Unintended Consequences of Immigration Enforcement:Household Services and High-Skilled Women’s Work∗

Chloe N. East†

Andrea Velasquez‡

September 21, 2020

Abstract

Immigration enforcement has intensified in the U.S., however, there is little

evidence on its effect on citizen’s labor outcomes. Exploiting the staggered roll-

out of a large, federal enforcement policy–Secure Communities (SC)–across local

areas, we estimate a difference-in-differences model with time and location fixed

effects. We find that SC reduced the labor supply of high-skilled citizen mothers.

This effect is larger for mothers with young children, and there are persistent

negative effects following the birth of a child. The increased cost of outsourcing

household production, due to reduced undocumented immigrant women’s labor

supply, is an important mechanism.

∗We are grateful to Patricia Cortes, Delia Furtado, Gaurav Khanna, Brian Kovak, Emily Lawler, MichelleMarcus, Terra McKinnish, Kathleen McGarry, Anita Mukherjee, Pia Orrenius, Analisa Packham, SandraRozo, Sarada, Na’ama Shenhav, Jenna Stearns, Jose Tessada, Alisa Tazhitdinova, and seminar participantsat the University of Colorado Denver, the Hawaii Applied Micro One-day Conference, the First AnnualColombian Economics Conference, the Census Bureau, the University of Houston, Carnegie Mellon Uni-versity, the University of Pittsburgh, Columbia University, the Institute of Behavioral Science and the CUPopulation Center, the Association for Public Policy Analysis and Management, the Southern EconomicAssociation, the CSWEP sessions at the Western Economic Association, and the Population Associationof America. We are also grateful to Reid Taylor, Tyler Collinson and Evan Generoli for excellent researchassistance. We obtained the TRAC data from Syracuse University as TRAC Fellows. Chloe East was sup-ported by funding from the Office of Research Services at the University of Colorado Denver. As always, allerrors are our own. A previous version of this work was circulated under the title: “The Effect of IncreasingImmigration Enforcement on the Labor Supply of High-Skilled Citizen Women”.†University of Colorado Denver and IZA - Institute of Labor Economics, email: chloe.east@ucdenver.edu‡University of Colorado Denver, email: andrea.velasquez@ucdenver.edu

1 Introduction

The costs and benefits of immigration on citizens’ outcomes are hotly debated by economists

and policy-makers, and they have been a major campaign topic in elections around the world

in recent years.1 In the U.S., where undocumented immigrants account for about 5% of the

U.S. work force (Krogstad, Passel and Cohn, 2017), undocumented immigration has received

particular attention in the public discourse.2 This has been accompanied by important

policy changes at the local, state, and national level aimed at addressing undocumented

immigration. As a consequence of these policy changes, over the last 15 years, interior

immigration enforcement has increased dramatically, which has in part led to the rapid rise

in immigrant detentions and deportations.3 Therefore, understanding the effects of these

changes, both on the immigrant population, and potential spillover effects on citizens, is

crucially important. While the previous literature has primarily focused on the effect of

immigration enforcement policies in the U.S. on the immigrant population,4 evidence on the

possible spillover effects of these policies on citizen’s labor outcomes is limited.

In this paper, we focus on the potential unintended consequences of Secure Commu-

nities (SC)–an immigration enforcement policy that increased the likelihood that undocu-

mented immigrants were identified, detained, and deported–on the labor supply of citizens

who are likely to outsource household production. Undocumented immigrants are overrepre-

sented in household service occupations–they make up 21% of maids and housekeepers and

1An extensive literature has focused on estimating the impact of migratory flows on labor outcomes ofcitizens. For excellent reviews of the literature see Friedberg and Hunt (1995), Longhi, Nijkamp and Poot(2005), and Longhi, Nijkamp and Poot (2006). For the current debate on immigration policies, see forexample: Menasce Horowitz (2014), Givens (2018), Thompson (2018), Economist (2018), Winders (2016).

2See for example: Felter and Renwickn (2019).3Between the early 2000s and mid 2010s, the number of people detained annually increased by 3,200% to

roughly 160,000 annual detentions in 2014, and the number of people deported annually increased by fourfold to roughly 200,000 annual deportations in 2014. Statistics from the Transactional Records AccessClearinghouse (TRAC) available at: https://trac.syr.edu/phptools/immigration/detainhistory/

and https://trac.syr.edu/phptools/immigration/removehistory/. This has far outpaced the growthin the estimated number of undocumented immigrants living in the U.S. in this time period: from roughly9 million in 2003 to 11 million in 2014 (Krogstad, Passel and Cohn, 2018).

4See, for example, Phillips and Massey (1999), Bansak and Raphael (2001), Orrenius and Zavodny (2009),Amuedo-Dorantes and Bansak (2014), Orrenius and Zavodny (2015) and Hansen (2019).

2

5.5% of workers in childcare services.5 Thus, decreased labor supply of undocumented immi-

grants through more restrictive policies is predicted to increase the cost of household services,

affecting time allocation decisions of citizens (Cortes, 2008; Cortes and Tessada, 2011). We

focus on high-skilled women, particularly with pre-school-aged children, who are the most

likely to be impacted by changes in the cost of outsourcing household production.6

The main contributions of this paper are twofold. First, we evaluate the effects of

a recent enforcement policy in the U.S. that led to the removal of immigrants, rather than

studying the effect of migratory inflows on the outcomes of interest, as in much of the previous

literature. Given that President Trump recently reinstated SC and expanded other interior

enforcement policies (Alvarez, 2017; Sakuma, 2017), understanding the effects of such policies

is important for policy-makers as they actively change immigration policy. Second, we add

to previous work documenting a positive relationship between the presence of low-skilled

immigrants and high-skilled women’s labor supply.7 In particular, our empirical strategy

overcomes methodological challenges faced by the previous literature, which used a shift-

share approach to estimate the effect of immigration inflows on high-skilled women’s labor

5Authors calculations using the American Community Survey (ACS). We do not observe documentationstatus in the ACS, so we follow the literature (e.g. Passel and Cohn (2014)) and assume that Hispanic, non-citizens, with high-school education or less, who arrived in the U.S. after 1980 are undocumented immigrants.

6High-skilled workers spend a larger fraction of their income outsourcing household work: on average, in2005, the fraction of income that college-educated households spend on household services is more than twiceas much the fraction spent for households with at most a high school degree in the Consumer ExpenditureSurvey: https://www.bls.gov/cex/2005/share/educat.pdf. Additionally, among high-skilled workers,women spend more time on these household tasks, and have a more elastic labor supply, when compared tomen (Blau and Kahn, 2007; Pew Research Center, 2013); high-skilled citizen women spend 45% more time onhousehold activities and 96% more time on child care relative to high-skilled citizen men in Table (2). Finally,households with young children (below school age) may be especially affected due to additional demands ofhousehold production tasks. Most states have a compulsory school age of 6 and, during our sample period,the vast majority of states did not provide full day kindergarten (National Center for Education Statistics,2018; Education Commission of the States, 2016). Married couples without children spend 1.3% of theirincome on household services, compared to 2% for married couples with a youngest child age 6-17, and 4.9%for married couples with a youngest child age 0-5. Information from the Consumer Expenditure Survey:https://www.bls.gov/cex/2005/share/educat.pdf.

7The literature has examined this relationship in the United States (Furtado and Hock, 2010; Cortes andTessada, 2011; Amuedo-Dorantes and Sevilla, 2014; Furtado, 2015, 2016), Italy (Barone and Mocetti, 2011;Peri, Romiti and Rossi, 2015), Hong Kong (Cortes and Pan, 2013), Spain (Farre, Gonzalez and Ortega,2011), the UK (Romiti, 2018), and in a cross-country approach (Forlani, Lodigiani and Mendolicchio, 2015).

3

supply in the spirit of Card (2001).8 In recent years, an increasing number of papers call into

question the assumptions behind this shift-share approach (Adao, Kolesar and Morales, 2018;

Borusyak, Hull and Jaravel, 2018; Goldsmith-Pinkham, Sorkin and Swift, 2018), including in

the context of immigration (Jaeger, Ruist and Stuhler, 2018). In contrast to this approach,

we use local enforcement policies as an exogenous driver of immigrants’ labor supply, and

we test the validity of this exogeneity assumption in several ways, discussed below.

SC increased information sharing between local law enforcement and Immigration and

Customs Enforcement (ICE), in order to identify and remove undocumented immigrants.

Specifically, SC required and automated a process by which arrested individuals would have

their fingerprints sent to ICE for immigration status screening. We expect this to affect the

labor supply of undocumented immigrant workers through three main channels. First, SC

affected the availability of low-skilled labor through direct removals. SC is credited with

more than 450,000 individuals deported, 96% of whom were men, over our sample period of

2005-2014.9 Of those deported in this time period, 20% were not convicted of a crime, and

26% were not convicted of a serious crime, so a broad population may have been directly

affected. Moreover, Hispanic and Latino immigrants were overrepresented among those

deported under SC, which could be due to racial profiling–advocacy groups have alleged

that SC provided a way for law enforcement to use minor violations to target the Hispanic

population (Kohli, Markowitz and Chavez, 2011).10 Second, voluntary migration may have

changed–either by increasing out-migration from the U.S. or decreasing in-migration to the

U.S., or both. Third, partially because of the broad nature of the deportations, and the

8The closest papers to ours are Furtado and Hock (2010), Cortes and Tessada (2011), and Furtado (2015).9Note that this over-representation of men is not reflected in the overall undocumented population, but

might reflect the fact that men are over-represented in the population that is incarcerated. Specifically,estimates suggest that in 2012, 53% of the undocumented population was male (Baker and Rytina, 2013)and 90% of the incarcerated population in 2001 was male (Bonczar, 2003).

10In 2007, 70% of all undocumented immigrants were estimated to come from Mexico and Central Americancountries, but 87% of the deportations through SC were among immigrants from these countries. AppendixTable (A1) shows the information about individuals who were deported under SC discussed here. Amuedo-Dorantes, Puttitanun and Martinez-Donate (2018) also find that broad enforcement policies, like SC, led toincreased detainment for minor violations.

4

fact that certain groups were overrepresented, fear and mistrust of local law enforcement,

and government more broadly, may have created a “chilling effect” among immigrants who

stayed in the U.S., causing them to reduce their labor supply.

SC is well-suited for our analysis because it was mandatory, and it was rolled out in

a staggered fashion across localities. We match data on the timing of implementation of

SC across local areas to high-skilled women’s labor outcomes in the American Community

Survey (ACS) from 2005-2014. This allows us to estimate a difference-in-differences model,

while controlling for local area and survey year fixed effects. Thus, our identification strat-

egy relies on the assumption that, conditional on observables and fixed effects, there are

not time-varying differences within local areas that are correlated with the timing of SC

adoption. We provide support for this assumption in several ways. We show SC start dates

are not correlated in a qualitatively meaningful way with pre-SC-implementation trends in

demographics and economic conditions in local areas, and the results are robust to control-

ling for these pre-trends. Moreover, event studies support the parallel pre-trend assumption

for high-skilled mother’s labor supply. Two other features of SC are worth noting: first,

local areas had little influence over the timing of policy adoption, and, once it was in place,

they had limited discretion in the operation of the program;11 second, because SC rolled out

quickly and eventually covered the entire country, internal migration is less likely to bias the

results (Borjas, 2003; Borjas and Katz, 2007; Cadena and Kovak, 2016).

Our primary finding is that working-age (20-64) college-educated citizen women signif-

icantly reduced their labor supply in response to SC exposure. These effects are particularly

large and robust for mothers, which is consistent with the fact that mothers will have more

11Some jurisdictions, known as “sanctuary cities”, refused to cooperate with ICE detainer requests byclaiming that the policy was unconstitutional under the Fourth Amendment. Following Alsan and Yang(2018), we use information from the U.S. Department of Homeland Security about whether a locality resistedsuch requests, to classify locations as sanctuary cities. Only 36 of roughly 1000 PUMAs fit this definitionprior to SC implementation; the vast majority of sanctuary city policies began in 2014, which, in part, led tothe replacement of SC with the Priority Enforcement Program in 2014. We do not examine heterogeneouseffects by sanctuary city status, given how rare this was during our sample period. Instead, we focus onother measures of program intensity discussed in more detail below.

5

household production responsibilities, and thus will be more sensitive to changes in the

price of outsourcing this production. Among mothers, the effects are driven by women with

children under age 6 (before children are likely to enter school); mothers of young chil-

dren experience a reduction in the likelihood of working by 0.8% and in hours worked by

1.2%. These estimates are about 1-2% of the effect of having a child for women in the U.S.

(Kuziemko et al., 2018; Kleven et al., 2019).12 However, it is important to note that our

estimates are for all high-skilled mothers, and many mothers in our sample may not respond,

so the effects on those who do respond are likely much larger. We also investigate the extent

to which having SC in place in the years following a child’s birth has lasting consequences

on mothers’ labor supply, since previous work finds the motherhood penalty is of similar

magnitude for many years after childbirth (Kuziemko et al., 2018; Kleven et al., 2019). We

find suggestive evidence that exposure to SC around a birth negatively impacts the labor

supply of college-educated women for several years after the birth.

We further provide evidence that changes in the price of market-provided household

services are an important mechanism driving the observed changes in high-skilled mothers’

labor market outcomes. Women constitute over 90% of the total employment in house-

hold services13, therefore, we examine whether SC affected the number of likely undocu-

mented Hispanic female workers in household services–defined as housekeeping and child-

care occupations–and find no evidence of an effect, which is consistent with most of the

removals under SC being men. However, among likely undocumented women who stayed

in the U.S., we find a negative effect on their labor supply at the intensive margin. These

effects are larger in places with more Hispanic deportations and more deportations for non-

serious crimes, where chilling effects are plausibly larger.14 These results highlight that when

12Kuziemko et al. (2018) estimate that having a first child decreases women’s labor force participation by46-96% and Kleven et al. (2019) estimate a decline in labor force participation of 43% following a first birthin the U.S. Note that both of these estimates include all women in the sample (unconditional on education)and both focus on much earlier time periods than we do.

13Calculations based on the 2005 American Community Survey. See Appendix Table (A2).14This is consistent with other literature that has documented chilling effects on other outcomes of Hispanic

immigrants. Specifically, Alsan and Yang (2018) find that SC reduced Hispanic citizens’ participation in

6

analyzing the spillover effects of immigration onto labor outcomes of citizens, it is crucially

important to consider not only the size of the immigrant population, which has been the fo-

cus of most of the existing literature, but also the political conditions that allow immigrants

to integrate in their destination country. Next, we examine the effect of SC on the cost

of outsourcing household services, as proxied for by hourly wages of female workers in this

sector. Given that production in these occupations is very labor intensive, wages are likely a

very close proxy for prices (Hock and Furtado, 2009), and we find a significant positive effect

on low-skilled female workers’ hourly wages. As further evidence of this mechanism, there

are no significant effects of SC on the labor outcomes of high-skilled women with no children,

whose labor supply should not be as responsive to changes in household service prices, and,

the effects are larger for high-skilled citizen mothers who live outside of their state of birth,

and are thus more likely to rely on market-provided household production, rather than in-

formal sources, like grandparents. Taken together, these results provide strong evidence that

changes in the price of outsourcing home production is an important mechanism behind the

labor supply effects on citizen mothers.

Finally, our results inform the debate about the distributional effects of immigration

and immigration enforcement on citizens. Specifically, we find enforcement policies directly

affecting low-skilled immigrants have negative spillover effects onto high-skilled citizen female

workers. The only existing evidence of these spillover effects of SC on citizen labor outcomes

is East et al. (2019), who document a different mechanism through which enforcement policies

can affect citizen’s labor outcomes.15 Their findings show that SC led to a reduction in

employment of low-skilled male immigrants, and a negative effect on the employment of

male citizens in higher-skilled occupations. These effects on citizen men are driven by those

safety net programs and attribute this to chilling effects. Wang and Kaushal (2018) find that enforcementcaused a worsening of self-reported mental health among Latino immigrants.

15Another literature examines the effect of other immigration policies on citizen’s labor market outcomes.For example, Bohn, Lofstrom and Raphael (2015) study the spillover effects of Arizona’s E-Verify law oncitizen workers who may substitute for undocumented workers. They find these citizens experience loweremployment, but higher earnings as a result of the policy.

7

working in sectors with a higher initial share of low-skilled likely undocumented workers,

which suggests that there may be complementarities in market production between workers

in low and high-skilled occupations that cause negative spillover effects onto citizen males.

In this paper, we find evidence of another important pathway through which SC affects the

labor market outcomes of citizens, explained by another type of complementarity: between

low-skilled female immigrants working in household services and high-skilled citizen mothers

working outside the home.16 We also show that the magnitude of the effects on high-skilled

mothers is not related to the initial share of likely undocumented workers in the sector the

mother works in, indicating that market complementarities are unlikely to be an important

mechanism that drives the effects we document here. Our paper also contributes to a growing

literature studying the effects of SC on local communities’ outcomes including spillover effects

to citizen safety net program participation (Alsan and Yang, 2018), local crime (Miles and

Cox, 2014; Hines and Peri, 2019), and immigrants’ health outcomes (Wang and Kaushal,

2018).

