Spillover Effects of Immigration Policies on Children’s Human Capital

Executive Summary

In this paper, authors Esther Arenas-Arroyo and Bernhard Schmidpeter study immigration and children. Specifically, they look at the spillover effects of immigration enforcement policies on US-born children with at least one undocumented parent. The authors find the English language skills of these children are negatively affected in areas where immigration enforcement laws have been introduced. 

The reduction in children’s English skills is caused by changes in their parents’ behavior. In areas of heightened immigration enforcement, parents are more likely to keep their children home instead of enrolling them in preschool. Parents also spend less time socializing, further limiting their children’s access to English-language speakers. 

These effects are pronounced enough to cancel out the positive effects of other policies that improve fluency among children of non-native speakers, such as Head Start and California’s Proposition 227. 

Language proficiency is an important skill. It is strongly associated with future success. It is essential to participating fully in society, including educational and labor market success. This spillover effect of immigration enforcement has long-lasting repercussions for the 8 percent of US children that are living with at least one undocumented parent. 

Growing up with immigrant parents can place a heavy economic and social burden on children, especially if the parents are undocumented. Over the last years, the growth of immigration enforcement might have deteriorated the situation for children further. Between 2009 and 2013, enacted immigration policies in the United States were responsible for the deportation of almost 2 million individuals (Vaughan, 2013). Immigration policies have led to the breakup of mixed-citizen families and generated fear in immigrant communities (Amuedo-Dorantes and Arenas-Arroyo, 2019; Capps et al., 2020). Fearing the reporting and deportation of family members, undocumented parents might reduce social contacts to a minimum. They might interact less with individuals outside their community and be less likely to enroll their children in non-mandatory educational programs (e.g., Gándara and Ee, 2018). These reductions in early parental human capital investments can have detrimental effects on children’s skills and their later success in the labor market. Given that around 8 percent of US children have at least one undocumented parent (Pew Research Center, 2019), understanding the spillover effects of immigration policies on children’s human capital accumulation and the role of parents is important. Research on this question is still scant, however.

In this paper, we follow two goals to better understand how immigration policies can affect children. First, we study the potential spillover effects of immigration enforcement policies on the English language proficiency of US-born children with at least one undocumented parent. We concentrate on language proficiency as an important skill, which is strongly associated with future success. Having a sufficient level of language proficiency is essential to participating fully in society (Arington, 1990). Higher verbal skills earlier during a child’s life cycle play a substantial role in explaining later educational success (e.g., college enrollment) even more so than math skills (Bleakley and Chin, 2010; Aucejo and James, 2021). Ultimately, language skills affect future labor market success, particularly for immigrants (Dustmann and Fabbri, 2003; Bleakley and Chin, 2004).

Our second goal is to explore how immigration enforcement policies can change parents’ human capital investment decisions, as an important underlying mechanism. Parents play an important role in shaping children’s language skills. Language proficiency is largely shaped by social interactions with peers and adults (Henry and Rickman, 2007; Weisleder and Fernald, 2014). Interactions with native speakers are particular important for the development of English proficiency for children of Spanish-speaking parents (e.g., Palermo and Mikulski, 2014; Villarreal and Gonzalez, 2016), who are often Hispanics and particularly affected by immigration policies in the United States. This gives immigrant parents a key role in directly and indirectly shaping their children’s language skills. Within the climate of fear following immigration enforcement, undocumented parents might limit social interactions for themselves and their kids. For example, parents may decide to not to enroll their US-born children in non-mandatory education programs to limit social contacts and exposure.

Exploiting the temporal and geographical variation in the enactment of the first police-based enforcement policy in a metropolitan statistical area (MSA), we find that the policy introduction had significant spillover effects on the English proficiency of US-born children with at least one undocumented parent. Our estimates show that the introduction of immigration enforcement policy reduces children’s likelihood of having high English language skills by a significant 3 percentage points. Investigating the dynamics of the effect, we find no impact prior to the enactment of immigration enforcement laws but a gradual decline in children’s language skills afterward. This pattern suggests a lack of intervention later in a child’s life cycle to compensate for the loss of early language skills.

Our estimated effects are quite sizeable when compared to policies aimed at improving language skills of children of non-native speakers. For example, our results are of similar magnitude but opposite sign than having access to Head Start, an early education intervention program aimed at disadvantaged children. Access to Head Start at age four increases children’s third grade reading and vocabulary skills by 11 percent and 8 percent of a standard deviation, respectively (Puma et al., 2012). In comparison, our estimates imply that immigration enforcement reduces the likelihood of children having high English language skills by around 8 percent of a standard deviation. Our results are also of similar size as those reported in, for example, Kuziemko (2014), who evaluates Proposition 227 in California, which mandates English as the language of instruction in schools. For districts with an average compliance rate, the introduction of the law increased the likelihood that children would have very good English skills by around 8 percent of a standard deviation.

We conduct several checks to assess the robustness of our results, such as using an alternative definition of likely undocumented immigrant, disregarding all parents without formal education from the sample, and allowing for heterogeneous treatment effects across treated cohorts. 1Recent research has shown that difference-in-differences based on two-way fixed effects regressions and the staggered rollout of a policy can be biased when treatment effects are not constant over cohort and time; see, for example, de Chaisemartin and D’Haultfœuille (2020), Borusyak et al. (2021), Callaway and Sant’Anna (2021), and Goodman-Bacon (2021). In all these cases, our estimates are very similar to our main results. We also conducted placebo regressions where we only included children of naturalized or native parents in our sample. The estimates for these samples are all close to zero and not statistically significant on any conventional level.

We then provide evidence that one important underlying mechanism for our results is the change of parental investment behavior caused by immigration enforcement. Immigration policies reduce the likelihood that likely undocumented parents enroll their US-born children in non-mandatory preschools by 2.2 percentage points or around 7 percent compared to the mean enrollment rate of Hispanics.

At the same time, our results also show that parents try to compensate for the reduction in preschool attendance and therefore time in formal educational by increasing their time investment in their children. The increase is, however, mostly concentrated in time spent on recreational activities like playing with the child. We do not find evidence that parents’ educational time investment is affected by immigration enforcement. One explanation may be that parents are not aware of the importance of early childhood investment (e.g., Boneva and Rauh, 2018). As recreational time spent with parents is less linguistically productive than time in preschool, this leads to a reduction in children’s English language skills.

We provide evidence that parents change their investment behavior out of fear of being detected and deported—a fear caused by immigration policies. This mirrors findings found for the take-up of government benefit programs (e.g., Watson, 2014; Aslan and Young, 2019).

Our estimates show that, in response to the introduction of immigration policies, parents reduce their time spent on activities such as attending events or socializing. In contrast, we do not find evidence that time spent on activities that take place predominantly at home are affected by the introduction of immigration enforcement policies. These results are in line with parents limiting the risk of detection by reducing children’s time spent outside their home to a minimum.

Our results are important in that they provide strong evidence for negative spillover effects of immigration policies on the human capital of US-born children of immigrant parents. Lower English language skill levels earlier in life reduce the likelihood that these children can participate fully in society. An underdeveloped English language skill set also reduces their chances of obtaining a university degree and lowers their future labor market success.

Ultimately, lower skill accumulation caused by immigration enforcement is likely to increase these children’s dependence on social security later in life and hamper the intergenerational mobility of migrant children. The overall future impact on the economy is substantial, given the amount of US-born children of immigrant parents. If a large share of the future workforce grows up accumulating fewer skills while young, this will ultimately reduce long-term growth prospects2While a child’s future contribution to the US economy is not, of course, the only reason to be concerned about his or her language acquisition, it does warrant concern if US policy is undermining economic health.

