pooled
Income (a) Survey, did not consent§ -3.83 2.03 -1.79 -0.69 -4.70 -7.21 -1.09 -2.40 -1.67
(Annual (4.04) (5.40) (4.40) (4.42) (6.65) (5.33) (2.72) (3.99) (3.00)
salary/wage 176 164 184 153 153 167 329 317 351
income in (b) Survey, consented§ 0.66 -0.59 -0.29 -1.29 -0.97 -1.37 -0.27 -1.18 -0.94
thousands (1.23) (1.42) (1.20) (1.16) (1.36) (1.14) (0.84) (0.96) (0.82)
of dollars) 1235 1226 1270 1260 1240 1299 2495 2466 2569
(c) UI, matched 0.49 -1.91 -0.73 -2.06 -3.87** -3.41** -0.41 -2.63** -1.64*
(1.08) (1.34) (1.16) (1.47) (1.76) (1.62) (0.87) (1.07) (0.94)
932 932 932 975 975 975 1907 1907 1907
(d) UI, matched + survey, 0.40 -1.36 -0.72 N/A N/A N/A N/A N/A N/A
consented and unmatched (1.01) (1.21) (1.05) (.) (.) (.) (.) (.) (.)
1262 1258 1288 - - - - - -
Pooling, (a) + (d) (IL) or (a) + 0.15 -1.20 -0.78 -1.93 -3.93** -3.73** -0.54 -2.09** -1.67**
(c) (TX) (0.98) (1.18) (1.02) (1.39) (1.70) (1.55) (0.80) (0.97) (0.85)

56
1438 1422 1472 1128 1128 1142 2566 2550 2614
Employment (a) Survey, did not consent§ 0.08 0.08 0.03 -0.03 -0.04 -0.03 0.01 0.04 0.01
(in (0.07) (0.06) (0.05) (0.07) (0.07) (0.07) (0.04) (0.04) (0.04)
percentage 191 187 196 168 166 172 359 353 368
points) (b) Survey, consented§ 0.01 -0.01 -0.01 -0.02 -0.05** -0.03* -0.01 -0.03* -0.02*†
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.02) (0.01)
1272 1268 1286 1287 1291 1308 2559 2559 2594
(c) UI, matched -0.04 -0.04 -0.03 -0.05 -0.08** -0.07**† -0.04** -0.06*** -0.04**†
(0.03) (0.03) (0.02) (0.03) (0.03) (0.03) (0.02) (0.02) (0.02)
932 932 932 975 975 975 1907 1907 1907
(d) UI, matched + survey, -0.02 -0.04** -0.04**† N/A N/A N/A N/A N/A N/A
consented and unmatched (0.02) (0.02) (0.02) (.) (.) (.) (.) (.) (.)
1287 1282 1295 - - - - - -
Pooling, (a) + (d) (IL) or (a) + -0.02 -0.03 -0.03*† -0.05 -0.07*** -0.06**† -0.03 -0.04*** -0.04***†
(c) (TX) (0.02) (0.02) (0.02) (0.03) (0.03) (0.03) (0.02) (0.02) (0.01)
1478 1469 1491 1143 1141 1147 2621 2610 2638
 Notes: This table compares the estimated impact of the guaranteed income program on income and employment for different data sources.
The rows marked (a) show the effects as estimated in SRC-enumerated survey data at midline and endline, for the few individuals who did
not consent to share administrative data; (b) shows the effects as estimated in SRC data for those who consented to share administrative
data; (c) shows the effects as estimated in the Unemployment Insurance data, for those who consented to share these data and could
be matched based on provided information (SSN in Illinois and SSN, name and age in Texas); (d) shows the effects as estimated in the
Unemployment Insurance data for those who consented to share these data and could be matched pooled with the SRC data for those
who consented but could not be matched (per the main text, this could not be run in Texas); (e) shows the aggregate effects, as combined
using fixed-effects meta-analysis. Results are presented separately for Illinois and Texas, as well as aggregated. The aggregation is done
across states using fixed-effects meta-analysis, except in the case of the rows indicated with a §, where the regressions can be run on the full
sample directly. All estimates are considered subgroup analyses for the FDR corrections except for those results that are both pooled across
data sources and aggregated across states. Standard errors are provided in parentheses, with the sample size below. Income is provided
in terms of thousands of dollars, and employment in percentage points. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value
thresholds.

57
Table 19: Estimates of the Marginal Propensity to Earn (MPE) by Net Change in Assets

MPE
Midline Endline Pooled
Observed Net change in assets
-2000 -0.07 -0.35 -0.28
0 -0.07 -0.37 -0.29
5000 -0.08 -0.43 -0.34
Model -0.24 -0.24 -0.24

Notes: This table shows how our estimates of reductions in total household income translate into
different MPEs depending on the net change in assets. The values of the net change in assets dis-
played in this table were selected to represent reasonable values for our data, with a change in
assets of $5000 representing an upper bound. The net asset accumulation is assumed for simplic-
ity to be constant over the time period as small changes will not affect the estimates much. The
pooled estimate places 30% weight on the midline results and 70% weight on the endline results,
for comparison to the main estimates reported in the paper. Total household income is adjusted as
described in the text by comparing the estimates of the aggregate total individual income reported
by participants with the calculated total individual income reported by participants when asked
about various specific types of income, after adjusting this to take into consideration the larger
estimate for salaried or wage income from the administrative data. The model estimates are pro-
vided for comparison. As the model does not put more emphasis on later time periods, the pooled
estimate is lower. The model-based estimates are also substantially smoothed.

58
 Table 20: Forecasts of NBER Affiliates and SSPP Forecasters
NBER Affiliates SSPP
Selected Fields Labor Studies
Median Mean N Median Mean N Median Mean N Result
Employed, in percentage points -0.5 -1.2 43 -1.0 -2.1 17 -0.7 0.3 94 -2.8
Work hours per week -0.9 -0.6 42 -1.2 -1.3 16 -1.4 -1.2 94 -1.6
Average hourly wage 1.0 1.2 42 0.5 0.9 16 0.7 0.8 94 -0.2
Duration of unemployment, in weeks 3.7 3.9 41 3.1 3.5 16 2.4 2.6 94 4.7
Participant is searching for work, in percentage points -2.8 -2.5 42 -4.8 -2.9 16 - - - 7.2
Enrollment in a post-secondary program 2.9 3.2 41 2.5 2.6 16 3.5 4.4 94 0.0
Individual salaried income, in thousands of dollars -0.7 -0.3 21 - - - 0 1.1 95 -1.6
Home production hours per week - - - - - - 0.8 1.8 93 0.8
Sleep, hours per week - - - - - - 0.7 0.6 93 -0.7
Social leisure, hours per week - - - - - - 4.7 5.0 92 0.9
Solitary leisure, hours per week - - - - - - 2.9 3.5 92 0.5

59
Notes: This table shows forecasts of NBER affiliates and users of the Social Science Prediction Platform (SSPP). As described in the text,
forecasts were elicited from NBER affiliates in several related Programs, and these forecasts were supplemented by forecasts from the
SSPP, including from members of its Superforecaster Panel. SSPP users were not asked the question about job search, to keep the survey
short, as they were asked to answer questions on a greater number of topics. Items forecast by fewer than 10 individuals for a subgroup
of forecasters are suppressed. All results are from endline data or year 3, as forecasters were asked to predict the effects at the end of the
study. Data for the employment and total household income results come from survey data, as this is what was specified in the questions
that forecasters saw. Data for the individual salaried income results comes from the administrative records, as this is what forecasters were
asked to predict, and data for the enrollment in a post-secondary program result comes from the combined NSC and survey data. All other
results come from the survey data.
 Figure 1: Location of Study

Notes: This figure plots the location of the sites in the study. Reproduced from Bartik et al. (2024).

60
 Figure 2: Flowchart of Recruitment Process

Notes: This figure shows a representation of the recruitment process.

61
 Figure 3: Timeline of Study

Notes: This figure shows a timeline of the program and study.

62
 Figure 4: Time Use Mobile App

Notes: This figure shows a screenshot of the mobile phone application participants used to fill in
time diaries on a randomly-selected weekday and weekend day each month.

63
 Figure 5: Time Use Results: Mobile App

Time Use Outcomes − Mobile App

Market Work q=0.308
Sleeping q=0.236
Child Care q=0.564
Other Income Generating Activities q=0.094
Community Engagement q=0.403
Caring for Others q=0.564
Self Improvement q=0.861
Searching for a Job q=0.861
Exercise q=0.861
Time with Others q=0.861
Self Care q=0.403
Home Production q=0.62
Solitary Leisure q=0.519
Non−Commuting Transportation q=0.022
Social Leisure q=0.371
Other Activities q=0.108
−20 −10 0 10 20
Minutes Per Day

Notes: This figure shows the main results from the time diaries.

64
References
Akee, Randall K Q, William E Copeland, Gordon Keeler, Adrian Angold and E Jane Costello. 2010.

“Parents’ Incomes and Children’s Outcomes: A Quasi-Experiment Using Transfer Payments from

Casino Profits.” American Economic Journal Applied Economics 2(1):86–115.

Allcott, Hunt, Luca Braghieri, Sarah Eichmeyer and Matthew Gentzkow. 2020. “The Welfare Effects

of Social Media.” American Economic Review 110(3):629–676.

Ashenfelter, Orley and Mark W. Plant. 1990. “Nonparametric Estimates of the Labor-Supply Effects

of Negative Income Tax Programs.” Journal of Labor Economics 8(1).

Atkinson, Anthony B and John Micklewright. 1991. “Unemployment Compensation and Labor Mar-

ket Transitions: A Critical Review.” Journal of Economic Literature 29(4):1679–1727.

Balakrishnan, Sidhya, Sewin Chan, Sara Constantino, Johannes Haushofer and Jonathan Morduch.

2024. “Household Responses to Guaranteed Income: Experimental Evidence from Compton, Cali-

fornia.” NBER Working Paper.

Banerjee, A, R Hanna, G Kreindler and B A Olken. 2017. “Debunking the Stereotype of the Lazy Wel-

fare Recipient: Evidence from Cash Transfer Programs.” The World Bank Research Observer 32(2):155–

184.

Bartik, Alex, David Broockman, Patrick Krause, Sarah Miller, Elizabeth Rhodes and Eva Vivalt. 2024.

“The Impact of Unconditional Cash Transfers on Consumption and Household Balance Sheets: Ex-

perimental Evidence from Two US States.”.

Benjamini, Yoav and Yosef Hochberg. 1995. “Controlling the false discovery rate: a practical and

powerful approach to multiple testing.” Journal of the Royal statistical society: series B (Methodological)

57(1):289–300.

Bernhardt, Annette, Christopher Campos, Allen Prohofsky, Aparna Ramesh and Jesse Rothstein. 2022.

“Independent Contracting, Self-Employment and Gig Work: Evidence from California Tax Data.”

NBER Working Paper.

Bertrand, M, S Mullainathan and D Miller. 2003. “Public Policy and Extended Families: Evidence from

Pensions in South Africa.” World Bank Economic Review 17(1):27–50.

65
Boswell, Wendy R, Ryan D Zimmerman and Brian W Swider. 2012. “Employee job search: Toward an

understanding of search context and search objectives.” Journal of Management 38(1):129–163.

Broockman, David, Elizabeth Rhodes, Alex Bartik, Karina Dotson, Patrick Krause, Sarah Miller and

Eva Vivalt. 2024. “The Causal Effects of Income on Political Attitudes and Behavior: A Randomized

Field Experiment.” NBER Working Paper.

Caliendo, Marco, Konstantinos Tatsiramos and Arne Uhlendorff. 2012. “Benefit Duration, Unemploy-

ment Duration and Job Match Quality: A Regression-Discontinuity Approach.” Journal of Applied

Econometrics 28(4).

Card, David, Raj Chetty and Andrea Weber. 2007. “The Spike at Benefit Exhaustion: Leaving the

Unemployment System or Starting a New Job?” American Economic Review 97(2).

Centeno, Mario. 2004. “The Match Quality Gains from Unemployment Insurance.” Journal of Human

Resources 39(3).

Cesarini, David, Erik Lindqvist, Matthew J Notowidigdo and Robert Ostling. 2017. “The Effect of

Wealth on Individual and Household Labor Supply: Evidence from Swedish Lotteries.” American

Economic Review 107(12):3917 – 3946.

Chetty, Raj and Nathaniel Hendren. 2018. “The Impact of Neighborhoods on Intergenerational Mo-

bility I: Childhood Exposure Effects.” Quarterly Journal of Economics 113(3).

Cox, James C and Ronald L Oaxaca. 1990. “Unemployment Insurance and Job Search.” Research in

Labor Economics 11:223–240.

DellaVigna, Stefano, Devin Pope and Eva Vivalt. 2019. “Predict Science to Improve Science.” Science

366(6464):428–429.

DellaVigna, Stefano and M. Daniele Paserman. 2005. “Job Search and Impatience.” Journal of Labor

Economics 23(3):527–588.

Eissa, N and J Liebman. 1996. “Labor Supply Response to the Earned Income Tax Credit.” Quarterly

Journal of Economics 111:605–637.

Eissa, Nada and Hilary Hoynes. 2004. “Taxes and the labor market participation of married couples:

the earned income tax credit.” Journal of Public Economics 88:1931–1958.

66
Feinberg, Robert M and Daniel Kuehn. 2018. “Guaranteed Nonlabor Income and Labor Supply:

The Effect of the Alaska Permanent Fund Dividend.” The B.E. Journal of Economic Analysis & Pol-

icy 18(3):350–13.

Ferguson, Joel, Rebecca Littman, Garret Christensen, Elizabeth Levy Paluck, Nicholas Swanson,

Zenan Wang, Edward Miguel, Birke David and John-Henry Pezzuto. 2023. “Survey of open sci-

ence practices and attitudes in the social sciences.” Nature Communications 14(5401).

Fiszbein, A, N Schady, F H G Ferreira, M Grosh, N Keleher, P Olinto and E Skoufias. 2009. “Condi-

tional Cash Transfers : Reducing Present and Future Poverty.” World Bank Policy Research Report.

Golosov, Mikhail, Michael Graber, Magne Mogstad and David Novgorodsky. 2023. “How Americans

Respond to Idiosyncratic and Exogenous Changes in Household Wealth and Unearned Income.”

Forthcoming, The Quarterly Journal of Economics .

Graham, Matthew, Erika McEntarfer, Kevin McKinney, Stephen Tibbets and Lee Tucker. 2022. “LEHD

Snapshot Documentation, Release S2021_R2022Q4.” Working Paper.

Greenberg, David and Harlan Halsey. 1983. “Systematic misreporting and effects of income mainte-

nance experiments on work effort: evidence from the Seattle-Denver experiment.” Journal of Labor

Economics 1(4):380–407.

Guess, Andrew M., Neil Malhotra, Jennifer Pan, Pablo Barberá, Hunt Allcott, Taylor Brown, Adri-

ana Crespo-Tenorio, Drew Dimmery, Deen Freelon, Matthew Gentzkow, Sandra González-Bailón,

Edward Kennedy, Young Mie Kim, David Lazer, Devra Moehler, Brendan Nyhan, Carlos Velasco

Rivera, Jaime Settle, Daniel Robert Thomas, Emily Thorson, Rebekah Tromble, Arjun Wilkins, Mag-

dalena Wojcieszak, Beixian Xiong, Chad Kiewiet de Jonge, Annie Franco, Winter Mason, Natalie Jo-

mini Stroud and Joshua A. Tucker. 2023. “Reshares on social media amplify political news but do

not detectably affect beliefs or opinions.” Science 381(6656):404–408.

Hausman, Jerry A and David A Wise. 1979. “Attrition bias in experimental and panel data: the Gary

income maintenance experiment.” Econometrica: Journal of the Econometric Society pp. 455–473.

Hoynes, Hilary and Jesse Rothstein. 2019. “Universal Basic Income in the United States and Advanced

Countries.” Annual Review of Economics 11:929–958.

67
Imbens, G W, D B Rubin and B I Sacerdote. 2001. “Estimating the effect of unearned income on labor

earnings, savings, and consumption: Evidence from a survey of lottery players.” American Economic

Review 91(4):778–794.

Jaroszewicz, Ania, Oliver P. Hauser, Jon M. Jachimowicz and Julian Jamison. 2023. “Cash Can Make

Its Absence Felt: Randomizing Unconditional Cash Transfer Amounts in the US.” Working Paper .

Jones, Damon and Ioana Marinescu. 2018. “The Labor Market Impacts of Universal and Permanent

Cash Transfers: Evidence from the Alaska Permanent Fund.” NBER Working Paper 24312.

Katz, Lawrence and Alan Krueger. 2019. “Understanding Trends in Alternative Work Arrangements

in the United States.” NBER Working Paper.

Kleven, Henrik. 2024. “The EITC and the Extensive Margin: A Reappraisal.” Journal of Public Economics

236.

Kling, Jeffrey, Jeffrey Liebman and Lawrence Katz. 2007. “Experimental Analysis of Neighborhood

Effects.” Econometrica 75(1):83–119.

Krueger, Alan B and Bruce D Meyer. 2002. Labor supply effects of social insurance. In Handbook of

Public Economics, ed. Alan J Auerbach and Martin Feldstein. Vol. 4 Elsevier B.V. chapter 33, pp. 2327–

2392.

Lalive, Rafael. 2007. “Unemployment Benefits, Unemployment Duration, and Post-Unemployment

Jobs: A Regression Discontinuity Approach.” American Economic Review 97(2).

Liebman, Jeffrey, Kathryn Carlsona, Eliza Novickc and Pamela Portocarreroa. 2022. “The Chelsea Eats

Program: Experimental Impacts.” Rappaport Institute for Greater Boston .

Meyer, Bruce D., Wallace K. C. Mok and James X. Sullivan. 2015. “Household Surveys Household

Surveys in Crisis.” Journal of Economic Perspectives 29(4):100–226.

Meyer, Bruce and Dan Rosenbaum. 2001. “Welfare, the Earned Income Tax Credit, and the Labor

Supply of Single Mothers.” Quarterly Journal of Economics 116(3).

