The effect of refugee resettlement on economic indicators in American Rust Belt cities

The effect of refugee resettlement on

economic indicators in American Rust

Belt cities

Alex Schwartz

S2132168

Leiden University

Masters of Public Administration

Concentration in Economics and Governance

Advisor: Dr. Alexander Afonso

Second Reader: Dr. Brendan Carroll

January, 2019

1

Table of Contents

Introduction ... 2

Theory ... 4

Hypotheses and Predicted Causal Mechanisms ... 11

Research Methods ... 13

Data Preparation... 20

Results ... 23

Limitations and Further Research ... 46

Conclusion and Policy Implications ... 48

Works Cited ... 50

Appendix 1- Data Coding Functions ... 53

2

Introduction

Rust Belt cities have faced significant hardships during the second half of the 20

th

century. The definition of a Rust Belt city somewhat fuzzy: they are indeed cities facing

post-industrial decline, but the clear parameters are more poorly defined. Baltimore and Washing DC

are adjacent to each other but only Baltimore would be considered in the Rust Belt. San

Bernardino, CA and Camden, NJ are both poor, fiscally struggling municipalities on the outskirts

of more prosperous cities but only Camden would be considered Rust Belt. Some authors define

the cities as those suffering from the steel and car manufacturing shock of the 1980s (Feyer, et al.

2007). Others, however, take a more “you know it when you see it” approach: economically

struggling cities, also struggling with post-industrialization in a country as mobile at the US,

suffer from population decline (Jamrisko and Englert 2017). A significant commonality among

Rust Belt cities is that, when jobs disappeared due to deindustrialization, unemployment rates

equilibrated with the rest of the country primarily through population decline (Feyer, et al. 2007).

It is interesting to note, this is not reflected in “Rust Belt”-like regions in other countries, such as

Northern England or Wallonia in Belgium, where labor force participation fell. As such,

population growth is considered one of the most significant challenges facing Rust Belt mayors,

counselors, and managers, in addition to job creation.

This is where refugees can be appealing to civic leaders. There is significant anecdotal

evidence that refugees are able to help revitalize some declining cities. To some degree, this can

seem a little obvious. If population decline is a problem, bringing in new people can be a

solution. Many municipalities, in the name of revitalization, have welcomed refugees. Various

news reports from Upstate New York to St. Louis discuss the perceived economic benefits of

refugees. A newspaper from Utica, New York noted that finally, the town is growing again,

presumably in part due to the influx of refugees (Burke 2017). Syracuse has started

entrepreneurial programs to assist refugees, perceived as being natural entrepreneurs.

Refugees have been credited with all sorts of local economic successes. “They’ve

stabilized neighborhoods, cushioned city coffers and, in the process, supported credit ratings and

bond sales,” reports Bloomberg News (Jamrisko and Englert 2017). Other researchers have cited

the same anecdotal evidence (Karam 2017). Many cities, including diverse, non-Rust Belt

locations such as Los Angeles, Dallas, Seattle, Louisville, Phoenix, and Boston have said

3

refugees are integral to the cities’ success. Mayors and governors have been willing to go

against the political winds to welcome refugees (Connors 2017). Michigan created an office to

aid refugees, despite having unified Republican control and voting for Trump in the 2016

election.

Unfortunately, there isn’t a lot of data to support these conclusions. Yolande

Pottie-Sherman found that a significant amount of discussion about refugee policy mixes values with

economics (Pottie-Sherman 2018). Many mayors invite refugees, citing economic reasons, but

often rely on more moralistic personal rational. There is further correlational data that some

mayors and newspapers cite, but they are indeed primarily correlations. It was noted, for

example, that the neighborhood with significant refugees in St. Louis was in the top 15

neighborhoods for property price appreciation (Jamrisko and Englert 2017). While this is

undoubtedly good for St. Louis, a city that has suffered significantly from population loss,

property value loss, and disinvestment, this statistic doesn’t provide much indication that the

refugees caused the appreciation. It was a simple correlation without any time-series

consideration and, indeed, refugees may have been placed in those particular neighborhoods

because they had optimistic outlooks. I will try to look deeper into substantiating these claims.

First, I will look at the theory and prior research on refugees and immigrants. Refugees

are a special case of immigrants, indeed, but they share some characteristics with them. As such,

some immigrant literature can be informative. This theory will focus not only on the economic

theory behind the effects of refugees but also the empirical research on refugee contribution to

local economies. Then, I will discuss how the theoretical framework underlying my hypotheses.

Then, I will outline the research methods and data I will use. Finally, I will provide an analysis

of the results, followed by conclusions and potential policy recommendations.

4

Theory

First, a note on terminology: different countries use different definitions for classifying

immigrants and their children. I will use the terminology used by the National Academy of

Sciences. “First generation” Americans are residents who were born outside of the United States

without American parents. “Second generation” Americans are citizens who have at least one

parent who is first generation. “Third generation” and greater refers to everyone else. About

75% of Americans are considered third generation (National Academies of Sciences,

Engineering, and Medicine 2017).

Most refugee and immigrant research has focused on refugees in major cities (Chicago,

New York, Los Angelis, etc.). This is, in large part, because refugees, along with first generation

immigrants more broadly, often settled (or were settled) in these larger, more diverse cities.

Granted, refugees often don’t choose their initial location, but do have mobility once settled.

Indeed, the larger the city, the more likely it is to have a significant population from any one

country, and people do benefit from settling with people of their own cultural and linguistic

background (Harris 2016). However, this does raise questions about the applicability of some of

this research to Rust Belt cities. Rust Belt cities are distinct from the large, flourishing

metropolitan areas that historically functioned as “gateways”. Economically, almost by

definition, they are different (Feyer, et al. 2007). They weathered the downturn of the 1970s

much more poorly than the gateway cities and have seen lower job and income growth.

Furthermore, their ethnic composition is different.

Interestingly, Rust Belt cities have comparable out-migration rates to other American

cities. In fact, Rust Belt out-migration levels were slightly lower than for other cities.

Population loss came almost entirely from a lack of in-migration (Feyer, et al. 2007). As such,

perhaps the intuition of the news reports is correct- increasing in-migration can stabilize the

population. While Rust Belt cities face a host of problems, stemming population decline can go

a long way. Population loss has been cited as one of the most significant challenges facing these

cities (Jamrisko and Englert 2017). As the population falls, there are fewer people paying for the

same amenities, infrastructure, and governing institutions, along with personnel costs, such as

5

pensions. As such, the cost burden on each remaining resident grows. This either forces cities to

raise taxes on existing residents or cut services, making the city less attractive to new residents.

Indeed, rust belt cities have poorer quality amenities than other cities and, furthermore, they are

still declining (Feyer, et al. 2007). Additionally, as residents leave, property values fall and

economic activity decreases, further hurting remaining residents. By simply attracting new

residents, many of these problems can be mitigated.

Rust Belt cities are also different in their ethnic composition and histories with refugees.