Our work is related to two important strands of literature. First, we contribute to the literature on the effects of immigration policies on US children with undocumented parents. Previous work has analyzed the effect of immigration enforcement on children’s Medicaid participation, living arrangements, foster care, and general access to economic resources (Watson, 2014; Amuedo-Dorantes and Arenas-Arroyo, 2019, 2018; Amuedo-Dorantes et al., 2018). Closer related to our project is the work by Amuedo-Dorantes and Lopez (2017), Dee and Murphy (2020), and Santillano et al. (2020), who study the impact of restrictive immigration policies on school enrollment, school dropout rates, and enrollment in Head Start. Amuedo-Dorantes and Lopez (2017) find that increasing immigration enforcement significantly increases both the likelihood of repeating a grade and the probability of dropping out of school for Hispanic children of likely unauthorized parents. Dee and Murphy (2020) find that local ICE partnerships reduce the number of Hispanic students in school. This effect is mostly concentrated on elementary school students. Santillano et al. (2020) finds that local immigration raids deter Hispanic parents from enrolling their children in Head Start.3Bellows (2019) finds small negative impacts of the introduction of Secure Communities in a county on average English Language Arts (ELA) scores using the Stanford Education Data Archive. While the results are important and insightful, given the aggregation of the data, possible selective participation in ELA test taking, and the lack of availability of the exact test taking dates make it difficult to deduce the real impact of immigration policies due to the likely presence of measurement errors, as also pointed out by Ho (2020).

We complement and extend this strand of the literature by providing a unifying picture of how immigration policies can affect children’s human capital accumulation. In our work, we first analyze the spillover effects of immigration policies on the language skills of US-born children, which is strongly associated with future success (Aucejo and James, 2021). Then, we carefully connect these spillover effects to potential changes in parental investment behavior caused by immigration policies as an important underlying mechanism. Our work therefore contributes to a better understanding of how and why immigration policies can affect children’s human capital, even if these children are not directly targeted by the policies.

Second, we contribute to the literature on determinants of parental human capital investment decisions (Baranov et al., 2020; Nicoletti and Tonei, 2020; Schmidpeter, 2020; Laffers and Schmidpeter, 2021) and, more specifically, investment decisions made by likely undocumented parents. Thus, to a certain extent, we also contribute to works investigating the intergenerational mobility of migrants (Chetty et al., 2020; Abramitzky et al., 2021)4See also Francesconi and Heckman (2016) for a review of the literature. These works do not, in general, investigate how (immigration) policies can have spillover effects on parents’ investment decisions. We analyze if and how immigration policies can change parental investment decisions. In our work, we also explore why parents may change their human capital input in their child. For example, we evaluate whether the fear of being detected and deported leads undocumented parents of US children to minimize social interactions outside their home and whether they are therefore less likely to enroll their children in non-mandatory formal education programs. While parents also increase the time spent with their children in some activities as a response to immigration policies, the extra time is not sufficient to compensate for the disadvantages caused by less time in formal educational childcare; see also, e.g., Bernal and Keane (2011) and Felfe and Lalive (2018) for the impact of formal childcare on children’s cognitive achievements. Ultimately, lower parental educational investments and language skills caused by immigration enforcement may reverse the improvements in intergenerational mobility Hispanics have made (e.g. Chetty et al., 2020).

The paper proceeds by first providing a conceptual framework to motivate how immigration policies can affect the skill accumulation of US-born children. The data for our analysis are described in section 3. We present our empirical strategy in section 4. In section 5, we discuss the spillover effects of immigration enforcement on children’s language skills. Changes in parental investment behavior as potential mechanism caused by heightened enforcement is explored in section 6. Finally, section 7 concludes the study.

Immigration Policies and Children’s Skills

To motivate our empirical analysis we consider a simple overlapping generations model with parental human capital investments to show how and through which channels immigration enforcement policies can affect children’s language skills. There are two agents in our model, undocumented parents and their child, and two time periods, childhood t_{1} and adulthood t_{2}. When the child is young, parents have three choices: how much to consume c^{p}, how much time to spend with their child l^{p}, and how much time k to send their child to formal childcare (preschool) at a per unit cost \kappa. Assume that parents’ time is normalized to one and that time not spent with the child is used for working at a per unit wage rate \omega, so that s^{p} = 1-l^{p}. Undocumented parents also face a probability p^{d} of being deported by the end of t_{1}, after parents’ investment decisions have been made.

The child receives its payoff in t_{2} when it is old and the parents are dead. Let the superscript d indicate that the family was deported by the end of t_{1}. Likewise, denote by nd if the family was not deported. The payoff function for the child in adulthood is given by

(1)   \begin{align*} c^{j,c} = h^{j}(l^{p},k^{p}) \end{align*}

where h^{j}(\cdot] is the human capital production function for j \in \{d,nd \}. We assume that h(\cdot) is increasing and concave in each of its two arguments. The child’s payoff when old depends on parental investments made in t_{1} during childhood. The payoff also depends on the deportation status, for example, to reflect that stress caused by deportation affects productivity of human capital, even if the child is not deported.5One could make such a distinction even more pronounced and allow the child also to make labor supply and human capital investment decisions when old. Such a model would not generate fundamentally different insights to what we discuss here, however, but would stress the long-lasting effects of initial parental human capital investment decisions. As we assume that the child takes parents’ inputs as given and the deportation status is realized before the child reaches adulthood, one could solve such a model backwards, starting with the decision of the child in t_{2}. Parents would then face a similar problem as discussed below, taking into account the maximum reachable utility level of their child given the deportation status.

Let parents’ instant utility function of family consumption be U(\cdot), which is concave in its argument, and denote by V^{c}(c^{j}) the child’s indirect utility function in adulthood. Then, abstracting from any discount factor, parents face the following maximization problem:6Notice that in our setting, parents are fully altruistic toward their child and fully convert the utility of the child in their own utility.

    \begin{equation*} \label{eq_max} \begin{aligned} \max_{c^{p},l^{p},k^{p}} \quad & U(c^{p}) + p^{d}V^{c}(c^{d,c}) + (1-p^{d})V^{c}(c^{nd,c})\\ \textrm{s.t.} \quad & c^{p} = \omega(1-l^{p}) - \kappa k^{p}\\ & c^{d,c} = h^{d}(l^{p},k^{p}) \\ & c^{nd,c} = h^{nd}(l^{p},k^{p}). \end{aligned} \end{equation*}

Denote by subscript the partial derivative. Then parents’ optimal investments in the child are given by the following two first-order conditions:

(2)   \begin{align*} U_{c}(c^{p}) \omega &=p^{d}V_{l}^{c}(c^{d,c})\frac{\partial h^{d}}{\partial l^{p}} + (1-p^{d})V_{l}^{c}(c^{nd,c})\frac{\partial h^{nd}}{\partial l^{p}} \\ U_{c}(c^{p}) \kappa &=p^{d}V_{k}^{c}(c^{d,c})\frac{\partial h^{d}}{\partial k^{p}} + (1-p^{d})V_{k}^{c}(c^{nd,c})\frac{\partial h^{nd}}{\partial k^{p}} \label{eq_optk} \end{align*}

where solving equation 2 for l yields the optimal time investment of parents in their child. Likewise, solving equation 3 for k gives the optimal allocation of preschool time.

The first-order conditions imply that there are three main factors through which a heightened risk of deportation affects parents’ investment decisions. First, how parents react depends on what type of investment is perceived as more valuable when not being deported. Second, the relative costs of consumption, the attainable wage \omega, and preschool fees \kappa all play a role in the adjustment process. Third, how parents change their investment decisions also depends on the relation between formal childcare and parental time investments and, more specifically, whether parents consider their own time investment and formal childcare to be substitutes or complements.