Miller, Sarah, Elizabeth Rhodes, Alex Bartik, David Broockman, Patrick Krause and Eva Vivalt. 2024.

“Does Income Affect Health? Evidence from a Randomized Controlled Trial of a Guaranteed In-

come.” NBER Working Paper.

68
Mostert, Cyprian M and Judit V Castello. 2020. “Long run educational and spillover effects of uncon-

ditional cash transfers: Evidence from South Africa.” Economics & Human Biology 36(C).

Nekoei, Arash and Andrea Weber. 2017. “Does Extending Unemployment Benefits Improve Job Qual-

ity?” American Economic Review 107(2).

Nichols, A. and J. Rothstein. 2016. The Earned Income Tax Credit. In Economics of Means-Tested Transfer

Programs in the United States, ed. R.A. Moffit. Chicago: University of Chicago Press pp. 137–218.

Sauval, Maria, Greg Duncan, Lisa A. Gennetian, Katherine Magnuson, Nathan Fox, Kimberly Noble

and Hirokazu Yoshikawa. 2024. “Unconditional Cash Transfers and Maternal Employment: Evi-

dence from the Baby’s First Years Study.” SSRN Working Paper.

Stillwell, Laura, Maritza Morales-Gracia, Katherine Magnuson, Lisa A. Gennetian, Maria Sauval,

Nathan A. Fox, Sarah Halpern-Meekin, Hirohazu Yoshikawa and Kimberly G. Noble. 2024. “Un-

conditional Cash and Breastfeeding, Child Care, and Maternal Employment among Families with

Young Children Residing in Poverty.” Social Science Review 98(2).

van Ours, Jan C. and Milan Vodopivec. 2008. “Does reducing unemployment insurance generosity

reduce job match quality?” Journal of Public Economics 92(3-4).

69
 The Employment Effects of a Guaranteed Income:
Experimental Evidence from Two U.S. States
Appendix
Eva Vivalt Elizabeth Rhodes Alexander Bartik David Broockman Patrick Krause
Sarah Miller

A Details on Recruitment
When targeting our mailers and ads, we aimed to generate a sample that was diverse along several

dimensions. First, we aimed to recruit a sample that was representative by geographic type (large

urban, medium-sized urban, rural, and suburban) based on the county of the applicant. We identified

1-5 counties of each type in each state that were demographically representative of this type. Nation-

ally, roughly 19% of households that meet the eligibility criteria for our program live in rural areas,

35% live in suburban areas, 17% live in medium-sized urban areas, and 28% live in large urban ar-

eas.2 Our goal was to recruit a sample that matched these population shares, although we ultimately

somewhat oversampled large urban areas to reduce recruitment costs. In the end, 13% of program

participants lived in rural counties, 18% in suburban, 16% in medium urban and 53% in large urban

areas.

We also aimed to over-represent low-income participants and to approximately match the eligible

population’s share of male and female individuals.

In additional to the geographically stratified sampling described above, we used stratified random

sampling to ensure that low-income individuals are over-represented in the sample of program partic-

ipants and the share of males and females is approximately proportionate to their shares of the eligible

population (which is roughly 62% female). Table 1 reports basic summary statistics of both eligible

mailer respondents and enrolled program participants and compares both groups to the population

mean characteristics computed using the American Community Survey for eligible households living

in study counties. We report estimates of the eligible population both unweighted and reweighted to

reflect the FPL group and county type stratification variables that were used.

B Detailed Blocking and Randomization Procedures

Strata were formed according to participants’ race/ethnicity (non-Hispanic White, Black, and His-

panic), income group (0-100% FPL, 101-200% FPL, 201-300% FPL), and state (IL or TX). A separate
2 Less than 1% live in small urban counties so we exclude this group.

70
strata contained all 20 clusters with more than one individual in them.

Participants were grouped within strata into blocks of three based on similarities across pre-

treatment covariates.3 One cluster per block was selected to be in the treatment group and the other

two in the control group.

All participants took up the treatment. Only one person was enrolled from the waitlist in order

to replace a participant in the treatment group who was removed from the program for violating

program rules regarding a threat of harm to another person. Since we had 99.9% compliance, we

analyze the experiment using intent-to-treat, following the original random assignment.

C Detailed Balance Tests and Simulations

We assigned a minimum critical p-value for each variable in a set of important baseline covariates,

such that any differences between the treatment and control group could not be significant at that

level. A randomization which failed to meet the p-value threshold for any baseline covariate was

rejected.

We also tested whether any set of baseline covariates within a given outcome area was jointly

significant. A randomization in which the p-value of any such F-test was over 0.25 was rejected.

In theory, our strategy could result in some participants being more likely to be assigned to the

treatment than others if they have particularly large or small values of some baseline variable. There-

fore, we conducted 1,000 simulations to check that our randomization process resulted in every cluster

having a 1 in 3 chance of being in the treatment group. A histogram of these simulations is provided in

Figure A11, and Figure A12 shows a quantile-quantile plot of this distribution against what one would

expect from Bernoulli coin flips with a 1 in 3 chance of being assigned to the treatment group. These

figures indicate that the observed distribution of treatment assignment probabilities is no different

from what we would expect by chance.

D False Discovery Rate

We compute false discovery rate (FDR) q-values within families of outcomes, following Benjamini

and Hochberg (1995). Our hypothesis tests are placed into tiers (denoted K0, K1, K2, K3, and K4) as

follows, corresponding with our prioritization of the tests:

3 After
blocking, some clusters were "left over" if the number of clusters in a strata did not divide evenly by three. A
second round of blocking was performed for these clusters, again forming blocks based on similarity across pre-treatment
covariates.

71
 • K0: Family-level estimates pooled across time. The q-values for these items will be computed

using all the K0 items across families in a paper.

• K1: Component-level estimates pooled across time. The q-values for these items are computed

using the K0 and K1 items in the outcome’s same family.

• K2: Primary item-level estimates pooled across time. The q-values for these items are computed

using the K0, K1, and K2 items in the outcome’s same family.

• K3: All other estimates (“exploratory” tier). This includes family-level, component-level, and

item-level estimates which are computed within each time period, estimates on items pre-specified

as secondary or tertiary, and all tests of heterogenous treatment effects, as well as descriptive

analyses. The q-values for these items are computed using the K0, K1, K2, and K3 items in the

outcome’s same family.

• K4: Any post hoc comparisons conducted after filing these pre-analysis plans (e.g., in response

to referee comments). The q-values for these items are computed using the K0, K1, K2, K3, and

K4 items in the outcome’s same family.

In some families, there is only one item pre-specified to be in the index for a given component,

or only one component in the family. In these cases, we use one fewer “level” in the FDR adjustment

(e.g., if there is only one item in a component, it would not be adjusted with K2, as it would already

have been adjusted at the K1 level for that component. If there is only one component in a family,

that component is counted as K0, primary items are counted as K1, secondary items are counted as

K2, etc.). For some families, we also distinguish between secondary and tertiary items; this effectively

pushes K3 items to K4 and K4 items to K5, so the distinct tertiary items can be in their own K3 tier.

These cases were flagged in the pre-analysis plan, which offers further details.

Table A7 summarizes the FDR tiers of our estimates.

E Changes from the Pre-Analysis Plan

The pre-specified analyses were closely followed, however, there were a few instances in which we

made a small change.

The first set of small changes were made prior to receiving midline survey data. At this stage, the

following changes were made:

72
– We specified a few supplementary tests, outside of the index, relating to considering whether to

model the household as following the unitary household model;

– If participants were looking for a job in the last 3 months was added as a primary item to the

active search component of the Employment Preferences and Job Search family. This was later

phrased in the pre-analysis plan as whether someone was looking for a job in the last year,

but this may be misleading as the question always asks about over the last 3 months, and the

responses are merely averaged to aggregate up to the year;

– We added more specificity as to how the descriptive conditions under which a respondent would

take a job measure would be treated for the purpose of multiple hypothesis corrections and

specified that a participant’s subjective expectations as to when they would find a job would be

a secondary outcome;

– We added as a primary measure whether the participant would be willing to take any job and

the reservation wage under the Selectivity of Job Search family;

– We specified that the items under the Employment Quality and Stability of Employment com-

ponents under the Quality of Employment family would all refer to both main and other jobs;

previously, some of the items had referred to the main job and some to any job;

– In the Stability of Employment component in the Quality of Employment family, we look at how

many jobs participants have held in the last 12 months, rather than any longer time period, given

that the longer time periods asked about could overlap with the pre-treatment time period;

– We added how hard it is to take time off and whether a scheduled shift was cancelled with

less than 24 hours notice in the last month as primary items under the Quality of Work Life

component under the Quality of Employment family;

– The index value for human capital formation was specified to, as an exception, be a binary

measure indicating receipt of any education or job training in the survey or National Student

Clearinghouse data (the National Student Clearinghouse data had not been collected yet, nor

any post-treatment survey data relevant to this question);

– We specified that informal educational outcomes would be considered exploratory;

73
 – We added more specificity to how we would combine outcomes into indices, specifying that

primary items would be combined into components using seemingly unrelated regression;

– We specified that we would use the false discovery rate (FDR), following Allcott et al. (2020),

rather than performing family-wise error rate corrections.

Additional exploratory analyses and robustness checks, including additional subgroup analyses,

were also specified.

After receiving the midline survey data, but before receiving the endline survey data, a few addi-

tional changes were made:

– We clarified the overall estimation approach that applied to all estimates in the paper, including:

– We specified that since only one person was enrolled from the waitlist, we would ignore

the waitlist in the estimation strategy and analyze the results using an intent-to-treat esti-

mation, given the compliance rate of 99.9%;

– We had previously pre-specified the weights we would place on the different time periods

and surveys in how they would be pooled, but we further specified how we would treat

missing observations (i.e., if we are missing a survey round for an individual, we replace

that measure for that individual at that time period with the treatment-arm-specific mean

following Kling, Liebman and Katz (2007));

– Though the previous version of the pre-analysis plan had specified that the FDR analysis

would follow the hierarchical nature of Guess et al. (2023), we more clearly specified the

structure of the outcomes with a table;

– We emphasized that the unconditional analyses would be preferred wherever possible. For

example, we cannot consider most aspects of quality of employment (such as whether one’s

manager treats one fairly) for those without jobs, so this family of outcomes is necessarily

conditional. However, in other cases we can run an unconditional analysis, such as in the

barriers to employment section where we can consider a respondent to miss 0 days of work

due to illness if they are unemployed.

– Given that the SRC survey version of job search questions were limited to having been asked

of those who were employed, and thus could be affected by selection into employment, we

74
 specified that we would instead focus on the Qualtrics version of these variables, which would

not be subject to this limitation;

– We excluded the reservation wage from the Selectivity of Job Search index given that it would

not be available for all individuals;

– There was a potential inconsistency within the Quality of Employment family, where in one

place we specified that we would prefer the SRC surveys if there were differential attrition in

the mobile surveys and in another place we specified that we would separately present a set

of results that were based only on the SRC data as a robustness check. Given that differential

attrition looked pretty minor, we kept to the latter rule;

– Under Formality of Employment, the percent of reported income not on W-2s using administra-

tive records for the W-2s and total income from the SRC survey was deemed a robustness check

rather than a primary item. No W-2 data had been obtained at this time;

– We widened the set of activities considered under informal education;

– We specified that total individual income would be considered the top-level index value for the

sake of FDR adjustments and that government transfers would be considered descriptive when

broken out separately under the Income family of outcomes;

Other than these changes, we added a few robustness checks and heterogeneity analyses, al-

though these were all pre-specified to be exploratory.

A few other changes were subsequently made based on feasibility/data availability:

– We originally specified an alternative measure of work hours (based off of part-time or full-time

employment) that we ultimately did not use as it was only asked once at midline;

– We originally specified an alternative measure of how many work hours the participant wanted,

under preferences for employment in the employment preferences and job search family, that

we ultimately did not use as it transpired participants could not indicate that they wanted less

work in the specified Qualtrics question;

– Income data for individuals paid per task or with tips was specified as exploratory, as both were

subject to error (e.g., if a respondent did not specify the right number of tasks per hour/shift

75
 or hours/shifts worked, we would not be able to calculate their total income from tasks). Tasks

data appeared more prone to error than tips data, so to avoid under-reporting income for the

few participants paid predominantly in tips, we included tips income in our total calculated

individual income measure;

– For duration of unemployment, we could not consider an unemployment-based version of the

average duration of non-employment, because we can only clearly distinguish between non-

employment and unemployment at the time of the SRC surveys, and the average duration of

non-employment variable was pre-specified to be based on both SRC and Qualtrics survey data.

As the next best alternative, we created a variable that captured unemployment at the time of

the survey, as well as a variable that captured non-employment at the time of the survey, for

comparison;

– One item in the quality of employment family was only asked to people who were pursuing

temp work. As this was answered by very few people, we decided it should be considered a

secondary rather than primary item.

F Relationship with Other Papers

It should be noted that the analyses in this paper come in part from three different pre-analysis plans

that focus, alternatively, on employment; income and financial health;, and housing and geographic

mobility. While we did not know at the time of registering the pre-analysis plans which outcome

variables would be included in which papers, we pre-specified that we would conduct our multiple

hypothesis corrections according to how the tests were originally registered. For example, if one fam-

ily of outcomes from the “income, expenditures and financial health” pre-analysis plan was included

in the paper based primarily off results from the “employment” pre-analysis plan, that family of out-

comes would be subject to false discovery rate corrections alongside the other tests in the “income and

financial health” pre-analysis plan. This measure ensured that there was no incentive to selectively

combine outcomes into papers in such a way as to make results appear more significant.

Readers are also referred to Bartik et al. (2024), Broockman et al. (2024), and Miller et al. (2024) for

information on household finance, political, and health outcomes.

76
G Time Use

G.1 Robustness Check: Secondary Activities

The mobile app’s time diary allowed participants to record if they were engaged in two activities

simultaneously (e.g., watching television while cooking dinner). Following the pre-analysis plan,

the estimates in the main text split this time equally between overlapping activities. For example, if

someone recorded cooking dinner from 6:00 - 6:30 and watching television from 6:00 - 7:00, this would

be counted as 15 minutes of home production (half of the 30 minutes from 6:00 - 6:30) and 45 minutes

of leisure (half of the 30 minutes from 6:00 - 6:30, and the entire 30 minutes from 6:30 - 7:00). Figure

5 in the main text uses this equal allocation method. Figure A6 shows that the results are similar

when we measure time use by the raw sum of all time and do not discount activities by the number

of simultaneous activities that occur.

G.2 Robustness Check: Recoding of “Other” Activities

Next, participants were able to select an “Other” category and write an open-ended description of how

they spent a particular block of time if they did not find any of the pre-existing categories suitable.

Figure 5 in the main text reported an imprecisely estimated 5 minutes/day increase in time spent on

these “Other” activities. We used ChatGPT-4 to recode these open-ended responses into one of our

pre-existing categories when possible. Figure A7 shows the results on this version of the measures.

G.3 Results from Enumerated and Quarterly Surveys

The enumerated midline and endline as well as the quarterly surveys also asked participants to report

the typical number of hours per week, hours per month, hours per year, or days per year, depend-

ing on the activity4 that they engaged in certain activities. Figure A3 shows the estimates on these

outcomes.

4 We rescale the estimates that are in terms of hours per month and days per year variables to be in terms of minutes per

day to match the scale used in the mobile app data.

77
 Table A1: Protection of Benefits

Benefit Illinois Texas

Medicaid Eligibility was not affected Eligibility was not affected

SNAP Eligibility was not affected First $300 per quarter did not affect SNAP,
but the remaining amount of the transfer was
considered unearned income for the purposes of
determining eligibility and the amount of the benefit

78
TANF Eligibility was not affected First $300 per quarter did not affect TANF,
but the remaining amount of the transfer was
considered unearned income for the purposes of
determining eligibility and the amount of the benefit

Housing Assistance Did not affect eligibility for Chicago Housing Not eligible to participate
Authority, other localities not eligible to participate

SSI Not eligible to participate Not eligible to participate

Notes: This table shows which major benefits were preserved or not preserved in Illinois and Texas.
Table A2: Baseline Characteristics of Respondents to Any Qualtrics Survey in Year 1 vs. Non-
Respondents

Respondents Non-Respondents
Control Treatment p-value Control Treatment p-value
Demographic
Age 30.078 30.203 0.574 28.196 27.933 0.847
Male 0.313 0.321 0.683 0.565 0.800 0.071
Female 0.684 0.676 0.685 0.435 0.200 0.071
Non-binary/other 0.003 0.003 0.990 0.000 0.000 0.661
Non-Hispanic Black 0.307 0.295 0.516 0.239 0.267 0.835
Non-Hispanic Asian 0.038 0.037 0.810 0.022 0.000 0.321
Non-Hispanic White 0.462 0.472 0.620 0.478 0.533 0.715
Non-Hispanic Native American 0.025 0.019 0.308 0.000 0.067 0.310
Hispanic 0.212 0.221 0.589 0.304 0.200 0.407
Household Size 2.999 2.947 0.445 2.848 2.667 0.705
Number of Other Adults in the Household 0.717 0.680 0.284 0.674 0.933 0.351
Any Children 0.573 0.570 0.851 0.457 0.467 0.946
Has Disability 0.312 0.338 0.157 0.256 0.352 0.427
Bachelor’s Degree 0.205 0.203 0.907 0.209 0.161 0.639
Employed 0.585 0.575 0.574 0.609 0.800 0.138
Income and employment
Total Household Income (1000s) 29.949 29.959 0.990 28.571 32.146 0.612
Total Individual Income (1000s) 21.235 21.319 0.916 20.459 23.716 0.490
Work Hours/Week 21.812 21.025 0.344 20.413 33.173 0.039
Has a Second Job 0.174 0.167 0.640 0.130 0.200 0.551
Months Employed in the Past Year 7.254 7.199 0.778 7.875 8.200 0.785
Number of Jobs in the Past 1 Year 1.433 1.395 0.437 1.705 1.933 0.579
Number of Jobs in the Past 3 Years 2.613 2.647 0.713 2.905 5.133 0.065
Searching for Work 0.508 0.495 0.504 0.587 0.467 0.424
Started or Helped to Start a Business 0.296 0.316 0.264 0.303 0.301 0.986
Housing
Lived Temporarily with Family or Friends 0.285 0.263 0.202 0.113 0.255 0.194
Stayed in Non-Permanent Housing 0.085 0.085 0.964 0.036 0.150 0.212
Housing Search Actions in Last 3 Months 0.241 0.251 0.582 0.276 0.532 0.052
Number of Times Moved in the Past 5 Years 1.363 1.321 0.316 1.147 1.759 0.097
Relationships
Is in a Romantic Relationship 0.622 0.626 0.829 0.565 0.667 0.482
Lives with a Romantic Partner 0.432 0.440 0.694 0.370 0.533 0.275
Married 0.222 0.220 0.912 0.217 0.267 0.708
Divorced 0.081 0.078 0.805 0.087 0.000 0.043

Notes: This table compares the baseline characteristics of participants who responded or did not
respond to a Qualtrics survey in Year 1 of the study.