The rest of the paper proceeds as follows: in the next section, we provide details about

SC and the data we use. Section 3 describes our empirical strategy and section 4 presents

our results. Section 5 concludes.

2 Policy Background and Data

The Secure Communities program (SC) is one of the largest interior immigration enforce-

ment programs in the United States.17 SC increased information sharing between local law

enforcement agencies and U.S. Immigration and Customs Enforcement (ICE). The goal of

16Because of this other mechanism that affects high-skilled men, we do not examine the effects on high-skilled men in this paper, except in looking at effects around the birth of a child.

17For comprehensive reviews of SC see Cox and Miles (2013), Miles and Cox (2014), and Alsan and Yang(2018). The information in this section comes primarily from these reviews and is similar to that discussedin East et al. (2019).

8

SC was to identify individuals eligible for removal from the U.S. Prior to SC, individuals’ fin-

gerprints would be taken upon being booked in state prisons or local jails and would be sent

to the Federal Bureau of Investigation (FBI) to conduct a criminal background investigation.

Under SC, these fingerprints would now also be sent to ICE, who would try to determine

an individual’s immigration status using their Automated Biometric Identification System

(IDENT).18 Based on this information, a detainer may be issued, and the law enforcement

agency would then be required to hold the individual for up to 48 hours in order for ICE

to obtain custody and start the deportation process. Importantly, detainers could be issued

for criminal reasons or for immigration-crime-related reasons, and they did not have to be

proceeded by a conviction.

Implementation of SC required establishing a partnership between local law enforce-

ment and local ICE officers, which took time and resources, and this led to the staggered

nature of the program roll-out over 2008-2013 that we exploit in our empirical approach.

Information on the date of implementation of SC in each county comes from ICE and Figure

(1) shows the pattern of the rollout across Public Use Microdata Areas (PUMAs). We focus

on the presence of SC by PUMA rather than by county, because the smallest consistent and

comprehensive geographic area available in the ACS is the PUMA.19 We describe this deci-

sion and variable construction in more detail below. The timing of adoption was determined

by the federal government. This is important for the assumptions underlying our empirical

model, since local areas had little discretion in the timing of implementation. Previous evi-

dence shows that early adopters were selected based on the size of their Hispanic population,

proximity to the U.S.-Mexico border, and presence of other local enforcement policies.20 The

18IDENT includes biometric and biographical information on non-U.S. citizens who have violated immi-gration law, or are lawfully present in the U.S., but have been convicted of a crime and are therefore subjectto removal, as well as naturalized citizens whose fingerprints were previously included in the database. Inaddition, the IDENT system includes biometric information on all travelers who enter or leave the U.S.through an official port, and when applying for visas at U.S. consulates.

19PUMAs are constructed as contiguous geographic areas that respect state borders and have at least100,000 people living in them. For this reason, PUMAs in rural areas cover much more geographic area thanPUMAs in urban areas.

20These other local enforcement policies are 287(g) agreements, which were similar in design to SC, but

9

timing of implementation among later adopters was more random (Cox and Miles, 2013),

because the government shifted to mass activations which led to waiting lists. Importantly,

this previous work has shown there is not a relationship between the timing of SC adoption

and the county’s pre-SC economic conditions, crime rates and potential political support for

SC. Our empirical specification described below includes PUMA fixed effects to control for

time-invariant unobserved heterogeneity at the local level, including pre-SC characteristics,

such as proximity to the border. So, of more concern is if the implementation of SC is

correlated with differential trends across PUMAs. We use multiple methods to show this

is not a likely source of bias in our estimates. First, we directly test whether the timing

of the rollout is correlated with pre-SC trends in demographics, immigration enforcement,

and economic conditions at the PUMA level. The results of this exercise are shown in Table

(1). Out of seventeen variables, only two variables are statistically significantly related to

the rollout timing: the change in the percentage of non-citizens and the change in housing

prices. However, these relationships are small in magnitude–the results imply that a one

standard deviation increase in the percentage of non-citizens reduces the year of adoption by

about 0.1, or roughly one month, and similarly, a one standard deviation increase in housing

prices reduces the year of adoption by 0.2 or 2.5 months.21 Moreover, the R-squared on this

model is very low (0.06), suggesting that pre-trends in observable characteristics do not do

well in predicting the timing of the rollout. Second, in our main specification we control for

pre-trends by interacting changes in PUMA characteristics pre-SC with linear time trends.

Third, we implement event studies that show that our outcomes of interest were not dif-

ferentially trending across PUMAs pre-SC. As a final check, we explore the robustness to

dropping early adopter places. As discussed in section 4, the results of these tests support

the main empirical assumption of our identification strategy.

were an optional policy that local areas and states could choose to adopt. We control for the presence of287(g) agreements over time by local area.

21Specifically, to calculate the magnitude of these effects, we multiply the standard deviation for eachvariable by the estimated coefficient for that variable. The outcome variable is the year of first SC imple-mentation in the PUMA. So, for example, for housing prices, a one standard deviation change in housingprices (31.217) results in a 0.22 increase in the year of SC implementation.

10

We merge the data on SC rollout dates to high-skilled citizen women’s labor supply

over the period 2005-2014 from the ACS (Ruggles et al., 2017). Since SC ended in December

2014 (before being reinstated in 2017) and our main analysis focuses on the period 2005-2014,

our results should be thought of as the effect of increasing immigration enforcement. The

ACS is a repeated cross-sectional dataset covering a 1% random sample of the U.S., and in

the publicly available data set, the smallest geographic area available is the PUMA, which

we use as the measure of geography, as in Alsan and Yang (2018). PUMAs are identifiable

beginning in 2005, so this is the first year of our sample. The advantages of using PUMAs are

that they allow us to most precisely measure policy exposure, to cover the entire U.S., and

to have a consistent measure of geography over time. However, a disadvantage of PUMAs

is that they do not map onto the level of policy variation one to one. In particular, some

PUMAs are equivalent to counties, whereas others include several counties, and still others

are smaller than individual counties. Hence, we calculate the population-weighted average

of the county values of the SC variable within each PUMA, similar to the approach taken

by Alsan and Yang (2018) and Watson (2013).22 We show an example of this geographic

mapping in Appendix Figure (A1). This figure displays the counties, county populations, and

associated weights for each county based on their percentage of the total PUMA population,

which we use to calculate the population-weighted measure of SC. Additionally, we have no

information about the month of survey within the ACS, only the year of survey, so we assign

to each observation the SC policy in January of the survey year and test the robustness of

this choice.

Our main sample includes women ages 20-64 with a four-year college degree or more,

22There are about 1000 PUMAs and 3000 counties in the U.S. If a PUMA is equivalent to a county, orsmaller than a county, the PUMA will get the value of the SC variable for that county. If multiple countiesare contained within a PUMA, we weight the value of the SC variable for each county by the fraction ofthe total PUMA population that each county represents. Additionally, the PUMA codes were revised afterthe 2011 ACS survey, so we use the time-consistent version of the PUMA codes provided by IPUMS. Wealso have checked the robustness of our results to using the county codes available in the ACS to clusterthe standard errors since this is the level of policy variation. County codes are only available for about 60%of the sample (those living in large counties), so this limits our sample size, however the results with thissubsample, not shown, are very similar whether we cluster by PUMA or county.

11

which we refer to as “high-skilled”. We further restrict the sample to only citizens, which

includes all U.S.-born and foreign-born who report being naturalized citizens. The primary

outcome variables in the ACS are high-skilled citizen women’s usual hours worked per week

in the past year (including 0 hours) and whether the woman worked any positive hours

usually in the past year.

We hypothesize that SC will reduce high-skilled women’s labor supply through in-

creases in the cost of services that substitute for household production (Cortes and Tessada,

2011). This price increase will be caused by a reduction in the labor supply of immigrants

through three potential channels, as discussed above: 1) forced out-migration of immigrants,

2) changes in cross-border migration (either reduced voluntary in-migration or increased

voluntary out-migration), and 3) reductions in immigrants’ labor supply among those that

remain in the U.S. due to fear (chilling effects).

Descriptive statistics of those deported in Appendix Table (A1) support the idea that

chilling effects may have existed due to the broad nature of the population affected, as well

as the over-representation of immigrants from Mexico and Central American countries.23

Anecdotal evidence further suggests that SC disrupted the social and economic relationships

of both citizens and non-citizens living in immigrant communities (Amuedo-Dorantes, Putti-

tanun and Martinez-Donate, 2018). For example, interviews of roughly 2,000 Latinos living

in Cook (Chicago), Harris (Houston), Los Angeles, and Maricopa (Phoenix) counties found

that 78% of undocumented immigrants think police officers stop Latin immigrants without

any reasonable cause, 61% are afraid of leaving their own home, and 62% feel more isolated

because local law enforcement is involved with immigration enforcement (Theodore, 2013).

Moreover, since we focus on spillover effects of SC on immigrant women, it is important

to note that anecdotal evidence suggests enforcements policies also had chilling effects on

women, even if they themselves were not deported or detained. Rhodes et al. (2015) find

23The TRAC data used in this table is described in the Appendix.

12

that after the implementation of 287(g) agreements in North Carolina, Hispanic and Latina

mothers used pre-natal care later in their pregnancy because of fear and mistrust related

to using health services. Additionally, women reported feeling scared of being targeted by

law enforcement while driving, and that this would prevent them from going to work: for

example, “I remember sometimes I would have to go to work, and I would say, ‘No, I’m

not going to be able to go to work right now because there’s a [DUI] checkpoint right in

front of my house. So I’m not going to put myself in a deportation situation.’” (Valdivia,

2019).24 This anecdotal evidence is also supported by the findings in Amuedo-Dorantes and

Antman (2019), who document that more deportations in a local-area reduce labor supply,

particularly among women, with even larger effects on women with young children, who

may especially fear issues of family separation. On the other hand, for women whose male

spouse/partner was deported, it is possible they increase their labor supply to make up for

the loss family income (Chaudry et al., 2010; Orrenius and Zavodny, 2015). As we discuss

in more detail below, we examine the net effects of SC on the labor supply of immigrant

women including any positive and negative effects.

To provide direct evidence that the cost of household services is an important mecha-

nism, we also look at employment and wages for workers in these services. We construct a

sample of individuals ages 20-64 who report that their occupation at their current or most

recent job was housekeeping or childcare services. We start by estimating the direct effect

of SC on the total number of likely undocumented women working in household childcare

services, as well as the total hours they supply. Then we also look at the average hourly

wage of female household service workers. Since the ACS does not have information about

undocumented status, we define likely undocumented workers as Hispanic non-citizens with

at most a high school diploma who entered the U.S. relatively recently. We also test the

24Concerns about domestic violence also particularly affect women. Amuedo-Dorantes and Arenas-Arroyo(2019) find that increased immigration enforcement reduces the likelihood women file for a VAWA petition,which is available to victims of domestic abuse to change their immigration status. Additionally, Grittner(2019) find increases in immigration enforcement reduced Hispanic women’s usage of domestic violenceservices.

13

robustness of the results to alternative definitions of likely undocumented.

Since our sample period spans the Great Recession, to account for changes in economic

conditions that may influence women’s labor supply, we add to the data several “Bartik-style”

measures of labor demand, as well as local-area housing price information.25 We prefer to

use these “Bartik-style” measures of labor demand, rather than more direct measures, such

as the local unemployment or labor force participation rates, because these direct measures

may be affected by SC. Our preferred specification adds additional controls for pre-trends in

local economic conditions, which we measure as the change in the local unemployment and

labor force participation rates between 2000-2005 interacted with a linear time trend. These

account for the fact that different areas experienced the recession differently. In addition,

during this period, another local interior enforcement policy–287(g) programs–changed across

locations, so we control for the presence of local 287(g) agreements.26 Details on these control

variables are included in the Appendix.

Summary statistics for the ACS sample are in Table (2). We use the survey-provided

person weights in all summary statistics and regressions. We have over 2.5 million ob-

servations of high-skilled women for the period between 2005 and 2014. We multiply the

dichotomous labor supply outcome variable by 100 to ease presentation of the results. So, for

example, 85.49% of high-skilled women worked at all in the past year, and this decreases to

78.98% for women with young children. At the bottom of this table we also show descriptive

statistics taken from the American Time Use Survey (ATUS) in 2005–before SC–for a sample

of citizens aged 20-64 with a college degree or more, for two measures of household produc-

tion: 1) time spent caring for household children (e.g. feeding them, socializing with them

and time spent on activities related to their education) and 2) time on household activities

25Controlling for economic conditions may also be important since they affect migratory decisions(Gonzalez, 2015). Therefore, accounting for the change in economic conditions is also important whenestimating the effect of SC on the labor supply of likely undocumented women.

26Two papers examine the labor market effects of 287(g) agreements and find evidence of reduced em-ployment in local areas after implementing these policies (Bohn and Santillano, 2017; Pham and Van, 2010).However, these papers do not separate these effects by citizenship status or gender.

14

(e.g. time spent on maintaining the respondent’s household, like housework, cooking, and

home maintenance). These statistics reinforce the idea that women, and especially mothers,

spend more time on average in these types of activities relative to men.

3 Empirical Strategy

Our identification strategy exploits both the geographic and temporal variation in the imple-

mentation of SC to identify its effect on labor market outcomes of high-skilled citizen women.

Our main analysis examining the effect of SC on high-skilled women’s contemporaneous labor

supply is estimated with the following model:

Yipt = α + βSCpt +X ′iptδ + Z ′ptγ + µp + φt + θ∆W ′p ∗ t+ εipt (1)

Where Yipt represents different measures of labor outcomes for a woman i, living in PUMA

p and observed in year t. SCpt is a continuous variable measuring exposure to SC at the

PUMA level, and takes values between 0 and 1. SCpt is equal to zero if SC has not been

implemented by January of the survey year in any of the counties in PUMA p, and is equal

to one once SC has been implemented in all counties in the PUMA by January of the survey

year. Since we focus on the roll-out period of SC, once SCpt takes a value equal to one, it

keeps that value for the remainder of the sample period. The coefficient of interest, β should

be interpreted as the effect of SC when the entire population in a PUMA is exposed to SC

by the beginning of the survey year.

We include fixed effects at the PUMA level, µp, that account for any time-invariant

unobserved heterogeneity at the PUMA level. Our initial specification also includes year

fixed effects (φt) to account for national shocks to labor outcomes over time. In order for

the difference-in-differences model to be valid, there should not be time-varying differences

within PUMAs that are correlated with the timing of the adoption of SC in those PUMAs.

15

In addition to directly testing this assumption as described above, we subsequently add in

controls at the PUMA-year level to further test this assumption. These controls (Z ′pt) in-

clude the Bartik-style measures of labor demand, and 287(g) agreements. Following Hoynes

and Schanzenbach (2009) and Almond, Hoynes and Schanzenbach (2011), in order to control

for pre-trends, we interact changes in PUMA characteristics between 2000 and 2005 (vec-

tor ∆W ′p) with linear time trends. Importantly, to account for any pre-trend in economic

conditions that could drive our results, we include changes in the PUMA-level labor force

participation rate, unemployment rate, and the size of the housing boom in ∆W ′p(Charles,

Hurst and Notowidigdo, 2018).27 Finally, we include individual level controls in X ′ipt: age and

age squared, race, marital status, educational attainment, number of children and number

of young children in the household.28

4 Results

We begin our analysis in Table (3) by showing the effects of SC on the labor supply of

all high-skilled citizen women (columns (1)-(2)), as well as women with children (columns

(3)-(4)), and women with young children (columns (5)-(6)). Focusing first on Panel A,

this model includes only PUMA and year fixed effects and the results indicate reductions in

high-skilled citizen women’s work, which are larger for mothers. Recall that the dichotomous

outcome variable has been multiplied by 100, so the results imply a reduction in the likelihood

of working at all by 0.26 percentage points (a 0.3% reduction off the sample mean) for all

women, 0.37 percentage points (0.4%) for all mothers, and this effect appears to be primarily

27The complete set of variables included are changes in the PUMA-level labor force participation rate,unemployment rate, housing prices, the share of the PUMA that are citizens, black, non-citizens, havechildren, have young children, work more than 50 and 60 hours, have a college degree, masters degree, ora Ph.D., as well as the same education categories just for women. The results are robust to using only thelevels in 2000, or in 2005, interacted with a time trend (results available upon request).

28Fertility may be directly affected by enforcement if the price of having children changes (Furtado, 2016).We directly test for this and find no evidence of changes in fertility as shown in Appendix Table (A3). Note,the sample size is slightly smaller in this model because this fertility question is only asked to women ages15-50 in the ACS.

16

driven by women with young kids, who experience a decline of 0.61 percentage points (0.8%).