Refugees have, historically, settled in “gateway” cities and, as such, Rust Belt cities tend to have

smaller refugee (and immigrant) communities. This does present a problem, as refugees, like

other immigrants, tend to concentrate in areas with people of similar backgrounds (Singer and

Wilson 2006). This also presents opportunities. In mid-sized and smaller cities, refugees have

been shown to have significant impacts because of their difference from the local population

(Singer and Wilson 2006). However, networks, institutions, and support have been shown to be

significantly important to the success of refugees. Refugees usually face linguistic and cultural

barriers, common with other immigrants, which can become significant issues (Best 2009).

Additionally, they often have lower levels human capital than native-born citizens (although, in a

weird statistical anomaly, Rust Belt regions with higher levels of human capital actually had

poorer unemployment outcomes around the time of deindustrialization) (Feyer, et al. 2007).

This presents a theoretical dilemma. Rust Belt cities are where refugees may be able to

have the greatest impact, but these cities may also be where they are less able to thrive. Yet,

while some research says the support and integration institutions are key, there is evidence that

refugees are able to make significant, positive impacts with even minimal support. Utica, the

city mentioned in the introduction, didn’t have many resources available and the refugees

reportedly flourished; as did those with the minimal support in Syracuse (Burke 2017).

Immigrants (notably, not exclusively refugees) were able to succeed in Detroit, a city which had

little money for support services, even to the point where their neighborhood gained population

while the remainder of the city experienced dramatic population decline (Wainer 2013).

Complicating things further, in recent years, the old patterns of refugee resettlement have

changed and refugees have begun being settled resettling in newer, less diverse cities (Portland,

Seattle, Atlanta, etc.) (Singer and Wilson 2006). The reasons for this are unclear. They could be

practical: “traditional” cities may be suffering disproportionally from the recent housing crunch.

6

They could be political. Nevertheless, these newer settlement patters create new questions about

the effects of refugees. It also presents opportunities for non-traditional “gateway” cities. Still,

these cities are likely not generalizable, particularly to Rust Belt cities. Seattle is, in many ways,

unique. Many cities dream of being “the next Seattle”. Additionally, what is true for large,

wealthy, growing cities may not apply for smaller, poorer, shrinking cities. Portland, Seattle, and

Atlanta are all considered very successful cities in the 21

st

century. They are likely several for

reasons that go far beyond their population of refugees, reasons that are likely difficult for Rust

Belt cities to reproduce.

Still, Rust Belt cities have many traits which seem to fit well with the research on

refugees. “Successful” Rust Belt cities (as defined by James Feyer- cities which seem to have

overcome their Rust Belt struggles) have managed to add jobs in manufacturing, health, and

business services (Feyer, et al. 2007). The healthcare industry is the largest employer of

refugees, followed by manufacturing (Kallick and Roldan 2018). While it is unknown whether

there is any causal link or if the correlation is even meaningful, it does provide some evidence

for a suitable skills match between Rust Belt cities and refugees. “Eds and Meds” is a commonly

cited “layman's development strategy”.

Additionally, there is research on the general characteristics of refugees which makes

them desirable. In St. Louis, refugees were 29% more likely to be entrepreneurs than the

native-born population (Jamrisko and Englert 2017). A joint study from the Fiscal Policy Institute and

the Center for American Progress found that Syrian refugees had a business ownership rate eight

percentage points higher than native-born citizens and had a median wage that was $7,000 higher

(Connors 2017). Additionally their crime rates are lower. And, while immigrants (NOT

exclusively refugees) tend to have lower incomes than the native-born population, their

children’s income are “virtually identical” to the native-born population, indicating significant

social mobility and highly effective integration (Morin 2013). These are all issues that Rust Belt

cities struggle with and characteristics from which any city would benefit.

David Dyssegaard Kallick and Cyierra Roland surveyed business owners for the Fiscal

Policy Institute to see their experiences with refugees (Kallick and Roldan 2018). They found

that refugees, generally, were a significant asset to a city’s labor force. Refugees were reported

to have lower turnover. Turnover is a significant cost to employers. Additionally, refugees were

found to be helpful in recruiting new employees. As firms develop relationships with refugees,

7

they develop relationships with the communities as well, allowing them access to the labor pool.

As labor markets have tightened, this is a significant asset for businesses struggling to find

workers. Furthermore, there are transitions that employers need to make, both with new

employees generally, but also specifically with refugees who have different needs. Working

through networks can reduce these transaction and transition costs. Some firms were also able to

find a new market for their products in these networks, further strengthening their businesses

prospects. Refugees were found to fail drug tests at rates much lower than the native born

population.

Additionally, there were some spillover effects reported. Economic theory would

indicate that by increasing the supply of labor, wages should fall. However, research indicates

that immigrants do not displace existing workers in the way an influx of domestic workers might

(Kallick and Roldan 2018). Additionally, businesses reported significant levels of patriotism and

civic pride among refugees. While these things aren’t quantified, investment, consumer

sentiment, and business sentiment are often “feeling” based indicators. Increasing local pride

could improve these too. Furthermore, positive experiences with refugees have led some

businesses to look for other “non-traditional” employees. One construction firm surveyed by

David Kallick and Cyierra Roldan reported looking to ex-offenders, having had such a good

experience with refugees. While the mental link between refugees and ex-offenders is unknown

(and perhaps questionably), it appears that the presence of refugees allowed for greater

opportunities for an additional economically-marginalized group. Furthermore, other researchers

have found that other people will move in after “urban pioneers” settle first (Wainer 2013).

Andrew Wainer also found this effect for businesses (“urban pioneer businesses”).

Indeed, some of these results were also positive indicators for the refugees themselves.

While low turnover can sometimes be an indication of labor immobility, it appeared that the low

rates of turnover were in fact indicators of job satisfaction. However, to be fully successful,

refugees also required some investment. In cities where both municipal governments and

businesses are struggling, this may be significant. Businesses voiced struggles with language

barriers (Kallick and Roldan 2018). Needless to say, the research on the role of local

governmental support of refugees is mixed.

Additionally, there has been a lot of research regarding the effects of immigrants more

broadly, both in specific cities and using panel data. Maggie Best analyzed the effects of

8

immigrants in west Buffalo (Best 2009). She found that immigration can help reduce some of

the effects of out-migration, although it’s a gradual and often politically uninspiring process.

Nevertheless, she found that immigrants helped “revitalize and reactivate gateway

neighborhoods” and helped “fill critical gaps in urban neighborhoods and key economic sectors”

(although the nature of those gaps and sectors was unspecified) (Best 2009, 19, 21). She also

found that investment in refugees produces a good return on investment of public funds.