79
Table A3: Baseline Characteristics of Respondents to Any Qualtrics Survey in Year 2 vs. Non-
Respondents

Respondents Non-Respondents
Control Treatment p-value Control Treatment p-value
Demographic
Age 30.113 30.163 0.823 28.337 30.467 0.094
Male 0.315 0.319 0.841 0.419 0.600 0.086
Female 0.682 0.678 0.843 0.581 0.400 0.086
Non-binary/other 0.003 0.003 0.984 0.000 0.000 0.013
Non-Hispanic Black 0.304 0.295 0.623 0.337 0.300 0.706
Non-Hispanic Asian 0.038 0.037 0.950 0.035 0.000 0.082
Non-Hispanic White 0.464 0.471 0.753 0.430 0.533 0.334
Non-Hispanic Native American 0.026 0.019 0.214 0.000 0.067 0.148
Hispanic 0.214 0.223 0.553 0.233 0.133 0.203
Household Size 3.018 2.948 0.317 2.512 2.833 0.347
Number of Other Adults in the Household 0.726 0.680 0.183 0.500 0.833 0.077
Any Children 0.575 0.569 0.741 0.465 0.567 0.339
Has Disability 0.311 0.337 0.154 0.281 0.375 0.302
Bachelor’s Degree 0.204 0.204 0.995 0.231 0.161 0.342
Employed 0.588 0.571 0.384 0.547 0.800 0.006
Income and employment
Total Household Income (1000s) 30.013 30.034 0.979 28.378 29.111 0.871
Total Individual Income (1000s) 21.264 21.203 0.939 20.281 26.052 0.182
Work Hours/Week 21.805 20.735 0.199 21.372 35.800 0.004
Has a Second Job 0.173 0.162 0.430 0.174 0.333 0.100
Months Employed in the Past Year 7.249 7.164 0.659 7.698 8.700 0.284
Number of Jobs in the Past 1 Year 1.420 1.376 0.356 1.872 2.267 0.241
Number of Jobs in the Past 3 Years 2.576 2.652 0.399 3.605 3.700 0.896
Searching for Work 0.510 0.494 0.412 0.500 0.500 1.000
Started or Helped to Start a Business 0.297 0.310 0.490 0.269 0.504 0.013
Housing
Lived Temporarily with Family or Friends 0.284 0.266 0.290 0.215 0.170 0.550
Stayed in Non-Permanent Housing 0.082 0.088 0.606 0.102 0.045 0.211
Housing Search Actions in Last 3 Months 0.242 0.252 0.552 0.235 0.358 0.175
Number of Times Moved in the Past 5 Years 1.360 1.323 0.377 1.343 1.425 0.719
Relationships
Is in a Romantic Relationship 0.627 0.630 0.891 0.500 0.567 0.530
Lives with a Romantic Partner 0.436 0.445 0.670 0.302 0.333 0.757
Married 0.228 0.223 0.751 0.093 0.167 0.330
Divorced 0.081 0.077 0.695 0.081 0.100 0.767

Notes: This table compares the baseline characteristics of participants who responded or did not
respond to a Qualtrics survey in Year 2 of the study.

80
Table A4: Baseline Characteristics of Respondents to Any Qualtrics Survey in Year 3 vs. Non-
Respondents

Respondents Non-Respondents
Control Treatment p-value Control Treatment p-value
Demographic
Age 30.140 30.222 0.714 28.803 28.100 0.541
Male 0.306 0.324 0.351 0.496 0.433 0.531
Female 0.691 0.673 0.341 0.496 0.567 0.485
Non-binary/other 0.003 0.003 0.848 0.007 0.000 0.319
Non-Hispanic Black 0.307 0.295 0.504 0.277 0.333 0.555
Non-Hispanic Asian 0.037 0.037 0.978 0.044 0.000 0.014
Non-Hispanic White 0.466 0.470 0.823 0.423 0.500 0.449
Non-Hispanic Native American 0.025 0.018 0.202 0.022 0.100 0.168
Hispanic 0.209 0.222 0.424 0.285 0.167 0.135
Household Size 3.019 2.950 0.324 2.737 2.867 0.702
Number of Other Adults in the Household 0.717 0.682 0.307 0.723 0.767 0.821
Any Children 0.578 0.572 0.748 0.482 0.533 0.610
Has Disability 0.312 0.339 0.156 0.271 0.287 0.854
Bachelor’s Degree 0.203 0.205 0.909 0.230 0.161 0.321
Employed 0.585 0.573 0.513 0.591 0.767 0.048
Income and employment
Total Household Income (1000s) 29.881 29.943 0.937 30.806 34.815 0.289
Total Individual Income (1000s) 21.273 21.237 0.965 20.751 26.775 0.109
Work Hours/Week 21.811 20.833 0.245 21.591 33.767 0.008
Has a Second Job 0.176 0.165 0.472 0.139 0.233 0.256
Months Employed in the Past Year 7.254 7.162 0.639 7.343 9.267 0.011
Number of Jobs in the Past 1 Year 1.434 1.377 0.243 1.514 2.333 0.004
Number of Jobs in the Past 3 Years 2.598 2.637 0.667 2.905 4.267 0.020
Searching for Work 0.510 0.492 0.362 0.489 0.533 0.662
Started or Helped to Start a Business 0.294 0.311 0.371 0.327 0.494 0.071
Housing
Lived Temporarily with Family or Friends 0.289 0.264 0.162 0.187 0.227 0.613
Stayed in Non-Permanent Housing 0.082 0.087 0.634 0.099 0.075 0.652
Housing Search Actions in Last 3 Months 0.241 0.255 0.404 0.242 0.241 0.987
Number of Times Moved in the Past 5 Years 1.367 1.317 0.234 1.268 1.613 0.104
Relationships
Is in a Romantic Relationship 0.626 0.630 0.844 0.562 0.633 0.468
Lives with a Romantic Partner 0.435 0.445 0.595 0.380 0.400 0.837
Married 0.226 0.221 0.742 0.168 0.267 0.259
Divorced 0.083 0.078 0.668 0.058 0.033 0.517

Notes: This table compares the baseline characteristics of participants who responded or did not
respond to a Qualtrics survey in Year 3 of the study.

81
Table A5: Baseline Characteristics of Respondents to the Enumerated Midline vs. Non-
Respondents

Respondents Non-Respondents
Control Treatment p-value Control Treatment p-value
Demographic
Age 30.075 30.149 0.741 29.160 31.300 0.123
Male 0.317 0.325 0.678 0.387 0.450 0.615
Female 0.680 0.672 0.681 0.613 0.550 0.615
Non-binary/other 0.003 0.003 0.978 0.000 0.000 .
Non-Hispanic Black 0.307 0.297 0.583 0.267 0.200 0.522
Non-Hispanic Asian 0.037 0.035 0.765 0.053 0.100 0.521
Non-Hispanic White 0.463 0.471 0.689 0.467 0.550 0.511
Non-Hispanic Native American 0.026 0.016 0.093 0.000 0.200 0.029
Hispanic 0.213 0.221 0.621 0.240 0.200 0.698
Household Size 3.002 2.947 0.423 2.867 2.850 0.969
Number of Other Adults in the Household 0.717 0.685 0.364 0.720 0.650 0.728
Any Children 0.573 0.569 0.851 0.520 0.550 0.813
Has Disability 0.310 0.337 0.149 0.283 0.416 0.269
Bachelor’s Degree 0.206 0.205 0.942 0.182 0.081 0.112
Employed 0.587 0.580 0.737 0.560 0.450 0.385
Income and employment
Total Household Income (1000s) 29.963 30.170 0.786 29.538 21.969 0.059
Total Individual Income (1000s) 21.227 21.481 0.753 21.097 14.862 0.132
Work Hours/Week 21.832 21.287 0.515 20.440 16.300 0.410
Has a Second Job 0.174 0.167 0.623 0.147 0.150 0.971
Months Employed in the Past Year 7.275 7.216 0.758 7.027 6.900 0.919
Number of Jobs in the Past 1 Year 1.443 1.401 0.385 1.347 1.500 0.665
Number of Jobs in the Past 3 Years 2.619 2.678 0.533 2.640 2.982 0.550
Searching for Work 0.508 0.492 0.399 0.533 0.600 0.594
Started or Helped to Start a Business 0.297 0.313 0.369 0.282 0.445 0.160
Housing
Lived Temporarily with Family or Friends 0.282 0.262 0.235 0.255 0.314 0.603
Stayed in Non-Permanent Housing 0.081 0.086 0.628 0.138 0.104 0.668
Housing Search Actions in Last 3 Months 0.241 0.256 0.370 0.263 0.212 0.620
Number of Times Moved in the Past 5 Years 1.364 1.325 0.347 1.237 1.385 0.525
Relationships
Is in a Romantic Relationship 0.625 0.631 0.729 0.533 0.450 0.511
Lives with a Romantic Partner 0.433 0.444 0.548 0.387 0.300 0.463
Married 0.224 0.223 0.918 0.173 0.150 0.800
Divorced 0.080 0.076 0.704 0.107 0.150 0.624

Notes: This table compares the baseline characteristics of participants who responded or did not
respond to the enumerated midline survey.

82
Table A6: Baseline Characteristics of Respondents to the Enumerated Endline vs. Non-
Respondents

Respondents Non-Respondents
Control Treatment p-value Control Treatment p-value
Demographic
Age 30.050 30.140 0.687 29.903 31.050 0.419
Male 0.313 0.326 0.478 0.447 0.350 0.415
Female 0.685 0.671 0.465 0.544 0.650 0.370
Non-binary/other 0.003 0.003 0.838 0.010 0.000 0.320
Non-Hispanic Black 0.308 0.297 0.515 0.243 0.250 0.945
Non-Hispanic Asian 0.038 0.037 0.892 0.029 0.000 0.083
Non-Hispanic White 0.466 0.470 0.836 0.417 0.550 0.280
Non-Hispanic Native American 0.025 0.019 0.261 0.019 0.100 0.244
Hispanic 0.208 0.222 0.393 0.320 0.150 0.068
Household Size 3.008 2.952 0.411 2.806 2.850 0.925
Number of Other Adults in the Household 0.715 0.692 0.492 0.748 0.350 0.016
Any Children 0.574 0.571 0.883 0.505 0.550 0.713
Has Disability 0.311 0.334 0.206 0.278 0.450 0.153
Bachelor’s Degree 0.204 0.205 0.970 0.220 0.141 0.299
Employed 0.585 0.582 0.871 0.602 0.400 0.096
Income and employment
Total Household Income (1000s) 29.951 30.171 0.774 29.865 25.930 0.337
Total Individual Income (1000s) 21.248 21.450 0.802 20.769 18.539 0.654
Work Hours/Week 21.750 21.286 0.580 22.400 18.050 0.468
Has a Second Job 0.173 0.168 0.749 0.184 0.150 0.699
Months Employed in the Past Year 7.263 7.229 0.858 7.272 6.850 0.703
Number of Jobs in the Past 1 Year 1.434 1.403 0.529 1.548 1.450 0.755
Number of Jobs in the Past 3 Years 2.586 2.676 0.325 3.226 3.150 0.920
Searching for Work 0.510 0.491 0.340 0.505 0.600 0.432
Started or Helped to Start a Business 0.293 0.312 0.300 0.351 0.480 0.269
Housing
Lived Temporarily with Family or Friends 0.286 0.262 0.170 0.195 0.314 0.278
Stayed in Non-Permanent Housing 0.082 0.086 0.749 0.100 0.154 0.528
Housing Search Actions in Last 3 Months 0.242 0.257 0.376 0.226 0.150 0.402
Number of Times Moved in the Past 5 Years 1.362 1.322 0.335 1.308 1.567 0.284
Relationships
Is in a Romantic Relationship 0.626 0.629 0.859 0.544 0.650 0.370
Lives with a Romantic Partner 0.432 0.443 0.579 0.408 0.500 0.453
Married 0.223 0.225 0.908 0.204 0.100 0.187
Divorced 0.081 0.077 0.724 0.078 0.050 0.620

Notes: This table compares the baseline characteristics of participants who responded or did not
respond to the enumerated endline survey.

83
 Table A7: FDR Tiers

Pooled Across Mid- Pooled Across Mid- Estimates At Each Time

line/Endline and Monthly line/Endline Surveys Period (e.g., at midline, in
Surveys Only (Omitting Monthly year 2, etc.)
Surveys)
Family K0 K0 K3

Primary Components K1 K1 K3

Primary Items K2 K2 K3

Secondary Items K3 K3 K3

Tertiary Items K3 K3 K3

Heterogeneous treat- K3 K3 Not calculated

ment effects

Any post-PAP tests K4 K4 K4

84
Table A8: Impact of Guaranteed Income on Annual Earned and Other Unearned Income:
Comparison of Impacts by Income at Baseline

Control Mean Entire Sample Below 100% FPL Above 100% FPL
Total household income (self-reported) 48.2 (33.9) -4.1***††† -2.8* -4.8***††
(1.0) (1.7) (1.2)
[0.001] [0.448] [0.011]
Total individual income 36.6 (27.0) -1.5* 0.2 -2.6**
(0.9) (1.6) (1.0)
[0.185] [1.000] [0.164]
Total individual income (self-reported) 33.5 (25.1) -2.5** -3.4** -2.3*
(1.0) (1.4) (1.2)
[0.105] [0.164] [0.375]
Individual salaried/wage income 26.0 (26.2) -1.1 -0.2 -1.7
(0.8) (1.3) (1.0)
[0.258] [1.000] [0.454]
Self-employment income 5.9 (13.7) -0.1 0.6 -0.8
(0.5) (0.8) (0.7)
[0.423] [0.897] [0.641]
Income from gig work 0.4 (1.3) -0.1 -0.0 -0.1
(0.0) (0.1) (0.1)
[0.263] [1.000] [0.582]
Passive income 0.0 (0.2) 0.0 0.0 0.0
(0.0) (0.0) (0.0)
[0.258] [0.910] [0.568]
Other income 4.7 (6.1) -0.1 -0.1 -0.1
(0.2) (0.3) (0.2)
[0.377] [1.000] [1.000]
Government transfers 3.6 (4.9) -0.2 -0.0 -0.2
(0.1) (0.3) (0.2)
[0.356] [1.000] [0.731]

Notes: This table compares results for income for participants by whether they were above or
below 100% of the FPL at baseline. Survey data are used for greater comparability across tables
with subgroup analyses, so the individual salaried/wage income is promoted to a primary item
for the sake of FDR corrections. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value
thresholds.

