Refugees to the United States pay $21,000 more in taxes than they receive in benefits, and other findings from a recent study of refugees

The paper was posted to NBER a few days ago and has gotten quite a bit of attention. Unfortunately there doesn’t seem to be an ungated version floating around right now, but lots of universities have access to NBER papers by email so you might be able to get to it by giving them your .edu email address if you have one.

The broad question that the authors set out to answer is: what are the economic outcomes for refugees in the United States? This question is split into roughly three parts.

1. How do economic outcomes for refugees differ according to age at the time of arrival?
2. How do economic outcomes for refugees change across time as they age in the US?
3. What are the lifetime fiscal costs and benefits to the United States of refugees?

To answer these questions, the researchers construct a somewhat novel dataset on refugees in the United States. As they point out, data on refugees is very hard to come by, which is probably why so little work on this topic has been completed to date.

Why it is so hard to research refugees

The authors point out that previous work on American refugees is scant due to difficulties in obtaining data on refugees. There has been lots of research on immigrants, of which refugees are a subset, but it is not straightforward to identify the refugees. Major data sets do not identify refugee status in the data, and data sets that do look at refugees are merely a snapshot and do not include followups from which time-path outcomes can be measured.

For these reasons, there have been few large-scale studies on refugees. The authors go through a few key papers investigating refugee outcomes, but in each of these the samples are small. Each study that the authors discuss concerns only a few hundred refugees, which is a challenge to their external validity — that is, the applicability of their results to refugees as a whole.

Getting data

One contribution of this paper is in implementing a method for getting data on refugees in the face of these challenges. The data comes from the American Community Survey, which is a large scale survey carried out by the US Census Bureau on an ongoing basis. The challenge with the ACS, of course, is that refugee status is not identified in the data — although immigrant status is. Respondents to the ACS who migrated from a particular country in a particular year are identified in the data.

To identify refugees, the authors make use of supplementary data from the Department of State (DOS). The DOS publicly tracks the number of refugees admitted from each country in each year. This information can be combined with the ACS data to construct a group of likely refugees. The basic idea is to identify intense periods of refugee resettlement from specific countries. For instance, the authors find that over 70% of immigrants from Libya in 1991 were refugees. Then, by taking ACS results that include only folks who immigrated from Libya in 1991, we have a dataset in which we are reasonably sure that 70% of respondents are refugees. By carrying this out over multiple periods of refugee resettlement — Afgnanistan in 1991–1992 and 2001–2003, Albania in 1991, and so on — the authors construct a dataset of 19,298 likely refugees, which is representative of around 588,000 refugees in total. This is the largest dataset on refugees that has been studied, and represents about a third of the total refugees that entered the U.S. between 1990 and 2015.

An interesting thing about this sample of refugees is that it probably oversamples refugees from nations with particularly egregious human rights records. Since the focus is on refugees who immigrate during periods of mass migration from their home countries, the refugees in the sample are especially likely to be victims of civil war, terrorism, or other human rights abuses which led to their migration.

How do economic outcomes for refugees differ according to age at the time of arrival?

A key finding of the paper is that there is a large and significant difference in outcomes for refugees who enter the US as young children and those who enter as older teenagers or adults. This difference is persistent across many dimensions.

Educational Attainment for Children, Teenagers, and Young Adults

The authors begin by investigating the rate of high school graduation by age 19 for refugees compared to US-born citizens. The immediate finding is that the graduation rate of refugees is two percentage points lower than the national average for US-born respondents. This number hardly tells the full story, however. It turns out that it really matters how old a refugee is when she or he enters the US. By conditioning on age of entry into the United States, the authors are able to show a much more nuanced picture.

Refugees who enter the US at age 12 or below graduate from high school at around the same rate or even slightly higher than their US-born counterparts. Within this group, there is no clear relationship between age and graduation rates: refugees who are 13 at entry graduate at around the same rate as refugees who are 4 at entry (the authors do not investigate refugees who are under 4 years of age at entry due to missing data).

Refugees who are 14 or older at arrival have a harder time. For refugees who are 14 years old at arrival graduate at a rate of around 2–3 percentage points lower than US born people, and the rate drops sharply as age of arrival increases, with a graduation rate of only 70% or 20 percentage points below the US born average for refugees aged 16 at arrival.