Rebecca Karam analyzed the effects of Mosques on neighborhood indicators in Detroit

(Karam 2017). Like Best, Karam found positive results from the presence of these houses of

worship. This, in large part, stems from the distinct needs of the new residents. These new

residents, being different from existing residents, need different amenities, such as ethnic food

store and houses of worship. Effectively, this means that new, different residents increase the

return on investment of new development because they are less likely to share with existing

residents. Directly, mosques fill spaces that were previously vacant, reducing the externalities of

blight. But, they also increase neighboring property values, encouraging neighborhood

investment and increasing tax revenue (even as houses of worship don’t pay taxes themselves).

Additionally, like most other religious organizations, mosques often provide charitable services

which benefit the surrounding community. It’s interesting to note, mosques seemed to have

greater benefits than other houses of worship, such as African American churches. It appears

that there are diminishing marginal benefits to houses of worship and, thus, the marginal effect

of mosques is greater since their numbers are significantly smaller. As one downside, Karam

noted that cultural tensions also arose, which could bring conflict and reduce development down

the road.

Andrew Wainer analyzed how low-skilled immigrants helped revitalize Rust Belt

economies (Wainer 2013). Some of his key findings were that immigration seemed to be

reducing or even reversing population decline in some rust-belt cities (although this isn’t terribly

shocking, as immigrants often populated these particular cities to begin with), immigrants were

more likely to be entrepreneurs, immigrants have a disproportionate contribution to economies,

particularly in the Rust Belt, and that unauthorized immigrants would be able to contribute more

if they received authorization. Significantly, Wainer presents an “economic impact ratio”,

showing the economic impact of immigrants compared to native-born residents. The ratio was

significantly greater than one for all Rust Belt cities shown, and they had the highest ratios

9

among the sample. Detroit and Baltimore had the highest percentage of foreign-born residents;

significantly, immigrants were twice as likely to business owners, compared to native-born

residents. Wainer also notes two policy effects. First, increased immigration enforcement seems

to have had negative effects on certain immigrant areas of the city. Second, it appears that in

Baltimore, support for immigration at the city level didn’t have much of an effect on immigrants.

The first isn’t relevant for refugees, as they are legal residents, but the second does call into

question the role of municipal support. Perhaps municipal refugee funding isn’t helpful and thus

overestimates the cost of refugees.

However, while the aforementioned research looks at immigrants in select cities, they do

not account for the effects of refugees specifically nor can the data be easily generalizable. First,

unlike most migrants, refugees come with significant funds from various government sources

(Singer and Wilson 2006). The vast majority of these benefits are federal or state: refugees

qualify for, at least for some time, TANF, Medicaid, and Social Security (Harris 2016). There

are questions about the ability of refugees to access all of these benefits, but that should not

increase municipal costs. This has the potential to change the fiscal and economic impact of

refugees. If they bring in significant out-of-region funds, this can increase the economic benefits

beyond what other immigrants can provide. If they, however, add additional costs to the local

government, this can increase fiscal expenditures, putting pressures on already strained

municipalities. There is some evidence that immigrants actually provide fiscal benefits to the

Federal Government while costing state and local governments, although this likely doesn’t

apply to refugees (National Academies of Sciences, Engineering, and Medicine 2017).

Refugees, qualifying for Federal support, are likely to be more costly to the Federal government

while thus relying less on local support, relative to other, similar immigrants. The same report

also noted that first-generation immigrants contribute less tax revenue while second-generation

immigrants contribute more, compared to third generation (native born) residents. It should be

noted, additionally, that Rust Belt cities typically have lower levels of social spending per capita

than wealthier coastal cities, thus potentially providing heterogeneity in local fiscal benefits. If

they spend less per capita, then the cost of refugees would also be lower. Additionally, refugees

have legal status to work, something not guaranteed with all immigrants (Kallick and Roldan

2018). This, naturally, increases the new resident’s economic potential.

10

Second, and perhaps more important, none of this research focuses on refugees on

aggregate, either on fiscal or economic sides. This is much in line with some municipal

development fiscal impact analyses. They can see how, individually, each new development

adds to the tax base of a city and how each new resident increases municipal costs, but they fail

to take into consideration what the synergistic effect of new development is. While, indeed, each

new apartment building may add more education costs, a city would be a far less dynamic place

without apartments. Likewise, while prior research points to individual costs and benefits from

refugees, it does not fully analyze how they, as an aggregate, interact with the city and what

costs/benefits increasing acceptance numbers causes a city and its economy to incur.

11

Hypotheses and Predicted Causal Mechanisms

Refugees appear to be a good economic fit for Rust Belt cities, even if they may not be

the preferred location for the refugees. Still, what is unclear is whether these new residents bring

about the fiscal and economic changes that these cities are seeking. This thesis will help provide

answers to the question, how do changes in refugee resettlement numbers affect the labor force,

employment levels, and the unemployment rate in US Rust Belt Cities?

Given the research so far, I hypothesize the following:

H

1

:

An increase in the number of refugees will correlate with a relative improvement in local

economic indicators.

H

0

:

Null hypothesis: an increase in refugees will have a negative effect or statistically

insignificant effect on the fiscal and economic health of cities.

I hypothesize that the causal mechanism works as follows. Population decline presents a

significant challenge to cities. As a result, the mere presence of refugees as “replacement”

population will have positive effects on cities, mitigating some of the effects of population

decline. On the economic side, mitigating population decline will help mitigate the economic

contraction which usually follows. Additionally, refugees have been shown to have direct

positive impacts on companies, both as entrepreneurs and employees. I have been unable to find

any research indicating any negative economic effects from refugees. Additionally, refugees

bring in outside funding, providing external money to the local economy.

There are three caveats to note. First, refugees are typically a very small portion of a

metropolitan area’s labor pool (Kallick and Roldan 2018). As such, the results may be small or

insignificant, not because the contributions of refugees are insignificant but simply because their

contributions are crowded out by the background noise of the cities. Regardless of refugees,

Rust Belt cities are facing numerous factors which affect their vitality.

Second, support systems have been shown, in some studies, to matter significantly.

Refugees have been found to have higher wages, better integration, and less social service

dependence when they are given support early (Harris 2016). The nature of social services has

been shown to matter significantly, as it in part determines the economic future of the new

residents. Granted, the research on this subject was completed in wealthy coastal cities and thus

may not apply to the more limited job markets in Rust Belt cities. Still, my results may

12

underestimate the potential for refugees. If governments and the private sector invested a little

more (or were able to invest more), the economic benefits may be greater. This may indicate a

tradeoff between economic and fiscal benefits; however, additional research could show that

expenditures have economic benefits great enough to increase revenue, offsetting costs.

Third, the effects may not be linear but proportional to the existing population of people

from same country. Settling refugees in clusters, rather than “broadcasting” them widely has

shown to enhance integration and thus increase the benefits from newcomers (Westermeyer

2011). As such, as the population of refugees from a particular country increases, the marginal

positive effects may get larger.

13

Research Methods

I will use panel data regressions to analyze if there is a link between refugee resettlement

and local economic indicators- the unemployment rate, the level of employment, and the size of

the labor force. This will analyze if there is a causal mechanism (over time) across several

rust-belt cities.