85
Table A9: Impact of Guaranteed Income on Annual Earned and Other Unearned Income:
Comparison of Impacts for Participants by Baseline Level of Education

No Bachelor’s Bachelor’s
Control Mean Entire Sample Degree Degree
Total household income (self-reported) 48.2 (33.9) -4.1***††† -4.3***†† -1.2
(1.0) (1.1) (3.4)
[0.001] [0.011] [1.000]
Total individual income 36.6 (27.0) -1.5* -1.5 -1.4
(0.9) (1.0) (2.4)
[0.185] [0.524] [0.934]
Total individual income (self-reported) 33.5 (25.1) -2.5** -2.8*** -2.1
(1.0) (1.0) (2.6)
[0.105] [0.145] [0.829]
Individual salaried/wage income 26.0 (26.2) -1.1 -1.5* 1.2
(0.8) (0.8) (2.4)
[0.258] [0.400] [1.000]
Self-employment income 5.9 (13.7) -0.1 0.4 -3.4***†
(0.5) (0.6) (1.2)
[0.423] [0.897] [0.098]
Income from gig work 0.4 (1.3) -0.1 0.0 -0.4***††
(0.0) (0.1) (0.1)
[0.263] [1.000] [0.016]
Passive income 0.0 (0.2) 0.0 0.0 0.0
(0.0) (0.0) (0.0)
[0.258] [0.897] [1.000]
Other income 4.7 (6.1) -0.1 -0.2 0.1
(0.2) (0.2) (0.5)
[0.377] [0.807] [1.000]
Government transfers 3.6 (4.9) -0.2 -0.3 0.1
(0.1) (0.2) (0.3)
[0.356] [0.525] [1.000]

Notes: This table compares results for income for participants by whether or not they had a bache-
lor’s degree at baseline. Survey data are used for greater comparability across tables with subgroup
analyses, so the individual salaried/wage income is promoted to a primary item for the sake of
FDR corrections. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

86
Table A10: Impact of Guaranteed Income on Annual Earned and Other Unearned Income:
Comparison of Impacts by Sex at Baseline

Control Mean Entire Sample Female/Other Male

Total household income (self-reported) 48.2 (33.9) -4.1***††† -3.6***† -4.9**
(1.0) (1.2) (1.9)
[0.001] [0.066] [0.164]
Total individual income 36.6 (27.0) -1.5* -1.2 -1.8
(0.9) (1.0) (1.8)
[0.185] [0.629] [0.752]
Total individual income (self-reported) 33.5 (25.1) -2.5** -1.9* -3.9**
(1.0) (1.1) (1.9)
[0.105] [0.425] [0.326]
Individual salaried/wage income 26.0 (26.2) -1.1 -1.2 -1.2
(0.8) (0.9) (1.6)
[0.258] [0.561] [0.861]
Self-employment income 5.9 (13.7) -0.1 0.3 -0.7
(0.5) (0.6) (1.1)
[0.423] [1.000] [0.897]
Income from gig work 0.4 (1.3) -0.1 -0.1 -0.0
(0.0) (0.0) (0.1)
[0.263] [0.554] [1.000]
Passive income 0.0 (0.2) 0.0 0.0** -0.0
(0.0) (0.0) (0.0)
[0.258] [0.166] [1.000]
Other income 4.7 (6.1) -0.1 0.0 -0.3
(0.2) (0.2) (0.3)
[0.377] [1.000] [0.626]
Government transfers 3.6 (4.9) -0.2 -0.1 -0.2
(0.1) (0.2) (0.2)
[0.356] [0.934] [0.728]

Notes: This table compares results for income for participants by sex at baseline. Survey
data are used for greater comparability across tables with subgroup analyses, so the individual
salaried/wage income is promoted to a primary item for the sake of FDR corrections. * p < 0.1, **
p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

87
Table A11: Impact of Guaranteed Income on Annual Earned and Other Unearned Income:
Comparison of Impacts by State

Control Mean Entire Sample Illinois Texas

Total household income (self-reported) 48.2 (33.9) -4.1***††† -3.0** -5.7***†††
(1.0) (1.4) (1.4)
[0.001] [0.125] [0.002]
Total individual income 36.6 (27.0) -1.5* -0.4 -1.7
(0.9) (1.3) (1.1)
[0.185] [0.899] [0.383]
Total individual income (self-reported) 33.5 (25.1) -2.5**† -1.6 -3.4**
(1.0) (1.4) (1.4)
[0.063] [0.519] [0.100]
Individual salaried/wage income 26.0 (26.2) -1.1 -0.4 -1.5
(0.8) (1.1) (1.1)
[0.258] [0.877] [0.427]
Self-employment income 5.9 (13.7) -0.1 0.1 -0.4
(0.5) (0.8) (0.7)
[0.423] [0.991] [0.733]
Income from gig work 0.4 (1.3) -0.1 -0.2***† 0.0
(0.0) (0.1) (0.1)
[0.263] [0.080] [0.717]
Passive income 0.0 (0.2) 0.0 -0.0 0.0***††
(0.0) (0.0) (0.0)
[0.258] [0.645] [0.044]
Other income 4.7 (6.1) -0.1 0.1 -0.2
(0.2) (0.3) (0.3)
[0.377] [0.980] [0.696]
Government transfers 3.6 (4.9) -0.2 -0.1 -0.2
(0.1) (0.2) (0.2)
[0.297] [0.733] [0.629]

Notes: This table compares results for income for participants by whether they lived in Illinois
or Texas at baseline. Survey data are used for greater comparability across tables with subgroup
analyses, so the individual salaried/wage income is promoted to a primary item for the sake of
FDR corrections. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

88
Table A12: Impact of Guaranteed Income on Annual Earned and Other Unearned Income:
Comparison of Impacts for Participants with and without Children at Baseline

Control Mean Entire Sample No Children Has Children

Total household income (self-reported) 48.2 (33.9) -4.1***††† -5.9***†† -2.6**
(1.0) (1.6) (1.3)
[0.001] [0.016] [0.333]
Total individual income 36.6 (27.0) -1.5* -2.4* -1.2
(0.9) (1.3) (1.1)
[0.185] [0.406] [0.670]
Total individual income (self-reported) 33.5 (25.1) -2.5** -3.1** -2.0
(1.0) (1.4) (1.3)
[0.105] [0.279] [0.495]
Individual salaried/wage income 26.0 (26.2) -1.1 -2.0 -0.5
(0.8) (1.3) (1.0)
[0.258] [0.464] [1.000]
Self-employment income 5.9 (13.7) -0.1 -0.8 0.4
(0.5) (0.8) (0.7)
[0.423] [0.777] [0.939]
Income from gig work 0.4 (1.3) -0.1 -0.0 -0.1
(0.0) (0.1) (0.1)
[0.263] [1.000] [0.464]
Passive income 0.0 (0.2) 0.0 0.0* -0.0
(0.0) (0.0) (0.0)
[0.258] [0.368] [1.000]
Other income 4.7 (6.1) -0.1 -0.0 0.0
(0.2) (0.3) (0.3)
[0.377] [1.000] [1.000]
Government transfers 3.6 (4.9) -0.2 -0.1 -0.1
(0.1) (0.2) (0.2)
[0.356] [0.833] [1.000]

Notes: This table compares results for income for participants by whether or not they had children
at baseline. Survey data are used for greater comparability across tables with subgroup analy-
ses, so the individual salaried/wage income is promoted to a primary item for the sake of FDR
corrections. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

89
Table A13: Impact of Guaranteed Income on Labor Supply: Comparison of Impacts by Base-
line Level of Education

No Bachelor’s Bachelor’s
Control Mean Entire Sample Degree Degree
Labor Supply Elasticity Index -0.06**† -0.08** 0.06
(0.03) (0.03) (0.06)
[0.076] [0.493] [1.000]
Labor Supply Elasticity Component -0.06**†† -0.08** 0.06
(0.03) (0.03) (0.06)
[0.043] [0.493] [1.000]
Whether the respondent is employed 0.74 (0.39) -0.02*† -0.04** 0.02
(0.01) (0.02) (0.02)
[0.072] [0.432] [1.000]
Hours worked per week 30.28 (19.83) -1.28**† -1.35* 0.78
(0.64) (0.79) (1.54)
[0.072] [0.814] [1.000]
Number of other household members which are employed 0.47 (0.61) -0.02 -0.02 -0.02
(0.02) (0.03) (0.05)
[1.000] [1.000] [1.000]
Total number of hours participant and spouse/partner works 40.69 (24.84) -2.16*** -2.36** -0.53
per week (0.78) (0.95) (2.24)
[0.316] [0.388] [1.000]
Total number of hours all household members (including the 48.22 (29.64) -2.21** -2.49** -0.69
participant) work per week (0.92) (1.12) (2.45)
[0.429] [0.526] [1.000]
Total number of hours participant’s parents in household 3.22 (12.07) -0.13 0.02 0.69
work per week (0.35) (0.41) (0.99)
[1.000] [1.000] [1.000]
Total number of hours participant’s adult children in 1.23 (6.75) 0.30 0.07 0.11
household work per week (0.29) (0.34) (0.26)
[1.000] [1.000] [1.000]

Notes: This table compares results for labor supply for participants by whether or not they had
a bachelor’s degree at baseline. Survey data are used for greater comparability across tables with
subgroup analyses, so whether an individual is employed is promoted to a primary item for the
sake of FDR corrections. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value
thresholds.

90
Table A14: Impact of Guaranteed Income on Labor Supply: Comparison of Impacts by Age
at Baseline

Control Mean Entire Sample Under 30 30+

Labor Supply Elasticity Index -0.06**† -0.10** -0.01
(0.03) (0.04) (0.04)
[0.076] [0.379] [1.000]
Labor Supply Elasticity Component -0.06**†† -0.10** -0.01
(0.03) (0.04) (0.04)
[0.043] [0.379] [1.000]
Whether the respondent is employed 0.74 (0.39) -0.02*† -0.04** 0.00
(0.01) (0.02) (0.02)
[0.072] [0.450] [1.000]
Hours worked per week 30.28 (19.83) -1.28**† -1.84** -0.59
(0.64) (0.88) (0.95)
[0.072] [0.613] [1.000]
Number of other household members which are employed 0.47 (0.61) -0.02 -0.04 0.01
(0.02) (0.03) (0.03)
[1.000] [1.000] [1.000]
Total number of hours participant and spouse/partner works 40.69 (24.84) -2.16*** -2.91*** -1.65
per week (0.78) (1.08) (1.19)
[0.316] [0.322] [0.970]
Total number of hours all household members (including the 48.22 (29.64) -2.21** -3.50*** -0.49
participant) work per week (0.92) (1.30) (1.35)
[0.429] [0.322] [1.000]
Total number of hours participant’s parents in household 3.22 (12.07) -0.13 -0.27 -0.18
work per week (0.35) (0.56) (0.33)
[1.000] [1.000] [1.000]
Total number of hours participant’s adult children in 1.23 (6.75) 0.30 0.19 0.38
household work per week (0.29) (0.12) (0.61)
[1.000] [0.940] [1.000]

Notes: This table compares results for labor supply for participants by age at baseline. Survey data
are used for greater comparability across tables with subgroup analyses, so whether an individual
is employed is promoted to a primary item for the sake of FDR corrections. * p < 0.1, ** p < 0.05,
*** p < 0.01; † refers to comparable q-value thresholds.

91
Table A15: Impact of Guaranteed Income on Labor Supply: Comparison of Impacts by Sex
at Baseline

Control Mean Entire Sample Female/Other Male

Labor Supply Elasticity Index -0.06**† -0.06* -0.05
(0.03) (0.03) (0.05)
[0.076] [0.763] [1.000]
Labor Supply Elasticity Component -0.06**†† -0.06* -0.05
(0.03) (0.03) (0.05)
[0.043] [0.763] [1.000]
Whether the respondent is employed 0.74 (0.39) -0.02*† -0.02 -0.01
(0.01) (0.02) (0.02)
[0.072] [0.961] [1.000]
Hours worked per week 30.28 (19.83) -1.28**† -1.37* -1.16
(0.64) (0.78) (1.20)
[0.072] [0.763] [1.000]
Number of other household members which are employed 0.47 (0.61) -0.02 0.00 -0.06
(0.02) (0.03) (0.04)
[1.000] [1.000] [0.940]
Total number of hours participant and spouse/partner works 40.69 (24.84) -2.16*** -2.38** -2.10
per week (0.78) (0.98) (1.39)
[0.316] [0.432] [0.940]
Total number of hours all household members (including the 48.22 (29.64) -2.21** -1.85 -4.26***
participant) work per week (0.92) (1.16) (1.62)
[0.429] [0.896] [0.322]
Total number of hours participant’s parents in household 3.22 (12.07) -0.13 -0.03 -0.17
work per week (0.35) (0.39) (0.67)
[1.000] [1.000] [1.000]
Total number of hours participant’s adult children in 1.23 (6.75) 0.30 0.64 -0.37*
household work per week (0.29) (0.42) (0.21)
[1.000] [0.940] [0.763]

Notes: This table compares results for labor supply for participants by sex at baseline. Survey data
are used for greater comparability across tables with subgroup analyses, so whether an individual
is employed is promoted to a primary item for the sake of FDR corrections. * p < 0.1, ** p < 0.05,
*** p < 0.01; † refers to comparable q-value thresholds.

92
Table A16: Impact of Guaranteed Income on Labor Supply: Comparison of Impacts by State

Control Mean Entire Sample Illinois Texas

Labor Supply Elasticity Index -0.06**†† -0.03 -0.08*
(0.03) (0.04) (0.04)
[0.041] [1.000] [0.900]
Labor Supply Elasticity Component -0.06**†† -0.03 -0.08*
(0.03) (0.04) (0.04)
[0.043] [1.000] [0.900]
Whether the respondent is employed 0.74 (0.39) -0.02*† -0.01 -0.03*
(0.01) (0.02) (0.02)
[0.072] [1.000] [0.900]
Hours worked per week 30.28 (19.83) -1.28**† -1.06 -1.55*
(0.64) (0.90) (0.93)
[0.072] [1.000] [0.900]
Number of other household members which are employed 0.47 (0.61) -0.02 -0.01 -0.03
(0.02) (0.04) (0.03)
[1.000] [1.000] [1.000]
Total number of hours participant and spouse/partner works 40.69 (24.84) -2.16*** -2.16** -2.12*
per week (0.78) (1.08) (1.18)
[0.703] [0.900] [0.900]
Total number of hours all household members (including the 48.22 (29.64) -2.21** -1.89 -3.36**
participant) work per week (0.92) (1.34) (1.34)
[0.703] [0.952] [0.900]
Total number of hours participant’s parents in household 3.22 (12.07) -0.13 -0.22 -0.07
work per week (0.35) (0.49) (0.49)
[1.000] [1.000] [1.000]
Total number of hours participant’s adult children in 1.23 (6.75) 0.30 0.60 0.00
household work per week (0.29) (0.41) (0.42)
[1.000] [0.952] [1.000]

Notes: This table compares results for labor supply for participants by whether they lived in Illinois
or Texas at baseline. Survey data are used for greater comparability across tables with subgroup
analyses, so whether an individual is employed is promoted to a primary item for the sake of FDR
corrections. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

93
Table A17: Impact of Guaranteed Income on Human Capital Formation: Comparison of
Impacts by Age

Control Mean Entire Sample Under 30 30+

Human Capital Index 0.22 (0.32) 0.02 0.02 0.01
(0.01) (0.02) (0.02)
[0.219] [1.000] [1.000]
Formal Education Component 0.02 0.06** -0.04
(0.02) (0.03) (0.03)
[0.404] [1.000] [1.000]
Completed a high school degree, GED or 0.92 (0.26) 0.01 0.01 0.01
post-secondary program (0.01) (0.01) (0.01)
[0.607] [1.000] [1.000]
Total years of post-secondary education completed 0.13 (0.33) 0.01 0.03 -0.02**
post-baseline (0.01) (0.02) (0.01)
[0.611] [1.000] [1.000]
Enrolled in post-secondary program 0.15 (0.29) 0.01 0.02* -0.01
(0.01) (0.01) (0.01)
[0.607] [1.000] [1.000]
Average hours of school per week (full-time, part-time, 3.80 (8.70) 0.29 1.08** -0.60*
withdrawn, etc.) in post-secondary program (0.31) (0.49) (0.33)
[0.936] [1.000] [1.000]
Participation in informal education 0.10 (0.21) 0.01 0.01 0.01
(0.01) (0.01) (0.01)
[0.936] [1.000] [1.000]
Extent of participation in informal education (full-time, 0.07 (0.18) -0.00 0.00 -0.01
part-time, not enrolled) (0.01) (0.01) (0.01)
[0.947] [1.000] [1.000]
Whether the participant plans to receive job training 0.03 (0.14) 0.01** 0.01* 0.02*
(0.01) (0.01) (0.01)
[0.405] [1.000] [1.000]

Notes: This table compares results for income for participants by employment at baseline. * p <
0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

94
Table A18: Impact of Guaranteed Income on Entrepreneurship: Comparison of Impacts by
Baseline Level of Education

No Bachelor’s Bachelor’s
Control Mean Entire Sample Degree Degree
Entrepreneurship Index 0.05***††† 0.04**† 0.05
(0.02) (0.02) (0.03)
[0.010] [0.091] [0.227]
Entrepreneurial Orientation Component 0.07***††† 0.07**† 0.09
(0.02) (0.03) (0.06)
[0.008] [0.076] [0.201]
The respondent’s self-reported willingness to take 4.52 (2.09) 0.08† 0.05 0.22
financial risks (1-10 scale) (0.06) (0.08) (0.14)
[0.092] [0.436] [0.201]
Midpoint of the constant relative risk aversion (CRRA) 1.82 (1.55) -0.16***†† -0.19***† -0.10
range implied by a participant’s coin flip gamble (0.06) (0.07) (0.14)
[0.025] [0.060] [0.435]
Entrepreneurial Intention Component 0.06**†† 0.06**† -0.02
(0.02) (0.03) (0.05)
[0.016] [0.094] [0.516]
Whether or not the respondent has an idea for a 0.58 (0.42) 0.03**†† 0.03* 0.03
business (0.01) (0.02) (0.04)
[0.027] [0.130] [0.406]
The respondent’s likelihood rating that they will start a 4.95 (3.05) 0.15*†† 0.18 -0.16
business in the next 5 years (1-10 scale) (0.08) (0.11) (0.20)
[0.040] [0.191] [0.406]
The respondent’s interest in starting a business (1-10 6.21 (2.96) 0.12† 0.18 -0.20
scale) (0.09) (0.12) (0.21)
[0.092] [0.209] [0.359]
Entrepreneurial Activity Component 0.01 -0.01 0.07
(0.02) (0.03) (0.05)
[0.189] [0.441] [0.229]
If a family member who started a business lives in the 0.06 (0.21) -0.01**†† -0.02***† 0.01
respondent’s household (0.01) (0.01) (0.02)
[0.037] [0.060] [0.433]
If the respondent knows someone who started or 0.60 (0.41) 0.03***†† 0.02 0.05
helped start a business (0.01) (0.02) (0.03)
[0.025] [0.310] [0.227]
If the respondent ever started or helped start a business 0.30 (0.40) 0.00 0.00 0.01
(0.01) (0.02) (0.03)
[0.291] [0.544] [0.516]

Notes: This table compares results for entrepreneurship for participants by whether or not they
had a bachelor’s degree at baseline. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable
q-value thresholds.