Parents’ change in investments as response to heightened enforcement ultimately spills over to the human capital accumulation of their child. To see this more clearly, consider the child’s payoff function in equation 1, which depends on parental investment. Differentiating it with respect to the risk of being deported p^{d} yields

(3)   \begin{align*} \frac{ \partial h}{\partial p^{d}} = \frac{ \partial h}{ \partial l^{p}}\frac{ \partial l^{p}}{\partial p^{d}} + \frac{ \partial h}{ \partial k^{p}}\frac{ \partial k^{p}}{\partial p^{d}}. \end{align*}

Assume that parents perceive their own time investments and preschool attendance as (weak) substitutes. There is strong evidence that this is the case and parents perceive their own inputs and educational inputs as substitutes (e.g. Das et al., 2013; Greaves et al., 2021). Also assume that skills learnt in preschool are in the future more valuable for the child when remaining in the US while skills derived from parental time inputs are equally valuable whether deported or not. Then, in our simple model, an increase in deportation risk caused by immigration policies leads parents to reduce their child’s preschool attendance. However, to compensate for the decrease in formal educational inputs, parents raise the time investment in their child as a response.

How parents’ reaction to immigration policies affects human capital accumulation of the child depends on both the productivity of each input and the magnitude of the change; see equation (4). If parental time investment is not as productive as formal early childhood education, for example, because parents are not aware of the educational benefits of certain activities, then children’s skills decrease as a response to immigration enforcement. These changes are amplified if parents reduce social contacts outside their own home to a minimum as a response to immigration policies.

While our conceptual framework is simple, it highlights how immigration policies and the fear of being detected and deported can spill over to children’s human capital. We analyze the spillover effects of immigration policies in section 5. In section 6 we investigate changes in parental investment behavior underlying the possible spillover effects. Compared to our simple model, we investigate the response to different types of parental inputs in our empirical analysis.


We use several data sets to identify the effect of heightened immigration enforcement on children’s human capital accumulation and parental investment decisions.

Data on Children’s Language Skills

Our analysis of the impact of immigration enforcement on children’s language skills is based on the 2005-2014 American Community Survey (ACS, see Ruggles et al., 2020).

Approximately 3.5 million randomly sampled households are interviewed on a yearly basis. The ACS provides rich demographic, social, economic, and housing information for a representative sample of individuals and their households.

We construct two measures of English proficiency based on the survey question: “How well does this person speak English?” The question has four possible responses: “very well,” “well,” “not well,” and “not at all.” Following Kuziemko (2014), we construct a categorical variable Proficiency 0-3 corresponding to “does not speak English”, “speaks English but not Well”, “speaks well”, and “speaks very well”. We also use a dummy variable which takes the value of one if the child speaks English “very well,” and zero otherwise.7Similar variables as a measure of language skills were also used in Bleakley and Chin (2004). A more objective measure for children’s language skills would be preferable, but the self-report skills are the only measure available in the data. In addition, Vikstrom et al. (2015) find that the self-reported skills are a valid measure to assess English ability, specifically when using our dummy variable indicating high English skills.

One limitation of the ACS is the lack of information about the legal status of immigrants. To proxy the legal status in our work, we follow the literature and use Hispanic non-citizens who have not completed high school and who have lived in the United States for at least five years as proxy for undocumented immigrants (e.g., Amuedo-Dorantes et al., 2018; Amuedo-Dorantes and Arenas-Arroyo, 2019).8As previous research shows, most undocumented immigrants have low education levels, and most of them are coming from Latin America (see, for example, Orrenius et al., 2018). Concentrating on Hispanic non-citizens as a proxy might include low-skilled immigrants or students with non-immigrant visas, however. We therefore follow Amuedo-Dorantes et al. (2018) and restrict our sample further to individuals without a high school diploma who have lived in the United States for at least five years. Then, we restrict our main sample to US-born children who are between 7 and 16 years old and have at least one likely undocumented parent, as previously defined. We will show as a robustness check that our results are similar when using an alternative proxy for the legal status of the parents, following the residual method used by Borjas (2017).

One might be concerned that undocumented immigrants affected by immigration enforcement may be less inclined to participate in the survey to avoid detection. While this is a valid concern, we do not think that it will lead to substantially biased estimates in our work. Previous works on the impact of immigration policies have found that the ACS is covering the population of likely undocumented immigrants well (Pope, 2016; Amuedo-Dorantes et al., 2018). The ACS interviews the resident population without regard to legal status or citizenship.9See the 2014 American Community Survey from the US Census Bureau, https://www.census.gov/history/pdf/acsdesign-methodology2014.pdf. During the interview, the ACS only asks individuals whether they are US citizens, naturalized, or hold any other citizenship. Hence, the group of non-citizens is a broad group comprised of all immigrants, including students and individuals on temporary visas. Given the sample design, all individuals have the same probability of being selected, regardless of their citizen status (Pope, 2016). But even in the unlikely case that our data suffered from selective non-response of likely undocumented parents, we would expect families with lower language proficiency to be less likely to participate in the survey. Therefore, our results would likely underestimate the true effect in this case.

While there is no evidence that likely undocumented parents are underrepresented in the ACS data, they may intentionally misreport the language proficiency of their children when interviewed. For example, parents might overstate the English proficiency of their children as a way to signal that they are legally in the country. If this were true, our results would also reflect a lower bound (in absolute terms) on the impact of immigration enforcement on children’s human capital.

One additional concern in our setting might be that if immigration enforcement reduced parents’ interactions with native speakers, this might lead to a lack of natural reference points parents can use to compare their children’s language skills. As a consequence, they might be less able to evaluate their children’s language proficiency, leading to a biased response. There is evidence, however, that parents tend to overestimate their children’s skills in situations where other children tend to have low skills (Kinsler and Pavan, 2021). In light of such biased parental beliefs, and as enforcement measures affect likely communities as a whole, if immigration enforcement reduces children’s language skills our estimates will likely reflect a lower bound (in absolute values) on the true effect. Our data do not allow us to assess such a potential bias in more detail, however.

Data on Parental Time Use

We are also interested in how immigration enforcement changes parental time investment in children. In our analysis, we make use of the American Time Use Survey (ATUS) from 2003 to 2018. The ATUS is an annual time use survey in the United States with the goal of measuring how people divide their time among different activities. Possible participants for the ATUS are randomly drawn from the pool of all Current Population Survey (CPS) interviewees who finish the CPS interview sequence. ATUS interviews are conducted by phone in either English or Spanish. The survey participants are asked about their activities starting at 4:00 a.m. on the designated day until 3:59 a.m. the following day, including the location of the activity and who else was present. While the ATUS has a much smaller sample size than the ACS, and while geographic information is only available at the state level, it provides detailed information on parental time investments.10The ATUS could be linked to the Current Population Survey (CPS) to get geographic information on the MSA level. The CPS is not representative for all MSAs, however, so we refrain from doing so.

When using the ATUS, we restrict the sample to low-skilled, Hispanic survey participants between 21 and 65 years, who lived in the US for at least five years, and with at least one child at preschool age, that is age 0 to 5, in the household. We concentrate on preschool children in the ATUS because language skills are largely shaped at a younger age (Palermo and Mikulski, 2014) and early parental time investment has likely persistent effects. By concentrating on preschool children, we therefore capture changes in parental investment decisions affected by immigration enforcement policies as one important underlying channel that can shape children’s language skills. To proxy the legal status of individuals in our sample, we follow the same definition as in the ACS discussed above.11Our results are robust to using alternative definitions such as the one in Borjas (2017).

As the ATUS contains detailed information about the nature of the activity and place, we can explore how immigration enforcement has changed patterns of parental time investments. To obtain a broad picture of how parental time investment may change due to heightened immigration enforcement, we use four different activity groups, closely following Fiorini and Keane (2014): general care, educational time, social activities, and recreational time.12These five categories are finer than the two broad categories of basic childcare and educational/recreational childcare used in Aguiar and Hurst (2007), who study long-term trends in time use for the United States.

Our general care measure includes activities such as eating and drinking or physical care. Educational time includes activities such as reading to the child and helping with homework.

Going to events, socializing, and participating in performances and plays are considered social time. Finally, we define activities such as playing, doing sports and relaxing as recreational time.13See appendix B for further details.