Unfortunately, while there are lot of questions regarding refugees and cities, the data

available is limited. This isn’t entirely a bad thing. Many cities have only a handful of refugees

resettled there each year. It could be an invasion of privacy to be able to access age, marital

status, family status, economic status, etc. of groups of refugees. If only two are resettled in one

particular city, it’s not hard to find out who is who. Still, this limits the ability to do research and

control for variation. After all, the marginal labor market effect of two single 30 year-old

refugees will be very different than a family with two parents and three children under the age of

five.

Additionally, there are limits to the ability to find economic data. Refugees are assigned

to particular municipalities, but much economic data is collected over an entire Core-Based

Statistical Area. It can be difficult to find the per-capita income specifically for a city. This is

further complicated by the significant presence of commuters. Agencies need to expend

significant resources to be able to calculate data when boundaries of home and working

environments overlap in confusing ways. As such, much of this data is only available with the

ten year census- not frequently enough for this analysis.

However, one of the great advantages of working with refugee resettlement data is that it

is nearly experimental in nature. A significant problem facing most migration studies when

analyzing economic conditions is that it can be hard to tell if migrants have an effect on the

economy or if the economy determines migration patterns. Refugees, under the US resettlement

system (discussed above), do not freely choose where they are settled but are assigned to a city

via a resettlement agency. This creates a nearly natural experiment. This also has the benefit of

reducing the number of controls needed. As a result, the regressions may show a causal link.

One main caveat here is a potential causal link between acceptance of refugees and local

political conditions. Potentially, political conditions which cause cities to accept more refugees

could also cause policy-makers to make decisions which affect economic conditions. This is not

14

expected to introduce bias for two reasons. First, elections are typically every two or four years

while the data is collected annually. Thus, at least half of the variation will not be due to

elections. Also, there does not appear to be any correlation between variance and the election

cycle- variance is stable over two and four year periods. This indicates the data doesn’t show a

correlation between elections and change in refugee policy. Additionally, while political

persuasions can very between elections and cities, the variation between cities is typically greater

than the variation between elections. This is to say, a city’s political leanings don’t change very

much, compared to the overall variation of political views between cities (Einstein and Kogan

2012) (de Benedictis-Kessne and Warshaw 2016). As such, given assumptions about a

trustee-style of government, policy shouldn’t vary too greatly within a city. Within any particular city,

there may be variance from election to election, but compared to other cities, policy should be

fairly stable. Additionally, any correlation between the implementation of refugee policies and

the effects from the implementation of other policies should be minimal, given likely time

delays.

The cities are analyzed relative to their states and their Core-Based Statistical Areas

(CBSAs). CBSAs have replaced the MSA (Metropolitan Statistical Area and Metropolitan

Statistical Area) as the primary unit of analysis for cities and the surrounding economic area. Do

note that names MSA and CBSA are used interchangeably throughout this thesis (particularly in

Appendix 2- Stata Code and Output, where shorter names are helpful). For more information on

CBSAs, please see the Office of Budget and Management (2010 Standards for Delineating

Metropolitan and Micropolitan Statistical Areas; Notice 2010). Cities need to be analyzed in

these contexts for two reasons. First, while central cities are typically large, the CBSA is

significantly larger. Most American cities are not like the Hague, where farmland begins at the

municipality’s boarder. As such, the economy of the CBSA can broadly have a significant

impact on cities themselves- the cities aren’t necessarily the overwhelming driver of local

economic activity. Secondly, the CBSAs can be very heterogeneous. Detroit, for example, is

very economically depressed while some of its suburbs are very affluent. White Flight led to a

segregation of residents and, to some degree, economic activity, in CBSAs. As such,

composition effects from resettlement may affect the city and the CBSA in surprising ways.

This could be looked at in further research.

15

The following are the primary data used. All data are annual.

- Count of refugees resettled (a scalar, for cities and States)

- Unemployment rates (a rate, for cities, CBSAs, and States)

- Annual Number of employed persons (a scalar, for cities, CBSA, and States)

- Annual Number of people in the labor force (a scalar, for cities, CBSAs, and States)

Each of the three dependent variables is regressed against the main independent variable-

the count of refugees resettled. The unemployment data are measured as a rate, so they are

analyzed differently than the number of people in the labor force and the number of employed

persons. The same regression equations are uses to analyze both the labor force and the

employment level.

Unemployment Rate

The unemployment rate U is a function of structural, frictional, and cyclical factors.

(1)

𝑈

_𝑡

= 𝑈

_𝑇𝑆

+ 𝑈

_𝑇𝐹

+ 𝑈

_𝑇𝐶

Cyclical unemployment is typically considered a random shock with lingering effects, but

every city should be hit at the same year. As such, cyclical unemployment is a function of a time

dummy variable. Frictional unemployment is typically fairly constant and is a result of

economy-wide factors (unemployment insurance, labor force mobility, the national housing

market, etc.). As such, the frictional rate of unemployment can be seen through other locally

relevant unemployment rates (primarily the State unemployment rate). The structural

unemployment rate is the result of the structure of the local economy. New refugees can affect

that, as they may be different than local workers and the number of residents, all else equal, may

change the structure of the economy. Otherwise, these structural effects can also be seen in the

State and CBSA unemployment rates.

(2)

𝑈

𝑡

= 𝑈

𝑇𝑆

(𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠, 𝑆𝑡𝑎𝑡𝑒 𝑎𝑛𝑑 𝐶𝐵𝑆𝐴 𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑅𝑎𝑡𝑒𝑠)

+𝑈

_𝑇𝐹

(𝑆𝑡𝑎𝑡𝑒 𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑟𝑎𝑡𝑒) + 𝑈

_𝑇𝐶

(𝑇𝑖𝑚𝑒)

I will assume that the mechanisms which translate the constituent parts into a total

unemployment rate are linear. Additionally, I will assume that the mechanisms which translate

the parts into the constituent unemployment rates are similar across unemployment rates. As

such, we can disaggregate the constituent unemployment rates (structural, frictional, cyclical)

into:

16

(3)

𝑈

𝑡

= 𝛼 + 𝛽

1

∗ 𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠 + 𝛽

2

∗ 𝑆𝑡𝑎𝑡𝑒 𝑎𝑛𝑑 𝐶𝐵𝑆𝐴 𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑅𝑎𝑡𝑒𝑠 + 𝛽

3

𝑇𝑖𝑚𝑒

One problem here is that all indicators are rates except time, which is a dummy, and

refugees resettled, which is a level. A time tummy variable isn’t a problem as it is bounded and

the coefficient calculated will indicate how much the cyclical unemployment rate for that year

added to the natural unemployment rate for the municipality. The refugees resettled indicator,

however, is an unbounded indicator. It is statistically problematic to have an unbounded

indicator regressed against a bounded indicator (the unemployment rate can only go between 0%

and 100%). As such, in a manner consistent with the unemployment rate, a new indicator will be

created for the analysis.