The effects on hours worked are less precisely estimated across the samples but suggest a

reduction that is again driven by women with young children, for whom there is a significant

reduction of 0.34 hours worked per week (p=0.04).

Moving through the subsequent panels we include additional controls. First, Panel B

adds controls for 287(g) agreements and economic conditions (Bartik-style variables). All

the point estimates across the different samples are robust to the inclusion of these controls.

Next, Panel C further includes interactions between changes in the PUMA characteristics

from 2000 to 2005 and a linear trend. The magnitude of the coefficients and their significance

are robust to these additional controls. Finally, the results are robust to the addition of

individual-level demographic characteristics shown in Panel D. The stability of the estimates

to these additional controls further supports the idea that the timing of SC adoption is

plausibly exogenous.

The specification, based on equation (1), and shown in Panel D of Table (3), is the

most conservative model because it includes all the PUMA and individual-level controls, so

we focus on this model in what follows. There is a significant reduction in the likelihood of

working at all for all mothers of 0.4% relative to the sample mean (p=0.05), and there are no

significant effects on hours worked, although the point estimate is negative and meaningful

in size (0.3% relative to the sample mean). These effects on all mothers appear to be driven

by mothers with young children; for mothers with children below school-age, SC reduces the

likelihood of working by 0.75% relative to the sample mean (p=0.09), and reduces usual hours

worked per week by 0.31 percentage points or by 1.08% relative to the mean (p=0.04). Recall

that the measure of hours worked includes zeros, so to get a sense of how much the change in

hours might be explained by a change in the likelihood of working at all, we conduct a back

of the envelope calculation multiplying the estimated change in the likelihood of working by

the average hours worked among those working. This predicts a decline in total hours worked

17

for all mothers of -0.13, and for mothers with young children of -0.21, which suggests that

the decline in working at all may explain a large part of the total change in hours worked.

However, this calculation relies on the assumption that the marginal mother dropping out

of the labor force worked the average number of hours prior to dropping out.29

To further test the validity of our identification strategy, we estimate an event study

approach using the model with the full set of controls (corresponding to Panel D of Table

(3)).30 Specifically, we estimate the following equation:

Yipt = α +4∑

j=−3

βj1(SCj = 1) +X ′iptδ + Z ′ptγ + µp + φt + θ∆W ′p ∗ t+ εipt (2)

The key difference between this and equation (1) is that we replace the continuous

measure of SC (SCpt) with dummy variables indicating how far each PUMA-year observation

is from the year the PUMA first implemented SC (1(SC1 = 1)). This is because we need to

have a single year that SC “turns on” to define event time, whereas in the baseline model

we can allow the implementation of SC to be continuous. For example, β1 represents the

outcome of interest in the year of SC implementation, and β−1 represents the outcome of

interest two years before implementation. We omit β0, which is the coefficient representing

the year before implementation. We define the year the PUMA implemented SC as the first

year that any county in the PUMA adopted SC. There are some PUMAs that experienced

a phase-in of SC over a period of multiple years, as SC rolled out across counties in the

PUMA, so we may see a phase-in of the effect of SC across event time as well. Because of

this, the magnitude of the estimates in the event study model are not directly comparable

to those in the difference-in-differences analysis in equation (1). However, the advantage of

the event study approach is that it allows us to test our key empirical assumption, which is

29Because of issues of selection into who stops working, we do not estimate the effects on hours of workconditional on working.

30While there may be concerns with including trends in the event study model, the results are very similarif we omit the trends from the model. Therefore, to be consistent with our main difference-in-differencesresults, we include trends in the event study model.

18

that, conditional on observable characteristics of the PUMAs and individuals, the timing of

SC adoption is unrelated to trends in outcomes across PUMAs. Figure (2) shows the results

of the event study model for high-skilled mothers of young children, for whom we found the

largest and most consistent effects in Table (3). The figures show no evidence that, prior to

SC adoption, the labor supply of high-skilled mothers of young children was differentially

trending across PUMAs. Moreover, there is strong evidence of significant negative effects of

SC on labor supply after implementation and this appears to phase-in over time.31 While this

phase-in of effects could represent dynamic treatment effects and indicate that the difference

in difference results may be biased (Goodman-Bacon, 2018), it is important to note this

could also be because SC is phased in across counties within the PUMS at different times.

In contrast, the treatment variable in equation (1) will account for this county-by-county

phase-in across counties.

Finally, we explore these results in two additional ways. First, the results above indicate

the biggest contemporaneous effects of SC are when the youngest child is ages 0-5, and we

further investigate heterogeneity in the effect by child age within this group in Table (4).

These results suggest SC has the largest negative effects while the youngest child is under age

3 (Panel A), which is consistent with the fact that childcare before age 3 is more expensive

and higher-quality care is harder to find for children younger than 3 (Jessen-Howard et al.,

2018; Workman and Jessen-Howard, 2018). Second, we examine the margins on which this

31Unfortunately 2005 is the first year that PUMAs are identifiable in the ACS, so we cannot extend ourpre-period back further. The event study model includes a dummy variable for -4 and earlier and, while weestimate this coefficient, we do not report it, because it is estimated on a unbalanced sample of PUMAs.Given the short rollout of SC, we do not restrict the post-period dummies to be estimated on a balancedsample. Instead, we check and confirm that the results of the event study are similar when we restrict thesample to only PUMAs that first adopted SC before 2013, which allows us to have a balanced sample for 3observations before and after implementation. For example, for counties that adopted in 2012, event time1 is the year 2012, event time 2 is 2013, and so on. Appendix Figure (A2) shows the event study resultswhen using the balanced sample–grey dots–and the unbalanced sample–black dots. The point estimatesare generally very similar in the balanced and unbalanced samples, and all the confidence intervals overlap.Additionally, as mentioned above, some PUMAs experience a gradual phase-in of SC across counties, so wetest the robustness of the event study results to defining the year the PUMA implemented SC as the firstyear that 50% of the PUMA had adopted SC. The results shown in Appendix Figure (A3) are also verysimilar to the baseline event study estimates.

19

decline in labor supply is operating in more detail by estimating the effect of SC on the

probability of working full-time (35+ hours), part-time (20-35 hours), and being marginally

employed (0-20 hours, inclusive of 0 hours worked). Full-time work may be more affected

because outsourcing of household production may be more important for women who work

longer hours (Cortes and Pan, 2019). The results in Table (5) suggest that indeed most of

the change in labor supply is coming from a reduction in the likelihood high-skilled women

work full-time and again these results seem to be driven by mothers with young children. For

mothers with young children, the estimates indicate a reduction in full-time work of 1.5%

(p=0.05). Overall, this suggests that SC may be particularly impactful for mothers of young

children working full-time jobs and may have important implications for the potential career

progression of mothers in very time-intensive jobs (Bertrand, Goldin and Katz, 2010).

4.1 Mechanisms

To better understand whether changes in the cost of outsourcing household work are driving

the negative effect of SC on high-skilled mothers, we conduct several tests. First, we directly

examine the labor market outcomes of likely undocumented workers in household services.

We start with a relatively broad definition of likely undocumented immigrants: low-skilled

(high-school education or less) Hispanic non-citizens, who arrived in the U.S. after 1980.32

We follow Passel and Cohn (2014) and Passel and Cohn (2016) in choosing 1980 as the

cutoff year.33 Then, we also explore further restricting the sample to the subset born in

Mexico, and the subset born in Mexico and Central American countries, given that most

32We show summary statistics for workers in household services in Appendix Table (A2). The first columnshows the percentage of Hispanic low-skilled non-citizens who entered the U.S. in 1980+ in each occupation;and the second column shows the percentage of female Hispanic low-skilled non-citizens who entered theU.S. in 1980+ in that occupation. Across both occupations, the vast majority of the low-skilled non-citizenswho work in that occupation are women (in both occupations women account for 94% of all low-skillednon-citizens). Additionally, likely undocumented workers are more over-represented in housekeeping relatedoccupations than child care, though they play an important role in both occupations.

33The results are very similar when we use the year of the Immigration Reform and Control Act (IRCA)legalization–1986–as the year of entry cutoff, instead of 1980. Results available upon request.

20

undocumented immigrants are from these countries. While we cannot perfectly identify

undocumented immigrants, it is important to note that the internal validity of these estimates

would only be biased if measurement error in our definition of “likely undocumented” is

related to the implementation of SC.34 Importantly, measurement error in our definition of

likely undocumented should not impact our estimates on the labor supply of high-skilled

citizen women.

We explore whether exposure to the policy affected the total number of likely undoc-

umented women working in household services, as well as the total number of hours they

work in household services, both scaled by the total PUMA population.35 Given our focus

on women, and that 97% of deportees were men, a negative effect on the number of work-

ers could be explained by cross-border migration, or dropping out from the labor force. A

negative effect on the total hours worked could be explained by these same mechanisms, as

well as chilling effects among those that remain in the U.S. in household service occupations.

Two other possible effects are worth mentioning. First, undocumented women may switch

between occupations as a result of SC. Second, women whose spouse/partner was deported

might work more to make up for the lost family income. Given the way we construct these

outcome variables, any positive effect on labor supply due to either of these other mechanisms

would be included in our estimated effect, so our results should be interpreted as the net

effect including any of these potential increases in labor supply.36 These analyses are at the

34A potential source of measurement error in the estimate on non-citizens is the undercount of undocu-mented immigrants in surveys conducted by the U.S. government (Passel and Cohn, 2011; Hoefer, Rytinaand Baker, 2012; Warren and Warren, 2013; Warren, 2014; Van Hook et al., 2014; Genoni et al., 2017; Brownet al., 2018). Unfortunately, although previous work has provided evidence about the size and selection ofthe undercount, we cannot assess whether the undercount varies in response to SC.

35Specifically, to construct the dependent variables, we sum the total number of working-age likely un-documented females in each PUMA and year working in household services, as well as the total hours thatthey work. We then divide each of these by the total working age population in each PUMA and year, andfinally multiply by 100 to ease the presentation of the results. We weight these models with the total PUMApopulation in 2000.

36Given that SC was rolled-out quickly and was eventually a nation-wide program, there is little incentivefor within-U.S. migration of immigrants. We also test directly whether exposure to SC affected likelyundocumented women’s decision to migrate across PUMAs and find no evidence that it did–results shownin Appendix Table (A4).

21

PUMA level, rather than the individual level, because changes in the composition of work-

ers in household services could affect any estimated effects within the likely undocumented

population; for example, if women who out-migrate or switch occupations in response to SC

work fewer hours on average, then the estimated effect on average hours worked among the

likely undocumented population in household services would be biased upward.

The results in Panel A of Table (6) show small and insignificant effects on the number

of likely undocumented women working in household services. As mentioned before, this

is an expected result given that the vast majority of deportations were male immigrants.

These results show, that the number of low-skilled immigrant female workers in household

services did not change due to voluntary cross-border migration, occupational switching, or

moving in or out of the labor force. We next examine the effects on the total number of

hours worked by this group in Panel B. The results show a consistent negative effect on the

number of hours worked among likely undocumented female workers (a reduction between 4

and 7% relative to the mean), and the effects are significant at the 10% level (p=0.08) when

considering the group of immigrant women born in Mexico and Central America.

These results suggest the negative effect of SC on the labor supply of undocumented

female workers is driven by an impact at the intensive margin among those who remain in U.S.

Given this pattern, we hypothesize that chilling effects among women in household service

occupations living in the U.S. are important. To test for this, we examine whether the effects

of SC on the total hours likely undocumented women work in household services are stronger

in places with more deportations of Hispanic individuals, and with more deportations for

non-serious crimes, similar to Alsan and Yang (2018).37 We expect that a greater fraction

37We define non-serious crimes as immigration-related crimes, marijuana-related crimes, DUIs, and trafficviolations. Note that the mean share of non-serious crimes in Table (6) does not perfectly match the meanshare reported in Appendix Table (A1) because we weight the mean in Table (6) with the survey weightsand the mean in Appendix Table (A1) is unweighted. We define deported individuals as Hispanic if they listtheir country of citizenship as a Spanish-speaking country. Finally, the sample size when using the TRACdata is slightly smaller because some PUMAs do not have information about deportations in the TRACdata.

22

of deportations that are Hispanic could indicate higher prevalence of racial profiling, which

would also lead to larger fear-related chilling effects. Additionally, we expect the fear-related

chilling effects will be larger in places where deportations of individuals without serious

criminal histories are more common. Empirically, we add to the model an interaction term

between SC and these measures. The results in Panel C shows that the negative effects

on total hours worked are significantly larger in places where more Hispanics were deported

under SC. To ease the interpretation of these results, we report the estimated effect evaluated

at the mean of the deportation intensity measures and at one standard deviation above its

mean. So, for example, in column (1), Panel C, we find a 0.30 percentage point reduction in

hours worked in household services at the mean share of Hispanic deportations, which is a

4.3% decline relative to the mean, and in a PUMA with deportation intensity one standard

deviation higher, the effect increases to a 0.53 percentage point (7.7%) decline in hours

worked. The results in Panel D also support the hypothesis of chilling effects operating

as a mechanism. We consistently find a larger negative effect in PUMAs with a higher

share of deportations related to non-serious crimes, although the estimations are imprecisely

estimated in this case. It is also important to highlight that we interpret this evidence as

suggestive, given that these intensity measures may be endogenous–for example, they may

be correlated with local attitudes about immigration–and the results should be interpreted

with this caveat in mind.

These results provide evidence of a negative effect of SC on the labor supply of likely

undocumented immigrant women in household services. However, it is possible citizen work-

ers are substitutes for undocumented workers, in which case the total effect on labor supply

in this market could be smaller than the direct effects on likely undocumented immigrants

we estimate above. We explore this possibility in Appendix Table (A5), and find that overall

there is a reduction in the total hours supplied in this market among all low-skilled female

workers, including citizen workers (Panel A, column (1)).38 We break down the sample into

38The effects for all low-skilled women (including citizens and non-citizens) in column 1 suggest there was

23

low-skilled citizen Hispanics and non-Hispanics in columns (3) and (4), respectively. On

the one hand, low-skilled Hispanic citizens may be the closest substitutes for likely undocu-

mented immigrants, so we may expect a positive effect on labor supply for this group, but

on the other hand, they may also be directly impacted by chilling effects which would mean

we would see a negative effect on their labor supply. We find evidence suggesting that the

latter effect is dominating–there is a significant negative decline in the hours supplied by

low-skilled Hispanic citizens. Importantly, there is no effect on the total hours supplied of

low-skilled non-Hispanic citizens, which would be expected if they are not perfect substitutes

for undocumented workers.39

Given this reduction in labor supply, both overall, and among likely undocumented

immigrant women, we expect to find an increase in female wages in these services, so we

investigate this in Table (7). Column (1) shows a positive, although insignificant, effect on

overall female hourly wages. When we restrict the sample to low-skilled female workers in

column (2), where we expect the effect to be larger, we find a significant increase in hourly

wages of 2.2%.40 Given that household and childcare services are relatively labor intensive,

we expect changes in wages to be a good proxy of the total cost of outsourcing household

production (Hock and Furtado, 2009). As points of comparison, Furtado (2016) finds that a

1% change in the low-skilled immigrant population in the U.S. reduced the median wage of

child care workers by about 4%, and Cortes (2008) finds that a 10% increase in low-skilled

immigrants reduced the price of immigrant-intensive services (mostly household services) by

roughly 2%.

a 2.6% reduction in working hours relative to the mean, with a p-value of 0.11. These results are consistentwith the increase in wages of low-skilled women (Table (7)) in this sector.

39The fact that the effect of the policy is larger for Hispanic citizens than for non-citizens would beconsistent with the idea that citizens may have a more elastic labor supply than non-citizens. For example,citizens may be less constrained than non-citizens, in terms of access to savings or safety nets (Bitler andHoynes, 2011), which allows them to react more to the chilling effect of the policy.

40In results not shown, we find very similar magnitude of effects on citizens’ and non-citizens’ wages,suggesting that changes in the composition of workers is not driving the effect on wage, but rather thedecrease in labor supply drives this effect. Since the ACS does not have a variable measuring hourly wages,we construct these wages by dividing annual income by total hours worked in the year, the latter of whichis the product of total weeks and hours worked in a usual week.

24

Building off of these findings, if a reduction in the labor supply of likely undocumented

female workers is associated with an increase in the cost of household services, we would

expect to see stronger effects on the labor supply of high-skilled citizen mothers in places

that have greater concentrations of immigrants affected by SC. To test this hypothesis, we

interact the SC variable with two different measures of immigrant intensity in Table (8):

the share of the PUMA working-age population that is likely undocumented in 2005 using

our baseline definition (Hispanic low-skilled non-citizens who entered the U.S. in 1980+)

(column 1); and, the share of the PUMA low-skilled non-citizen population that is Hispanic

and entered the U.S. in 1980+ in 2005 (column 2).41 We expect the effects on high-skilled

citizens mothers to be larger (more negative) for higher values of these intensity measures,

and this is what we find across all outcomes and measures of intensity for all mothers (Panels

A-B) and mothers of young children (Panels C-D). For example, looking at the effect on

working hours for mothers with young children (Panel D, column 2), the effect of SC in a

PUMA with the average share of low-skilled non-citizens that are Hispanic and entered the

U.S. in 1980+ (56%) is -0.31 hours (1% relative to the sample mean), whereas the effect in a

PUMA with one standard deviation higher share (82%) is -0.61 hours (2.2% relative to the

sample mean).