Notice that we explicitly distinguish between social and recreational time spent with the child. We do so as we are particularly interested in the behavioral response of parents. If immigration enforcement leads parents to spend less time outside the home with their children and reduce social activities, it is likely as a response to heightened fear. As a majority of social activities take place outside of participants’ homes, putting those two categories together would mask important differences in parental time use.

From our sample, we exclude observations where the survey participant reports to have spent an unusually large amount of time with the child. Specifically, we choose to exclude observations where the total time spent with the child is above 17 hours per day. Similar restrictions were also applied in Fiorini and Keane (2014).

One might also be concerned about non-response caused by immigration enforcement in the ATUS data. While the non-response rate in the ATUS is substantially higher in comparison to the CPS and also the ACS, there is no evidence that the higher non-response rate is driven by the refusal of likely undocumented immigrant to answer. First, if immigrants selectively take part in the ATUS, this should also be reflected in the CPS. The CPS covers the population of likely unauthorized immigrants reasonably well, however.1414For a discussion, see, for example, Jeffrey S. Passel and D’Vera Cohn, “US Unauthorized Immigrant Total Dips to Lowest Level in a Decade,” Pew Research Center, November 27, 2018, https://www.pewresearch.org/hispanic/2018/11/27/unauthorized-immigration-estimate-methodology/. Second, if the introduction of immigration enforcement had affected the response rate, one would expect to see large changes over time. However, the non-response rate in the ATUS follows a similar trend over time as other household surveys which do not include sensitive questions on citizenship status, such as the Consumer Expenditure Survey.1515See the Bureau of Labor Statistics notes on household and establishment survey response rates: https://www.bls.gov/osmr/response-rates/home.htm. Lastly, the results in Abraham et al. (2006) do not point to any differences in the general propensity of Hispanics to respond to the ATUS in comparison to non-Hispanic whites once background characteristics, such as age and sex of the participants, are taken into account.

While the above points do not suggest a large bias in our estimation caused by non-response when using the ATUS, we acknowledge that being unable to directly investigate any possible selectivity is a limitation in our empirical setting. Nevertheless, as with the ACS, we would expect our estimates to understate the true effect if non-response was an issue.

Data on Immigration Enforcement

We hand-collected historical and current data about different local police-based interior immigration policies. Specifically, we gathered data on 287(g) agreements from the ICE bureau’s 287(g) Fact Sheet website. These policies are directly linked to apprehension and deportation. Information about the enactment of Secure Communities (SC) programs are obtained from the ICE Activated Jurisdictions document.1616The 287(g) and Secure Communities enactment dates can be accessed through https://www.ice.gov/factsheets/287g and https://www.hsdl.org/?view&did=682236 respectively. Similar policy data were also used, for example, in Amuedo-Dorantes et al. (2018) and Amuedo-Dorantes and Arenas-Arroyo (2021). In appendix A we provide further discussions about the 287(g) and the SC program.

Our hand-gathered data allow us to identify the date and name of the county enacting any 287(g) or SC measures. To merge the information on immigration policies available on the county level to the ACS data which are available on the MSA level, we use the cross-walk provided by the US Census Bureau.17Reference files are available at https://www.census.gov/geographies/reference-files/time-series/demo/metro-micro/delineation-files.html. We use the 2013 definitions for metropolitan statistical areas (MSAs) from the US Office of Management and Budget (OMB). This definition has two advantages. First, it provides a consistent identification of the MSAs from 2005 to 2014. Second, the delineations are entirely county-based. The latter allows us to merge our policy data directly to the ACS without any need for adjustments, such as additional weighting. Using information on the enactment date of the first immigration policy within an MSA, we construct a dummy variable I E_{m, t} taking a value equal from the first year an MSA adopted an immigration policy, and zero otherwise. In appendix A, we show the roll-out of the policies over time. By the end of 2013, the whole United States was covered by at least one immigration policy.

Empirical Approach

To identify the effects of heightened immigration enforcement on children’s English language proficiency, we rely on difference-in-difference and event-study approaches. To quantify average effects, we first estimate the following equation by exploiting the geographic and temporal variation in the enactment of our immigration policies on the sample of US-born children with at least one likely undocumented parent:

(4)   \begin{eqnarray*} y_{i,m,t}= \alpha^{DiD}_{m}+\beta IE_{m,t}+X'_{i,m,t}\Gamma+\gamma^{DiD}_{m}+\theta^{DiD}_{t}+\epsilon^{DiD}_{i,m,t} \end{eqnarray*}

where { y_{m,t}} is the outcome variable, children’s English proficiency, for a child i observed at time t and living in the metropolitan statistical area (MSA) m. IE_{m,t} is an indicator variable equal to one if the MSA m has adopted a measure of interior immigration enforcement policy in year t, and zero otherwise. Thus, \beta represents the coefficient of interest in our analysis. It captures how immigration enforcement affects children’s English language proficiency.

We also include children and household characteristics summarized by the vector X_{i,m,t}. Children’s characteristics include age, gender, and grade level attendance. Household characteristics include the household head’s marital status, years the United States, education level, gender, and total number of children in the household. Additionally, we also include geographic and temporal fixed effects. The geographic fixed effects \gamma_{m} address unobserved and time-invariant area-specific characteristics potentially correlated with the outcome. The temporal fixed effects, captured by \theta_{t}, account for aggregate level shocks potentially impacting children’s English language proficiency. We cluster all standard errors at the local MSA level.

In our analysis, we use two different measures for language skills. Our first measure is overall English language proficiency, which ranges from zero (does not speak English) to three (speaks very well), so more proficient skills have a higher value. Therefore, a negative (positive) impact of heightened immigration enforcement on our outcome would imply that heightened immigration enforcement decreases (increases) the overall language proficiency of the US-born children of likely unauthorized parents.

As an alternative language skill measure, we also use a dummy variable which takes a value of one if the child speaks English “very well,” and zero otherwise. As discussed in section 3, this binary variable captures the English proficiency of non-native speakers well, even under the presence of self-reporting bias, and captures high English skills.

We also estimate dynamic effects of the impact of immigration policies within an event-study framework:

(5)   \begin{align*} \begin{split} y_{i,m,t}= & \alpha^{ES}_{m}+ \sum_{\substack{a=-4\\ a \neq -1}}^{5}\delta_{a} \mathbbm{1}(t - C_{m} = a) + \delta_{-5}\mathbbm{1}(t - C_{m} < -4) \\ & + \delta_{6}\mathbbm{1}(t - C_{m} > 5) + X'_{i,m,t}\Gamma^{ES} + \theta^{ES}_{t}+\epsilon^{ES}_{i,m,t} \end{split} \end{align*}

where C_{m} is the year when the first immigration policy was introduced in MSA m. As we only have a limited number of observations for years distant from the actual treatment year, we bin all time periods with a relative treatment time further away than four years prior or five years after the introduction of the first policy. In our event study, we include the same set of control variables as in equation 6.

A dynamic specification as in equation 6 allows us to investigate any possible persistence in the effect of immigration policies on children’s skill accumulation. It also helps us to identify possible delays until the effects of immigration policies on children’s language skills materialized and to gauge when such policies have the most impact on children. For example, if parents minimized activities outside their home as a response to heightened immigration enforcement, a lack of social contacts would affect children’s language skills likely only gradually. At the same time, we would also see a long-term decline in language skills with little sign of reversal if there are no interventions to compensate for the lack of social interaction.

The dynamic specification also enables us to examine possible differences in the outcomes prior to the adoption of the laws. If we do not find evidence that outcomes differed prior to the adoption of immigration enforcement, this will lend support to the so-called parallel trend assumption, which is necessary for identification in our models.18In appendix C we provide additional estimates using alternative difference-in-difference specifications robust to treatment effect heterogeneity.