95
Table A19: Impact of Guaranteed Income on Entrepreneurship: Comparison of Impacts by
Age at Baseline

Control Mean Entire Sample Under 30 30+

Entrepreneurship Index 0.05***††† 0.05**† 0.06***†
(0.02) (0.02) (0.02)
[0.010] [0.087] [0.064]
Entrepreneurial Orientation Component 0.07***††† 0.07**† 0.06*
(0.02) (0.03) (0.04)
[0.008] [0.087] [0.168]
The respondent’s self-reported willingness to take 4.52 (2.09) 0.08† 0.07 0.10
financial risks (1-10 scale) (0.06) (0.08) (0.09)
[0.092] [0.377] [0.326]
Midpoint of the constant relative risk aversion (CRRA) 1.82 (1.55) -0.16***†† -0.17**† -0.12
range implied by a participant’s coin flip gamble (0.06) (0.08) (0.09)
[0.025] [0.087] [0.227]
Entrepreneurial Intention Component 0.06**†† 0.05 0.09**†
(0.02) (0.03) (0.04)
[0.016] [0.201] [0.076]
Whether or not the respondent has an idea for a 0.58 (0.42) 0.03**†† 0.05**† 0.04*
business (0.01) (0.02) (0.02)
[0.027] [0.087] [0.127]
The respondent’s likelihood rating that they will start a 4.95 (3.05) 0.15*†† 0.12 0.30**†
business in the next 5 years (1-10 scale) (0.08) (0.13) (0.13)
[0.040] [0.374] [0.087]
The respondent’s interest in starting a business (1-10 6.21 (2.96) 0.12† 0.07 0.20
scale) (0.09) (0.13) (0.15)
[0.092] [0.441] [0.227]
Entrepreneurial Activity Component 0.01 0.01 0.02
(0.02) (0.03) (0.03)
[0.189] [0.479] [0.406]
If a family member who started a business lives in the 0.06 (0.21) -0.01**†† -0.00 -0.03***††
respondent’s household (0.01) (0.01) (0.01)
[0.037] [0.485] [0.042]
If the respondent knows someone who started or 0.60 (0.41) 0.03***†† 0.03 0.05**†
helped start a business (0.01) (0.02) (0.02)
[0.025] [0.218] [0.065]
If the respondent ever started or helped start a business 0.30 (0.40) 0.00 -0.00 0.03
(0.01) (0.02) (0.02)
[0.291] [0.566] [0.201]

Notes: This table compares results for entrepreneurship for participants by age at baseline. * p <
0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

96
Table A20: Impact of Guaranteed Income on Quality of Employment: Comparison of Impacts
by Baseline Level of Education, Summary Measures

No Bachelor’s Bachelor’s
Control Mean Entire Sample Degree Degree
Quality of Employment Index -0.01 -0.02 -0.02
(0.01) (0.02) (0.03)
[0.449] [1.000] [1.000]
Adequacy of Employment Component 0.01 0.01 0.00
(0.03) (0.03) (0.05)
[1.000] [1.000] [1.000]
Employment Quality Component -0.01 -0.01 0.02
(0.02) (0.03) (0.05)
[1.000] [1.000] [1.000]
Single-item Component: Whether the respondent 0.24 (0.37) -0.00 -0.00 0.02
reports working any informal job (0.01) (0.02) (0.03)
[1.000] [1.000] [1.000]
Single-item Component: Average hourly income from 17.26 (9.72) -0.18 -0.37 -0.75
all jobs, weighted by hours worked at each job (0.37) (0.43) (0.99)
[1.000] [1.000] [1.000]
Stability of Employment Component -0.02 -0.01 -0.04
(0.02) (0.03) (0.04)
[1.000] [1.000] [1.000]
Quality of Work Life Component -0.02 -0.03 0.02
(0.02) (0.02) (0.03)
[1.000] [1.000] [1.000]

Notes: This table compares summary-level results for quality of employment for participants by
whether or not they had a bachelor’s degree at baseline. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers
to comparable q-value thresholds.

97
 Table A21: Impact of Guaranteed Income on Quality of Employment: Comparison of Impacts
by Baseline Level of Education, Expanded Measures
Entire No Bachelor’s Bachelor’s
Control Mean Sample Degree Degree
Adequacy of Employment
The respondent is employed part-time in their main job and would prefer 0.24 (0.39) -0.00 0.00 0.03
to work full-time (0.02) (0.02) (0.03)
[1.000] [1.000] [1.000]
The respondent would prefer to work more hours in their current main 0.21 (0.36) 0.01 0.01 -0.03
job (0.02) (0.02) (0.03)
[1.000] [1.000] [1.000]
The number of jobs held by the respondent apart from their main job 0.38 (0.70) -0.03 -0.05 0.01
(0.03) (0.03) (0.06)
[1.000] [1.000] [1.000]
Employment Quality
Whether training is offered by the respondent’s main employer 0.53 (0.45) 0.01 0.00 0.02
(0.02) (0.02) (0.04)
[1.000] [1.000] [1.000]
Whether training is offered during work hours by the respondent’s main 0.49 (0.45) 0.01 -0.00 0.04
employer (0.02) (0.02) (0.04)
[1.000] [1.000] [1.000]
Whether formal training is offered by the respondent’s main employer 0.13 (0.29) -0.00 -0.00 0.01
(0.01) (0.01) (0.03)
[1.000] [1.000] [1.000]
Number of non-wage benefits at respondent’s job(s), weighted by hours 3.62 (2.90) -0.12 -0.14 -0.14
worked at each job (0.11) (0.13) (0.25)
[1.000] [1.000] [1.000]
Whether the respondent must work an irregular shift at each job, 0.19 (0.34) 0.01 0.00 0.01
weighted by hours worked at each job (0.01) (0.02) (0.03)
[1.000] [1.000] [1.000]
Number of non-wage benefits at respondent’s job(s), alternate measure 4.53 (2.97) -0.17 -0.22 -0.15
(0.11) (0.14) (0.26)
[1.000] [1.000] [1.000]
Informality of Employment
Whether the respondent reports any gig economy jobs such as Uber, TaskRabbit, 0.09 (0.25) -0.00 -0.01 -0.00
or online surveys (0.01) (0.01) (0.02)
[1.000] [1.000] [1.000]
Stability of Employment
How many months the respondent has been employed in the past year 10.69 (2.66) -0.03 -0.01 -0.17
(0.10) (0.13) (0.19)
[1.000] [1.000] [1.000]
How long the respondent has spent at their current main job and other 24.88 (34.85) 1.70 3.13** -1.13
jobs (months), weighted by hours worked at each job (1.15) (1.41) (2.10)
[1.000] [1.000] [1.000]
How many jobs the respondent has held in the past 12 months 1.76 (1.60) -0.12** -0.15** -0.17
(0.05) (0.06) (0.17)
[1.000] [1.000] [1.000]
How many jobs the respondent has held in the past two years 2.33 (3.67) -0.17* -0.20** -0.19
(0.09) (0.09) (0.23)
[1.000] [1.000] [1.000]
Whether the respondent’s main job is a temp job 0.10 (0.26) 0.01 0.00 0.02
(0.01) (0.01) (0.03)
[1.000] [1.000] [1.000]
Whether each of the respondent’s jobs is salaried, weighted by hours 0.23 (0.39) -0.01 -0.02 0.01
worked at each job (0.01) (0.01) (0.04)
[1.000] [1.000] [1.000]
Whether the respondent is performing contract or freelance work at each 0.25 (0.38) 0.00 0.01 -0.01
job, weighted by hours worked at each job (0.01) (0.02) (0.03)
[1.000] [1.000] [1.000]
How many months the respondent expects to remain in their main job 8.97 (6.56) -1.30* -1.10 -1.38
(0.70) (0.87) (1.66)

98
 [1.000] [1.000] [1.000]
Quality of Work Life
Advance notice of schedule provided at the respondent’s main job (1-4 2.52 (1.24) -0.03 -0.01 -0.11
scale) (0.05) (0.06) (0.11)
[1.000] [1.000] [1.000]
The work activities are not boring at the respondent’s main job (1-5 scale) 3.11 (1.05) -0.01 -0.08 0.18*
(0.04) (0.06) (0.09)
[1.000] [1.000] [1.000]
Satisfaction with compensation at the respondent’s main job (1-5 scale) 3.51 (1.06) -0.02 -0.05 0.03
(0.04) (0.05) (0.10)
[1.000] [1.000] [1.000]
Whether the respondent faces age discrimination at work 0.06 (0.21) 0.00 -0.00 0.02
(0.01) (0.01) (0.02)
[1.000] [1.000] [1.000]
Whether the respondent faces sex discrimination at work 0.08 (0.25) 0.00 0.00 -0.01
(0.01) (0.01) (0.02)
[1.000] [1.000] [1.000]
Whether the respondent faces racial or ethnic discrimination at work 0.08 (0.25) 0.01 -0.00 0.00
(0.01) (0.01) (0.02)
[1.000] [1.000] [1.000]
Whether the respondent experienced fair treatment by their supervisor 4.05 (0.91) 0.04 0.03 0.12
(1-5 scale) (0.04) (0.05) (0.08)
[1.000] [1.000] [1.000]
Whether job demands do not interfere with family life (1-4 scale) 2.91 (0.87) 0.02 -0.02 0.01
(0.03) (0.04) (0.07)
[1.000] [1.000] [1.000]
Whether the job is a good fit with the respondent’s experience and skills 4.19 (0.92) -0.05 -0.03 -0.08
(1-5 scale) (0.04) (0.05) (0.08)
[1.000] [1.000] [1.000]
Flexibility of schedule at the respondent’s main job (1-4 scale) 1.91 (0.91) 0.01 -0.02 0.11
(0.04) (0.05) (0.09)
[1.000] [1.000] [1.000]
Overall satisfaction with the respondent’s main job (1-5 scale) 3.96 (0.96) 0.03 -0.01 0.03
(0.04) (0.05) (0.09)
[1.000] [1.000] [1.000]
Whether the respondent has decision-making input in their job (1-4 scale) 2.67 (0.98) -0.03 -0.06 0.12
(0.04) (0.05) (0.08)
[1.000] [1.000] [1.000]
Satisfaction with non-wage aspects of respondent’s main job (1-5 scale) 3.69 (1.12) 0.02 -0.01 0.14
(0.04) (0.06) (0.10)
[1.000] [1.000] [1.000]
Whether the respondent does not plan to leave their job in the next year 2.27 (0.72) -0.04 -0.05 0.07
(1-3 scale) (0.03) (0.04) (0.07)
[1.000] [1.000] [1.000]
Opportunities for promotion at the respondent’s main job (1-5 scale) 3.41 (1.27) -0.10* -0.09 0.08
(0.05) (0.07) (0.11)
[1.000] [1.000] [1.000]
Safety and health conditions at the respondent’s main job (1-5 scale) 4.22 (0.79) 0.02 0.01 0.03
(0.03) (0.04) (0.07)
[1.000] [1.000] [1.000]
Whether a scheduled shift was canceled with less than 24 hours notice in 0.09 (0.26) 0.02* 0.02 0.03
the last month (0.01) (0.01) (0.02)
[1.000] [1.000] [1.000]
Number of stressors in their work environment at respondent’s main job 1.25 (1.24) 0.09* 0.07 0.09
(0.05) (0.06) (0.11)
[1.000] [1.000] [1.000]
How easy is it to take time off from the respondent’s main job? (1-4 scale) 3.18 (0.81) -0.06* -0.07* -0.09
(0.03) (0.04) (0.07)
[1.000] [1.000] [1.000]
Notes: This table compares item-level results for quality of employment by participants’ baseline level of education. *
p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

99
 Table A22: Robustness checks for Impact of Guaranteed Income on Annual Earned and Other Unearned Income (in $1,000s)

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Total household income -4.1*** -4.6*** -4.2*** -4.1*** -4.1*** -5.4*** -3.8***
(self-reported) (1.0) (1.3) (1.0) (1.0) (1.0) (0.9) (1.0)
Total individual income -1.5* -1.8* -2.4*** -1.4* -1.5* -3.1*** -1.1
(0.9) (1.1) (0.9) (0.9) (0.9) (0.8) (0.9)
Total individual income -2.5** -2.5** -3.1** -3.3*** -2.5** -4.5*** -1.6*
(self-reported) (1.0) (1.0) (1.3) (0.9) (1.0) (0.9) (1.0)
Individual salaried/wage -1.1 -1.7* -1.3 -1.1 -1.1 -2.2*** -1.0
income (0.8) (1.0) (0.9) (0.8) (0.8) (0.7) (0.8)
Self-employment income -0.1 0.1 -0.0 -0.1 -0.1 -1.2*** 0.0
(0.5) (0.6) (0.0) (0.5) (0.5) (0.4) (0.5)

100
Income from gig work -0.1 -0.1 N/A -0.1 -0.1 -0.2*** -0.1
(0.0) (0.1) (.) (0.1) (0.0) (0.0) (0.0)
Passive income 0.0 0.0 N/A 0.0 0.0 -0.0 0.0
(0.0) (0.0) (.) (0.0) (0.0) (0.0) (0.0)
Other income -0.1 -0.2 -0.1 -0.1 -0.1 -0.3* -0.1
(0.2) (0.2) (0.1) (0.2) (0.2) (0.2) (0.2)
Government transfers -0.2 -0.2 N/A -0.2 -0.2 -0.3** -0.1
(0.1) (0.2) (.) (0.1) (0.1) (0.1) (0.1)

Notes: This table presents robustness checks for the estimates of impact on income, using survey data. The columns, in turn, present
the main estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences
approach; results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness
check can necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those
questions asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases
in which there is a binary dependent variable.
 Table A23: Robustness checks for Impact of Guaranteed Income on Employment

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Labor Supply Elasticity Index -0.06** -0.08** -0.06** -0.05** -0.06** -0.07*** -0.05*
(0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03)
Labor Supply Elasticity -0.06** -0.08** -0.06** -0.05** -0.06** -0.07*** -0.05*
(0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03)
Whether the respondent is employed -0.02* -0.03* N/A -0.03 -0.02* -0.02* -0.02
(0.01) (0.02) (.) (0.02) (0.01) (0.01) (0.01)
Hours worked per week -1.28** -1.77** N/A -1.28** -1.28** -1.87*** -1.07*
(0.64) (0.80) (.) (0.64) (0.64) (0.62) (0.64)
Number of other household members which are -0.02 -0.02 N/A 0.03 -0.02 -0.04** -0.02
employed (0.02) (0.02) (.) (0.02) (0.02) (0.02) (0.02)

101
Total number of hours participant and -2.16*** -2.42** N/A -2.49*** -2.16*** -2.60*** -1.95**
spouse/partner works per week (0.78) (1.00) (.) (0.85) (0.78) (0.77) (0.78)
Total number of hours all household members -2.21** -2.94** -1.67 -2.21** -2.21** -2.81*** -2.01**
(including the participant) work per week (0.92) (1.17) (1.13) (0.92) (0.92) (0.90) (0.92)
Total number of hours participant’s parents in -0.13 -0.21 N/A -0.34 -0.13 -0.68** -0.12
household work per week (0.35) (0.46) (.) (0.39) (0.35) (0.30) (0.35)
Total number of hours participant’s adult children 0.30 0.30 N/A 0.07 0.30 -0.23 0.32
in household work per week (0.29) (0.29) (.) (0.27) (0.29) (0.22) (0.29)

Notes: This table presents robustness checks for the estimates of impact on employment outcomes, using survey data. The columns, in
turn, present the main estimate; a version run without any covariates; results from median regression; results from using a difference-
in-differences approach; results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not
every robustness check can necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Mid-
line/Endline") for those questions asked only on web-based surveys. Additionally, median regression will occasionally not converge, and
we do not run it in cases in which there is a binary dependent variable.
 Table A24: Robustness checks for Impact of Guaranteed Income on Enumerated and Quarterly Time Use

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Finances Hrs/Mo 0.31* 0.44** 0.33*** 0.31* 0.41** 0.22 0.32*
(0.17) (0.20) (0.11) (0.17) (0.20) (0.16) (0.17)
Helping Hrs/Mo 0.38 0.42 0.37*** 0.38 0.37 0.13 0.40
(0.26) (0.28) (0.14) (0.26) (0.31) (0.24) (0.26)
Medical Hrs/Mo 0.35 0.82 0.07 0.35 0.59 -0.01 0.35
(0.63) (0.76) (0.06) (0.63) (0.70) (0.60) (0.63)
Meetings Hrs/Mo -0.01 0.03 -0.00 -0.01 0.01 -0.07 -0.01
(0.07) (0.08) (0.01) (0.07) (0.09) (0.07) (0.08)
Religion Hrs/Mo 0.05 0.17 -0.01 0.05 0.13 -0.02 0.05
(0.12) (0.16) (0.02) (0.12) (0.14) (0.12) (0.12)
Childcare Hrs/Wk -0.51 -0.68 0.06 -0.51 -0.69 -0.95 -0.46
(0.88) (1.41) (0.24) (0.88) (1.01) (0.86) (0.88)
Chores Hrs/Wk -0.21 0.01 0.18 -0.21 -0.09 -0.36 -0.17
(0.27) (0.33) (0.25) (0.27) (0.32) (0.27) (0.27)

102
Communicating Hrs/Wk -0.16 -0.15 0.01 -0.16 -0.26 -0.41 -0.11
(0.35) (0.40) (0.20) (0.35) (0.43) (0.34) (0.35)
Commuting Hrs/Wk -0.05 -0.00 -0.10 -0.05 0.06 -0.16 -0.04
(0.14) (0.17) (0.12) (0.14) (0.18) (0.13) (0.14)
Education Hrs/Wk -0.03 0.13 -0.01 -0.03 -0.01 -0.19 -0.01
(0.22) (0.25) (0.10) (0.22) (0.26) (0.21) (0.22)
Eldercare Hrs/Wk -0.13 -0.07 -0.01 -0.13 -0.24 -0.32 -0.13
(0.21) (0.25) (0.01) (0.21) (0.25) (0.19) (0.21)
Entertainment Hrs/Wk -0.01 0.13 -0.05 -0.01 0.06 -0.22 0.04
(0.36) (0.45) (0.36) (0.36) (0.42) (0.35) (0.36)
Family Hrs/Wk -0.66 -0.79 0.03 -0.66 -1.25 -1.01 -0.62
(0.73) (0.93) (0.70) (0.73) (0.87) (0.72) (0.73)
Friends Hrs/Wk 0.09 0.28 -0.02 0.09 0.09 -0.05 0.11
(0.26) (0.31) (0.22) (0.26) (0.33) (0.25) (0.27)
Hobbies Hrs/Wk -0.05 -0.00 -0.06 -0.05 -0.08 -0.19 -0.04
(0.15) (0.17) (0.09) (0.15) (0.18) (0.14) (0.15)
Reading Hrs/Wk -0.12 -0.02 -0.05 -0.12 -0.29 -0.23 -0.09
(0.19) (0.22) (0.14) (0.19) (0.23) (0.19) (0.19)
 Recreation Hrs/Wk -0.47*** -0.39** -0.37** -0.47*** -0.57*** -0.61*** -0.45**
(0.18) (0.20) (0.15) (0.18) (0.22) (0.17) (0.18)
Sleeping Hrs/Wk 0.19 0.49 0.15 0.19 -0.09 0.06 0.33
(0.37) (0.47) (0.44) (0.37) (0.43) (0.37) (0.37)
Working Hrs/Wk -1.40*** -1.67** N/A -1.40*** -1.59*** -1.57*** -1.32***
(0.50) (0.66) (.) (0.50) (0.58) (0.50) (0.50)
Vacation Days/Yr 0.12 0.15 0.03 0.12 -0.12 -0.01 0.14
(0.31) (0.37) (0.22) (0.31) (0.32) (0.30) (0.31)
Volunteer Hrs/Yr 0.35 1.70 0.01 0.35 1.00 -1.03 0.38
(1.48) (1.73) (0.21) (1.48) (1.76) (1.37) (1.48)

Notes: This table presents robustness checks for the estimates of impact on time use from the enumerated and quarterly time use surveys.
The columns, in turn, present the main estimate; a version run without any covariates; results from median regression; results from using a
difference-in-differences approach; results restricting attention to administrative data or data from the enumerated surveys; and the lower
and upper Lee bound. Not every robustness check can necessarily be run for every item: we cannot restrict attention to administrative
data or results from enumerated surveys ("Midline/Endline") for those questions asked only on web-based surveys. Additionally, median
regression will occasionally not converge, and we do not run it in cases in which there is a binary dependent variable.