(4)

As 𝑈 =

𝑃𝑒𝑜𝑝𝑙𝑒 𝑢𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑

𝐿𝑎𝑏𝑜𝑟 𝐹𝑜𝑟𝑐𝑒

→ 𝑅 =

𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠 𝑟𝑒𝑠𝑒𝑡𝑡𝑙𝑒𝑑 𝐿𝑎𝑏𝑜𝑟 𝐹𝑜𝑟𝑐𝑒

Additionally, we are working with panel data using a random effects model (there is no

reason to believe that the effect is the same among heterogeneous cities and therefore no reason

to use a fixed effects model). As such, the constant will be different for each city (i), reflecting a

potential different natural rate of unemployment, relative to the surrounding areas (after all, the

CBSA unemployment rate should account for structural effects). As such, we have a final

equation of:

(5)

𝑈

_𝑖𝑡

= 𝛼 + 𝛽

₁

∗ 𝑅

_𝑖𝑡

+ 𝛽

₂

∗ 𝑈

_{𝐶𝐵𝑆𝐴(𝑖𝑡)}

+ 𝛽

₃

∗ 𝑈

_{𝑆𝑡𝑎𝑡𝑒(𝑖𝑡)}

+ 𝛽

₄

𝑇𝑖𝑚𝑒

_𝑡

+ 𝛽

₅

∗ 𝐶𝑖𝑡𝑦

_𝑖

+ 𝜀

The equation, as with all following ones, will be analyzed using a panel, random effects

model.

Employment and Labor Force

The equations for the labor force and the employment levels are derived identically.

They are both scalars which analyze highly related aspects of the labor market. As such, I will

not duplicate the derivations. Everything under this section (“Employment and Labor Force”)

applies to both the labor force and employment levels. The derivation will be done for the

employment, but in all cases below, until the “Data Preparation” section, the “employment” can

be replaced with “labor force”.

The employment level E in one period is the function of the employment in the last

period, plus some change. This change is partly the result of a trend (whether a city is growing

17

or not). This is also the result of other changing factors- the local economy, state finances, etc.

This is also the result of known new entrants to the area- newly resettled refugees.

(6)

𝐸 = 𝐸

𝑡−1

+ 𝑇𝑟𝑒𝑛𝑑 + 𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠 + 𝐿𝑜𝑐𝑎𝑙 𝑎𝑛𝑑 𝑟𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝑓𝑎𝑐𝑡𝑜𝑟𝑠

Operationalizing this with the data available, we get the following. The trend is the

constant. CBSA and STATE employment indicators are proxies for things which cause people

to enter or leave employment- if there is a state factor which would cause migration, it should

show up in the state employment level. Likewise with the CBSA employment, which could

capture elements of the local economy. Time is added for annual shocks which may affect

migration (perhaps people move less during a recession, etc.):

(7)

𝐸 = 𝐸

𝑡−1

+ 𝛼 + 𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠 + 𝐶𝐵𝑆𝐴 𝑎𝑛𝑑 𝑆𝑡𝑎𝑡𝑒 𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 + 𝑌𝑒𝑎𝑟 𝑑𝑢𝑚𝑚𝑦

Unfortunately, this equation is problematic. The CBSA and State employment levels

cannot be a component of the local employment level- the local level is a component of the

CBSA and State levels! Furthermore, to have a constant attached to those levels implies that a

municipality has a fixed proportion of the State and CBSA’s employment level. This is

obviously not true, given how part of the problem facing Rust Belt cities was “White Flight” to

the suburbs. Therefore, in levels this cannot work. Therefore, I propose three regressions to test

the relationship between the number of refugees resettled and the number of people employed in

a city.

Equation 1

If the problem is that the State and CBSA data cannot be components of a city’s

employment level, then they can be left out. While adding explanatory variables increases the

explanatory potential of the equation, the State and CBSA employment data should not be

correlated with the number of refugees resettled (and correlation analyses with the data do

support this claim). Thus, we can use:

(8)

𝐸

_𝑇

= 𝐸

_𝑡−1

+ 𝛼 + 𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠 + 𝑌𝑒𝑎𝑟 𝑑𝑢𝑚𝑚𝑦

This is panel data, so city i must also be accounted for as a dummy variable. In this case,

the city dummy captures the individual trend for each city while the constant captures the general

employment growth of the collective of the Rust Belt cities analyzed.

18

Equation 2

The initial equation can also be done in log differences. Logs are important because

populations usually don’t grow linearly. Thus, logged differences give an approximate rate of

change. This means that the change in employment is explained by the change in employment of

the past period, the trend of employment, the change in the new number of refugees, some

annual factors (economy, national politics, etc.), and the changes in State and CBSA

employment. Technically, State and CBSA employment data here are more instruments than

independent variables. However, given that the model already uses panel time series data, it is

best now not to further complicate it with instrumental variables. That can be done through

further research. We thus get this as a final equation, including the city dummy. Note that, in

line with Stata coding, “d.” indicates a first difference.

(10)

𝑑. ln(𝐸

_𝑖𝑡

) = 𝛼 + 𝛽

₁

∗ 𝑑. ln(𝐸

_𝑖𝑡−1

) + 𝛽

₂

∗ 𝑑. ln(𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠

_𝑖𝑡

) + 𝛽

₃

∗ 𝑑. ln(𝐸

_𝑖𝑡𝑆𝑡𝑎𝑡𝑒

) + 𝛽

₄

∗

𝑑. ln(𝐸

_𝑖𝑡𝐶𝐵𝑆𝐴

) + 𝛽

₅

∗ 𝑑. ln(𝑇𝑖𝑚𝑒

_𝑡

) + 𝛽

₆

∗ 𝑑. ln(𝐶𝑖𝑡𝑦

_𝑖

) + 𝜀

Equation 3

A third option is to take the assumption rejected from the original equation: that the city

is, all else equal, a fixed proportion of the state or CBSA’s employed population. This could be

true in the short-run. A city’s employment level, relative to the state, may change relatively little

with respect to short-run fluctuations. If this is true, however, then there is no theoretical reason

to including a lagged dependent variable. If the relationship between the state/ CBSA and the

city is fixed across years, then adding a lagged dependent variable introduces a paradox. To use

fictional numbers, it could imply that the employment level today is 5% of the state’s working

population plus 90% of last period’s employed workers. This would count workers twice. There

is a large groups of workers present in both the lagged dependent variable and the percentage of

the State’s employment numbers. Most of the people in the %5 today were in the 90% in the

prior period (those who didn’t move). Therefore, we cannot use the lagged dependent variable.

Also, this means we cannot regress the employment level against the State’s employment level

and the CBSA’s employment level at the same time (since the CBSA is also in the State).