While these findings strongly suggest that changes in the cost of household services

are an important mechanism through which enforcement policies affect high-skilled women’s

labor supply, there are other channels through which changes in the labor supply of undoc-

umented immigrants could affect high-skilled individuals’ work. For example, there could

be complementarities in the production process of market work (Chassamboulli and Peri,

2015; East et al., 2019). We therefore look at the effect of enforcement on different groups

in the population whose labor supply should not be as highly affected through changes in

the price of outsourcing household production. First, we estimate the effects for high-skilled

41The sample size in column (2) is slightly smaller than our baseline models because there are a fewPUMAs with no low-skilled non-citizens, so the denominator in these intensity measures is 0.

25

women with no resident children, since the presence of a child at home affects the demand

of household services. Panel A of Table (9) shows these results. For this sample, the effects

are smaller both in absolute levels and in percentage terms than the effects for mothers, and

none of the estimates are statistically different from zero. However, we note that the confi-

dence intervals on these estimates overlap with those for mothers in Table (3). Due to these

smaller and statistically insignificant findings for high-skilled women without children, we

continue to focus only on mothers for the remainder of the analysis. Importantly, we do not

examine the results for men or fathers in this contemporaneous analysis given the findings

in East et al. (2019) that indicate there may be negative effects on middle-to-higher skilled

men due to market production complementarities. Similarly, we do not look at mothers with

lower levels of education since they are more likely to be affected by substitution in market

production.

We also expect stronger effects for mothers who do not have access to informal help

with household production. To proxy for this, we explore the heterogeneity of the results

by whether a woman lives in the state of her birth. Women that live in their state of birth

might be more likely to live in proximity to their own families, and thus have more access to

informal household production outsourcing (Compton and Pollak, 2014), so we expect the

effects of SC to be larger for women who have moved out of their state of birth (this includes

U.S.-born women not living in their state of birth and all foreign-born citizen women). In

Panels B and C, we show the results of the model that interacts SC, as well as all the

individual-level demographic controls (not shown), with an indicator variable equal to one if

a mother does not live in her state of birth. As expected, the results are stronger for mothers

who are less likely to have access to this type of informal childcare.42 Finally, we verify in

42It is possible that mothers who live in their state of birth are systematically different in other importantways. We show descriptive statistics for both samples in Appendix Table (A6). The samples are very similaron all observable characteristics. The sample that lives out of their state of birth is slightly more likely tobe married, have a Ph.D., and works slightly less on average compared to women still living in their stateof birth. Additionally, we find no evidence that SC affects whether a mother lives in her state of birth andthese results are shown in Appendix Table (A7). We conduct two additional tests: first, we estimate themodel reported in Panels B and C of Table (9), including all the demographic control variables interacted

26

Appendix D that the effects on high-skilled mothers are unlikely to be driven by market

complementarities. To do this we stratify the sample by sector, and we do not find the effect

size to be related to the undocumented employment intensity in the sector. Taken together,

these results all suggest that an important mechanism through which SC affected the labor

supply of high-skilled citizen women is the increased price of household services.

4.2 Lasting Impacts of SC Exposure Around Childbirth

In the contemporaneous results above, we show that SC has the largest negative effect

on mother’s labor supply when their youngest child is under age 3. Previous evidence on

the effects of motherhood have found persistent effects of having a child on women’s labor

market outcomes (Kuziemko et al., 2018; Kleven et al., 2019). So, we test whether having

enforcement policies in place around the time of a child’s birth may have lasting negative

consequences on women’s labor market outcomes. Specifically, we examine the longer-run

effects of exposure to SC when the youngest child is between 0 and 2. We do this on a

sample of high-skilled citizen women whose youngest child was born between 2000 and 2011,

and whose youngest child is observed at ages 3-5 and 6-7 in the 2005-2018 ACS samples.

There are two reasons we chose 2011 as the last birth cohort in this sample: 1) children

who were born after 2012 would be exposed to the ending of SC in the first two years of

life, and we focus only on the program roll-out, and, 2) restricting the last birth year to

be 2011 means we will observe all birth cohorts at ages 3-7. It is important to note that

because of these differences in sample definitions and survey years included compared to

the contemporaneous model, we are using slightly different policy variation than in the main

contemporaneous model, so we do not directly compare these results to the contemporaneous

with SC, and the results are qualitatively similar to those reported in Table (9). Second, it is also possiblethat, since early adopter locations are often in larger states, which might have different state migration ratesthan smaller states, that this test is picking up differences across early and later adopting locations. Toexplore this possibility, we re-estimate these models dropping early-adopter PUMAs; once again the resultsare very similar. Both of these results are available upon request.

27

results.

We estimate the following regression:

Yipts = α + β2SC02ps + β3SC3plusps +X ′iptδ + Z ′ptγ + µp + φt + λs + θ∆W ′p ∗ t+ εipts (3)

where Yipts represents the labor outcomes for a woman i, living in PUMA p and year t,

who had their youngest child in year s. SC02ps is the sum of annual PUMA-level exposure

to SC when the youngest child was 0, 1 and 2, so it can take on a value between 0 and 3.

Therefore, β2 should be interpreted as the effect of one additional year of exposure to SC

before the youngest child turns 3. We also control for SC3plusps, which is the sum of annual

PUMA-level exposure to SC when the youngest child is age 3 to the age when surveyed.

Note, the limitation of this analysis is we only observe PUMA of residence, rather than

PUMA of child’s birth, which may introduce measurement error, and we return to this issue

below. In addition to the controls specified in equation (1), we add a youngest child birth

year fixed effect, λs.43

Table (10) shows the results for the probability of working positive hours in column

(1) and the usual hours worked in column (2). We show in the table both the birth cohorts

included in each sample and the survey years included. Panel A shows the effect of SC

exposure during the first two years of life of the youngest child, when the youngest child

is observed between 3 and 5 years old. We find negative, lasting, effects of exposure to SC

around child birth at these ages. When we change the sample to observe children at older

ages (and thus even longer-run effects) in Panel (B), the point estimates become insignificant

and are mixed in sign, which could indicate a fading out of the effect when children reach

school age, or could be due to more measurement error over time because the likelihood the

family has moved away from the child’s PUMA of birth increases over time.

43The only controls measured based on the year of the youngest child’s birth are the controls for 287(g)exposure at ages 0-2 to mimic the measure of SC exposure.

28

To get a sense of the severity of this measurement error due to migration, we do a

number of things. First, we tabulate the migration rates among these samples to get a sense

of how many mothers might have moved from their child’s PUMA of birth. To do this,

we exploit information from the ACS about the place of residence in the year prior to the

survey year.44 The migration rate for each group is defined as the number of migrants in a

given demographic group relative to the PUMA population in 2005, and we multiply this

number by 100. This rate is quite low–for all mothers with children under age 6, 0.13%

moved PUMAs since the year prior to the survey, and for mothers with children of all ages,

0.27% moved PUMAs. We also directly estimate whether SC impacted the likelihood high-

skilled mothers moved, using a modification of the model in equation (1), where we assign

SC exposure based on the PUMA of residence in the year before the survey (t − 1), and

estimate the effect of SC on the likelihood the mother moves PUMAs between t − 1 and t.

The results in Appendix Table (A4) show that SC did not have a significant effect on the

migration rates of high-skilled citizen mothers (or likely undocumented immigrant women

shown in the third column). Finally, we estimate equation (3) on the sample who report

living in the same PUMA in the year before the survey (about 96% of the full sample),

and therefore may be subject to less measurement error issues. The results in Appendix

Table (A8) show that the long-run effects are quantitatively very similar with and without

migrants.45 Therefore, we do not view this measurement error as severe.

We also estimate the same model for fathers. Appendix Table (A9) shows that in this

case there are no significant effects, and the point estimates are smaller than those estimated

for women–for example, there is significant 1.3% reduction in hours for mothers of children

3 to 5, compared to an insignificant 0.3% reduction in hours worked for fathers of children

44The ACS provides migration information at the MIGPUMA level (slightly larger than the PUMA levelin our main analysis), which identifies the place of residence the year prior to the interview. We generatemigration rates at the consistent MIGPUMA level using this information.

45We have also re-estimated this model looking at the effect around child birth omitting the early adopterPUMAs, since duration of SC exposure may be correlated with early adopter status, and again the resultsare quantitatively very similar. These results are available upon request.

29

3 to 5–although some estimates have overlapping confidence intervals with the effects on

women. It is important to highlight that our analysis does not explore the contemporaneous

effects of SC on men, since previous evidence has found contemporaneous effects of SC

on high-skilled men due to market complementarities (East et al., 2019). In the long-run

analysis, however, we do not expect the effect of market complementarities to be differential

by the age of the youngest child, so we view this analysis as testing a different mechanism–

the cost of outsourcing household production–than in East et al. (2019), and therefore not

contradictory to their findings. These results, together with previous findings that gender

gaps in labor market outcomes seem to be at least partially explained by the “motherhood

gap” (Juhn and McCue, 2017), and that the convergence in the gender wage gap has been

slowest for high-skilled workers (Blau and Kahn, 2017), suggest that an increase in the cost

of household services may be an important avenue that affects the gender gap in the labor

market, particularly for high-skilled workers. Future research should further explore this

possibility with data that more precisely measures exposure to enforcement in a child’s early

life.

4.3 Robustness Checks

We test the robustness of our main results on the labor supply of high-skilled mothers. First,

we test the sensitivity of the findings to alternative timing assumptions. In the baseline

results, we code SC as being in place in a given survey year if it was in place in January

of that year. Appendix Table (A10) Panels A and C replicate the results from the main

specification. In Panels B and D we show the results coding the enforcement policy as

the fraction of the year before the survey year, since the ACS interviews are conducted

continuously throughout the year but we do not know the survey month, and the results are

very similar.

Second, because early adopters of SC may be more highly selected, we test the robust-

30

ness of the results to dropping PUMAs that had at least one county that adopted SC in

2008-2009 in Appendix Table (A11). We define “early adopters” as PUMAs that adopted

SC by the end of 2009, and these PUMAs are shown in Appendix Figure (A4). There

are 155 early-adopter PUMAs and they are mostly along the U.S.-Mexico border, so, while

24% of the total U.S. population live in these PUMAs, 44% of the likely undocumented

population (Hispanic low-skilled non-citizens, who entered the U.S. after 1980) live in these

regions.46 The intensity results shown in Table (8) suggest that the effects of SC are driven

by places with a large population of likely undocumented immigrants, therefore, the over-

all effect may be weakened when we drop the early adopter PUMAs for this reason, and

not because of anything else unique (or endogenous) about these PUMAs. As a result, in

addition to re-estimating the main model (column (1)), we estimate the model interacted

with the percentage of low-skilled non-citizens who are Hispanic low-skilled non-citizens who

entered the U.S. in 1980+ (column (2)). In column (1), for the main model, when we drop

early adopters, the point estimates shrink somewhat and the standard errors increase, so the

estimates become insignificant, although the confidence intervals are overlapping with the

baseline estimates on the full sample. The results in column (2) indicate that this drop in

the coefficient in column (1) is due to the fact that the early adopters are also places with

large likely undocumented populations. In particular, the effects on high-skilled mothers,

when interacted with likely undocumented population intensity, are nearly identical with or

without the early adopter PUMAs included. Therefore, while the results are slightly weaker

when we drop early adopter places, this is explained by the fact they have large likely un-

documented populations, and not because of anything else about these PUMAs that would

suggest the effects in these particular regions are endogenously driven.

Finally, to examine how likely estimates of the magnitude we find would be due to

chance, we implement a placebo test. To do this, we randomly assign county-level SC

46One reason these PUMAs include such a large share of the total population is that Los Angeles countyis included in this sample.

31

implementation dates 1000 times. After each assignment, we aggregate the county-level

data up to the PUMA-level following the method described in section (2), and then re-

estimate equation (1) using this alternative SC variable. We plot the kernel density of the

estimated βs from equation (1) from each of these repetitions in Appendix Figure (A5) for

mothers with young children. We plot our baseline coefficients from Table (3) in the red

vertical lines. Only 4.3% of the distribution is to the left of the estimate on any work, and

only 2.4% is to the left on the number of hours worked, so estimated effects of the magnitude

we find are very unlikely to occur due to chance.

4.4 Discussion

Low-skilled immigrants are over-represented in household services, and a policy-driven de-

crease in immigrant labor supply may result in an increase in the price of these services,

which has important consequences for workers who outsource household production. Our

results support this hypothesis; they indicate a statistically significant negative effect of the

roll-out of SC on high-skilled mother’s labor supply, driven by mothers with children below

school-age. When interpreting our results, it is important to remember that our estimates

are the average effect of SC, and the effects among mothers who change their outsourcing of

household production may be much larger. Directly comparing our estimates to those in the

related literature is not easy, as other papers look at how high-skilled women’s labor supply

is related to the quantity of immigrants in a local area, whereas we demonstrate that the

policy-driven changes in immigrant labor supply here are caused by changes in labor supply

decisions among immigrants living in the U.S. Nevertheless, it is informative to frame our

results in light of these past findings. Cortes and Tessada (2011) use the closest sample to

ours, but use a shift-share approach to identification, and find that a 10% increase in the

low-skill immigrant population in the U.S. was associated with an increase in hours of work

by 0.3% among women earning wages at the top of the distribution. Moreover, Cortes (2008)

32

finds that a 10% increase in low-skill immigration decreased prices of immigrant-intensive

services by 2%. Combining these estimates, this suggests that, due to changes in the im-

migrant population size, there is an elasticity of high-skilled women’s hours worked with

respect to prices of -0.15. Comparing this to our findings, we find SC reduced hours work

among all high-skilled mothers by 0.3% and increased low-skilled household service wages

by 2.2%. Thus, our estimated elasticity of high-skilled mother’s labor supply with respect

to the price of household services is -0.14, and, as expected this elasticity is much larger

for mothers with young children: -0.5.47 Moreover, it is important to point out that our

estimates show that it is not only the number of immigrants that affect high-skilled women’s

decisions, as has been the focus of most of the prior literature, but also the political climate

surrounding immigrants that can have these spillover effects.

In a different context, Farre, Gonzalez and Ortega (2011) find that, in Spain, a 10

percentage point increase in the predicted number of female immigrants living in a local

area increases the likelihood women with children or elderly dependents living with them

work by about 2 percentage points. In the paper using an empirical approach more similar

to ours, but in a very different setting, Cortes and Pan (2013) examine the effect of a

series of policy changes in the 1970s to 2000s regarding foreign domestic workers in Hong

Kong on high-skill women’s labor supply. They identify the effects of these policy changes

in several ways, including comparing long-run changes in the labor supply of women with

and without children over the period of these policy changes in Hong Kong, and find that

women with young children increase the likelihood of working by 12-13 percentage points

over time. Finally, Monras, Vazquez-Grenno et al. (2019) document that a large wave of

immigrant legalization and increased restrictions on informal work in Spain led to a decline

in high-skilled women’s work; for each one newly legalized immigrant, roughly 0.05 high-

skilled citizen women stopped working. The authors argue this is due to the roughly 22%

47These elasticity calculations ignore the large confidence intervals on our estimates, and assume that theentire change in high-skilled mother’s labor supply is due to this change in price.

33

increase in the cost of household services as a result of these two policy changes.

A final point of comparison to our estimates is the literature studying the effect of

childcare prices on mother’s labor supply. Our results are consistent with two findings from

this literature. First, this literature often finds larger elasticities for mothers with children

younger than age 6 (Morrissey, 2017), consistent with theoretical expectations about greater

household production demands before children reach school age. And, second our estimates

imply employment elasticities with respect to wages of household service workers of 0.20

for all mothers, and 0.34 for mothers with young children, which is well within the range

of estimated elasticities of mother’s employment with respect to child care costs from this

literature (Anderson and Levine, 1999; Blau, 2003; Morrissey, 2017).

5 Conclusion

This paper examines the spillover effects of a federal immigration enforcement policy on

the labor outcomes of high-skilled citizen women. Given the over-representation of female

undocumented workers in household services, a negative effect on their labor supply can

affect the cost of outsourcing household production, and thus the labor supply of those most

likely to outsource these services. Our empirical analysis supports this hypothesis.