103
 Table A25: Robustness checks for Impact of Guaranteed Income on Mobile App-Based Time Use

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Caring for others Min/Day -1.03 -2.09 N/A -1.12 N/A -3.88*** -0.97
(1.18) (1.30) (.) (1.18) (.) (0.90) (1.20)
Childcare Min/Day -3.94 -4.13 N/A -3.99 N/A -15.11*** -3.55
(4.31) (6.08) (.) (4.32) (.) (3.63) (4.39)
Community Engagement Min/Day -1.17 -0.44 0.00 -1.04 N/A -3.44*** -1.03
(0.89) (1.01) (0.02) (0.90) (.) (0.72) (0.90)
Exercise Min/Day -0.12 -0.07 0.29 -0.17 N/A -2.76*** 0.19
(0.87) (1.04) (0.24) (0.88) (.) (0.64) (0.88)
Home Production Min/Day 2.52 4.51 4.12 2.69 N/A -4.77 5.70*
(3.43) (4.10) (3.15) (3.48) (.) (3.08) (3.42)
Market Work Min/Day -8.60* -12.20* -9.59* -8.78* N/A -17.77*** -5.68
(5.20) (6.57) (5.68) (5.23) (.) (4.88) (5.27)
Non-Commuting Transportation Min/Day 4.84*** 5.34*** N/A 4.80*** N/A 0.71 5.67***
(1.50) (1.66) (.) (1.51) (.) (1.18) (1.52)

104
Other Min/Day 5.99** 5.21* 1.44*** 5.99** N/A -2.20 6.53**
(2.59) (2.97) (0.50) (2.59) (.) (1.92) (2.63)
Other Income Min/Day -2.79** -2.50** N/A -2.74** N/A -5.74*** -2.68**
(1.10) (1.15) (.) (1.10) (.) (0.86) (1.11)
Search for a job Min/Day -0.30 0.18 N/A -0.38 N/A -3.64*** -0.19
(1.03) (1.12) (.) (1.03) (.) (0.71) (1.05)
Self-care Min/Day 1.62 1.21 2.42** 1.62 N/A -1.37 2.56**
(1.24) (1.41) (1.22) (1.24) (.) (1.05) (1.26)
Self-Improvement Min/Day -0.41 -0.53 0.47 -0.57 N/A -6.16*** 0.30
(2.24) (2.78) (0.75) (2.30) (.) (1.89) (2.27)
Sleep Min/Day -7.55* -4.52 -4.80 -8.92** N/A -13.20*** -0.34
(3.99) (5.05) (3.56) (4.28) (.) (3.87) (3.73)
Social Leisure Min/Day 5.85 5.27 5.07 5.82 N/A -1.58 9.17**
(3.92) (4.65) (4.02) (3.96) (.) (3.67) (3.93)
Solo Leisure Min/Day 3.61 4.87 1.04 3.07 N/A -3.23 5.17
(3.34) (5.16) (2.78) (3.35) (.) (3.03) (3.39)
Time with Others Min/Day 0.59 -1.07 3.13 0.59 N/A -13.35** 4.33
 (6.29) (8.34) (5.50) (6.29) (.) (5.63) (6.35)

Notes: This table presents robustness checks for the estimates of impact on time use from the mobile app-based time diaries. The columns,
in turn, present the main estimate; a version run without any covariates; results from median regression; results from using a difference-
in-differences approach; results restricting attention to administrative data or data from the enumerated surveys; and the lower and upper
Lee bound. Not every robustness check can necessarily be run for every item: we cannot restrict attention to administrative data or results
from enumerated surveys ("Midline/Endline") for those questions asked only on web-based surveys. Additionally, median regression will
occasionally not converge, and we do not run it in cases in which there is a binary dependent variable.

105
 Table A26: Robustness checks for Impact of Guaranteed Income on Duration of Unemployment

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Duration of Unemployment Index -0.10*** -0.10** -0.10*** N/A -0.10*** -0.11*** -0.07*
(0.04) (0.04) (0.04) (.) (0.04) (0.04) (0.04)

106
Notes: This table presents robustness checks for the estimates of impact on duration of unemployment. The columns, in turn, present
the main estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences
approach; results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness
check can necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those
questions asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases
in which there is a binary dependent variable.
 Table A27: Robustness checks for Impact of Guaranteed Income on Employment Preferences and Job Search

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Employment Preferences and Job Search 0.02 0.02 0.02 0.01 0.04 0.00 0.04**
Index (0.02) (0.02) (0.02) (0.01) (0.02) (0.02) (0.02)
Active Search 0.03 0.03 0.03 0.01 N/A 0.00 0.04*
(0.02) (0.03) (0.02) (0.02) (.) (0.02) (0.02)

107
Preferences for Employment 0.01 0.01 0.01 0.01 0.01 -0.00 0.04
(0.02) (0.03) (0.02) (0.02) (0.02) (0.02) (0.02)

Notes: This table presents robustness checks for the estimates of impact on employment preferences and job search. The columns, in
turn, present the main estimate; a version run without any covariates; results from median regression; results from using a difference-
in-differences approach; results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not
every robustness check can necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Mid-
line/Endline") for those questions asked only on web-based surveys. Additionally, median regression will occasionally not converge, and
we do not run it in cases in which there is a binary dependent variable.
 Table A28: Robustness checks for Impact of Guaranteed Income on Selectivity of Job Search

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Selectivity of Job Search Index -0.00 -0.02 -0.00 0.01 0.02 -0.08*** 0.08***
(0.02) (0.03) (0.02) (0.02) (0.06) (0.02) (0.02)
Selectivity -0.00 -0.02 -0.00 0.01 N/A -0.08*** 0.08***

108
(0.02) (0.03) (0.02) (0.02) (.) (0.02) (0.02)

Notes: This table presents robustness checks for the estimates of impact on selectivity of job search. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A29: Robustness checks for Impact of Guaranteed Income on Quality of Employment

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Quality of Employment Index -0.01 -0.02 -0.01 -0.01 -0.01 -0.04** 0.01
(0.01) (0.02) (0.01) (0.01) (0.01) (0.01) (0.01)
Adequacy of Employment 0.01 -0.01 0.01 0.01 0.01 -0.03 0.01
(0.03) (0.03) (0.03) (0.02) (0.03) (0.03) (0.03)
Employment Quality -0.01 -0.03 -0.01 -0.01 -0.01 -0.02 0.01
(0.02) (0.03) (0.02) (0.02) (0.02) (0.02) (0.02)
Whether the respondent reports working any -0.00 0.01 N/A -0.00 -0.00 0.01 -0.00
informal job (0.01) (0.02) (.) (0.01) (0.01) (0.01) (0.01)

109
Average hourly income from all jobs, -0.18 -0.18 -0.24 -0.18 -0.18 -0.36 0.13
weighted by hours worked at each job (0.37) (0.44) (0.37) (0.37) (0.37) (0.37) (0.37)
Stability of Employment -0.02 -0.01 -0.02 -0.01 -0.02 -0.06*** 0.00
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
Quality of Work Life -0.02 -0.02 -0.02 -0.02 -0.02 -0.04** -0.00
(0.02) (0.02) (0.02) (0.01) (0.02) (0.02) (0.02)

Notes: This table presents robustness checks for the estimates of impact on quality of employment. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A30: Robustness checks for Impact of Guaranteed Income on Entrepreneurship

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Entrepreneurship Index 0.05*** 0.07*** 0.05*** 0.04*** N/A 0.04*** 0.05***
(0.02) (0.02) (0.02) (0.01) (.) (0.02) (0.01)
Entrepreneurial Orientation 0.07*** 0.10*** 0.07*** 0.06*** N/A 0.07*** 0.08***
(0.02) (0.03) (0.02) (0.02) (.) (0.02) (0.02)
Entrepreneurial Intention 0.06** 0.08** 0.06** 0.06** N/A 0.05** 0.06***

110
(0.02) (0.03) (0.02) (0.02) (.) (0.02) (0.02)
Entrepreneurial Activity 0.01 0.04 0.01 0.01 N/A -0.00 0.02
(0.02) (0.03) (0.02) (0.02) (.) (0.02) (0.02)

Notes: This table presents robustness checks for the estimates of impact on entrepreneurship. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A31: Robustness checks for Impact of Guaranteed Income on Disability

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Disability Index -0.09*** -0.10** -0.09*** -0.09*** -0.09*** -0.10*** -0.07**
(0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03)
Disability -0.09*** -0.10** -0.09*** -0.09*** -0.09*** -0.10*** -0.07**

111
(0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03)

Notes: This table presents robustness checks for the estimates of impact on disability. The columns, in turn, present the main estimate;
a version run without any covariates; results from median regression; results from using a difference-in-differences approach; results
restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can necessarily
be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions asked only
on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which there is a
binary dependent variable.
 Table A32: Robustness checks for Impact of Guaranteed Income on Barriers to Employment

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Barriers to Employment Index -0.03 -0.03 -0.03 -0.02 -0.03 -0.03 0.03
(0.02) (0.03) (0.02) (0.02) (0.02) (0.02) (0.02)
Barriers to Employment -0.03 -0.03 -0.03 -0.02 -0.03 -0.03 0.03

112
(0.02) (0.03) (0.02) (0.02) (0.02) (0.02) (0.02)

Notes: This table presents robustness checks for the estimates of impact on barriers to employment. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A33: Robustness checks for Impact of Guaranteed Income on Human Capital

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Human Capital Index 0.02 0.02 N/A 0.02 N/A 0.01 0.02
(0.01) (0.01) (.) (0.01) (.) (0.01) (0.01)
Formal Education 0.02 0.04 0.02 0.03 -0.00 -0.00 0.03

113
(0.02) (0.03) (0.02) (0.02) (0.01) (0.02) (0.02)

Notes: This table presents robustness checks for the estimates of impact on human capital. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A34: Robustness checks for Impact of Guaranteed Income on Relationship Status

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Relationship Status Index -0.013 -0.002 -0.013 -0.027 0.006 -0.019 0.012
(0.017) (0.027) (0.017) (0.020) (0.019) (0.017) (0.016)
Relationship Stability -0.037 -0.028 -0.037 -0.065** N/A -0.048* 0.008
(0.025) (0.033) (0.025) (0.026) (0.025) (0.023)

114
Relationship Status 0.011 0.024 0.011 0.011 0.007 0.010 0.016
(0.016) (0.027) (0.016) (0.022) (0.016) (0.016) (0.016)

Notes: This table presents robustness checks for the estimates of impact on relationship status. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A35: Robustness checks for Impact of Guaranteed Income on Moving Labor Markets

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Move Labor Market Index 0.087*** 0.096*** 0.087*** 0.072*** 0.041 0.040 0.096***
(0.026) (0.029) (0.026) (0.022) (0.038) (0.025) (0.026)
Moved labor markets since 0.018 0.023* N/A N/A 0.013 0.017 0.019*
baseline (0.011) (0.012) (.) (.) (0.012) (0.011) (0.011)

115
Search new labor market 0.113*** 0.116*** 0.111*** 0.072*** N/A 0.025 0.128***
(0.034) (0.038) (0.034) (0.022) (0.030) (0.035)

Notes: This table presents robustness checks for the estimates of impact on moving labor markets. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A36: Robustness checks for Impact of Guaranteed Income on Labor Market Quality

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Labor Market Quality Index -0.01 0.00 -0.01 -0.01 N/A -0.01* -0.00
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Labor Quality 0.00 0.02 0.00 0.00 N/A -0.00 0.00
(0.01) (0.02) (0.01) (0.02) (0.01) (0.01)

116
Labor Market Amenities -0.02 -0.02* -0.02 -0.02 N/A -0.03** -0.01
(0.01) (0.01) (0.01) (0.02) (0.01) (0.01)

Notes: This table presents robustness checks for the estimates of impact on labor market quality. The columns, in turn, present the main
estimate; a version run without any covariates; results from median regression; results from using a difference-in-differences approach;
results restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can
necessarily be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions
asked only on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which
there is a binary dependent variable.
 Table A37: Robustness checks for Impact of Guaranteed Income on Benefits

Main No Median Diff-in-Diff Midline/ Lower Upper

Estimate Covariates Regression Endline Lee Bound Lee Bound
Benefits Index -0.02 -0.03 -0.02 -0.02 -0.02 -0.05** -0.02
(0.02) (0.04) (0.02) (0.03) (0.02) (0.02) (0.02)
Take-Up Benefits -0.02 -0.03 -0.02 -0.02 -0.02 -0.05** -0.02

117
(0.02) (0.04) (0.02) (0.03) (0.02) (0.02) (0.02)

Notes: This table presents robustness checks for the estimates of impact on benefits. The columns, in turn, present the main estimate;
a version run without any covariates; results from median regression; results from using a difference-in-differences approach; results
restricting attention to data from the enumerated surveys; and the lower and upper Lee bound. Not every robustness check can necessarily
be run for every item: we cannot restrict attention to results from enumerated surveys ("Midline/Endline") for those questions asked only
on web-based surveys. Additionally, median regression will occasionally not converge, and we do not run it in cases in which there is a
binary dependent variable.
Table A38: Impact of Guaranteed Income on Employment: Reasons for Not Working

Control Mean Treatment Effect N

Not working due to inability to find child care 0.07 (0.22) 0.01 0.38 2941
(0.01)
[0.80]
Not working due to attending school 0.04 (0.15) 0.01 0.18 2941
(0.01)
[0.88]
Not working due to caring for elderly 0.02 (0.13) 0.01 0.76 2941
(0.01)
[0.81]
Not working due to have given up looking for work 0.04 (0.17) -0.01 -1.64 2941
(0.01)
[0.80]
Not working due to illness 0.07 (0.23) 0.01 0.36 2941
(0.01)
[0.80]
Not working due to lack in necessary skills 0.08 (0.24) 0.00 0.14 2941
(0.01)
[1.00]
Not working due to other reasons 0.06 (0.19) 0.01 0.13 2941
(0.01)
[0.88]
Not working due to personal or family responsibilities 0.13 (0.29) 0.01 0.13 2941
(0.01)
[1.00]
Not working due to preferring to stay at home 0.09 (0.26) 0.00 0.10 2941
(0.01)
[1.00]
Not working due to lack in transportation to/from work 0.06 (0.20) 0.01 0.29 2941
(0.01)
[0.88]
Not working due to suitable work being unavailable 0.13 (0.29) 0.01 0.16 2941
(0.01)
[0.94]

Notes: This table provides exploratory analysis of self-reported reasons participants provided for
why they were not working. As usual, unconditional estimates are presented for the sake of main-
taining the causal interpretation of the estimate, so if someone is employed they would be treated
as having answered no to a question. These questions were secondary items in the Labor Supply
family and have been adjusted for multiple hypothesis testing accordingly. * p < 0.1, ** p < 0.05,
*** p < 0.01; † refers to comparable q-value thresholds.

118
Table A39: Impact of Guaranteed Income on Employment: Second/Third/Fourth Jobs

Control Mean Treatment Effect N

Whether the respondent has a second job 0.20 (0.35) -0.01 -0.11 2939
(0.01)
[1.00]
Whether the respondent has a third job 0.06 (0.20) 0.00 0.01 2939
(0.01)
[1.00]
Whether the respondent has a fourth job 0.02 (0.10) -0.01 -0.84 2939
(0.00)
[0.70]
Hours per week worked at 1st job 27.27 (17.98) -1.31** -0.14 2939
(0.57)
[0.38]
Hours per week worked at 2nd job 2.41 (5.69) -0.08 -0.10 2937
(0.22)
[1.00]
Hours per week worked at 3rd job 0.49 (2.37) -0.01 -0.04 2938
(0.09)
[1.00]
Hours per week worked at 4th job 0.10 (0.94) -0.03 -0.53 2939
(0.03)
[1.00]
Hours per week worked at 1st job (conditional on having 1st 36.39 (12.95) -0.97* -0.06 2404
job) (0.52)
[0.46]
Hours per week worked at 2nd job (conditional on having 12.88 (11.48) -0.09 -0.02 795
2nd job) (0.80)
[1.00]
Hours per week worked at 3rd job (conditional on having 8.94 (8.23) -0.28 -0.08 259
3rd job) (1.06)
[1.00]
Hours per week worked at 4th job (conditional on having 4th 7.78 (7.19) -1.58 -0.29 58
job) (1.81)
[1.00]
Maximum number of hours worked in a typical week 33.12 (21.68) -1.64** -0.15 2940
(0.68)
[0.47]
Minimum number of hours worked in a typical week 22.18 (16.70) -0.69 -0.10 2940
(0.52)
[0.80]

Notes: This table provides exploratory analysis of impacts on whether participants reduced hours
in particular at first/second/third/fourth jobs. As usual, unconditional estimates are presented for
the sake of maintaining the causal interpretation of the estimate, so for example if someone does
not have a third job they would be coded as working 0 hours at that third job. These questions
were secondary or exploratory post-pre-analysis plan items in the Labor Supply family and have
been adjusted for multiple hypothesis testing accordingly. * p < 0.1, ** p < 0.05, *** p < 0.01; †
refers to comparable q-value thresholds.