Therefore, equation three actually becomes 3.1 and 3.2. Therefore, we get, with panel variables

included:

19

(11)

𝐸

𝑖𝑡

= 𝛼 + 𝛽

1

∗ 𝐸

𝑖𝑡𝐶𝐵𝑆𝐴

+ 𝛽

2

∗ 𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠

𝑖𝑡

+ 𝛽

3

∗ 𝑌𝑒𝑎𝑟

𝑡

+ 𝛽

4

𝐶𝑖𝑡𝑦

𝑖

+ 𝜀

(12)

𝐸

_𝑖𝑡

= 𝛼 + 𝛽

₁

∗ 𝐸

_𝑖𝑡𝑆𝑡𝑎𝑡𝑒

+ 𝛽

₂

∗ 𝑅𝑒𝑓𝑢𝑔𝑒𝑒𝑠

_𝑖𝑡

+ 𝛽

₃

∗ 𝑌𝑒𝑎𝑟

_𝑡

+ 𝛽

₄

𝐶𝑖𝑡𝑦

_𝑖

+ 𝜀

For reader reference, the primary equations for this research then will be:

- 5- Unemployment

- 9, 10, 11, 12- Employment and Labor Force

Additional Significant Regressions

In addition to the regressions above, I also added lags of the number of refugees resettled.

This was for two reasons. First, refugees are resettled over the course of a year. Therefore, one

may not see an affect from the “total refugees resettled” during the same time period as the other

data were collected because the refugees are not having an effect over the whole time period.

Additionally, there may be some “settling” in. It may take them some time to enter into the labor

force, start consuming products, and affect the local economy.

I also did other, additional regressions not discussed above. They were primarily for

statistical curiosities or to test alternative methods. They were not used in the final results, but

can be found in Appendix 2- Stata Code and Output.

20

Data Preparation

Data preparation proved to be surprisingly time consuming. Data collection across

agencies isn’t standardized, so a significant amount of time was consumed linking data from

various sources, along with selecting appropriate cities. I automated as much of the data linking

as possible to reduce the risk of human error. The majority of the coding for that “automation”

can be found in Appendix 1, along with descriptions of the functioning of the coding.

Refugee resettlement data was drawn from the Refugee Reprocessing Center, a non-profit

which focuses on data on refugee resettlement (Refugee Reprocessing Center 2018). The

datasets included data on country of origin, destination state, destination city, and numbers from

any one origin to any particular state and city. Two datasets were downloaded- one for the fiscal

year and one for the calendar year. Data was for the fiscal year included years 2003-2018; the

calendar year data, naturally, only went through 2017. Data for the fiscal year could have been

used for analysis of the fiscal effects of refugees. However, the economic data follows the

calendar year and as such, the analysis was eventually completed using only the calendar year

dataset.

Given the size of the dataset and the computing power needed to do calculations,

appropriate cities were selected first to minimize computations needed later in joining different

data sources. In the initial dataset, there were over 1,500 city names (not counting cities whose

names are shared with cites in other states (ex. Springfield, IL and Springfield, IN)) and over

160,000 rows of data. First, cities were selected on states typically considered to be in the Rust

Belt. As discussed in the introduction, the definition of Rust Belt is neither consistent nor purely

scientific. I used seventeen states frequently cited as holding Rust Belt cities: Connecticut,

Delaware, Illinois, Indiana, Maine, Maryland, Massachusetts, Michigan, Missouri, New

Hampshire, New Jersey, New York, Ohio, Pennsylvania, Rhode Island, Vermont, and West

Virginia.

Then, I needed to calculate the total number of refugees settling in any particular city.

The data sets did not include a total to each city- only a total to each city from any one particular

country. Cities were selected based on the number of refugees resettled there. For the data to

produce meaningful time-series results, there needs to be variation in the time-series. Many

cities had no new refugees for a majority of the years. Thus, cities were discarded if they had no

21

new refugees for a majority of the years available. After selecting for states and refugee

resettlement numbers, the number of unique city names was reduced from 1531 to 298. Again,

do note that this is the number of unique names, not unique cities. Springfield Illinois and

Springfield Missouri are only counted as one unique city name (it appears that some city

founders were not terribly creative when selecting names).

Then CBSA identifier data was joined and central CBSA cities were selected. Note,

identifier data is data indicating which CBSA a city was in. CBSA identifier data was pulled

from the Census Bureau (The Census Bureau 2018). For more information on CBSAs, please

refer to the Office of Budget and Management (2010 Standards for Delineating Metropolitan and

Micropolitan Statistical Areas; Notice 2010).

First, cities were selected for being a central city in their CBSA. This was done as part of

the data joining process. This is necessary because many of the cities which received refugees

are suburbs of other, more prominent cities in the same CBSA. While there are surely interesting

effects from refugee resettlement to suburbs, that is outside the scope of this thesis. Suburbs are

frequently very different, in many ways, from their central cities and this would certainly affect

the applicability of the research to the target of Rust Belt cities. Thus, cities were discarded if

they were not central to the CBSA. Because cities were looked-up based on whether their names

were listed among the central cities, any city which couldn’t be joined (because it wasn’t in the

CBSA name) was discarded.

This is more complicated than it seems at face-value. There is no unique identifier which

ties the city present in the refugee resettlement data to CBSA data. CBSAs are identified by

clusters of cities and states (such as Chicago-Naperville-Elgin, IL-IN-WI). It should be noted,

there is no necessary connection between the city names and the states (in the prior example, the

CBSA extends from Illinois into Indiana and Wisconsin, but all of the principle cities are in

Illinois). Additionally, city names are often repeated in different states. Therefore, I had to use a

lookup function to search for the city and state in the refugee spreadsheet within the up to six

different columns of city and states within the CBSA table. See Appendix 1 for more details

about the lookup function. To facilitate easier reference in the future, a unique CBSA number

was added, along with the CBSA name. This allowed faster joining of data in the future. Both

the name and number were required because not all sources use the same names for CBSAs.

22

Unfortunately, if the names aren’t spelled exactly the same, lookup functions can return some

undesirable results.

During this process, several smaller, more trivial variables also had to be modified. This

includes things like state names- some use the name, others use the state code (i.e. Illinois vs.

IL). This doesn’t change the meaning of the data nor does it add any value to it. Still, it’s

necessary for the lookup function to work property. Furthermore, it runs the risk of introducing

errors and provides rational for standardizing data entry practices across Federal agencies.

Once the CBSA indicators were added to the refugee data and principle cities were

selected, State and CBSA economic data was added. State, CBSA, and municipal

unemployment rate, labor force, employment, and unemployment numbers were taken from the

Bureau of Labor Statistics (Bureau of Labor Statistics n.d.). State data was joined from 17

different spreadsheets from Bureau of Labor Statistics (one for each state). CBSA data was then

joined (76 different spreadsheets). The state data was easily ordered- the states were

downloaded in alphabetical order, despite the file names having no connection to the state

names. CBSA and municipal data, however, were not easily sortable and thus was joined

manually.