Spillover Effects on Children’s Language Skills

Main Results

The results from estimating equation (5) using ordinary least squares are shown in table 2. We estimate two different specifications without and with a full set of background characteristics to see how sensitive our results are.

Columns (1) and (2) show the estimation results using the proficiency score as the outcome variable. Immigration enforcement decreases the English proficiency score significantly by around 0.028 points when we only include year and meta area fixed effects; see column (1). Including additional controls barely changes our estimates. The results in column (2) indicate that immigration enforcement decreases children’s English proficiency by around 0.023 points. That our results do not depend on the inclusion of additional household characteristics in our model is reassuring and gives us confidence in our identification assumption.

One might be concerned that our results are affected by the fact that survey participants are asked to rate the proficiency of the child. To minimize the risk that our estimation is driven by self-reporting bias and to better understand whether immigration enforcement affects children with high English skills, we also report the estimates for our binary outcome variable in columns (3) and (4). The outcome variable takes now a value of one if the child speaks English “very well,” and zero otherwise.

Our estimates are in line with the findings using the proficiency score. If we only control for year and meta area fixed effects, the introduction of immigration enforcement policies reduces children’s likelihood of speaking English “very well” by around 3 percentage points or around 8 percent of a standard deviation; see column (3).19Using our estimates and the reported standard deviation of 0.41 from table 1, the effect size is −0.0323/0.41. As before, our results remain virtually unchanged when we include a wide range of household characteristics in our estimation equation; see the results in column (4).20Our results are also robust to including potentially endogenous covariates in the estimation, such as English proficiency of the survey respondent and employment status.

It is interesting to compare these effect sizes to those in the literature on policy interventions aimed to improve children’s language skills. Our effect sizes are comparable, but of opposite sign, to the impact of having access to the Head Start program at age four on children’s third grade language and literacy skills. The program’s primary goal is to boost the school readiness of low-income children. Having access to Head Start at age four increases reading comprehension as measured by the ECLS-K Reading Assessment by 11 percent and vocabulary knowledge by 8 percent of a standard deviation in third grade, although the latter estimates are not statistically significant (see the results in Puma et al., 2012).21The Head Start program was launched in 1965. The program’s primary goal is to boost the school-readiness of low-income children by providing preschool education, healthcare support, nutrition services, and help for parents to foster their child’s development. The estimates in Puma et al. (2012) are based on the Head Start Impact Study launched in 2012.
Given the similar but opposite signed effect sizes and the evidence that Head Start improves children’s long-term educational outcomes (e.g., Deming, 2009), our results also imply that immigration enforcement likely lowers the future success of US-born children of likely undocumented parents.

Our effect sizes are also comparable to those of Kuziemko (2014), who evaluates how mandating school instructions to be in English in California (Proposition 227) impacts children’s (and parents’) language skills. Evaluating her estimates at the average compliance rate on the school district level with Proposition 227, her results imply an increase in the likelihood of speaking English “very well” for immigrant children by 8 percent of a standard deviation.22Kuziemko (2014) estimates a coefficient of 0.246 on the interaction of Proposition 227 with the compliance rate. Taking the average compliance rate of 13 percent reported in the text and the standard deviation of 0.414 of the outcome “speaks very well” in the children-parent sample, the standardized effect size is (0.246 · 0.13)/0.414 = 0.077.

Overall, we find that the spillover effects of immigration enforcement on US-born children with likely undocumented parents are of comparable magnitude but opposite sign to policies intended to improve the language skills of disadvantaged children. Therefore, immigration policies not only counteract the purposes of policies aimed at increasing children’s education, but they also may induce an inefficient allocation of resources. On the one side, substantial funding is allocated to improve the language skills of disadvantaged children, a large share of whom are of Hispanic origin.23For example, Hispanic children are the majority of participants in the group of four-year-olds in the Head Start Impact Study. On the other side, by introducing strict immigration enforcement measures that spill over to US-born children, any possibly positive effects of education policies are diminished or even entirely erased.

Dynamic Effects

Having established that immigration enforcement lowers the language skills of US-born children of likely undocumented parents, we investigate any dynamics of our effects next. Such an exploration allows us to explore any persistence in the effect of immigration policies on children’s skill accumulation. It also allows us to assess if there are any trends in our outcome variable, prior to the enactment of immigration enforcement policies.

Figures 1 and 2 plot the coefficient δa from equation (6) from five years prior to six years after the enactment of immigration enforcement for the English proficiency score and our binary indicator if the child speaks English “Very well” as an outcome, together with 95 percent confidence intervals.24Remember that the estimates for t − 5 and t + 6 represent binned estimates. From the estimated pattern in the figures, two important features emerge.

First, all of our estimates prior to the enactment of any immigration enforcement are both economically small and statistically insignificant. This is true both when using our proficiency score as outcome or our binary indicator if the child speaks English “very well.” This lack in pre-trends considering both outcomes gives reassurance in our estimation strategy.

Second, we see a strong and significant decline in children’s language proficiency after the enactment of immigration enforcement for both of our measures. The negative impact of immigration enforcement policies on children’s language skills is also very persistent and does not show any sign of reversal. For example, within five years after the introduction of the first immigration enforcement policy, our results show that US-born children of likely undocumented parents have a more than 5 percentage points lower likelihood of speaking English “very well.”

The gradual and persistent decline also suggests that there is no intervention to compensate for the loss of language skills caused by heightened immigration enforcement. As shown in our framework in section 2, this suggests that immigration enforcement also affects parental human capital investment decisions as an important underlying mechanism. Before exploring parents’ responses to immigration enforcement in section 6, we discuss the robustness of our results first.


Despite the absence of any detectable pre-trends, one might still be concerned that our estimates capture effects unrelated to the enactment of immigration policy. For example, schools in areas that saw a drop in test scores also enact immigration enforcement laws earlier. If this was true, our estimates would not reflect the impact of immigration enforcement on children’s language skills but would instead pick up differences in school quality, at least partly.

In order to investigate such a possibility, we conduct a placebo check where we estimate equation (5) on a sample of US-born children with low-skilled documented parents (naturalized or native).25As in our main specification we restrict our sample to US-born children with low-skilled parents. The sample differs only by parents’ citizen status; only US citizens are included in our placebo. Given that these families are citizen and therefore reside legally in the United States, they should not be affected by any immigration enforcement policies. The estimation results from our placebo regression are reported in columns (1) and (2) in table 3, using our two measures of children’s English language proficiency.

As one can see from the results, we do not find evidence that immigration enforcement has any impact on the English proficiency of US-born children with documented parents. Our results are not only statistically insignificant but are also very small. This also allows us to rule out meaningful impacts for children of documented parents in general.

Additionally, we investigate the robustness of our results to how we proxy the legal status of parents. Remember that in our data, we do not directly observe whether an individual resides legally in the United States. We therefore use an alternative approach to proxy an individual’s status using the residual method.26The residual method was initially proposed by (Passel et al., 2014) and subsequently applied by others (e.g., Borjas, 2017). We first define who is living legally in the United States. Persons are considered to be legally in the United States if they satisfy any of the following criteria: they were born in Cuba, arrived before 1980, have US citizenship, receive public benefits, have a spouse who is a legal immigrant or US citizen, or work in the government sector. Then, according to the residual method, any person who does not fulfil this requirement is likely to be undocumented. The results when using this alternative proxy are shown in columns (3) and (4) in table 3.

Using the residual method to define likely undocumented parents leaves our results virtually unchanged. We still find that immigration enforcement policies lower both English proficiency and high English skills substantially and significantly.27 In appendix C, we also provide estimates from an event study design using the alternative definition to proxy for parents’ legal status. The estimates are similar in terms of size and significance as those in figures 1 and 2. That our estimates do not depend on the exact definition we use is reassuring.

We also investigate whether the negative impact we find is driven by children who drop out of school. Children in our sample are in general required to attend school, but some parents might pull their children out of school as a response to immigration enforcement. Thus, lower formal education might in the end explain children’s lower language skills.