Exploiting the roll-out of Secure Communities, we estimate a difference-in-differences

model with time and location fixed effects. We find that SC reduced the labor supply of

high-skilled citizen mothers and that these effects are driven by mothers with young children,

who have the greatest demands on their household production time. These results are robust

to including a wide variety of controls including time-varying local economic conditions,

local-area characteristic trends, and individual demographics. Additionally, these results

are supported by an event study model. Importantly, we find several pieces of evidence

that suggest potential lasting effects of exposure to SC around the birth of a child: 1) the

34

reduction in contemporaneous working hours seems to be driven by a decline in full-time

work, and an increase in part-time work, which can harm potential career progression of

women in time-intensive jobs (Bertrand, Goldin and Katz, 2010); and 2) exposure to SC

when child is younger than 3 has negative and persistent effects on the labor supply of

mothers.

To provide support for the hypothesis that changes in the price of outsourcing household

services is an important mechanism behind the labor supply effects on high-skilled citizen

mothers, we look directly at the total labor supply of likely undocumented workers in these

occupations. We find that, while SC did not affect the number of likely undocumented

women working in these occupations, it did reduce their total hours worked. These results

suggest the effects of SC on female labor supply of likely undocumented women operate

through chilling effects, and we provide evidence suggesting this is the case. Then, we show

that this reduction in working hours was accompanied by an increase in the wages of low-

skilled female workers in these occupations. Finally, to further support the increase in the

cost of outsourcing household production as an important mechanism, we document that

there were no similar effects for high-skilled women without children, and that the effects

were stronger for women less likely to have access to informal childcare–those living away

from their extended family.

This paper shows an important spillover effect of immigration enforcement policies onto

citizen workers. Understanding the full effects of enforcement policies is crucially important

today, as immigration policy is being actively debated and changed. For example, recent

policy proposals plan to give priority to high-skilled immigrant workers (Holland and Ramp-

ton, 2019), but our results indicate that the labor supply of low-skilled immigrants can have

positive spillover effects on the employment of high-skilled citizen workers as well. Our pa-

per also speaks to broader literatures that examines how policy can influence women’s labor

supply and time spent in household production, especially around the birth of a child (see

35

for example: Baker, Gruber and Milligan (2008); Baker and Milligan (2008); Cascio (2009);

Havnes and Mogstad (2011); Rossin-Slater, Ruhm and Waldfogel (2013)). The decline in

mother’s labor supply as a result of SC may have far-reaching consequences to the gender

gap in work and wages, as well as children’s well-being. We view this paper as a first step

to analyzing the full impact of immigration enforcement policies on high-skilled women and

their families’ well-being.

36

References

Adao, Rodrigo, Michal Kolesar, and Eduardo Morales. 2018. “Shift-share designs: Theoryand inference.” National Bureau of Economic Research.

Almond, Douglas, Hilary W Hoynes, and Diane Whitmore Schanzenbach. 2011. “Insidethe war on poverty: The impact of food stamps on birth outcomes.” The Review of Economicsand Statistics, 93(2): 387–403.

Alsan, Marcella, and Crystal Yang. 2018. “Fear and the Safety Net: Evidence from SecureCommunities.” National Bureau of Economic Research Working Paper 24731.

Alvarez, Priscilla. 2017. “How Trump Is Changing Immigration Enforcement.” The Atlantic.

Amuedo-Dorantes, Catalina, and Almudena Sevilla. 2014. “Low-Skilled Immigration andParenting Investments of College-Educated Mothers in the United States Evidence from Time-Use Data.” Journal of Human Resources, 49(3): 509–539.

Amuedo-Dorantes, Catalina, and Cynthia Bansak. 2014. “Employment verification man-dates and the labor market outcomes of likely unauthorized and native workers.” ContemporaryEconomic Policy, 32(3): 671–680.

Amuedo-Dorantes, Catalina, and Esther Arenas-Arroyo. 2019. “Police Trust and DomesticViolence: Evidence from Immigration Policies.”

Amuedo-Dorantes, Catalina, and Francisca M. Antman. 2019. “Immigration Enforcement,Awareness, and Worker Engagement.” Working Paper.

Amuedo-Dorantes, Catalina, Thitima Puttitanun, and Ana P Martinez-Donate. 2018.“Deporting Bad Hombres? The Profile of Deportees under Widespread Versus Prioritized En-forcement.” International Migration Review, 0197918318764901.

Anderson, Patricia M, and Philip B Levine. 1999. “Child care and mothers’ employmentdecisions.” National bureau of economic research.

Baker, Bryan, and Nancy Rytina. 2013. “Estimates of the Unauthorized Immigrant Popula-tion Residing in the United States: January 2012.” Pew Research Center.

Baker, Michael, and Kevin Milligan. 2008. “How does job-protected maternity leave affectmothers employment?” Journal of Labor Economics, 26(4): 655–691.

Baker, Michael, Jonathan Gruber, and Kevin Milligan. 2008. “Universal child care, ma-ternal labor supply, and family well-being.” Journal of political Economy, 116(4): 709–745.

Bansak, Cynthia, and Steven Raphael. 2001. “Immigration reform and the earnings of Latinoworkers: Do employer sanctions cause discrimination?” ILR Review, 54(2): 275–295.

Barone, Guglielmo, and Sauro Mocetti. 2011. “With a little help from abroad: the effect oflow-skilled immigration on the female labour supply.” Labour Economics, 18(5): 664–675.

Bertrand, Marianne, Claudia Goldin, and Lawrence F Katz. 2010. “Dynamics of thegender gap for young professionals in the financial and corporate sectors.” American EconomicJournal: Applied Economics, 2(3): 228–255.

37

Bitler, Marianne, and Hilary W Hoynes. 2011. “Immigrants, welfare reform, and the USsafety net.” National Bureau of Economic Research.

Blau, David. 2003. “Child care subsidy programs.” In Means-tested transfer programs in theUnited States. 443–516. University of Chicago Press.

Blau, Francine D, and Lawrence M Kahn. 2007. “Changes in the labor supply behavior ofmarried women: 1980–2000.” Journal of Labor Economics, 25(3): 393–438.

Blau, Francine D, and Lawrence M Kahn. 2017. “The gender wage gap: Extent, trends, andexplanations.” Journal of Economic Literature, 55(3): 789–865.

Bohn, Sarah, and Robert Santillano. 2017. “Local Immigration Enforcement and LocalEconomies.” Industrial Relations: A Journal of Economy and Society, 56(2): 236–262.

Bohn, Sarah, Magnus Lofstrom, and Steven Raphael. 2015. “Do E-verify mandates improvelabor market outcomes of low-skilled native and legal immigrant workers?” Southern EconomicJournal, 81(4): 960–979.

Bonczar, Thomas P. 2003. “Prevalence of Imprisonment in the U.S. Population, 1974-2001.”U.S. Department of Justice.

Borjas, George J. 2003. “The Labor Demand Curve is Downward Sloping: Reexamining the Im-pact of Immigration on the Labor Market.” The Quarterly Journal of Economics, 118(4): 1335–1374.

Borjas, George J., and Lawrence Katz. 2007. “The Evolution of the Mexican-Born Workforcein the United States.” In Mexican Immigration to the United States, edited by George Borjas.

Borusyak, Kirill, Peter Hull, and Xavier Jaravel. 2018. “Quasi-experimental shift-shareresearch designs.” National Bureau of Economic Research.

Brown, J. David, Misty L. Heggeness, Suzanne M. Dorinski, Lawrence Warren, andMoises Yi. 2018. “Understanding the Quality of Alternative Citizenship Data Sources for the2020 Census.” Center for Economic Studies, U.S. Census Bureau Working Papers 18-38.

Cadena, Brian C., and Brian K. Kovak. 2016. “Immigrants Equilibrate Local Labor Mar-kets: Evidence from the Great Recession.” American Economic Journal: Applied Economics,8(1): 257–290.

Card, David. 2001. “Immigrant inflows, native outflows, and the local labor market impacts ofhigher immigration.” Journal of Labor Economics, 19(1): 22–64.

Cascio, Elizabeth U. 2009. “Maternal labor supply and the introduction of kindergartens intoAmerican public schools.” Journal of Human resources, 44(1): 140–170.

Charles, Kerwin Kofi, Erik Hurst, and Matthew J Notowidigdo. 2018. “Housing boomsand busts, labor market opportunities, and college attendance.” American Economic Review,108(10): 2947–94.

Chassamboulli, Andri, and Giovanni Peri. 2015. “The labor market effects of reducing thenumber of illegal immigrants.” Review of Economic Dynamics, 18(4): 792–821.

38

Chaudry, Ajay, Randy Capps, Juan Manuel Pedroza, Rosa Maria Castaneda, RobertSantos, and Molly M Scott. 2010. “Facing Our Future: Children in the Aftermath of Immi-gration Enforcement.” Urban Institute (NJ1).

Compton, Janice, and Robert A Pollak. 2014. “Family proximity, childcare, and women?slabor force attachment.” Journal of Urban Economics, 79: 72–90.

Cortes, Patricia. 2008. “The effect of low-skilled immigration on US prices: evidence from CPIdata.” Journal of political Economy, 116(3): 381–422.

Cortes, Patricia, and Jessica Pan. 2013. “Outsourcing household production: Foreign domes-tic workers and native labor supply in Hong Kong.” Journal of Labor Economics, 31(2): 327–371.

Cortes, Patricia, and Jessica Pan. 2019. “When Time Binds: Substitutes for HouseholdProduction, Returns to Working Long Hours, and the Skilled Gender Wage Gap.” Journal ofLabor Economics, 37(2): 351–398.

Cortes, Patricia, and Jose Tessada. 2011. “Low-skilled immigration and the labor supply ofhighly skilled women.” American Economic Journal: Applied Economics, 3(3): 88–123.

Cox, Adam B, and Thomas J Miles. 2013. “Policing immigration.” The University of ChicagoLaw Review, 80(1): 87–136.

East, Chloe N, Annie Hines, Philip Luck, Hani Mansour, and Andrea Velasquez. 2019.“The Labor Market Effects of Immigration Enforcement.” Working Paper.

Economist, The. 2018. “Immigration worries loom over Sweden’s election campaign.”

Education Commission of the States. 2016. “50-State Review; Full Day Kindergarten: Alook across the states.”

Farre, Lıdia, Libertad Gonzalez, and Francesc Ortega. 2011. “Immigration, family respon-sibilities and the labor supply of skilled native women.” The BE Journal of Economic Analysis& Policy, 11(1).

Felter, Claire, and Danielle Renwickn. 2019. “The U.S. Immigration Debate.” [Council onForeign Relations; posted online 25-July-2019].

Forlani, Emanuele, Elisabetta Lodigiani, and Concetta Mendolicchio. 2015. “Impactof low-skilled immigration on female labour supply.” The Scandinavian Journal of Economics,117(2): 452–492.

Friedberg, Rachel M., and Jennifer Hunt. 1995. “The Impact of Immigrants on Host CountryWages, Employment and Growth.” Journal of Economic Perspectives, 9(2): 23–44.

Furtado, Delia. 2015. “Immigrant labor and work-family decisions of native-born women.” IZAWorld of Labor.

Furtado, Delia. 2016. “Fertility Responses of High-Skilled Native Women to Immigrant Inflows.”Demography, 53(1): 27–53.

Furtado, Delia, and Heinrich Hock. 2010. “Low Skilled Immigration and Work-FertilityTradeoffs Among High Skilled US Natives.” The American economic review, 100(2): 224–228.

39

Genoni, Maria, Gabriela Farfan, Luis Rubalcava, Graciela Teruel, Duncan Thomas,and Andrea Velasquez. 2017. “Mexicans in America.” BREAD Working Paper.

Givens, Terri. 2018. “The Most Significant Global Debate Is Over Immigration.” [Arc Digital;posted online 23-July-2018].

Goldsmith-Pinkham, Paul, Isaac Sorkin, and Henry Swift. 2018. “Bartik instruments:What, when, why, and how.” National Bureau of Economic Research.

Gonzalez, A. 2015. “More Mexicans leaving than coming to the US.” Washington, DC, PewHispanic Center, 19.

Goodman-Bacon, Andrew. 2018. “Difference-in-differences with variation in treatment tim-ing.” National Bureau of Economic Research.

Grittner, Amanda M. 2019. “Local Immigration Enforcement and Domestic Violence.”

Hansen, Collin. 2019. “Sweet Casa Alabama (and Arizona, and...): Examining the EconomicOutcomes of State Immigration Reform.” Available at SSRN 3377360.

Havnes, Tarjei, and Magne Mogstad. 2011. “Money for nothing? Universal child care andmaternal employment.” Journal of Public Economics, 95(11): 1455–1465.

Hines, Annie Laurie, and Giovanni Peri. 2019. “Immigrants’ Deportations, Local Crime andPolice Effectiveness.”

Hock, Heinrich, and Delia Furtado. 2009. “Female Work and Fertility in the United States:Effects of Low-Skilled Immigrant Labor.” University of Connecticut Department of EconomicsWorking Paper, 20.

Hoefer, Michael, Nancy Francis Rytina, and Brian Baker. 2012. Estimates of the unau-thorized immigrant population residing in the United States: January 2011. Citeseer.

Holland, Steve, and Roberta Rampton. 2019. “Trump to propose plan to make U.S. immi-gration more merit-based.” [Reuters; posted online 15-May-2019].

Hoynes, Hilary W, and Diane Whitmore Schanzenbach. 2009. “Consumption responsesto in-kind transfers: Evidence from the introduction of the food stamp program.” AmericanEconomic Journal: Applied Economics, 1(4): 109–39.

Jaeger, David A, Joakim Ruist, and Jan Stuhler. 2018. “Shift-share instruments and theimpact of immigration.” National Bureau of Economic Research.

Jessen-Howard, Steven, Rasheed Malik, Simon Workman, and Katie Hamm. 2018.“Understanding Infant and Toddler Child Care Deserts.” Center for American Progress.

Juhn, Chinhui, and Kristin McCue. 2017. “Specialization then and now: Marriage, children,and the gender earnings gap across cohorts.” Journal of Economic Perspectives, 31(1): 183–204.

Kleven, Henrik, Camille Landais, Johanna Posch, Andreas Steinhauer, and JosefZweimuller. 2019. “Child penalties across countries: Evidence and explanations.” Vol. 109,122–26.

40

Kohli, Aarti, Peter L Markowitz, and Lisa Chavez. 2011. “Secure communities by thenumbers: An analysis of demographics and due process.” The chief justice earl warren instituteon law and social policy, 1–20.

Kostandini, Genti, Elton Mykerezi, and Cesar Escalante. 2013. “The impact of immi-gration enforcement on the US farming sector.” American Journal of Agricultural Economics,96(1): 172–192.

Krogstad, Jens Manuel, Jeffrey S Passel, and D’Vera Cohn. 2017. “5 facts about illegalimmigration in the U.S.” Pew Research Center.

Krogstad, Jens Manuel, Jeffrey S Passel, and D’Vera Cohn. 2018. “5 facts about illegalimmigration in the U.S.” Pew Research Center.

Kuziemko, Ilyana, Jessica Pan, Jenny Shen, and Ebonya Washington. 2018. “TheMommy Effect: Do Women Anticipate the Employment Effects of Motherhood?” NationalBureau of Economic Research.

Longhi, Simonetta, Peter Nijkamp, and Jacques Poot. 2005. “A Meta-Analytic Assessmentof the Effect of Immigration on Wages.” Journal of Economic Surveys, 19(3): 451–477.

Longhi, Simonetta, Peter Nijkamp, and Jacques Poot. 2006. “The impact of immigrationon the employment of natives in regional labour markets: A meta-analysis.” IZA Working Paper.

Menasce Horowitz, Juliana. 2014. “Widespread anti-immigrant sentiment in Italy.”

Miles, Thomas J, and Adam B Cox. 2014. “Does immigration enforcement reduce crime?evidence from secure communities.” The Journal of Law and Economics, 57(4): 937–973.

Monras, Joan, Javier Vazquez-Grenno, et al. 2019. “Understanding the effects of legalizingundocumented immigrants.”

Morrissey, Taryn W. 2017. “Child care and parent labor force participation: a review of theresearch literature.” Review of Economics of the Household, 15(1): 1–24.

National Center for Education Statistics. 2018. “Types of state and district requirementsfor kindergarten entrance and attendance, waivers and exemptions for kindergarten entrance,by state.”

Orrenius, Pia M, and Madeline Zavodny. 2009. “The effects of tougher enforcement on the jobprospects of recent Latin American immigrants.” Journal of Policy Analysis and Management,28(2): 239–257.

Orrenius, Pia M, and Madeline Zavodny. 2015. “The impact of E-Verify mandates on labormarket outcomes.” Southern Economic Journal, 81(4): 947–959.

Passel, Jeffrey, and D’Vera Cohn. 2016. “Overall Number of U.S. Unauthorized ImmigrantsHolds Steady Since 2009.” Pew Research Center.

Passel, Jeffrey S, and D’Vera Cohn. 2011. Unauthorized immigrant population: National andstate trends, 2010. Pew Hispanic Center Washington, DC.

Passel, Jeffrey S, and D’Vera Cohn. 2014. “Unauthorized Immigrant Totals Rise in 7 States,Fall in 14.” Pew Research Center.