The authors are able to explain a great deal of this heterogeneity by examining the English speaking ability of older refugees, as well as the presence of parents at the time of arrival. By controlling for English speaking ability and the presence of a biological mother, the negative effects on graduation rates by age for the older teens are essentially cut in half. The method used to control for these variables is somewhat imprecise, but the effect of controlling for them ends up being quite large, so it does seem like there is something there. It is also quite intuitive: folks who arrive in America with families intact and good English speaking skills are more likely to graduate from high school than those without those assets. It would have been interesting to include these controls for the US born population as well, to investigate the marginal effect of refugee status holding English proficiency and family status constant, but unfortunately the authors did not investigate this.

The good news for the older teens is that they do seem to catch up to their US born counterparts at later ages. The rate of both high school and college enrollment among refugees is significantly higher than that of US born citizens, and by ages of mid 20s, the difference in high school graduation rates essentially vanishes. This shows that refugees who arrive as older teenagers are not simply dropping out of high school — they are just taking longer to finish. For me, this is quite an encouraging finding, especially with respect to the sometimes-held view that refugees who arrive as older teenagers and young adults resist assimilating and participating in the culture. These findings show that the vast majority of refugees who arrive as older teenagers with poor English skills work hard to overcome these disadvantages and obtain basic levels of educational attainment by young adulthood.

How do economic outcomes for refugees change across time as they age in the US?

An important point made by the authors throughout the paper is that a simple snapshot of a refugee at a moment in time will not capture the full picture of his or her resettlement experience. It takes time for refugees to acclimatize to a new environment, and the economic outcomes of a single refugee at a point in time may be very different from the outcomes of that same person a few years later.

For this reason, the authors create a synthetic panel to measure outcomes across time. While it is not possible due to the nature of the data set to follow individuals across time, by making some assumptions they can try to compute time varying outcomes by designing a sample that carefully includes a group of refugees who have been in the US for 1 year, a group who have been in the US for 2 years, and so on, up to year twenty. This data set allows the authors to estimate some facts about “the average refugee”, which can shed light on how the refugee experience changes across time.

Employment rates and other labor statistics for refugees who arrive as adults

It is shown that the employment rates and labor force participation rates for new refugees are significantly below the corresponding rates for the US born population. This is unsurprising given that on average, refugees who arrive as adults have significantly lower levels of educational attainment and English proficiency.

What is surprising however is that only a few years after resettlement, the employment and labor force participation rates jump up to above their corresponding US baseline rates. After only four years, refugees have a higher labor force participation rate than natives, and after seven years, refugees have a higher employment rate.

Interestingly, this result is not affected much at all by language ability. Regardless of English proficiency, refugees start out with low employment rates for the first few years of resettlement, and eventually end up with higher employment rates than natives.

Refugee earnings remain lower on average that earnings for native-born citizens for life. On average, there is an annual earnings gap of $10,000 or more between refugees and US born citizens within the first few years of arrival. While this gap closes slightly over time, it remains at around $5,000 even at 20 years since resettlement. Controlling for English proficiency does narrow the gap, suggesting that learning English has large returns for refugees who arrive as adults. Nonetheless, the gap never closes to less than about $2,500 per year when language ability is taken into account.

Welfare programs

Since refugees enter the country with relatively low levels of education and experience, we would expect that they would use assistance programs like welfare and the Supplemental Nutrition Assistance Program (SNAP, often referred to as food stamps) at a higher rate than US-born people, and this turns out to be right. An average of 9.3% of refugees aged 18–45 at arrival use welfare and 38.7% use SNAP, compared with averages of 3% and 4.8% respectively for US-born adults in the same age range.

However, a snapshot of this kind masks the temporal changes that occur in individual refugees across time as they age in the US. Examining assistance program usage as a function of years since arrival, the authors find that the bulk of assistance usage is in the early years of a refugee’s time in the US, with SNAP receipt over 70% within the first year of resettlement, but falling to 20% a decade later. The usage of the measured assistance programs never drops to the averages for US born respondents, however most of the difference between refugees and non-refugees in welfare usage is explained by education and language differences — controlling for these variables, there is almost no statistically significant difference in welfare or SNAP usage.

What are the lifetime fiscal costs and benefits to the United States of refugees?