Even if we disregard children who drop out of school, we still find strong evidence on the negative impact of immigration enforcement on children’s language skills; see columns (5) and (6). That our main results are unaffected by disregarding school dropouts also lends support to our motivation to have a closer look at how immigration enforcement policies can change early parental investment behavior.

In appendix C, we provide results from additional robustness checks we conduct. We evaluate whether children’s English language proficiency scores predict the first year when an immigration policy is enacted. Such a correlation would likely indicate a violation of the no-anticipation assumption of the policy. We do not find any evidence for a systematic introduction of immigration enforcement policies as a response to children’s language skills.

We also check if migrants are moving as a response to tougher immigration policies. To evaluate whether this is the case, we first look at the impact of the immigration policy on the population composition within the MSA. We do not find evidence that immigration enforcement impacted the composition of the MSAs, which we interpret to mean that any bias introduced in our estimates by mobility is likely small. Second, we restrict the sample to those US children who did not move over the preceding year. We find similar results to those reported in table 2. Nevertheless, we would expect that migrants with more success in the labor market, and thus those with likely higher investments in their children, are more mobile. Therefore, any mobility bias in our estimates would likely lead to an understatement of the true spillover effects of immigration policies on children’s human capital.

Finally, we also investigate whether heterogeneity in the estimated treatment effect might bias our results. The recent econometric literature on difference-in-differences has raised concerns that in settings with staggered treatment adoption, as in our case, standard estimates might be biased if treatment effects are heterogeneous (de Chaisemartin and D’Haultfœuille, 2020; Borusyak et al., 2021; Goodman-Bacon, 2021). Using the robust approach proposed by Sun and Abraham (2021), we do not find evidence that this is a concern in our estimation. Our results obtained from the robust approach are very similar both in terms of dynamics and magnitude as the results discussed above (see figure C.2 and figure C.1 in appendix C).

Parents’ Response to Immigration Enforcement

To better understand our results, we explore changes in parental investment behavior caused by immigration enforcement as one important underlying channel. An increase in the deportation risk after the introduction of immigration policies may change parents’ inputs in terms of formal education (e.g., preschool) and may also change their time investment; see our framework in section 2.

Preschool Enrollment Decision

We first investigate how immigration enforcement affects parents’ decisions to enroll their children in non-mandatory preschool. Attending preschool as a form of formal educational care can improve children’s language skills, specifically for children of disadvantaged backgrounds. Social interactions at a younger age with native speakers is particularly important for the development of language skills (e.g., Palermo and Mikulski, 2014; Villarreal and Gonzalez, 2016). An increase in the deportation risk caused by immigration enforcement might deter likely undocumented parents from enrolling their children in non-mandatory preschool programs. This decision might ultimately lead to lower language skills in the children.28Parents might not enroll their children in preschool for financial reasons, if immigration enforcement reduces their earnings possibilities. While we cannot investigate such a channel directly, we will show in the next section that their investment decisions are closely linked to parents’ heightened fear of being detected and deported.

In our analysis, we concentrate on children between the ages of three and four to capture the impact of immigration enforcement on enrollment in non-mandatory formal early childhood education programs. Preschool attendance is reported in the ACS only from age three onward. We choose the upper bound to be three years, as in some states compulsory schooling starts at age five.29Our results are virtually identical when considering children between three years and five years. In some states, for example in Maryland, children have to attend a mandatory year of Kindergarten at age five before starting school at age six, however. We present our estimation results using the difference-in-different approach form equation (5) in table 4.

Looking at column (1) of the table, we find that undocumented parents are less likely to enroll their US children in non-mandatory preschool as a response to immigration enforcement. The enrollment probability drops by around 2.19 percentage points or around 7 percent as a response to immigration enforcement.30In the ACS, 32 percent of US children with likely undocumented parents attend non-mandatory preschool programs. This drop is quite substantial. We find similar results using the residual method as proxy for parents’ citizenship status; see column (2). Using the residual methods, however, leads to slightly more precisely estimated effects.

Our effect is of similar magnitude as those found in Santillano et al. (2020), who investigate the impact of immigration raids on the Head Start enrollment of Hispanics.31Related are also the findings in Watson (2014), who shows that Medicaid enrollment decreases if migrant apprehension in a region rises. They find that an immigration raid reduces enrollment by approximately 10 percent. Our results show that immigration enforcement policies can reduce voluntary general preschool enrollment in the population of US-born children with likely undocumented parents. Such lower enrollment propensity ultimately leads to lower skill accumulation of US-born children of likely undocumented parents.

One concern might be that areas enacting immigration policies saw a drop in preschool quality, for example. If this was the case, one would expect that parents residing in those areas have in general a different enrollment propensity, regardless of the enacted immigration policies. To investigate this concern further, we also estimate the impact of immigration enforcement on the non-mandatory preschool enrollment of children of native or naturalized parents. The results are reported in column (3) of table 4.

Our results show that immigration enforcement has no effect on native or naturalized parents’ enrollment decisions. The estimates for this group are very small and not statistically significant on any conventional level. These null effects are also quite precisely estimated, and we can rule out any meaningful impact of immigration enforcement. Overall, there is no evidence that systematic unobserved differences in parents’ general enrollment decisions between MSAs with and without immigration enforcement can explain our results.

Time Investment Decision

We find that immigration policies reduce non-mandatory preschool enrollment of children of likely undocumented parents. Given the predictions in our model, it is interesting to see whether parents try to compensate for the likely disadvantageous effect by adjusting their time spent with the child. To do so, we look at the impact of immigration enforcement on how parents with children of preschool age divide their time among general, educational, social, and recreational activities, using our ATUS sample. Notice that in comparison to our conceptual framework, we allow for different types of time investment in our empirical analysis. For example, a separate activity “social” in our analysis allows us to obtain further evidence of whether immigration enforcement causes heightened fear among immigrants by observing reduced time spent on social activities.

Given that the ATUS is only representative on the state level and the significantly smaller sample size compared to the ACS, we estimate a slightly modified version of our difference-in-difference model. In our modified model, we explicitly use the proxied immigration status of the parent (see, for example, Kuka et al., 2020, for a similar approach using DACA eligibility status):

(6)   \begin{align*} y_{i,a,s,t}= \alpha^{PT}+\Lambda^{PT}_{1} IE_{s,t}+ \Lambda^{PT}_{2}LU_{i} + \beta^{PT} IE_{s,t}*LU_{i}+X'_{i, s,t}\Gamma^{PT}+\gamma^{PT}_{s}+\theta^{PT}_{t}+\epsilon^{PT}_{i,a,s,t} \end{align*}

where y_{i,a,s,t} is the time spent (in minutes) of individual i living in state s on activity a at time t. The vector {X_{i,s,t}} includes children and household characteristics. Similar as before, IE_{s,t} is our indicator variable equal to one if the state s has adopted a measure of interior immigration enforcement policy in year t, and zero otherwise. The variable LU_{i} is an indicator that takes the value of one if the individual in the sample is a likely undocumented immigrant, as discussed in section 3.

The estimates of our coefficient of interest, \beta^{PT}, are reported in table 5. In each column of the table we present the impact of immigration enforcement on parental time spent with the child in one of our five activities. The effects are expressed in minutes per day. Notice that we consider parents with preschool-age children only.

We do not find evidence that parents directly compensate for lower preschool enrollment and increase their educational time with the child as a response to heightened enforcement; see the results in column (2) of the table. We find, however, evidence that parents spend considerably more recreational time with their children by playing, doing sports, or spending leisure time with them. The estimates presented in column (3) indicate that immigration policies increase recreational time with the child by more than 13 minutes per day. This is quite a substantial increase compared to the mean and implies that immigration enforcement leads parents to double their time spent on recreational activities with their children.