41

Peri, Giovanni, Agnese Romiti, and Mariacristina Rossi. 2015. “Immigrants, domesticlabor and women’s retirement decisions.” Labour Economics, 36: 18–34.

Pew Research Center. 2013. “Another Gender Gap: Men Spend More Time in Leisure Activi-ties.” Pew Research Center.

Pham, Huyen, and Pham Hoang Van. 2010. “Economic impact of local immigration regula-tion: an empirical analysis.” Immigr. & Nat’lity L. Rev., 31: 687.

Phillips, Julie A, and Douglas S Massey. 1999. “The new labor market: Immigrants andwages after IRCA.” Demography, 36(2): 233–246.

Rhodes, Scott D, Lilli Mann, Florence M Siman, Eunyoung Song, Jorge Alonzo, MarioDowns, Emma Lawlor, Omar Martinez, Christina J Sun, Mary Claire O’Brien,et al. 2015. “The impact of local immigration enforcement policies on the health of immigrantHispanics/Latinos in the United States.” American Journal of Public Health, 105(2): 329–337.

Romiti, Agnese. 2018. “The effects of immigration on household services, labour supply andfertility.” Oxford Bulletin of Economics and Statistics, 80(4): 843–869.

Rossin-Slater, Maya, Christopher J Ruhm, and Jane Waldfogel. 2013. “The effects ofCalifornia’s paid family leave program on mothers leave-taking and subsequent labor marketoutcomes.” Journal of Policy Analysis and Management, 32(2): 224–245.

Ruggles, Steven, Katie Genadek, Ronald Goeken, Josiah Grover, and Matthew Sobek.2017. “Integrated Public Use Microdata Series: Version 7.0. [dataset].”

Sakuma, Amanda. 2017. “Donald Trumps Plan to Outsource Immigration Enforcement to LocalCops.” The Atlantic.

Theodore, Nik. 2013. “Insecure communities: Latino perceptions of police involvement in im-migration enforcement.” Chicago, IL: Department of Urban Planning and Policy, University ofIllinois at Chicago.

Thompson, Derek. 2018. “How immigration became so controversial.”

Valdivia, Carolina. 2019. “Expanding Geographies of Deportability: How Immigration Enforce-ment at the Local Level Affects Undocumented and Mixed-Status Families.” Law & Policy,41(1): 103–119.

Van Hook, Jennifer, Frank D Bean, James D Bachmeier, and Catherine Tucker. 2014.“Recent trends in coverage of the Mexican-born population of the United States: Results fromapplying multiple methods across time.” Demography, 51(2): 699–726.

Wang, Julia Shu-Huah, and Neeraj Kaushal. 2018. “Health and mental health effects oflocal immigration enforcement.” International Migration Review, 0197918318791978.

Warren, Robert. 2014. “Democratizing Data about Unauthorized Residents in the UnitedStates: Estimates and Public-use Data, 2010 to 2013.” Journal on Migration and Human Secu-rity, 2(4): 305–328.

Warren, Robert, and John Robert Warren. 2013. “Unauthorized Immigration to the UnitedStates: Annual Estimates and Components of Change, by State, 1990 to 2010.” InternationalMigration Review, 47(2): 296–329.

42

Watson, Tara. 2013. “Enforcement and immigrant location choice.” National Bureau of Eco-nomic Research.

Winders, Jamie. 2016. “Immigration and the 2016 Election.” southeastern geographer,56(3): 291–296.

Workman, Simon, and Steven Jessen-Howard. 2018. “Understanding the true cost of childcare for infants and toddlers.” Center for American Progress.

43

6 Figures

Figure 1: Rollout of Secure Communities at the PUMA level by Year

Notes: PUMAs that had adopted Secure Communities based on January of each year are shaded. See text for source.

44

Figure 2: Effect of SC on the Labor Supply of High-Skilled Mothers with Kids Under 6, Event Study

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with children under age 6,and with a college degree or more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMAcharacteristics trends and individual demographic controls. The PUMA-year controls include: labor demand controls, housing pricecontrols, and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with the change in the followingPUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housing prices, the share of thePUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and have a collegedegree, masters degree, or a Ph.D., as well as the same education categories just for women. The demographic controls include: age,number of kids, number of kids under age 6, educational attainment, marital status, and race. The results are weighted using theindividual-level weights in the ACS. Standard errors are clustered at the PUMA level and the 95% confidence intervals are shown by thedashed lines. The horizontal axis denotes “event time” where the omitted year is the year before the first SC policy in the PUMA wasimplemented.

45

7 Tables

Table 1: Correlation of 2000-2005 Changes in PUMA Characteristics and SC Adoption Year

Mean of Change in Characteristic Standard Deviation of Change in Characteristic Regression Estimate

Change % Citizen 0.005 0.023 -1.083(2.216)

Change % Black 0.001 0.025 -1.532(1.429)

Change % Labor Force Participation 0.588 2.543 -0.003(0.015)

Change % Non-Citizen 0.009 0.024 -4.868**(2.047)

Change % with Children Under 6 -0.006 0.024 -1.966(1.666)

Change % with Children -0.008 0.030 0.124(1.310)

Change % Work > 50 Hours if Work -1.022 2.113 0.028(0.023)

Change % Work > 60 Hours if Work -0.432 1.242 -0.019(0.038)

Change % with College 0.166 0.021 2.587(2.781)

Change % with Masters 0.010 0.013 7.489(4.677)

Change % with Ph.D. or Professional Degree 0.001 0.008 7.377(6.464)

Change % Women with College 0.010 0.014 -1.372(4.127)

Change % Women with Masters 0.007 0.008 4.823(7.066)

Change % Women with Ph.D. 0.001 0.005 -10.011(10.732)

Change Unemployment Rate 1.10 1.011 -0.068(0.042)

Change Housing Prices 47.47 31.217 -0.007***(0.001)

Change Detentions 5.18 55.07 -0.001(0.001)

Mean Y 2011.71R-Squared 0.06N 1071

Notes: Data are from the American Community Survey, 2000 Census, and TRAC Detentions Data. We estimate the following regression:yearp = α + θ∆W ′

p + εp where yearp is the first year SC was implemented in the PUMA. ∆W ′p includes changes in PUMA-level

demographics and economic conditions between 2000 and 2005. The detention data begin in 2002, so for this variable we use the changefrom 2002 to 2005.

46

Table 2: Summary Statistics

High-Skilled Women High-Skilled Men

All With Kids With Kids Under 6 All

ACS

Age 41.86 42.07 34.28 43.30Black 0.09 0.09 0.08 0.07Married 0.61 0.81 0.89 0.65# Children Under 6 0.20 0.43 1.30 0.19# All Children 0.86 1.83 1.95 0.84College Degree 0.66 0.66 0.66 0.66Masters Degree 0.26 0.26 0.26 0.22Ph.D. or Professional Degree 0.08 0.08 0.08 0.11Work >0 Hours (*100) 85.49 83.00 78.98 93.01Usual Hours Worked per Week 33.10 31.12 28.78 41.30Secure Communities 0.36 0.36 0.35 0.35N 2558751 1213275 392637 2216524

ATUS (2005)

Hours Spent Caring for Children in Household per Week 5.57 14.01 21.52 2.84Hours Spent on Household Activities per Week 14.80 18.01 16.93 10.24N 766 407 188 664

Notes: Data are from the 2005-2014 American Community Survey and the 2005 American Time Use Survey. Thesample includes all U.S. citizens with a college degree or more, aged 20-64. The results are weighted using individual-level weights in the ACS and in the ATUS.

47

Table

3:

Eff

ect

of

SC

on

Hig

h-S

kil

led

Wom

en’s

Lab

or

Su

pp

lyby

Pre

sen

ceof

Ch

ild

ren

All

Wom

enW

ith

Kid

sof

Any

Age

Wit

hK

ids

Under

6

Wor

k>

0H

ours

Usu

alH

ours

Wor

ked

Wor

k>

0H

ours

Usu

alH

ours

Wor

ked

Wor

k>

0H

ours

Usu

alH

ours

Wor

ked

A:PUMA

FE,YearFE

Sec

ure

Com

munit

ies

-0.2

57∗∗

-0.1

17∗

-0.3

65∗∗

-0.1

31-0

.606∗

-0.3

35∗∗

(0.1

21)

(0.0

65)

(0.1

79)

(0.0

89)

(0.3

58)

(0.1

59)

Mea

nY

85.4

933

.11

83.0

131

.13

78.9

928

.79

P-V

alue

0.03

0.07

0.04

0.14

0.09

0.04

B:Add

PUMA-Y

earCon

trols

Sec

ure

Com

munit

ies

-0.2

82∗∗

-0.1

20∗

-0.4

08∗∗

-0.1

40-0

.664∗

-0.3

53∗∗

(0.1

24)

(0.0

66)

(0.1

83)

(0.0

90)

(0.3

62)

(0.1

61)

Mea

nY

85.4

933

.11

83.0

131

.13

78.9

928

.79

P-V

alue

0.02

0.07

0.03

0.12

0.07

0.03

C:Add

PUMA

CharacteristicTrends

Sec

ure

Com

munit

ies

-0.2

39∗

-0.0

90-0

.366∗∗

-0.1

10-0

.636∗

-0.3

30∗∗

(0.1

25)

(0.0

66)

(0.1

84)

(0.0

90)

(0.3

62)

(0.1

60)

Mea

nY

85.4

933

.11

83.0

031

.12

78.9

928

.79

P-V

alue

0.06

0.17

0.05

0.22

0.05

0.02

D:Add

Dem

ographics

Sec

ure

Com

munit

ies

-0.2

77∗∗

-0.1

04-0

.361∗∗

-0.1

04-0

.595∗

-0.3

10∗∗

(0.1

23)

(0.0

66)

(0.1

82)

(0.0

88)

(0.3

55)

(0.1

53)

Mea

nY

85.4

933

.11

83.0

031

.12

78.9

828

.78

P-V

alue

0.02

0.11

0.05

0.24

0.09

0.04

N25

5875

125

5875

112

1327

512

1327

539

2637

3926

37

Note

s:D

ata

are

from

the

2005-2

014

Am

eric

an

Com

mu

nit

yS

urv

ey.

Th

esa

mp

lein

clu

des

all

U.S

.ci

tize

nw

om

enw

ith

aco

lleg

ed

egre

eor

more

aged

20-6

4.

All

mod

els

incl

ud

eP

UM

Afi

xed

effec

ts,

an

dyea

rfi

xed

effec

ts.

Th

eP

UM

A-y

ear

contr

ols

incl

ud

e:la

bor

dem

an

dco

ntr

ols

,h

ou

sin

gp

rice

contr

ols

,an

d287(g

)p

rogra

ms.

Th

eP

UM

Ach

ara

cter

isti

ctr

end

sin

clu

de

inte

ract

ion

sof

ati

me

tren

dw

ith

the

chan

ge

inth

efo

llow

ing

PU

MA

chara

cter

isti

csb

etw

een

2000

an

d2005:

lab

or

forc

ep

art

icip

ati

on

rate

,u

nem

plo

ym

ent

rate

,an

dh

ou

sin

gp

rice

s,th

esh

are

of

the

PU

MA

that

are

citi

zen

s,b

lack

,n

on

-cit

izen

s,h

ave

child

ren

,h

ave

you

ng

child

ren

,w

ork

more

than

50

an

d60

hou

rs,

an

dh

ave

aco

lleg

ed

egre

e,m

ast

ers

deg

ree,

or

aP

h.D

.,as

wel

las

the

sam

eed

uca

tion

cate

gori

esju

stfo

rw

om

en.

Th

ein

div

idu

al

dem

ogra

ph

icco

ntr

ols

incl

ud

e:age,

nu

mb

erof

kid

s,nu

mb

erof

kid

su

nd

erage

6,

edu

cati

on

al

att

ain

men

t,m

ari

tal

statu

s,an

dra

ce.

Th

ere

sult

sare

wei

ghte

du

sin

gth

ein

div

idu

al-

level

wei

ghts

inth

eA

CS

.S

tan

dard

erro

rscl

ust

ered

at

the

PU

MA

level

an

dsh

ow

nin

pare

nth

eses

.*

p<

0.1

0,

**

p<

0.0

5,

***

p<

0.0

1

48

Table 4: Effect of SC on High-Skilled Mothers’ Labor Supply, By Age of Youngest Child

Work > 0 Usual HoursHours Worked

A: Youngest Kid 0-2Secure Communities -0.736∗ -0.421∗∗

(0.439) (0.200)Mean Y 78.85 28.73P-Value 0.09 0.04N 268649 268649

B: Youngest Kid 3-5Secure Communities -0.257 -0.149

(0.501) (0.230)Mean Y 79.65 29.05P-Value 0.61 0.52N 179519 179519

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with a college degree or moreaged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trends and demographiccontrols. The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs. The PUMA characteristictrends include interactions of a time trend with the change in the following PUMA characteristics between 2000 and 2005: labor forceparticipation rate, unemployment rate, and housing prices, the share of the PUMA that are citizens, black, non-citizens, have children, haveyoung children, work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D., as well as the same educationcategories just for women. The individual demographic controls include: age, number of kids, number of kids under age 6, educationalattainment, marital status, and race. The results are weighted using the individual-level weights in the ACS. Standard errors clustered atthe PUMA level and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

Table 5: Effect of SC on High-Skilled Women’s Probability of Working Full-Time and Part-Time by Presence ofChildren

=1 if Working 35+ Hours = 1 if Working 20−35 Hours = 1 if Working <20 Hours

A: Full SampleSecure Communities -0.302∗ 0.131 0.171

(0.174) (0.120) (0.144)Mean Y 67.25 12.45 20.30P-Value 0.08 0.28 0.24N 2558751 2558751 2558751

B: Kids of Any AgeSecure Communities -0.213 0.078 0.135

(0.249) (0.179) (0.206)Mean Y 62.24 13.78 23.98P-Value 0.39 0.66 0.51N 1213275 1213275 1213275

C: Kids Under 6Secure Communities -0.858∗∗ 0.133 0.725∗∗

(0.431) (0.316) (0.366)Mean Y 57.44 13.65 28.92P-Value 0.05 0.67 0.05N 392637 392637 392637

Notes: Data are from the 2005-2014 American Community Survey. The sample includes all U.S. citizen women with a college degree ormore aged 20-64. All models include PUMA fixed effects, and year fixed effects. The PUMA-year controls include: labor demand controls,housing price controls, and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with the change inthe following PUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housing prices, theshare of the PUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and have acollege degree, masters degree, or a Ph.D., as well as the same education categories just for women. The individual demographic controlsinclude: age, number of kids, number of kids under age 6, educational attainment, marital status, and race. The results are weightedusing the individual-level weights in the ACS. Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10, **p<0.05, *** p<0.01

49

Table 6: Effect of SC on Female Non-Citizens’ Labor Supply in Household Services

Hisp 80+ MX 80+ CA/MX 80+

A: (Total # Work in Household Services / Total PUMA Pop) *100

Secure Communities 0.002 -0.001 -0.002(0.008) (0.006) (0.007)

Mean Y 0.22 0.14 0.19P-Value SC 0.82 0.85 0.76N 10710 10710 10710

B: (Total # Hours Work in Household Services / Total PUMA Pop) *100

Secure Communities -0.279 -0.247 -0.425∗

(0.266) (0.203) (0.242)Mean Y 6.97 4.27 6.02P-Value SC 0.29 0.23 0.08N 10710 10710 10710

C: (Total # Hours Work in Household Services / Total PUMA Pop) *100

Secure Communities 1.482 1.179∗ 1.127(0.993) (0.693) (0.836)

SC * (Share Dep Hispanic) -1.940∗ -1.548∗∗ -1.718∗

(1.052) (0.759) (0.908)Mean Y 6.90 4.27 6.02Mean Intensity 0.92 0.92 0.92SD Intensity 0.12 0.12 0.12β–Mean Int -0.30 -0.24 -0.45β–1 SD Higher Int -0.53 -0.42 -0.65P-Value SC 0.14 0.09 0.18P-value for F-Test: SC + SC * (Share Dep Hispanic)=0 0.10 0.09 0.05N 10500 10500 10500

D: (Total # Hours Work in Household Services / Total PUMA Pop) *100

Secure Communities 0.029 -0.123 -0.311(0.383) (0.280) (0.332)

SC * (Share Dep Non-Serious Crimes) -1.234 -0.448 -0.519(0.946) (0.734) (0.854)

Mean Y 6.90 4.27 6.02Mean Intensity 0.27 0.27 0.27SD Intensity 0.12 0.12 0.12β–Mean Int -0.30 -0.24 -0.45β–1 SD Higher Int -0.45 -0.30 -0.51P-Value SC 0.94 0.66 0.35P-value for F-Test: SC + SC * (Share Dep Non-Serious Crimes)=0 0.19 0.42 0.15N 10500 10500 10500