119
Table A40: Impact of Guaranteed Income on Employment Preferences and Job Search: Ac-
tions Taken to Search for Work

Control Mean Treatment Effect N

Whether participant looked at any job postings in the last 3 0.54 (0.39) 0.06***††† 2942
months (0.01)
[0.001]
Whether participant directly contacted any employers for a 0.36 (0.38) 0.03** 2942
job in the last 3 months (0.01)
[0.243]
Whether participant contacted any job centers in the last 3 0.28 (0.35) 0.01 2942
months (0.01)
[0.521]
Whether participant contacted friends or relatives to find 0.36 (0.37) 0.04***† 2942
work in the last 3 months (0.01)
[0.065]
Whether participant contacted professional network to find 0.22 (0.32) 0.00 2942
work in the last 3 months (0.01)
[0.845]
Whether participant posted a resume online in the last 3 0.38 (0.38) 0.02* 2942
months (0.01)
[0.267]
Whether participant took other actions to find work in the 0.03 (0.13) 0.01* 2942
last 3 months (0.01)
[0.250]

Notes: This table provides exploratory analysis of self-reported actions participants took to search
for work. As usual, unconditional estimates are presented for the sake of maintaining the causal
interpretation of the estimate, so if someone is not searching for work they would be treated as
having answered that they did not take that action. These questions were secondary items in the
Employment Preferences and Job Search family and have been adjusted for multiple hypothesis
testing accordingly. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

120
Table A41: Impact of Guaranteed Income on Employment Preferences and Job Search: Ad-
ditional Regressions

Control Mean Treatment Effect N

Whether the respondent is seeking a new, additional, or any 0.37 (0.40) 0.02 2939
job (alternate measure) (0.01)
[0.386]
Number of job applications sent (alternate measure) 5.76 (12.92) -0.25 2980
(0.43)
[0.475]
Number of job applications sent, conditional on having 11.47 (17.85) -2.16***††† 2488
applied for a job (0.61)
[0.009]
Number of jobs interviewed for, conditional on having 1.58 (2.63) -0.25***† 2491
interviewed for a job (0.09)
[0.055]
Whether the participant applied for a job that they were 0.37 (0.42) -0.01 2064
unqualified for (0.02)
[0.817]
Proportion of jobs the participant applied to that the 0.19 (0.29) -0.01 2064
participant was unqualified for (0.01)
[0.551]

Notes: This table provides exploratory analysis of the impact of the transfers on alternative mea-
sures of job search and/or the types of jobs that participants applied for. As usual, unconditional
estimates are presented for the sake of maintaining the causal interpretation of the estimate, so
if someone did not apply for a job they would be treated as having not applied for any jobs for
which they were unqualified. These questions were secondary or exploratory post-pre-analysis
plan items in the Employment Preferences and Job Search family and have been adjusted for mul-
tiple hypothesis testing accordingly. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable
q-value thresholds.

121
Table A42: Impact of Guaranteed Income on Selectivity of Job Search: Work Requirements

Control Mean Treatment Effect N

Work requirement: chances for advancement 0.73 (0.43) -0.00 965
(0.03)
[1.000]
Work requirement: comfortable workstation or physical environment 0.80 (0.38) 0.02 965
(0.02)
[1.000]
Work requirement: flexible hours 0.74 (0.41) 0.03 965
(0.02)
[1.000]
Work requirement: high income potential 0.78 (0.39) -0.01 964
(0.02)
[1.000]
Work requirement: interesting or meaningful work 0.70 (0.43) 0.06** 965
(0.03)
[0.286]
Work requirement: convenient location 0.81 (0.37) -0.02 964
(0.02)
[1.000]
Work requirement: secure, regular earnings 0.89 (0.29) -0.01 965
(0.02)
[1.000]
Work requirement: consistent, predictable schedule 0.81 (0.37) -0.02 965
(0.02)
[1.000]
Participant is not willing to work under any conditions 0.00 (0.04) 0.00 1106
(0.00)
[1.000]
Work requirement: other 0.21 (0.38) -0.01 966
(0.02)
[1.000]

Notes: This table provides exploratory analysis of self-reported requirements participants stated
that a job would have in order for them to be willing to take it. These questions were only asked of
those seeking a job and were secondary items in the Selectivity of Job Search family and have been
adjusted for multiple hypothesis testing accordingly. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to
comparable q-value thresholds.

122
 Table A43: Impact of Guaranteed Income on Quality of Employment: Specific Benefits

Control Mean Treatment Effect N

Receives health insurance (100% of premium covered by employer) 0.20 (0.37) -0.00 2065
(0.02)
[1.000]
Receives health insurance (Less than 100% of premium covered by employer) 0.39 (0.46) -0.02 2123
(0.02)
[1.000]
Receives dental and/or vision insurance 0.55 (0.47) -0.02 2165
(0.02)
[1.000]
Receives traditional pension plan (defined benefit plan) 0.31 (0.43) -0.01 2092
(0.02)
[1.000]
Receives retirement account without employer contribution 0.27 (0.41) -0.03* 2100
(0.02)
[1.000]
Receives employer contribution to a retirement account 0.34 (0.44) 0.02 2100
(0.02)
[1.000]
Receives health care or dependent care Flexible Spending Account 0.34 (0.45) -0.02 2108
(0.02)
[1.000]
Receives housing or housing subsidy 0.03 (0.16) -0.01 2017
(0.01)
[1.000]
Receives life or disability insurance 0.48 (0.47) -0.01 2153
(0.02)
[1.000]
Receives commuter benefits 0.12 (0.31) -0.02 2065
(0.01)
[1.000]
Receives childcare assistance 0.09 (0.26) -0.01 2030
(0.01)
[1.000]
Receives paid vacation 0.63 (0.45) 0.00 2193
(0.02)
[1.000]
Receives tuition reimbursement 0.31 (0.43) -0.02 2099
(0.02)
[1.000]
Can work from home 0.45 (0.48) -0.03 2200
(0.02)
[1.000]
Receives other non-wage benefit 0.15 (0.33) 0.00 2071
(0.01)
[1.000]

Notes: This table provides exploratory analysis of self-reported benefits participants reported re-
ceiving as part of their jobs. These questions were secondary items in the Quality of Employment
family and have been adjusted for multiple hypothesis testing accordingly. These questions were
only asked of people who were employed. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to compara-
ble q-value thresholds.

123
Table A44: Impact of Guaranteed Income on Human Capital: Programs and Fields of Study

Control Mean Treatment Effect N

Studied liberal arts in post secondary education 0.10 (0.30) -0.00 2932
(0.00)
[1.000]
Studied business in post secondary education 0.04 (0.20) 0.00 2932
(0.00)
[0.947]
Studied education in post secondary education 0.02 (0.14) -0.00 2932
(0.00)
[0.947]
Studied health in post secondary education 0.05 (0.22) -0.01* 2932
(0.00)
[0.829]
Studied social sciences in post secondary education 0.08 (0.26) -0.00 2932
(0.00)
[0.947]
Studied STEM in post secondary education 0.06 (0.22) 0.01 2932
(0.00)
[0.829]
Studied a vocational major in post secondary education 0.03 (0.17) 0.00 2932
(0.00)
[0.947]
Whether the participant has an Associate’s degree 0.15 (0.36) -0.01 2593
(0.00)
[1.000]
Whether the participant has a Bachelor’s degree 0.23 (0.42) -0.00 2593
(0.01)
[1.000]
Whether the participant has a Master’s or Doctoral degree 0.08 (0.26) -0.00 2593
(0.01)
[1.000]
Whether the participant has a Master’s degree 0.07 (0.25) -0.01 2593
(0.00)
[1.000]
Whether the participant has a Doctoral degree 0.02 (0.12) 0.00 2593
(0.00)
[1.000]

Notes: This table provides exploratory analysis of programs and fields of study that participants
pursued, according to the National Student Clearinghouse data. These questions were secondary
items in the Human Capital family and have been adjusted for multiple hypothesis testing accord-
ingly. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value thresholds.

124
Table A45: Impact of Guaranteed Income on Relationship Status: Reasons for Relationships
Ending

Control Mean All Surveys N

Whether relationship ended because of abuse 0.031 (0.128) 0.001 2903
(0.005)
[1.000]
Whether relationship ended because of distance 0.028 (0.120) 0.009* 2903
(0.005)
[1.000]
Whether relationship ended because of drugs 0.027 (0.120) 0.004 2903
(0.005)
[1.000]
Whether relationship ended because of family 0.019 (0.096) -0.000 2903
(0.003)
[1.000]
Whether relationship ended because of financial issues 0.033 (0.127) 0.006 2903
(0.005)
[1.000]
Whether relationship ended because of illness 0.006 (0.057) 0.001 2903
(0.002)
[1.000]
Whether relationship ended because of relationship issues 0.123 (0.241) 0.022** 2903
(0.010)
[0.386]
Whether relationship ended because of religion 0.005 (0.049) 0.004* 2903
(0.002)
[1.000]
Whether relationship ended because of other reasons 0.019 (0.094) 0.003 2903
(0.004)
[1.000]
Participant’s relationship was ended by participant 0.078 (0.200) 0.024*** 2903
(0.008)
[0.227]
Participant’s relationship was ended by partner 0.037 (0.137) 0.005 2903
(0.005)
[1.000]
Participant’s relationship ended mutually 0.045 (0.146) 0.002 2903
(0.006)
[1.000]

Notes: This table provides exploratory analysis of reasons why relationships ended. These ques-
tions were secondary items in the Relationship Status family and have been adjusted for multiple
hypothesis testing accordingly. * p < 0.1, ** p < 0.05, *** p < 0.01; † refers to comparable q-value
thresholds.

125
 Figure A1: Illinois Bill SB 1735

Notes: This figure provides a synopsis of the bill that was passed to protect benefits in Illinois.

126
 Figure A2: Response Rates Over Time

Notes: This figure shows response rates for the Qualtrics surveys and enumerated surveys over
time.

127
 Figure A3: Time Use Results: Enumerated and Quarterly Surveys

Time Use Outcomes − Enumerated and Quarterly Surveys

Working q=0.115
Family q=1
Childcare q=1
Recreation q=0.115
Chores q=1
Communicating q=1
Eldercare q=1
Read q=1
Hobbies q=1
Commuting q=1
Education q=1
Entertain q=1
Meetings q=1
Volunteer q=1
Religion q=1
Vacation q=1
Finances q=0.685
Medical q=1
Helping q=1
Friends q=1
Sleeping q=1
−20 −10 0 10
Minutes Per Day

Notes: This figure shows the results from the enumerated and quarterly time use surveys.

128
 Figure A4: Time Use Results: Mobile App - By Children in Household at Baseline

Time Use Outcomes

Mobile App
Children in Household No Children in Household
at Baseline at Baseline
Caring for Others q=1 q=0.475
Child Care q=0.91 q=0.861
Community Engagement q=0.28 q=0.94
Exercise q=0.857 q=0.966
Home Production q=0.968 q=0.923
Market Work q=1 q=0.085
Non−Commuting Transportation q=0.048 q=0.535
Other Activities q=0.038 q=1
Other Income Generating Activities q=0.097 q=0.606
Searching for a Job q=1 q=0.843
Self Care q=0.678 q=0.842
Self Improvement q=0.998 q=1
Sleeping q=0.068 q=0.981
Social Leisure q=0.252 q=1
Solitary Leisure q=1 q=0.3
Time with Others q=0.839 q=0.531
−25 0 25 −25 0 25
Minutes Per Day

Notes: This figure shows the results from the mobile phone app, by whether participants had
children in the household at baseline.

129
 Figure A5: Time Use Results: Mobile App (Time Spent With Others)

Time Use Outcomes − Mobile App

(Time With Others)
With My Spouse/Partner q=0.544

With My Parents q=0.779

With My Boss q=0.272

Alone q=0.975

With My Co−Workers q=0.46

With Other People q=0.74

With My Customers/Clients q=0.444

With My Children q=1

With My Friends q=0.952

With My Relatives q=0.756

−10 0 10 20
Minutes Per Day

Notes: This figure shows the results from the mobile phone app for time spent with others.

130
 Figure A6: Time Use Results: Mobile App (Raw Times)

Time Use Outcomes − Mobile App

(Raw Numbers)
Social Leisure q=0.828
Market Work q=0.309
Sleeping q=0.387
Child Care q=0.839
Other Income Generating Activities q=0.201
Caring for Others q=0.777
Community Engagement q=0.365
Self Improvement q=1
Exercise q=1
Searching for a Job q=0.924
Home Production q=1
Self Care q=0.483
Solitary Leisure q=0.527
Non−Commuting Transportation q=0.053
Other Activities q=0.251
−40 −20 0 20
Minutes Per Day

Notes: This figure shows the results from the mobile phone app, without adjusting for simultane-
ous activities.

131
 Figure A7: Time Use Results: Mobile App (ChatGPT-4 Recoded)

Time Use Outcomes − Mobile App

(GPT−4 Recategorized)
Sleeping q=0.172
Market Work q=0.366
Child Care q=0.853
Other Income Generating Activities q=0.117
Community Engagement q=0.526
Caring for Others q=0.778
Self Improvement q=1
Searching for a Job q=1
Exercise q=1
Other Activities q=0.562
Self Care q=0.387
Home Production q=0.703
Solitary Leisure q=0.435
Non−Commuting Transportation q=0.022
Social Leisure q=0.36
−20 −10 0 10 20
Minutes Per Day

Notes: This figure shows the results from the mobile phone app, using GPT to recode open-ended
responses.

132
 Figure A8: Labor, Consumption, and Asset Paths of Fitted Model

12000

9000

Asset Control
US Dollar

Asset Treatment
6000 Consumption Control
Consumption Treatment
Labor Supply Control
Labor Supply Treatment

3000

−1.25 0 1.25 2.5 3.75 5

Years since start of program

Notes: This figure shows the fitted labor supply, consumption, and net asset paths over time. A
few things are worth noting. First, labor and consumption values are presented per quarter, while
asset accumulation is cumulative. The light blue and green vertical lines represent the start and
end of the transfers, respectively. Assets continue to go up for one period after the end of the
transfers because the agent decided to save a portion of their transfer in the preceding period.

133
 Figure A9: Forecasts of Employment Outcomes
Employed, in percentage points Work hours per week
15
20

15
10

10
5
5

0 0
−20 −10 0 10 20 −10 −5 0 5 10

Treatment = −2.8 Mean = −0.2 Median = −0.5 Treatment = −1.6 Mean = −1.0 Median = −1.0

Average hourly wage Duration of non−employment, in weeks

15

30

10
20

5
10

0 0
−5 0 5 −10 0 10

Treatment = −0.2 Mean = 0.9 Median = 0.8 Treatment = 4.7 Mean = 3.0 Median = 2.9

Participant is searching for work, in percentage points Enrollment in a post−secondary program, in percentage points
6 20

15
4

10

2
5

0 0
−10 −5 0 5 10 15 −10 0 10

Treatment = 7.2 Mean = −2.5 Median = −2.8 Treatment = −0.0 Mean = 4.0 Median = 3.4

Individual salaried income, in thousands of dollars

10

0
−5 0 5 10

Treatment = −2.6 Mean = 0.9 Median = 0.0

Notes: These figures show the full distribution of forecasts provided by NBER affiliates and users
of the Social Science Prediction Platform. A few rare outliers more than two SD from the mean are
omitted from these figures for the sake of legibility.

134
 Figure A10: Forecasts of Time Use Outcomes
Home production, hours per week Sleep, hours per week
25
15
20

10 15

10
5
5

0 0
−50 −25 0 25 −30 −20 −10 0 10 20

Treatment = 0.8 Mean = 2.2 Median = 1.2 Treatment = −0.7 Mean = 0.6 Median = 0.7

Social leisure, hours per week Solitary leisure, hours per week

20 15

15
10
10

5
5

0 0
−40 −20 0 20 −20 0 20 40

Treatment = 0.9 Mean = 4.8 Median = 4.4 Treatment = 0.5 Mean = 3.2 Median = 2.8

Notes: These figures show the full distribution of forecasts provided by NBER affiliates and users
of the Social Science Prediction Platform. A few rare outliers more than two SD from the mean are
omitted from these figures for the sake of legibility.

135
 Figure A11: Histogram of Treatment Assignment Probabilities

Note: This graph displays a frequency distribution of participants’ average treatment assignments, based on 1,000
simulated runs of the assignment process. The vertical line on the graph is positioned at 0.33333, representing the 1 in 3
probability of assignment.

136
 Figure A12: QQ-plot of Treatment Probability against Bernoulli Distribution

Note: This graph compares the actual distribution of treatment assignments with the theoretical distribution expected from
a random assignment process where each participant has a one in three chance of being assigned to the treatment group.
The x-axis shows the quantiles of the observed treatment assignments, while the y-axis represents the quantiles of the
expected distribution under random assignment. A Kolmogorov-Smirnov test was conducted to compare these
distributions. The test result (p=0.5226) indicates that there is not sufficient evidence to conclude that the observed
distribution differs significantly from what would be expected by chance.

137
 Figure A13: Results for Income by Time Period

Income

Total
Total household −2500
individual
income
income
−5000

Estimated Effect in Dollars

Total
Individual
individual −2500
salaried/wage
income,
income
self−reported
−5000

Self−employment Income from gig −2500

income work
−5000

Passive income Other income −2500

−5000

e e d e e d
lin dlin ole lin dlin ole
id id
M En Po M En Po

Notes: This figure plots the results for treatment effects on income over time from the survey data,
showing a clear time trend in the major categories of income.