There is additional data available and additional data was prepared that may of interest to

other researchers. The Census Bureau does an Annual Survey of State and Local Government

Finances which provides a wealth of knowledge on the finances of local governing bodies. The

data is highly coded and downloadable yearly. For more information, please see the Census

Bureau (The Census Bureau n.d.).

23

Results

For reference, all of the Stata code and the full output were put into appendix 2. Below, I

will discuss the main results from the main equations and some interesting additional results that

were additionally found. There are many additional tests performed and interesting detours that

were taken. The Stata code does include notes, for interested readers. Below are the results

pertinent to the main equations. The tests and detours are only included here if they provide

substantially different results from the initial equations.

Unemployment Rate

𝑈

𝑖𝑡

= 𝛼 + 𝛽

1

∗ 𝑅

𝑖𝑡

+ 𝛽

2

∗ 𝑈

𝐶𝐵𝑆𝐴(𝑖𝑡)

+ 𝛽

3

∗ 𝑈

𝑆𝑡𝑎𝑡𝑒(𝑖𝑡)

+ 𝛽

4

𝑇𝑖𝑚𝑒

𝑡

+ 𝛽

5

∗ 𝐶𝑖𝑡𝑦

𝑖

+ 𝜀

The primary equation analyzing the relationship between the unemployment rate and the

incoming refugees as a proportion of the labor force finds no significant correlation between

refugees and the unemployment rate. Note, in many following tables, the dummies for

individual cities have been removed for visual clarity. To see the full results, please see

Appendix 2.

When lagged resettlement indicators are included, refugees as a proportion of the labor

force remain insignificant. Interestingly, when the non-lagged refugee indicator is removed from

the regression, the significance of the lagged indicator goes down. Now, they are both

insignificant in both circumstances, so this provides no evidence for anything, but it is an

interesting statistical phenomenon.

There are a few interesting things to be found in the other coefficients. While they don’t

provide any evidence on the relationship between refugees and unemployment, they do provide

some insight into the functioning of the local economies and to the strength of the model. First,

the constant is positive, significant, and around 3. The constant can be considered the “baseline”

level of unemployment- the amount of unemployment that exists independent of state and local

factors. This is lower than the typically used natural rate of unemployment, but this isn’t

surprising, given that this analysis looks at the states with some of the highest levels of

unemployment. If the states with the highest unemployment are removed, then the national

natural rate is likely to be lower.

24

Interestingly, with the exception of a small number of cities, most of the city coefficients

are significantly negative while the CBSA rate is significantly positive. This is unusual as the

city unemployment rate is typically higher than the CBSA unemployment rate. While one would

imagine that, if roughly half of the cities had positive coefficients and half had negative

coefficients, this would be an indication of where cities fell with respect to the average.

However, the vast majority of cities are negative. This could indicate that there is a key

explanatory variable of the unemployment rate which is missing- one that greatly raises the

unemployment rate, ceteris paribus, for most cities. However, the R

2

is very high, reducing the

likelihood of a significant missing explanatory variable. This isn’t significant for refugee

analysis, however, as this variables likely isn’t correlated with accepted refugees.

It is also interesting that the state unemployment rate is insignificant. This may be

because of imbalances in the data- some states had many cities (like Illinois or Pennsylvania)

while other states only had one (like Rhode Island). This could also be due to boarder effects.

CBSAs do indeed cross state boundaries. Cities near boarders may be just as strongly affected

by the economies of neighboring states as their own. Thus, the CBSA becomes significant while

the State isn’t. This can also question whether the state unemployment rate can effectively

capture the frictional unemployment rate. Recall, the frictional unemployment rate was

primarily captured by the state rate. This may not be important, however, as the frictional

unemployment rate may stable across states and across time, and thus simply captured in the

constant, effectively making it part of the natural unemployment rate.

Finally, the recession is visible among the year dummies. This is as expected, both in

terms of sign and timing. What is curious is that the coefficients are both fairly small and not

always significant. One would expect the 2009 financial crisis to have a stronger effect on the

unemployment rate, given how it roughly doubled the unemployment rate nationally.

25

Initial unemployment equation:

Random-effects GLS regression Number of obs = 1065 Group variable: City Number of groups = 71 R-sq: within = 0.9397 Obs per group: min = 15 between = 1.0000 avg = 15.0 overall = 0.9684 max = 15 Wald chi2(17) = . corr(u_i, X) = 0 (assumed) Prob > chi2 = .

(Std. Err. adjusted for 71 clusters in City) --- --- | Robust

City Unemployment Rate| Coef. Std. Err. z P>|z| [95% Conf. Interval] ---+--- Refugees/Labor Force | 12.61152 7.037721 1.79 0.073 -1.182161 26.4052 CBSA Unemployment Rate| 1.089847 .0831042 13.11 0.000 .9269659 1.252728 State Unemp. Rate| -.0190412 .0893361 -0.21 0.831 -.1941368 .1560544 | Year | 2004 | -.0614079 .0444125 -1.38 0.167 -.1484548 .025639 2005 | -.2495177 .1065337 -2.34 0.019 -.4583199 -.0407155 2006 | -.2386581 .1069133 -2.23 0.026 -.4482044 -.0291118 2007 | -.2933896 .1252203 -2.34 0.019 -.5388169 -.0479623 2008 | -.2264807 .1087084 -2.08 0.037 -.4395452 -.0134161 2009 | -.1022672 .2484772 -0.41 0.681 -.5892736 .3847392 2010 | .0653879 .2704887 0.24 0.809 -.4647602 .595536 2011 | .1745302 .2183659 0.80 0.424 -.2534591 .6025195 2012 | .129917 .2012041 0.65 0.518 -.2644359 .5242699 2013 | .0485599 .1683229 0.29 0.773 -.2813469 .3784668 2014 | -.1015601 .1190734 -0.85 0.394 -.3349397 .1318194 2015 | -.2393017 .1141313 -2.10 0.036 -.462995 -.0156085 2016 | -.3264804 .1402296 -2.33 0.020 -.6013254 -.0516353 2017 | -.2490333 .1437016 -1.73 0.083 -.5306833 .0326166 | City | CT,Hartford | 1.567089 .0391203 40.06 0.000 1.490415 1.643764 CT,New Haven | -1.146375 .0333852 -34.34 0.000 -1.211808 -1.080941 CT,Stamford | -3.944393 .0049893 -790.58 0.000 -3.954171 -3.934614 DE,Dover | -2.327769 .0583697 -39.88 0.000 -2.442172 -2.213366 IL,Bloomington | -3.268543 .1185879 -27.56 0.000 -3.500972 -3.036115 IL,Freeport | -2.396403 .0925028 -25.91 0.000 -2.577705 -2.215101 IL,Moline | -2.984067 .0854015 -34.94 0.000 -3.151451 -2.816683 IL,Rock Island | -2.316363 .0852482 -27.17 0.000 -2.483447 -2.14928 IL,Rockford | -2.35148 .1944331 -12.09 0.000 -2.732562 -1.970398 IL,Springfield | -3.015869 .0917377 -32.87 0.000 -3.195671 -2.836066 IN,Bloomington | -3.468822 .0348639 -99.50 0.000 -3.537154 -3.400491 IN,Carmel | -5.350871 .0175613 -304.70 0.000 -5.385291 -5.316452 IN,Columbus | -3.73173 .3023959 -12.34 0.000 -4.324415 -3.139045 IN,Fort Wayne | -2.828477 .0275482 -102.67 0.000 -2.88247 -2.774483 IN,Goshen | -3.415213 .0761396 -44.85 0.000 -3.564443 -3.265982 IN,Indianapolis | -2.635164 .020217 -130.34 0.000 -2.674789 -2.595539 IN,Mishawaka | -3.413372 .0690803 -49.41 0.000 -3.548766 -3.277977 IN,Muncie | -2.917282 .1054416 -27.67 0.000 -3.123944 -2.710621 IN,South Bend | -2.134575 .0690002 -30.94 0.000 -2.269813 -1.999337 MA,Springfield | -.5773397 .0746537 -7.73 0.000 -.7236582 -.4310211 MA,Worcester | -2.785748 .0578613 -48.15 0.000 -2.899154 -2.672342 MD,Baltimore | -.8195626 .0708661 -11.56 0.000 -.9584576 -.6806677 MD,Hagerstown | -1.797446 .082185 -21.87 0.000 -1.958526 -1.636367 ME,Auburn | -3.528929 .0447106 -78.93 0.000 -3.61656 -3.441298 ME,Lewiston | -2.957837 .0447754 -66.06 0.000 -3.045595 -2.870079 ME,Portland | -3.288938 .1058348 -31.08 0.000 -3.496371 -3.081506 MI,Battle Creek | -2.225125 .1317232 -16.89 0.000 -2.483297 -1.966952 MI,Dearborn | -5.66343 .1714798 -33.03 0.000 -5.999525 -5.327336 MI,Detroit | 4.010362 .1717399 23.35 0.000 3.673758 4.346966 MI,Grand Rapids | -1.203485 .135798 -8.86 0.000 -1.469644 -.9373254 MI,Holland | -2.65889 .12399 -21.44 0.000 -2.901906 -2.415874 MI,Lansing | -.463482 .1418976 -3.27 0.001 -.7415962 -.1853678