Having estimated employment employment rates, income, and welfare usage for refugees within their first 20 years of resettlement, it is now possible to use these estimates to obtain an estimate of the total fiscal costs and benefits to the U.S. government of admitting refugees. It is clear that resettlement of refugees imposes a fiscal cost at the outset, both in the direct costs of resettlement and in subsequent assistance programs. However, as the authors have made clear in the preceding sections, these costs shift over time: as time goes on, refugees use social assistance programs less, and work more. Earnings go up, and with them, income taxes. For this reason, the problem of determining the real cost to the host nation of resettlement is not as straightforward as it might seem.

To produce an estimate of the costs, the authors obtain average direct resettlement costs from the Administration for Children and Families (ACF) and the Bureau of Population, Refugees, and Migration (PRM), the government organizations concerned with resettling refugees. The average direct cost of resettling a refugee is estimated at $10,194 per refugee.

These direct costs are added to the imputed costs of the receipt of six main social assistance programs: welfar, SSI, Social Security, SNAP, Medicare, and Medicaid. By combining the numbers with estimates from the ACS, they are able to estimate approximate usage patterns and costs of these programs for refugees. The assumption is, of course, that the patterns of usage for each of these programs for refugees are similar to those of the overall population; there doesn’t seem to be a reason to doubt that this assumption mostly holds.

Next, using responses from ACS and an NBER software program called TAXSIM, the authors estimate the total taxes contributed over the 20 years span across which the data exists.

Their finding is that on average, the U.S. spends $15,148 on relocation costs and $92,217 in assistance over the 20 years, while the average refugee pays $128,689 in taxes during that same period. This yields a net benefit to the U.S. government of $21,324 over 20 years per refugee. This may be an unexpected result — the common wisdom is that refugees are a drain on the system, but these findings show that while the initial costs of resettlement are high, refugees pay for themselves over time.

Last Remarks

This work will draw a lot of criticism, with varying levels of associated fairness. One is that there is a huge number of assumptions that go in to forming many of these estimates — particularly the last one. The authors do address this partly by providing a table in the appendix providing some estimates of the net cost of resettlement under varying assumptions, and find that tweaking the parameters of their model yields a positive benefit to the government under any configuration that they tried. Included in these tests are variations that include only looking at older refugees, as well as perturbations to the discount rate.

Another interesting point that the authors do address is that the cost of interaction with the criminal justice system, which could be substantial if refugees are likely to incur those costs. Obtaining these costs would not be straightforward, although the authors do present an interesting finding about incarceration rates for refugees. Since the ACS is administered to people in prisons, they are able to compare the fraction of refugees in prisons and similar quarters to the general population. They find that only 0.5 percent of refugees were in prisons and similar facilities, compared with 2.34% for the overall US-born population. By this measure, refugees are about 80% less likely than the general population to be incarcerated.

Another tricky thing is that the sample is constructed to include people who are 65 years old or younger, and so the sample does not include people who are eligible for Medicare and Social Security. The authors explain that including respondents older than that would introduce complications as mortality rates at ages 65 increase. Since there data set only includes those who are alive to fill out the survey, estimates of Social Security and Medicare payments would be inaccurate — perhaps wildly so. For instance, a refugee who dies at age 64 costs Medicare and Social Security zero, whereas as refugee who dies at age 120 likely costs these programs a lot. If mortality rates for refugees differ from the broader population, this could impact estimates significantly, and the authors were not able to obtain this mortality data for their estimates.

One thing that I’ve noticed about the coverage of this article is that the focus is very much on the $21,000 net benefit to the government for refugees, despite this being a relatively small part of the overall research. The major contribution of the work is in finding a way to study the experience of a large number of refugees across time, and in my view the findings that refugees are enrolled in school and employed at higher rates than the overall US born population is of greater public interest than an estimate of the cost of resettlement. We shouldn’t forget that refugees are refugees because they are escaping something horrible, and even if it costs a little bit to help them do that, it may well be worth it on humanitarian grounds. If it is a net economic benefit to the host nation, hey, that’s a cherry on the sundae, but even if the true benefit is less than $21,000, or 0, or negative, that should not in principle dissuade us from helping those in need. And on the flip side, what this research does show is that those who are helped respond and work hard to build better lives for themselves. Those who are brought in as young children achieve success at around the same levels as the U.S. born population, but even those who arrive at later stages are able to catch up and eventually find themselves in comparable life situations to the native population.