138
 Figure A14: Results for Employment by Time Period

Employment

Estimated Effect in Raw Units

0.02

0.00

Whether the
respondent is
−0.02
employed

−0.04

−0.06
e e ed
lin d lin ol
id
M En Po

Work Hours

in Hours Worked
Estimated Effect
Hours worked −1
per week

−2

−3

e e d
in in le
id
l dl o
M En Po

Notes: This figure plots the results for employment over time from the survey data, showing
treatment effects on employment trending more negative towards the end of the study.

139
 Figure A15: Results for Income and Employment Over Time: Quarterly UI Data

Notes: This figure plots the results for income and employment over time, using data from UI data
in each state for those who could be matched. The data points in this figure represent estimated
effects on individual salaried income or employment for the preceding quarter and are formed via
regressions within each quarter (i.e., the value for the treatment group is the estimated treatment
effect added to the constant term, with confidence intervals). No controls are included in these
regressions. The last three quarters use data for Texas only as administrative data for the end of
the study has yet to be made available in Illinois.

140
 Figure A16: Results for Employment Over Time: Quarterly Survey Data

Notes: This figure plots the results for employment over time, using survey data. Employment his-
tory is constructed from a several different survey questions. First, the baseline/midline/endline
are used as tentpoles, as participants are asked which months they were employed over the past
year. Second, we fill in the gaps using data from the more frequent web-based surveys. The
data points in this figure represent whether participants were ever employed during the preced-
ing quarter and are formed via regressions within each quarter (i.e., the value for the treatment
group is the estimated treatment effect added to the constant term, with confidence intervals). No
controls are included in these regressions.

141
 Figure A17: Results for Time Use by Time Period: Mobile App (1)

Time Use: Mobile App

10

0
Caring for
Child Care
Others
−10

−20

10

Estimated Effect in Minutes per Day

0
Community Non−Commuting
Engagement Transportation
−10

−20

10

0
Exercise Home Production
−10

−20

10

0
Search for a
Market Work
Job
−10

−20

1 r2 r3 d r1 r2 r3 d
ar a a ole a a a ole
Ye Ye Ye Po Ye Ye Ye Po

Notes: This figure plots the results for time use over time, using data from the mobile app.

142
 Figure A18: Results for Time Use by Time Period: Mobile App (2)

Time Use: Mobile App

20

10

Other Other Income 0

−10

−20

20

Estimated Effect in Minutes per Day

10

Self−Care Self−Improvement 0

−10

−20

20

10

Sleep Social Leisure 0

−10

−20

20

10

Time with
Solo Leisure 0
Others

−10

−20

ar
1
ar2 ar3 l ed ar1 ar2 ar3 l ed
o o
Ye Ye Ye Po Ye Ye Ye Po

Notes: This figure plots the results for time use over time, using data from the mobile app.

143
Figure A19: Results for Time Use by Time Period: Enumerated and Quarterly Surveys (1)

Time Use: Enumerated and Quarterly Surveys

0.25

Finances Helping 0.00

−0.25

0.25

Medical Meetings 0.00

−0.25

Estimated Effect in Hours per Day

0.25

Religion Childcare 0.00

−0.25

0.25

Chores Communicating 0.00

−0.25

0.25

Commuting Education 0.00

−0.25

e e d e e d
ar1 ar
2
lin ar3 dl
in ole ar1 ar2 lin ar3 dl
in ole
Ye Ye id Ye Ye Ye id Ye
M En Po M En Po

Notes: This figure plots the results for time use over time, using data from enumerated and quar-
terly surveys.

144
Figure A20: Results for Time Use by Time Period: Enumerated and Quarterly Surveys (2)

Time Use: Enumerated and Quarterly Surveys

0.4

0.2

0.0
Eldercare Entertainment
−0.2

−0.4

0.4

0.2

0.0
Family Friends
−0.2

−0.4

0.4

Estimated Effect in Hours per Day

0.2

0.0
Hobbies Reading
−0.2

−0.4

0.4

0.2

0.0
Recreation Sleeping
−0.2

−0.4

0.4

0.2

0.0
Working Vacation
−0.2

−0.4

0.4 e e
r1 r2 r3 d
a a lin a dlin ole
Ye Ye id Ye
0.2 M En Po

0.0
Volunteer
−0.2

−0.4

1 2 e e d
ar ar lin ar3 dl
in ole
Ye Ye id Ye
M En Po

Notes: This figure plots the results for time use over time, using data from enumerated and quar-
terly surveys.

145
 Figure A21: Results for Duration of Unemployment by Time Period

Duration of Unemployment

Estimated Effect in Months

2.0

1.5
Average length
of continuous
spells of
non−employment 1.0
in months,
over the study
duration 0.5

0.0

i ne
dl
En

Notes: This figure plots the results of the estimates of the transfers on duration of unemployment
over time.

146
 Figure A22: Results for Employment Preferences and Job Search by Time Period

Employment Preferences and Job Search

0.2

Whether the
Whether the 0.1
respondent is
participant
seeking a new,
searched for a
additional, or 0.0
job
any job

−0.1

0.2

Number of 0.1
different Number of job

Estimated Effect in SD
actions taken applications
to search for a sent 0.0
job

−0.1

0.2

How many work 0.1

hours the
Number of jobs
respondent
interviewed for
wants (less, 0.0
same, more)

−0.1

e e d
ar1 ar2 lin ar3 dl
in ole
0.2 Ye Ye id Ye
M En Po
Whether a
respondent is 0.1
employed or,
if unemployed,
would prefer to 0.0
be working
−0.1

1 e e d
ar ar2 lin ar3 dl
in ole
Ye Ye id Ye
M En Po

Notes: This figure plots the results of the estimates of the transfers on employment preferences
and job search over time.

147
 Figure A23: Results for Selectivity of Job Search by Time Period

Job Search and Selectivity

0.2

0.1

Natural log of
Whether the
average income 0.0
respondent is
of jobs which
willing to take
the respondent
any job offered

Estimated Effect in SD
applied to −0.1

−0.2

0.2

0.1

If searching Number of
for a job, how sacrifices
0.0
long respondent participants
is willing would be
to search in willing to make
months to secure a job −0.1

−0.2

e e d e e d
ar1 ar2 lin ar3 dl
in ole ar1 ar2 lin ar3 dl
in ole
Ye Ye id Ye Ye Ye id Ye
M En Po M En Po

Notes: This figure plots the results of the estimates of the transfers on selectivity of job search over
time.

148
 Figure A24: Results for Adequacy of Employment by Time Period

Quality of Employment: Adequacy of Employment

The respondent 0.15
The respondent

Estimated Effect in SD
is employed 0.10
would prefer
part−time in 0.05
to work more
their main 0.00
hours in their
job and would −0.05
current main
prefer to work
job −0.10
full−time
0.15 e e
lin in ed
0.10 id dl ol
The number of M En Po
jobs held by 0.05
the respondent 0.00
apart from −0.05
their main job
−0.10

e e ed
lin in ol
id dl
M En Po

Notes: This figure plots the results of the estimates of the transfers on adequacy of employment
over time.

149
 Figure A25: Results for Employment Quality by Time Period

Quality of Employment: Employment Quality

0.10
Is training
Is training offered 0.05
offered by the during work
respondent's hours by the 0.00
main employer respondent's −0.05
main employer
−0.10

Estimated Effect in SD
Number of 0.10
non−wage
Is formal 0.05
benefits at
training
respondent's
offered by the 0.00
job(s),
respondent's
weighted by −0.05
main employer
hours worked at
each job −0.10

e e ed
lin in ol
Whether the id dl
0.10 M En Po
respondent
must work an 0.05
irregular shift
at each job, 0.00
weighted by −0.05
hours worked at
each job −0.10

e e ed
lin in ol
id dl
M En Po

Notes: This figure plots the results of the estimates of the transfers on employment quality over
time.

150
 Figure A26: Results for Informality and Hourly Wage by Time Period

Quality of Employment: Informality of Employment

Estimated Effect in Raw Units

0.06

0.03
Whether the
respondent
reports working 0.00
any informal
job
−0.03

−0.06
e e ed
in in ol
i dl dl
M En Po

Quality of Employment: Hourly Wage

Estimated Effect in Dollars

0.5

Average hourly 0.0

income from all
jobs, weighted
by hours worked
at each job −0.5

−1.0

e e ed
lin in ol
id dl
M En Po

Notes: This figure plots the results of the estimates of the transfers on informality and hourly wage
over time.

151
 Figure A27: Results for Stability of Employment by Time Period

Quality of Employment: Stability of Employment

How long the
respondent has 0.1
How many months spent at their
the respondent current main
has been job and other 0.0
employed in the jobs (months),
past year weighted by
hours worked at −0.1
each job

Estimated Effect in SD
0.1
How many jobs Whether the
the respondent respondent's
has held in the main job is a 0.0
past 12 months temp job
−0.1

Whether the
Whether
respondent 0.1
each of the
is performing
respondent's
contract or
jobs is
freelance work 0.0
salaried,
at each job,
weighted by
weighted by
hours worked at
hours worked at −0.1
each job
each job

e e ed e e ed
lin in ol lin in ol
id dl id dl
M En Po M En Po

Notes: This figure plots the results of the estimates of the transfers on employment stability over
time.

152
 Figure A28: Results for Quality of Work Life by Time Period (1)

Quality of Employment: Quality of Work Life

0.1
The work
Advance notice
activities
of schedule
are not
provided at the
boring at the 0.0
respondent's
respondent's
main job (1−4
main job (1−5
scale)
scale)
−0.1

0.1
Satisfaction
with Whether the
compensation respondent
at the faces age 0.0
respondent's discrimination
main job (1−5 at work
scale)
−0.1

Estimated Effect in SD
0.1
Whether the
Whether the
respondent
respondent
faces racial
faces sex 0.0
or ethnic
discrimination
discrimination
at work
at work
−0.1

0.1
Whether the
respondent Whether job
experienced demands do not
fair treatment interfere with 0.0
by their family life
supervisor (1−5 (1−4 scale)
scale)
−0.1

0.1
Whether the
job is a good Flexibility of
fit with the schedule at the
respondent's respondent's 0.0
experience and main job (1−4
skills (1−5 scale)
scale)
−0.1

e e d e e ed
lin in le lin in ol
id dl o id dl
M En Po M En Po

Notes: This figure plots the results of the estimates of the transfers on quality of work life over
time.

153
 Figure A29: Results for Quality of Work Life by Time Period (2)

Quality of Employment: Quality of Work Life

0.3

Overall 0.2
Whether the
satisfaction
respondent has 0.1
with the
decision−making
respondent's 0.0
input in their
main job (1−5
job (1−4 scale)
scale) −0.1

−0.2
0.3
Whether the 0.2
Satisfaction
respondent
with non−wage
does not plan 0.1
aspects of
to leave their
respondent's 0.0
job in the
main job (1−5
next year (1−3
scale) −0.1
scale)

Estimated Effect in SD
−0.2
0.3
Safety 0.2
Opportunities
and health
for promotion
conditions 0.1
at the
at the
respondent's 0.0
respondent's
main job (1−5
main job (1−5
scale) −0.1
scale)
−0.2
0.3
Whether a 0.2
Number of
scheduled shift
stressors in
was canceled 0.1
their work
with less than
environment at 0.0
24 hours notice
respondent's
in the last
main job −0.1
month
−0.2
0.3 e e
ar
1
lin lin ed
id d ol
How easy is it 0.2 Ye M En Po
to take time
0.1
off from the
respondent's 0.0
main job? (1−4
scale) −0.1

−0.2
1 e e d
ar lin in ol
e
id dl
Ye M En Po

Notes: This figure plots the results of the estimates of the transfers on quality of work life over
time.

154
 Figure A30: Results for Entrepreneurship by Time Period

Entrepreneurship
Midpoint of
The the constant 0.1
respondent's relative
self−reported risk aversion
willingness to (CRRA) range 0.0
take financial implied by a
risks (1−10 participant's
scale) coin flip −0.1
gamble

The
respondent's 0.1
likelihood
Whether or not
rating that
the respondent

Estimated Effect in SD
they will start 0.0
has an idea for
a business
a business
in the next
5 years (1−10 −0.1
scale)

If a family 0.1
The
member who
respondent's
started a
interest in
business 0.0
starting a
lives in the
business (1−10
respondent's
scale) −0.1
household

0.1
If the
If the
respondent
respondent
knows someone
ever started or 0.0
who started or
helped start a
helped start a
business
business −0.1

1 2 r3 d r1 r2 r3 d
ar ar a ole a a a ole
Ye Ye Ye Po Ye Ye Ye Po

Notes: This figure plots the results of the estimates of the transfers on entrepreneurship over time.

155
 Figure A31: Results for Disability by Time Period

Disability
0.20

0.15
Participant
has a health
Participant
problem/disability 0.10
has a health
that limits the
problem/disability
work they can

Estimated Effect in SD
do 0.05

0.00

0.20
How long the
participant's
How much the 0.15
health
participant's
problem/disability
worst health
has affected
problem/disability 0.10
the work they
limits the
can do (no
amount of work
disability, 0.05
they can do
less than 1
(0−7 scale)
year, more than
1 year) 0.00

e e d e e d
lin in le lin in le
id dl o id dl o
M En Po M En Po

Notes: This figure plots the results of the estimates of the transfers on disability over time.

156
 Figure A32: Results for Barriers to Employment by Time Period

Barriers to Employment
0.050

Whether the

Estimated Effect in Raw Units

Whether the
respondent 0.025
respondent
missed work
missed work
due to lack of
due to lack of
trasportation 0.000
childcare in
in the last
the last month
month
−0.025
0.050 e e d
lin d lin ole
id
M En Po
Whether the 0.025
respondent
missed work due
to illness in 0.000
the last month

−0.025

lin
e i ne le
d
id dl o
M En Po

Notes: This figure plots the results of the estimates of the transfers on barriers to employment over
time.

157
 Figure A33: Results for Human Capital by Time Period

Human Capital

Estimated Effect in Raw Units

Total years of 0.02
Completed
post−secondary
a GED or
education
postsecondary 0.00
completed
program
post−baseline
−0.02

r1 r2 r3 d
a a a ole
0.02 Ye Ye Ye Po
Enrolled in
post−secondary
0.00
program
−0.02

r1 ar
2 r3 ed
a a ol
Ye Ye Ye Po

Notes: This figure plots the results for human capital over time, showing the point estimates for
completion of a GED or post-secondary program trending upwards by the end of the study. There
is no value for this variable for Year 1 because participants were only asked about whether they had
completed a high school degree or GED in the midline and endline SRC survey. For all outcome
variables, data from the National Student Clearinghouse (NSC) were preferred to survey data for
those participants that consented to their administrative records being used. For example, for
completion of a GED or postsecondary program, GED completion was captured in survey data
as it is not in the NSC data, postsecondary program completion was captured in the NSC data for
those participants who consented to share these data, and postsecondary program completion was
captured in survey data for those participants who did not consent to share NSC data.

158
 Figure A34: Results for Relationship Status by Time Period

Relationship Status
0.2

Whether the Whether the 0.1

respondent is respondent has
divorced a spouse
0.0

Estimated Effect in SD
−0.1

0.2

How long the Whether the 0.1

respondent has respondent is
been in their in a romantic
relationship relationship 0.0

−0.1
1 2 e 3 e d
0.2 ar ar lin ar d lin ol
e
Ye Ye id Ye
Number of M En Po
times the
respondent said 0.1
they started
or ended a
relationship in 0.0
the last year
−0.1
1 2 e 3 e ed
ar ar lin ar in ol
id dl
Ye Ye M Ye En Po

Notes: This figure plots the results of the estimates of the transfers on relationship status over time.

159
 Figure A35: Results for Moving Labor Markets by Time Period

Move Labor Market

0.25

0.20

0.15
Moved labor Any active
markets since area−search 0.10
baseline behaviors

Estimated Effect in SD
0.05

0.00

−0.05
0.25

0.20

0.15
Number of
Interested in active labor
0.10
moving areas market−search
behaviors
0.05

0.00

−0.05
e e ed e e ed
ar1 ar
2
lin ar
3
dl
in ol ar1 ar2 lin ar3 d lin ol
Ye Ye id Ye Ye Ye id Ye
M En Po M En Po

Notes: This figure plots the results of the estimates of the transfers on moving labor markets over
time.

160
 Figure A36: Results for Quality of Labor Market by Time Period

Labor Market Quality

0.1

Consumption Mean wage

amenities index by education 0.0
(PCA log scale, in 2022 (in
ranges from −2 dollars per
to 2) hour) −0.1

−0.2

0.1
Employment BLS projected
to population job−growth for
ratio for ages respondent...s 0.0
25 to 64 for education
respondent...s group (percent
education group change) −0.1

−0.2

Mean percentile 0.1

household
Median annual
income rank for
income for 0.0
children whose
respondent...s
parents were
education group
in the 25th
(in dollars) −0.1
percentile of
income
−0.2

Estimated Effect in SD
0.1

Natural Pollution index

amenities index (mean PM2.5, 0.0
(ranges from −5 RSEI, and AQI
to 9) z−score)
−0.1

−0.2

0.1
Recent
Mean wage
population
growth
growth for 0.0
2019−2022
respondent...s
by education
education
(percent
group (percent −0.1
change)
change)

−0.2

0.1

Annual property
Annual
crime rate 0.0
per−pupil
(crimes
school spending
per 100,000
(in dollars)
residents) −0.1

−0.2

ar
1
ar
2
ar
3 ed
ol
0.1 Ye Ye Ye Po

Annual violent
crime rate 0.0
(crimes
per 100,000
residents) −0.1

−0.2

ar
1
ar
2
ar
3 ed
ol
Ye Ye Ye Po

161 of the transfers on quality of labor markets over

Notes: This figure plots the results of the estimates
time.
 Figure A37: Results for Benefits by Time Period

Estimated Effect in Dollars

Benefits
200
Total amount
of government 0
benefits −200
received during
the previous −400
year
−600

−800
e e d
in in ol
e
idl dl
M En Po

Estimated Effect in Count

0.10
Number of
government 0.05
benefits
programs
0.00
received during
the previous
year −0.05

e e d
lin in le
id dl o
M En Po

Notes: This figure plots the results of the estimates of the transfers on benefits over time.

162