One can infer from these results that parents try to compensate for not sending their child to preschool by increasing the time spent with their children. As the increase is concentrated on recreational activities, the quality of the time investment is likely not sufficient to fully offset and compensate for the disadvantages of not sending their child to preschool. One explanation for such a behavior may be that undocumented parents tend to underestimate the importance of early human capital investments (e.g., Boneva and Rauh, 2018).

Another explanation might be that the positive effect of immigration enforcement on recreational time spent with the child is employment-related and simply mechanical. After the introduction of immigration enforcement measures, prospective employers might be more reluctant to hire likely undocumented immigrants (see, for example, the discussion in Amuerdo-Dorantes et al., 2021). Facing fewer employment possibilities, parents spend more time at home, which also leads them to mechanically increase the time interacting with their children.32Using the ATUS, we find that immigration policies reduce weekly hours worked by 1.5 hours/week. Employment is reduced by 6 percentage points. Both of these estimates are very noisy, however, and not statistically significant on any conventional level.

Our estimates on parental time spent socializing with the child, presented in the last column of table 5 do not support such an explanation. Immigration enforcement reduces time spent on daily social activities by 37 minutes per day, or a drop of roughly 50 percent compared to the mean. If parents really did spend more time with their children only because they spent less time at work, then we also would expect to see them spend more time on social activities. At least, we would not expect such a large drop as we estimate.

We interpret the reduction in social activities as support for our hypothesis that fear of deportation caused by immigration enforcement leads to social isolation of likely undocumented parents and their US-born children.33Fear of being detected and deported is also in line with our suggestive findings that immigration enforcement reduced employment and hours worked for likely undocumented parents. See also East et al. (n.d.) for evidence of such “chilling” effects in the labor market.

Our results show two important channels through which immigration enforcement can lower children’s language skills. On the one side, immigration enforcement policies lead to a reduction in time spent on social activities, likely caused by an increased fear of being detected and deported. On the other side, fear of being detected and deported leads parents to reduce children’s time in formal childcare. Parents try to compensate for this reduction in formal education with an increase in recreational time spent with the child. This additional time, however, is less productive.

The findings presented here complement and extend the results of Kuka et al. (2020), who show that granting temporary work authorization and deferral from deportation for undocumented, high school-educated youth through the Deferred Action for Childhood Arrivals (DACA) program increases human capital investment, likely through higher perceived returns to education. We show that parents change their human capital investment decisions as a response to immigration enforcement. Unlike in Kuka et al. (2020), however, parents change their investment not (only) because immigration enforcement leads to changes in the expected return to education, but (likely) because of heightened fears of detection and deportation. These fears lead parents to substitute formal childcare with parental care at home. However, such parental care is not as productive, leading to lower human capital accumulation in children; this can explain the lower language proficiency we have documented in US-born children of likely undocumented parents.


Considerably more resources have been devoted to immigration and customs enforcement in the United States since 2001. Many of the introduced policies are aimed at identifying, apprehending, and ultimately deporting undocumented immigrants in the country. While they primarily target undocumented immigrants, the negative consequences of immigration enforcement can spill over to US-born children of those undocumented individuals.

Using the temporal and spatial variations in the introduction of immigration enforcement policies, we evaluate how immigration enforcement policies affect language skills of US-born children of likely undocumented parents. We concentrate on language proficiency as one important skill. On the one side, it is an important determination for future educational and labor market outcomes. On the other side, language skills of children are largely shaped by social interactions, which immigration enforcement policies might reduce.

Our difference-in-difference estimates show a large and significant impact of immigration enforcement on children’s English language skills. Our estimated effects are of similar size but have the opposite sign as important programs to improve skills of disadvantaged children, such as Head Start. Investigating the dynamics of the effect, we find that immigration enforcement leads to a gradual deterioration of children’s English language skills over time without a sign of reversal. This suggests a role for changes in parental human capital investment behavior caused by immigration enforcement as an important underlying channel.

We find that immigration enforcement indeed changes parents’ human capital investment in their children. Heightened immigration enforcement deters undocumented parents from enrolling their younger children in non-mandatory preschool. We also show that parents try to compensate for the likely negative consequences of not enrolling their children in preschool. While parents respond by increasing recreational time or play time spent with their children, this extra time cannot fully compensate for reduced formal education; this ultimately leads to a decrease in English language skills. Our estimates also show that immigration enforcement reduces time spent on social interactions. We interpret this finding as evidence that an increasing fear of being detected and deported leads parents to change their behavior.

Overall, our results show substantial negative spillovers of immigration enforcement on US-born children. As US-born children of undocumented parents can legally stay in the country, lower accumulated human capital arising from immigration enforcement will likely lower these children’s education and labor market prospects. It also likely increases their dependence on the social security system later in life. Ultimately, immigration enforcement will hamper intergenerational mobility for these children, reversing the slow progress these disadvantaged groups have made.


Abraham, K. G., Maitland, A. and Bianchi, S. M. (2006), ‘Nonresponse in the American time use survey: Who is missing from the data and how much does it matter?’, The Public Opinion Quarterly 70(5), 676–703.

Abramitzky, R., Boustan, L., Jacome, E. and Perez, S. (2021), ‘Intergenerational Mobility of Immigrants in the United States over Two Centuries’, American Economic Review 111(2), 580–608.

Aguiar, M. and Hurst, E. (2007), ‘Measuring trends in leisure: The allocation of time over five decades’, Quarterly Journal of Economics 122(3), 969–1006.

Amuedo-Dorantes, C. and Arenas-Arroyo, E. (2018), ‘Split Families and the Future of Children: Immigration Enforcement and Foster Care Placements’, AEA Papers and Proceedings 108, 368–372.

Amuedo-Dorantes, C. and Arenas-Arroyo, E. (2019), ‘Immigration Enforcement and Children’s Living Arrangements’, Journal of Policy Analysis and Management 38(1), 11–40.

Amuedo-Dorantes, C. and Arenas-Arroyo, E. (2021), ‘Immigration policy and fertility: Evidence from undocumented migrants in the u.s.’, Journal of Economic Behaviour and Organization 189, 274–297.

Amuedo-Dorantes, C., Arenas-Arroyo, E. and Sevilla, A. (2018), ‘Immigration enforcement and economic resources of children with likely unauthorized parents’, Journal of Public Economics 158(December 2017), 63–78. URL: https://doi.org/10.1016/j.jpubeco.2017.12.004.

Amuedo-Dorantes, C. and Lopez, M. J. (2017), ‘The Hidden Educational Costs of Intensified Immigration Enforcement’, Southern Economic Journal 84(1), 120–154.

Amuerdo-Dorantes, C., Arenas-Arroyo, E. and Schmidpeter, B. (2021), ‘Immigration enforcement and the hiring of low-skilled labor’, AEA Papers & Proceedings 111, 593–597.

Arington, M. (1990), ‘English-Only Laws and Direct Legislation: The Battle in the States over Language Minority Rights’, Journal of Law and Politics 1.

Aslan, M. and Young, C. S. (2019), ‘Fear and the safty net: Evidence from secure communities’, mimeo.

Aucejo, E. and James, J. (2021), ‘The path to college education: The role of math and verbal skills’, Journal of Political Economy 129(10), 2905–2946.

Baranov, V., Bhalotra, S., Biroli, P. and Maselko, J. (2020), ‘Maternal Depression, Women’s Empowerment, and Parental Investment: Evidence from a Randomized Controlled Trial’, American Economic Review 110(3), 824–859.

Bellows, L. (2019), ‘Immigration enforcement and student achievement in the wake of secure communities’, AERA Open 5(4), 1–20.

Bernal, R. and Keane, M. P. (2011), ‘Child care choices and children’s cognitive achievement: The case of single mothers’, Journal of Labor Economics 29(3), 459–512.

Bleakley, H. and Chin, A. (2004), ‘Language skills and earnings: Evidence from childhood immigrants’, The Review of Economics and Statistics 86(2), 481–496.