Notes: Data are from the 2005-2014 American Community Survey. The sample includes women aged 20-64 who report being non-citizens,who have at most a high school degree, and who report their current or most recent occupation as household services. All models includePUMA fixed effects, year fixed effects, PUMA-year controls, and PUMA characteristics trends. The PUMA-year controls include: labordemand controls, housing price controls, and 287(g) programs. The PUMA characteristic trends include interactions of a time trendwith the change in the following PUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, andhousing prices, the share of the PUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and60 hours, and have a college degree, masters degree, or a Ph.D., as well as the same education categories just for women. The results areweighted using PUMA population in 2000. Standard errors clustered at the PUMA level and shown in parentheses. ”MX”=individualsborn in Mexico. ”MX/CA”=individuals born in Mexico or Central America. * p<0.10, ** p<0.05, *** p<0.01

50

Table 7: Effect of SC on Average Hourly Wages of Female Workers in Household Services

All Low-SkillSC 0.010 0.022∗

(0.011) (0.013)Mean Y 2.30 2.29P-Value 0.34 0.10N 10670 10438

Notes: Data are from the 2005-2014 American Community Survey. The sample includes women aged 20-64 who report their currentor most recent occupation as household services. All models include PUMA fixed effects, year fixed effects, PUMA-year controls, andPUMA characteristics trends. The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs.The PUMA characteristic trends include interactions of a time trend with the change in the following PUMA characteristics between2000 and 2005: labor force participation rate, unemployment rate, and housing prices, the share of the PUMA that are citizens, black,non-citizens, have children, have young children, work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D.,as well as the same education categories just for women. The results are weighted using PUMA population in 2000. Standard errorsclustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

51

Table 8: Effect of SC on High-Skilled Mothers’ Labor Supply by Intensity of Treatment

Intensity of Treatment Measure

% of PUMA pop that is % of PUMA LSNC pop that islikely undoc in 2005 likely undoc in 2005

A: Work >0 Hours, Women with KidsSecure Communities -0.244 0.148

(0.213) (0.285)SC * Intensity -2.768 -0.893∗∗

(2.702) (0.382)Mean Y 83.00 82.99Mean Intensity 0.04 0.56SD Intensity 0.04 0.26β–Mean Int -0.35 -0.35β–1 SD Higher Int -0.47 -0.59P-Value SC 0.25 0.60P-value for F-Test: SC + SC * (Intensity)=0 0.09 0.01N 1213275 1210447

B: Hours Worked, Women with KidsSecure Communities -0.051 0.231∗

(0.103) (0.137)SC * Intensity -1.263 -0.596∗∗∗

(1.268) (0.185)Mean Y 31.12 31.11Mean Intensity 0.04 0.56SD Intensity 0.04 0.26β–Mean Int -0.10 -0.10β–1 SD Higher Int -0.16 -0.26P-Value SC 0.62 0.09P-value for F-Test: SC + SC * (Intensity)=0 0.31 0.00N 1213275 1210447

C: Work >0 Hours, Women with Kids Under 6Secure Communities -0.162 0.791

(0.424) (0.549)SC * Intensity -10.121∗∗ -2.444∗∗∗

(4.695) (0.733)Mean Y 78.98 78.97Mean Intensity 0.04 0.56SD Intensity 0.04 0.26β–Mean Int -0.54 -0.59β–1 SD Higher Int -1.00 -1.23P-Value SC 0.70 0.15P-value for F-Test: SC + SC * (Intensity)=0 0.02 0.00N 392637 391786

D: Hours Worked, Women with Kids Under 6Secure Communities -0.091 0.351

(0.182) (0.226)SC * Intensity -5.104∗∗ -1.165∗∗∗

(2.294) (0.327)Mean Y 28.78 28.78Mean Intensity 0.04 0.56SD Intensity 0.04 0.26β–Mean Int -0.28 -0.31β–1 SD Higher Int -0.51 -0.61P-Value SC 0.62 0.12P-value for F-Test: SC + SC * (Intensity)=0 0.01 0.00N 392637 391786

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with a college degree ormore aged 20-64. Each column shows the results of the model interacted with a different intensity measure. Column 1 shows the resultsof the model interacted with the share of the PUMA working-age population that is likely undocumented in 2005 using our baselinedefinition (Hispanic low-skilled non-citizens who entered the U.S. in 1980+). Column 2 shows the results of the model interacted with theshare of the PUMA low-skilled non-citizen population that is Hispanic and entered the U.S. in 1980+ in 2005. The model includes PUMAfixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trends and demographic controls. The PUMA-year controlsinclude: labor demand controls, housing price controls, and 287(g) programs. The PUMA characteristic trends include interactions of atime trend with the change in the following PUMA characteristics between 2000 and 2005: labor force participation rate, unemploymentrate, and housing prices, the share of the PUMA that are citizens, black, non-citizens, have children, have young children, work morethan 50 and 60 hours, and have a college degree, masters degree, or a Ph.D., as well as the same education categories just for women.The individual demographic controls include: age, number of kids, number of kids under age 6, educational attainment, marital status,and race. The results are weighted using the individual-level weights in the ACS. Standard errors clustered at the PUMA level and shownin parentheses. * p<0.10, ** p<0.05, *** p<0.01

52

Table 9: Effect of SC on Labor SupplyFurther Tests of Mechanisms

Work > 0 Usual HoursHours Worked

A: High-Skilled Women with No ChildrenSecure Communities -0.174 -0.079

(0.159) (0.085)Mean Y 87.68 34.86N 1345476 1345476

B: Any Kids, Split by State of ResidenceSecure Communities -0.074 0.061

(0.210) (0.099)SC * State Birth 6= State Resid -0.548∗∗∗ -0.314∗∗∗

(0.185) (0.084)Mean Y 83.00 31.12Fraction State Birth 6= State Resid 52.07 52.07N 1213275 1213275

C: Kids Under 6, Split by State of ResidenceSecure Communities -0.342 -0.185

(0.381) (0.168)SC * State Birth 6= State Resid -0.496 -0.243∗

(0.342) (0.146)Mean Y 78.98 28.78Fraction State Birth 6= State Resid 49.29 49.29N 392637 392637

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen women with a college degree ormore aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trends anddemographic controls. The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs. ThePUMA characteristic trends include interactions of a time trend with the change in the following PUMA characteristics between 2000and 2005: labor force participation rate, unemployment rate, and housing prices, the share of the PUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D., aswell as the same education categories just for women. The individual demographic controls include: age, number of kids, number of kidsunder age 6, educational attainment, marital status, and race. The results are weighted using the individual-level weights in the ACS.Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

53

Table 10: Lasting Effects of SC around Birth on Labor Supply of High-Skilled Citizen Mothers

Work > 0 Usual HoursHours Worked

A: Youngest Child Age 3−5SC when Youngest Aged 0−2 -0.537∗ -0.383∗∗∗

(0.291) (0.133)Mean Y 79.98 29.25P-Value 0.07 0.00Survey Years Included 2005-2016 2005-2016Birth Years Included 2000-2011 2000-2011N 197016 197016

B: Youngest Child Age 6−7SC when Youngest Aged 0−2 0.041 -0.063

(0.324) (0.160)Mean Y 84.14 31.18P-Value 0.90 0.69Survey Years Included 2006-2018 2006-2018Birth Years Included 2000-2011 2000-2011N 131124 131124

Notes: Data are from the 2005-2018 American Community Survey. The sample includes U.S. citizen mothers with a college degree ormore aged 20-64 who gave birth to their youngest child between 2000-2011. The model includes PUMA fixed effects, year (of survey) fixedeffects, year of birth of the youngest child fixed effects, PUMA-year controls, PUMA characteristics trends and demographic controls.The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs. The PUMA characteristictrends include interactions of a time trend with the change in the following PUMA characteristics between 2000 and 2005: labor forceparticipation rate, unemployment rate, and housing prices, the share of the PUMA that are citizens, black, non-citizens, have children,have young children, work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D., as well as the same educationcategories just for women. The results are weighted using PUMA population in 2000. The demographic controls include: age, number ofkids, number of kids under age 6, educational attainment, marital status, and race. The results are weighted using the individual-levelweights in the ACS. Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

54

A TRAC Data DescriptionWe use data obtained from the Transactional Records Access Clearinghouse (TRAC) on indi-viduals deported under SC between 2008 and 2014. For each individual we have demographicinformation (e.g. age, gender, country of citizenship), as well as the county of apprehension, anddate of removal (not date of apprehension). TRAC obtained this data through Freedom of In-formation Requests to U.S. Immigration and Customs Enforcement. We aggregate to the PUMAlevel using a similar weighting process as described in the main text for the SC variable. We usethe data to generate summary statistics of all those deported under SC shown in Appendix Table(A1) as well as measures of policy intensity described in more detail in the text. Additionally,we use a very similar data set on detentions from TRAC in Table (1) to look at pre-trends indetentions.

B Control Variables DescriptionIn some regressions, we include controls for labor demand, housing prices, and other enforcementpolicies described here. First, we construct four Bartik-style measures of labor demand thatcorrespond to the following four demographic groups: 1) all working-age adults, 2) foreign-bornworking-age adults, 3) working-age women with a college degree or more, and 4) working-agemen with a college degree or more. For each of these four demographic groups, we calculate thePUMA-level group-specific employment by industry, as a fraction of total group-specific PUMAemployment in 2005. We then apply to these group-specific industry shares the changes in nationalgroup-specific employment for the group-specific national sample of working age adults for eachindustry over time, to obtain a measure of predicted changes in local labor demand. Second, thehousing prices information comes from the Federal Housing Finance Agency and is available atthe county by year level. Finally, start and end dates for all 287(g) agreements came from reportspublished by ICE, the Department of Homeland Security, the Migration Policy Institute, as wellas Kostandini, Mykerezi and Escalante (2013), and various news articles. We focus on county-level 287(g) agreements only and ignore state-level agreements. We aggregate up the county-levelhousing and 287(g) information to the PUMA level using the same weighting process as describedin the main text for the SC variable.

55

C Additional Results

Figure A1: County to PUMA Matching Example

Notes: This figure shows the four counties that make up one PUMA. Additionally, the counties’ populations are shown, and associatedweights that are used to aggregate the county-level data to the PUMA-level.

56

Figure A2: Effect of SC on High-Skilled Mothers with Kids Under 6 Labor Supply, Event StudyEstimated on Balanced Subsample of PUMAs

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with children under age 6,and with a college degree or more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMAcharacteristics trends and individual demographic controls. The PUMA-year controls include: labor demand controls, housing pricecontrols, and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with the change in the followingPUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housing prices, the share of thePUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and have a collegedegree, masters degree, or a Ph.D., as well as the same education categories just for women. The demographic controls include: age,number of kids, number of kids under age 6, educational attainment, marital status, and race. The results are weighted using theindividual-level weights in the ACS. The horizontal axis denotes “event time” where the omitted year is the year before the first SCpolicy in the PUMA was implemented. The point estimates in black reflect our main estimates including all PUMAs, and the pointestimates in gray reflect the estimates for a subsample of PUMAs that first implemented SC before 2013, in order to observe a balancednumber of observations in both the pre and post periods.

57

Figure A3: Effect of SC on High-Skilled Mothers with Kids Under 6 Labor Supply, Event StudyDefined as 50% of PUMA covered by SC

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with children under age 6,and with a college degree or more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMAcharacteristics trends and individual demographic controls. The PUMA-year controls include: labor demand controls, housing pricecontrols, and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with the change in the followingPUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housing prices, the share of thePUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and have a collegedegree, masters degree, or a Ph.D., as well as the same education categories just for women. The demographic controls include: age,number of kids, number of kids under age 6, educational attainment, marital status, and race. The results are weighted using theindividual-level weights in the ACS. Standard errors are clustered at the PUMA level and the 95% confidence intervals are shown by thedashed lines.The horizontal axis denotes “event time” where the omitted year is the year before the 50% of the PUMA has enacted SC.

58

Figure A4: Early Adopter PUMAs

Notes: PUMAs that had adopted Secure Communities “early” (before the end of 2009) are shaded. See text for source.

59

Figure A5: Effect of SC on High-Skilled Mothers with Kids Under 6 Labor Supply, Placebo Tests(a) Work >0 Hours

(b) Hours Worked

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with children under age 6,and with a college degree or more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMAcharacteristics trends and individual demographic controls. The PUMA-year controls include: labor demand controls, housing pricecontrols, and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with the change in the followingPUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housing prices, the share of thePUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and have a collegedegree, masters degree, or a Ph.D., as well as the same education categories just for women. The demographic controls include: age,number of kids, number of kids under age 6, educational attainment, marital status, and race. The results are weighted using theindividual-level weights in the ACS. We plot the density of the 1000 estimated βs from equation (1) after randomizing SC adoption dates.The red lines shows the baseline estimates of βs from Table (3).

60

Table A1: Characteristics of Deportees under SC, 2008-2014

Characteristic Share of Deportees (percent)

Most Serious Criminal ConvictionNone 20.63Traffic 7.01Immigration 5.46DUI 10.93Marijuana 2.38Other 53.59GenderMale 95.61Country of CitizenshipSpanish-Speaking Countries (proxy for Hispanic) 90.34Central America 85.60Mexico 62.58Central America Excluding Mexico 23.02

Notes: Data on deportees comes from individual listings of all deportations under SC from TRAC records described inAppendix A. The most serious criminal conviction may be, but does not have to be, the crime for which the deporteewas initially apprehended.

Table A2: Employment in Household Services in 2005

Occupation % of Occupation Emp % of Occupation Empthat is Hisp LSNC 80+ that is Female Hisp LSNC 80+

housekeepers, maids, butlers, stewards 20.77 19.46child care workers 5.46 5.38

Notes: Data are from the 2005 American Community Survey. The sample includes all individuals aged 20-64 who report working in thetwo household service occupations. The columns show the percent of employment in the given occupation that is Hispanic low-skillednon-citizens who entered the U.S. in 1980+ and female Hispanic low-skilled non-citizens who entered the U.S. in 1980+, respectively.The results are weighted using individual survey weights.

61

Table A3: Effect of SC on High-Skilled Women’s Fertility

Birth in Last12 Months (*100)

A: Enforcement- JanuarySecure Communities 0.128

(0.100)Mean Y 5.81N 1770283

B: Enforcement- Fraction Last YearSecure Communities 0.120

(0.119)Mean Y 5.81N 1770283

Notes: Data are from the 2005-2014 American Community Survey. The sample includes all U.S. citizen women with a college degreeor more aged 20-50. All models include PUMA fixed effects, and year fixed effects. The PUMA-year controls include: labor demandcontrols, housing price controls, and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with thechange in the following PUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housingprices, the share of the PUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60hours, and have a college degree, masters degree, or a Ph.D., as well as the same education categories just for women. The individualdemographic controls include: age, educational attainment, marital status, and race. The results are weighted using the individual-levelweights in the ACS. Standard errors clustered at the PUMA level and shown in parentheses. Panel A includes our main measure of SCexposure–measured as of January of the survey year. Panel B includes an alternative measure of SC exposure to account for the lag infertility decisions and births–specifically, we measure the fraction of the year prior to the survey year that SC was in place in the PUMA.* p<0.10, ** p<0.05, *** p<0.01

Table A4: Effect of SC on PUMA-by-Demographic-Group Level Migration Rates

High-Skilled High-Skilled LSNCCitizen Women Citizen Women Hispanic

Kids of Any Age Kids <6 Women

Secure Communities 0.014 0.006 -0.003(0.009) (0.006) (0.002)

Mean Y 0.27 0.13 0.02N 9639 9639 9639

Notes: Data are from the 2005-2014 American Community Survey. The sample is based on U.S. citizen mothers with a college degree ormore aged 20-64 in columns 1 and 2, and on likely undocumented females in column 3. The data is collapsed to the PUMA by year level andthe model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trends and demographic controls. ThePUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs. The PUMA characteristic trends includeinteractions of a time trend with the change in the following PUMA characteristics between 2000 and 2005: labor force participation rate,unemployment rate, and housing prices, the share of the PUMA that are citizens, black, non-citizens, have children, have young children,work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D., as well as the same education categories just forwomen. The demographic controls include cell-level averages of: age, number of kids, number of kids under age 6, educational attainment,marital status, and race. The results are weighted using the population in each PUMA by year cell. Standard errors clustered at the PUMAlevel and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

Table A5: Effect of SC on Female Labor Supply in Household Services, Overall and for Citizens

All Low-Skill Low-Skill Cit LS Cit Hisp LS Cit Not Hisp

A: (Total # Hours Work in Household Services / Total PUMA Pop) *100

Secure Communities -0.854 -0.573 -0.620∗∗∗ 0.020(0.541) (0.424) (0.211) (0.365)

Mean Y 32.56 22.98 4.55 18.39P-Value 0.11 0.18 0.00 0.96N 10710 10710 10710 10710

Notes: Data are from the 2005-2014 American Community Survey. The sample includes women aged 20-64 who have at most a highschool degree, and who report their current or most recent occupation as household services. All models include PUMA fixed effects,year fixed effects, PUMA-year controls, and PUMA characteristics trends. The PUMA-year controls include: labor demand controls,housing price controls, and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with the change inthe following PUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housing prices, theshare of the PUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and havea college degree, masters degree, or a Ph.D., as well as the same education categories just for women. The results are weighted usingPUMA population in 2000. Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

62

Table A6: Summary Statistics: Split by Whether Woman Lives in Her State of Birth

High-Skilled Women with Kids of Any Age High-Skilled Women with Kids Under 6

Out of State of Birth In State of Birth Out of State of Birth In State of Birth

ACS

Age 42.86 41.22 34.95 33.63Black 0.09 0.10 0.07 0.08Married 0.82 0.80 0.91 0.88# Children Under 6 0.41 0.45 1.30 1.30# All Children 1.83 1.84 1.96 1.94College Degree 0.64 0.68 0.63 0.68Masters Degree 0.26 0.26 0.26 0.26Ph.D. or Professional Degree 0.09 0.06 0.10 0.06Work >0 Hours (*100) 80.72 85.48 75.72 82.16Usual Hours Worked per Week 30.26 32.06 27.54 29.99Secure Communities 0.37 0.35 0.35 0.34N 628463 584812 192593 200044

Notes: Data are from the 2005-2014 American Community Survey. The sample includes all U.S. citizens with a collegedegree or more, aged 20-64. The results are weighted the using individual-level weights in the ACS.