Bleakley, H. and Chin, A. (2010), ‘Age at arrival, english proficiency, and social assimilation among us immigrants’, American Economic Journal: Applied Economics 2(1), 165–192.

Boneva, T. and Rauh, C. (2018), ‘Parental beliefs about returns to educational investments-the later the better?’, Journal of the European Economic Association 16(6), 1669–1711.

Borjas, G. (2017), ‘The Labor Supply of Undocumented Immigrants’, Labour Economics 46, 1–13.

Borusyak, K., Jaravel, X. and Spiess, J. (2021), ‘Revisiting event study designs: Robust and efficient estimation’, mimeo.

Callaway, B. and Sant’Anna, P. (2021), ‘Difference-in-differences with multiple time periods’, Journal of Econometrics 225(2), 200–230.

Capps, R., Berger Cardoso, J., Brabeck, K., Fix, M. and Ruiz Soto, A. G. (2020), ‘Immigration Enforcement and the Mental Health of Latino High School Students’, MPI Report.

Chetty, R., Hendren, N., Jones, M. R. and Porter, S. R. (2020), ‘Race and Economic Opportunity in the United States: an Intergenerational Perspective’, Quarterly Journal of Economics 135(2), 711–783.

Das, J., Dercon, S., Habyarimana, J., Krishnan, P., Muralidharan, K. and Sundararaman, V. (2013), ‘School inputs, household substitution, and test scores’, American Economic Journal: Applied Economics 5(2), 29–57.

de Chaisemartin, C. and D’Haultfœuille, X. (2020), ‘Two-way fixed effects estimators with heterogeneous treatment effects’, American Economic Review 110(9), 2964–96.

Dee, T. S. and Murphy, M. (2020), ‘Vanished Classmates: The Effects of Local Immigration Enforcement on School Enrollment’, American Educational Research Journal 57(2), 694–727.

Deming, D. (2009), ‘Early childhood intervention and life-cycle skill development: Evidence from head start’, American Economic Journal: Applied Economics 1(3), 111–34.

Dustmann, C. and Fabbri, F. (2003), ‘Language Proficiency and Labour Market Performance of Immigrants in the UK’, Economic Journal 113(489), 695–717.

East, C. N., Hines, A. L., Luck, P., Mansour, H. and Velásquez, A. (n.d.), ‘The labor market effects of immigration enforcement’, mimeo.

Felfe, C. and Lalive, R. (2018), ‘Does early child care affect children’s development?’, Journal of Public Economics 159, 33–53.

Fiorini, M. and Keane, M. P. (2014), ‘How the allocation of children’s time affects cognitive and noncognitive development’, Journal of Labor Economics 32(4), 787–836.

Francesconi, M. and Heckman, J. J. (2016), ‘Child Development and Parental Investment: Introduction’, Economic Journal 126, F1–F27.

Gándara, P. and Ee, J. (2018), U.S. Immigration Enforcement Policy and Its Impact on Teaching and Learning in the Nation’s Schools, Technical report.

Goodman-Bacon, A. (2021), ‘Difference-in-differences with variation in treatment timing’, Journal of Econometrics 225(2), 254–277.

Greaves, E., Hussain, I., Rabe, B. and Rasul, I. (2021), ‘Parental responses to information about school quality: Evidence from linked survey and administrative data’, forthcoming Economic Journal.

Henry, G. T. and Rickman, D. K. (2007), ‘Do peers influence children’s skill development in preschool?’, Economics of Education Review 26(1), 100–112.

Ho, A. D. (2020), ‘What is the stanford education data archive teaching us about national educational achievement?’, AERA Open 6(3), 1–4.

Kinsler, J. and Pavan, R. (2021), ‘Local distortions in parental beliefs over child skill’, Journal of Political Economy 129(1), 81–100.

Kuka, E., Shenhav, N. and Shih, K. (2020), ‘Do Human Capital Decisions Respond to the Returns to Education? Evidence from DACA’, American Economic Journal: Policy 12(1), 293–324.

Kuziemko, I. (2014), ‘Human capital spillovers in families: Do parents learn from or lean on their children?’, Journal of Labor Economics 32(4), 755–786.

Laffers, L. and Schmidpeter, B. (2021), ‘Early Child Development and Parent’s Labor Supply’, Journal of Applied Econometrics 36(2), 190–208.

Nicoletti, C. and Tonei, V. (2020), ‘Do parental time investments react to changes in child’s skills and health?’, European Economic Review 127, 103491.

Orrenius, P. M., Zavodny, M. and Gutierrez, E. (2018), ‘Do state employment eligibility verification laws affect job turnover?’, Contemporary Economic Policy 36(2), 394–409.

Palermo, F. and Mikulski, A. M. (2014), ‘The role of positive peer interactions and English exposure in Spanish-speaking preschoolers’ English vocabulary and letter-word skills’, Early Childhood Research Quartlery 29(4), 625–635.

Passel, J. S., Lopez, M., Cohn, D. and Rohal, M. (2014), ‘As growth stalls, unauthorized immigrant population becomes more settled.’. URL: http://www.pewhispanic.org/2014/09/03/as-growth-stalls-unauthorized-immigrant-population-becomes-more-settled/.

Pew Research Center (2019), ‘U.S. Unauthorized Immigrant Total Dips to Lowest Level in Decade’, Report – November.

Pope, N. G. (2016), ‘The Effects of DACA Amentation: The Impact of Deferred Action for Childhood Arrivals on Unauthorized Immigrants’, Journal of Public Economics 143, 98–114.

Puma, M., Bell, S., Cook, R., Heid, C., Broene, P., Jenkins, F., Mashburn, A. and Downer, J. (2012), ‘Third grade follow-up to the head start impact study’, OPRE Report 2012-45.

Ruggles, S., Flood, S., Goeken, R., Grover, J., Meyer, E., Pacas, J. and Sobek, M. (2020), ‘IPUMS USA: Version 10.0 [dataset]’.

Santillano, R., Potochnik, S. and Jenkins, J. (2020), ‘Do Immigration Raids Deter Head Start Enrollment?’, AEA Papers and Proceedings 110(5), 419–423.

Schmidpeter, B. (2020), ‘The Long-Run Labor Market Effects of Parental Unemployment’, Upjohn Institute Working Paper 20-322.

Sun, L. and Abraham, S. (2021), ‘Estimating dynamic treatment effects in event studies with heterogeneous treatment effects’, Journal of Econometrics 225(2), 175–199.

Vaughan, J. M. (2013), ‘Deportation numbers unwrapped raw statistics reveal the real story of ice enforcement in decline’, Center for Immigration Studies. URL:  https://cis.org/Report/Deportation-Numbers-Unwrapped.

Vikstrom, E. R., Shin, H. B., Collazo, S. G. and Bauman, K. J. (2015), ‘How Well—Still Good? Assessing the Validity of the American Community Survey English-Ability Question’, SEHSD Working Paper No. 2015-18.

Villarreal, V. and Gonzalez, J. E. (2016), ‘Extracurricular activity participation of Hispanic students: Implications for social capital outcomes’, International Journal of School and Educational Psychology 4(3), 201–212.

Watson, T. (2014), ‘Inside the refrigerator: Immigration enforcement and chilling effects in medicaid participation’, American Economic Journal: Economic Policy 6(3), 313–338.

Weisleder, A. and Fernald, A. (2014), Social environments shape children’s language experiences, strengthening language processing and building vocabulary, in I. Arnon, Casillas, C. Kurumada and B. Estigarriba, eds, ‘Language in Interaction. Studies in honor of Eve V. Clark’, John Benjamins, pp. 29–50.

CGO scholars and fellows frequently comment on a variety of topics for the popular press. The views expressed therein are those of the authors and do not necessarily reflect the views of the Center for Growth and Opportunity or the views of Utah State University.