Table A7: Effect of SC on High-Skilled Women’s Likelihood of Moving from their State of Birth

State Birth 6= State Resid

A: Any KidsSecure Communities -0.002

(0.002)Mean Y 0.52N 1213275

B: Kids Under 6Secure Communities -0.001

(0.004)Mean Y 0.49N 392637

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with a college degreeor more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trendsand demographic controls. The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs. ThePUMA characteristic trends include interactions of a time trend with the change in the following PUMA characteristics between 2000 and2005: labor force participation rate, unemployment rate, and housing prices, the share of the PUMA that are citizens, black, non-citizens,have children, have young children, work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D., as well asthe same education categories just for women. The individual demographic controls include: age, number of kids, number of kids underage 6, educational attainment, marital status, and race. The results are weighted using the individual-level weights in the ACS. Standarderrors clustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

63

Table A8: Lasting Effects of SC around Birth on Labor Supply of High-Skilled Citizen Mothers,Robustness to Dropping Migrants

Work > 0 Usual HoursHours Worked

A: Youngest Child Age 3−5SC when Youngest Aged 0−2 -0.537∗ -0.383∗∗∗

(0.291) (0.133)Mean Y 79.98 29.25P-Value 0.07 0.00Min Survey Year 2005 2005Max Survey Year 2016 2016Min Birth Year 2000 2000Max Birth Year 2011 2011N 197016 197016

B: Youngest Child Age 3−5, Drop MigrantsSC when Youngest Aged 0−2 -0.476 -0.340∗∗

(0.291) (0.133)Mean Y 80.20 29.30P-Value 0.10 0.01Min Survey Year 2005 2005Max Survey Year 2016 2016Min Birth Year 2000 2000Max Birth Year 2011 2011N 188578 188578

C: Youngest Child Age 6−7SC when Youngest Aged 0−2 0.041 -0.063

(0.324) (0.160)Mean Y 84.14 31.18P-Value 0.90 0.69Min Survey Year 2006 2006Max Survey Year 2018 2018Min Birth Year 2000 2000Max Birth Year 2011 2011N 131124 131124

D: Youngest Child Age 6−7, Drop MigrantsSC when Youngest Aged 0−2 0.123 -0.046

(0.335) (0.161)Mean Y 84.34 31.25P-Value 0.71 0.77Min Survey Year 2006 2006Max Survey Year 2018 2018Min Birth Year 2000 2000Max Birth Year 2011 2011N 126661 126661

Notes: Data are from the 2005-2018 American Community Survey. The sample includes U.S. citizen mothers with a college degree or moreaged 20-64 whose youngest child was born between 2000-2011. The model includes PUMA fixed effects, year fixed effects, year of birth ofthe youngest child fixed effects. PUMA-year controls, PUMA characteristics trends and demographic controls. The PUMA-year controlsinclude: labor demand controls, housing price controls, and 287(g) programs. The PUMA characteristic trends include interactions of a timetrend with the change in the following PUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate,and housing prices, the share of the PUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and60 hours, and have a college degree, masters degree, or a Ph.D., as well as the same education categories just for women. The individualdemographic controls include: age, number of kids, number of kids under age 6, educational attainment, marital status, and race. The resultsare weighted using the individual-level weights in the ACS. Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10,** p<0.05, *** p<0.01

64

Table A9: Lasting Effects of SC around Birth on Labor Supply of High-Skilled Citizen Fathers

Work > 0 Usual HoursHours Worked

A: Youngest Child Age 3−5SC when Youngest Aged 0−2 -0.073 -0.130

(0.122) (0.099)Mean Y 97.78 45.15P-Value 0.55 0.19Survey Years Included 2005-2016 2005-2016Birth Years Included 2000-2011 2000-2011N 162265 162265

B: Youngest Child Age 6−7SC when Youngest Aged 0−2 -0.149 -0.108

(0.154) (0.124)Mean Y 97.56 44.87P-Value 0.33 0.38Survey Years Included 2006-2018 2006-2018Birth Years Included 2000-2011 2000-2011N 106017 106017

Notes: Data are from the 2005-2018 American Community Survey. The sample includes U.S. citizen fathers with a college degree or moreaged 20-64 whose youngest child was born between 2000-2011. The model includes PUMA fixed effects, year fixed effects, year of birth ofthe youngest child fixed effects. PUMA-year controls, PUMA characteristics trends and demographic controls. The PUMA-year controlsinclude: labor demand controls, housing price controls, and 287(g) programs. The PUMA characteristic trends include interactions of a timetrend with the change in the following PUMA characteristics between 2000 and 2005: labor force participation rate, unemployment rate,and housing prices, the share of the PUMA that are citizens, black, non-citizens, have children, have young children, work more than 50 and60 hours, and have a college degree, masters degree, or a Ph.D., as well as the same education categories just for women. The individualdemographic controls include: age, number of kids, number of kids under age 6, educational attainment, marital status, and race. The resultsare weighted using the individual-level weights in the ACS. Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10,** p<0.05, *** p<0.01

65

Table A10: Effect of SC on High-Skilled Mothers’ Labor Supply, Robustness to Timing

Work > 0 Usual HoursHours Worked

A: Any Kids, JanuarySecure Communities -0.361∗∗ -0.104

(0.182) (0.088)Mean Y 83.00 31.12N 1213275 1213275

B: Any Kids, Fraction Last YearSecure Communities -0.258 -0.126

(0.242) (0.118)Mean Y 83.00 31.12N 1213275 1213275

C: Kids Under 6, JanuarySecure Communities -0.595∗ -0.310∗∗

(0.355) (0.153)Mean Y 78.98 28.78N 392637 392637

D: Kids Under 6, Fraction Last YearSecure Communities -0.787∗ -0.362∗

(0.439) (0.193)Mean Y 78.98 28.78N 392637 392637

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with a college degreeor more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trendsand demographic controls. The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs. ThePUMA characteristic trends include interactions of a time trend with the change in the following PUMA characteristics between 2000 and2005: labor force participation rate, unemployment rate, and housing prices, the share of the PUMA that are citizens, black, non-citizens,have children, have young children, work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D., as well asthe same education categories just for women. The individual demographic controls include: age, number of kids, number of kids underage 6, educational attainment, marital status, and race. The results are weighted using the individual-level weights in the ACS. Standarderrors clustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05, *** p<0.01

66

Table A11: Effect of SC on High Skill Mothers’ Labor Supply, Robustness to Dropping Early SC Adopters

Main % of PUMA LSNC pop that is likely undoc in 2005

A: Work >0 Hours, Women with Kids, Full SampleSecure Communities -0.361∗∗ 0.148

(0.182) (0.285)SC * Intensity -0.893∗∗

(0.382)Mean Y 83.00 82.99N 1213275 1210447

B: Work >0 Hours, Women with Kids, Drop Early AdopterSecure Communities -0.239 0.275

(0.237) (0.322)SC * Intensity -0.934∗∗

(0.401)Mean Y 83.41 83.40N 939848 937020

C: Hours Worked, Women with Kids, Full SampleSecure Communities -0.104 0.231∗

(0.088) (0.137)SC * Intensity -0.596∗∗∗

(0.185)Mean Y 31.12 31.11N 1213275 1210447

D: Hours Worked, Women with Kids, Drop Early AdopterSecure Communities -0.117 0.241

(0.114) (0.155)SC * Intensity -0.664∗∗∗

(0.195)Mean Y 31.10 31.09N 939848 937020

E: Work >0 Hours, Women with Kids Under 6, Full SampleSecure Communities -0.595∗ 0.791

(0.355) (0.549)SC * Intensity -2.444∗∗∗

(0.733)Mean Y 78.98 78.97N 392637 391786

F: Work >0 Hours, Women with Kids Under 6, Drop Early AdopterSecure Communities -0.186 1.117∗

(0.455) (0.613)SC * Intensity -2.397∗∗∗

(0.788)Mean Y 79.39 79.38N 305340 304489

G: Hours Worked, Women with Kids Under 6, Full SampleSecure Communities -0.310∗∗ 0.351

(0.153) (0.226)SC * Intensity -1.165∗∗∗

(0.327)Mean Y 28.78 28.78N 392637 391786

H: Hours Worked, Women with Kids Under 6, Drop Early AdopterSecure Communities -0.255 0.393

(0.204) (0.256)SC * Intensity -1.192∗∗∗

(0.353)Mean Y 28.76 28.75N 305340 304489

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with a college degree or moreaged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trends and demographiccontrols. The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs. The PUMA characteristicstrends include interactions of time trends with the change in PUMA characteristics between 2000 and 2005. The individual demographiccontrols include: age, number of kids, number of kids under age 6, educational attainment, marital status, and race. The results are weightedusing the individual-level weights in the ACS. Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05,*** p<0.01

67

D Exploration of Alternative MechanismWhile the pattern of results we find is strongly suggestive that changes in the cost of outsourcinghousehold production is the primary mechanism driving the decline in high-skilled mother’s laborsupply, it is still possible other mechanisms are contributing to this effect. In particular, Eastet al. (2019) find that SC leads to reductions in the labor supply of both likely undocumentedmen, and middle-to-high-skilled citizen men. These spillover effects on citizen men are driven bymen working in sectors which relied on undocumented labor heavily before the implementation ofSC, indicating that complementarities in market production drive these negative spillover effects.Therefore, we examine whether the effects we find here for high-skilled citizen mothers might bealso driven by market complementarities, by investigating heterogeneity in the effect by sector. InAppendix Figure (D1), we plot the effect by sector for high-skilled citizen mothers. The horizontalaxis indicates the percent of total sector employment that was likely undocumented immigrantsin the base year of 2005. The vertical axis displays β from equation (1) estimated on samplesstratified by sector of the mother’s current or most recent employment.48 The solid gray line plotsthe average effect from Table (2) estimated on the sample pooled across all sectors, and the dashedgray line shows the line of best fit. If complementarities in market production were driving theresults, we would expect to see more negative effects as the percent of total sector employmentthat was likely undocumented immigrants increases (moving to the right on the figures). Forall outcomes and subsamples, there is no evidence of this type of relationship. Moreover, as anadditional check, we drop sectors one at a time from our sample, and re-estimate equation (1) toensure that no single sector is driving our results. As shown in Appendix Table (D1), the results forhigh-skilled mothers with young children are very consistent when we omit each sector, includingsectors that relied heavily on undocumented labor before SC. Taken as a whole, we believe this isstrong evidence that complementarities in market production are not driving the results we findfor high-skilled citizen mothers.

48For women who have not worked in the past 5 years, information about their sector of work is missing. Weincluded these women in our main analysis, but do not include them when we split the sample by sector ofemployment.

68

Figure D1: Effect of SC on High-Skilled Mothers Labor Supply, by Sector

(a) All High-Skilled Mothers

(b) High-Skilled Mothers with Children Under 6

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers, and with a college degreeor more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMA characteristics trends andindividual demographic controls. The PUMA-year controls include: labor demand controls, housing price controls, and 287(g) programs.The PUMA characteristic trends include interactions of a time trend with the change in the following PUMA characteristics between2000 and 2005: labor force participation rate, unemployment rate, and housing prices, the share of the PUMA that are citizens, black,non-citizens, have children, have young children, work more than 50 and 60 hours, and have a college degree, masters degree, or a Ph.D.,as well as the same education categories just for women. The demographic controls include: age, number of kids, number of kids underage 6, educational attainment, marital status, and race. The results are weighted using the individual-level weights in the ACS. Thehorizontal axis denotes the percent of total sector employment that was likely undocumented immigrants in the base year of 2005. Thevertical axis denotes the estimated effect for the overall samples and split by sector. The solid gray line plots the estimates from the fullsample from Table (3), the dashed gray line is the line of best fit, and the black line is at 0. We omit Active Duty Military from thesefigures.

69

Table D1: Effect of SC on High-Skilled Mothers’ with Kids Under 6 Labor Supply, Robustness to DroppingSectors One at a Time

Work >0 Hours Usual Hours Worked

Drop AGRICULTURESecure Communities -0.638∗ -0.307∗∗

(0.357) (0.154)Mean Y 78.93 28.78P-Value 0.07 0.05N-Omitted Sector 3142 3142N-Sample 389501 389501

Drop MININGSecure Communities -0.599∗ -0.310∗∗

(0.355) (0.153)Mean Y 78.97 28.77P-Value 0.09 0.04N-Omitted Sector 531 531N-Sample 392113 392113

Drop CONSTRUCTIONSecure Communities -0.596∗ -0.309∗∗

(0.355) (0.156)Mean Y 78.94 28.78P-Value 0.09 0.05N-Omitted Sector 3697 3697N-Sample 388951 388951

Drop MANUFACTURINGSecure Communities -0.633∗ -0.341∗∗

(0.367) (0.156)Mean Y 78.63 28.50P-Value 0.08 0.03N-Omitted Sector 20646 20646N-Sample 372015 372015

Drop TRANS, COMM, UTILSecure Communities -0.627∗ -0.331∗∗

(0.362) (0.154)Mean Y 78.82 28.66P-Value 0.08 0.03N-Omitted Sector 9971 9971N-Sample 382695 382695

Drop WHOLESALE, RETAILSecure Communities -0.589 -0.320∗∗

(0.373) (0.161)Mean Y 78.76 28.86P-Value 0.12 0.05N-Omitted Sector 33414 33414N-Sample 359321 359321

Drop FIN, INS, REAL ESTSecure Communities -0.553 -0.272∗

(0.369) (0.158)Mean Y 78.50 28.43P-Value 0.13 0.09N-Omitted Sector 30631 30631N-Sample 362065 362065

Drop BUSINESS, REPAIR SERVSecure Communities -0.516 -0.247

(0.360) (0.156)Mean Y 78.75 28.71P-Value 0.15 0.11N-Omitted Sector 16979 16979N-Sample 375702 375702

Drop ENTERT, RECREA SERVSecure Communities -0.582 -0.282∗

(0.361) (0.155)Mean Y 78.87 28.91P-Value 0.11 0.07N-Omitted Sector 11017 11017N-Sample 381658 381658

Drop EDU, HEALTH SERVSecure Communities -0.819 -0.453∗

(0.587) (0.243)Mean Y 68.56 25.36P-Value 0.16 0.06N-Omitted Sector 212334 212334N-Sample 180900 180900

Drop PUB ADMINSecure Communities -0.601∗ -0.319∗∗

(0.364) (0.156)Mean Y 78.41 28.45P-Value 0.10 0.04N-Omitted Sector 16694 16694N-Sample 376004 376004

Drop MILITSecure Communities -0.606∗ -0.317∗∗

(0.354) (0.153)Mean Y 78.98 28.78P-Value 0.09 0.04N-Omitted Sector 257 257N-Sample 392381 392381

Notes: Data are from the 2005-2014 American Community Survey. The sample includes U.S. citizen mothers with children under age 6,and with a college degree or more aged 20-64. The model includes PUMA fixed effects, year fixed effects, PUMA-year controls, PUMAcharacteristics trends and demographic controls. The PUMA-year controls include: labor demand controls, housing price controls,and 287(g) programs. The PUMA characteristic trends include interactions of a time trend with the change in the following PUMAcharacteristics between 2000 and 2005: labor force participation rate, unemployment rate, and housing prices, the share of the PUMAthat are citizens, black, non-citizens, have children, have young children, work more than 50 and 60 hours, and have a college degree,masters degree, or a Ph.D., as well as the same education categories just for women. The individual demographic controls include:age, number of kids, number of kids under age 6, educational attainment, marital status, and race. The results are weighted using theindividual-level weights in the ACS. Standard errors clustered at the PUMA level and shown in parentheses. * p<0.10, ** p<0.05, ***p<0.01

70