26

MI,Warren | -2.162643 .1714181 -12.62 0.000 -2.498616 -1.82667 MI,Wyoming | -2.984239 .1317804 -22.65 0.000 -3.242524 -2.725955 MO,Kansas City | -2.134534 .007428 -287.36 0.000 -2.149092 -2.119975 MO,Saint Louis | -1.68812 .0412887 -40.89 0.000 -1.769044 -1.607196 NH,Concord | -3.164503 .1762799 -17.95 0.000 -3.510005 -2.819001 NH,Manchester | -2.762119 .147298 -18.75 0.000 -3.050818 -2.47342 NH,Nashua | -2.550878 .1466712 -17.39 0.000 -2.838349 -2.263408 NJ,Atlantic City | -.2004402 .2210932 -0.91 0.365 -.6337749 .2328945 NJ,Camden | 4.169994 .0205355 203.06 0.000 4.129746 4.210243 NJ,Trenton | 1.007899 .0389258 25.89 0.000 .9316061 1.084192 NJ,Vineland | -3.865885 .2622681 -14.74 0.000 -4.379921 -3.351849 NY,Albany | -2.519961 .0707884 -35.60 0.000 -2.658704 -2.381218 NY,Binghamton | -2.721103 .02846 -95.61 0.000 -2.776884 -2.665322 NY,Buffalo | -1.886887 .0641425 -29.42 0.000 -2.012604 -1.76117 NY,Rochester | -1.349648 .0340821 -39.60 0.000 -1.416448 -1.282849 NY,Schenectady | -1.751511 .0633239 -27.66 0.000 -1.875624 -1.627398 NY,Syracuse | -2.392088 .0820339 -29.16 0.000 -2.552872 -2.231305 NY,Troy | -1.719746 .0962233 -17.87 0.000 -1.90834 -1.531152 NY,Utica | -2.610578 .1047459 -24.92 0.000 -2.815876 -2.40528 OH,Akron | -2.534913 .0533277 -47.53 0.000 -2.639433 -2.430392 OH,Canton | -2.183421 .0858934 -25.42 0.000 -2.351769 -2.015073 OH,Cincinnati | -2.735337 .033256 -82.25 0.000 -2.800518 -2.670156 OH,Cleveland | -1.405535 .036149 -38.88 0.000 -1.476385 -1.334684 OH,Dayton | -1.889008 .070612 -26.75 0.000 -2.027405 -1.750611 OH,Springfield | -2.963504 .0877565 -33.77 0.000 -3.135503 -2.791504 OH,Toledo | -2.440038 .1067102 -22.87 0.000 -2.649186 -2.23089 OH,Warren | -1.823562 .1423146 -12.81 0.000 -2.102494 -1.544631 PA,Allentown | -.9436873 .039111 -24.13 0.000 -1.020343 -.8670312 PA,Erie | -2.783949 .0864495 -32.20 0.000 -2.953387 -2.614512 PA,Harrisburg | -.397882 .0722342 -5.51 0.000 -.5394583 -.2563056 PA,Lancaster | -.8386402 .1101593 -7.61 0.000 -1.054548 -.6227318 PA,Pittsburgh | -3.325758 .0157034 -211.79 0.000 -3.356536 -3.29498 PA,Scranton | -3.385428 .0954013 -35.49 0.000 -3.572411 -3.198445 PA,Wilkes-Barre | -2.336324 .0957392 -24.40 0.000 -2.523969 -2.148678 PA,York | .9012974 .0277839 32.44 0.000 .846842 .9557528 RI,Providence | -2.184305 .0926206 -23.58 0.000 -2.365838 -2.002772 VT,Burlington | -3.232881 .161899 -19.97 0.000 -3.550198 -2.915565 WV,Charleston | -3.635403 .0151522 -239.93 0.000 -3.665101 -3.605705 Constant | 2.991793 .4366312 6.85 0.000 2.136012 3.847575 ---+--- sigma_u | 0 sigma_e | .56240992

rho | 0 (fraction of variance due to u_i)

27

Initial unemployment equation with lagged refugee indicator (cities suppressed):

Random-effects GLS regression Number of obs = 994 Group variable: City Number of groups = 71 R-sq: within = 0.9462 Obs per group: min = 14 between = 1.0000 avg = 14.0 overall = 0.9712 max = 14 Wald chi2(17) = . corr(u_i, X) = 0 (assumed) Prob > chi2 = .

(Std. Err. adjusted for 71 clusters in City) --- | Robust

City Unemployment Rate| Coef. Std. Err. z P>|z| [95% Conf. Interval] ---+--- Refugees/Labor Force |

--. | 5.66437 4.259324 1.33 0.184 -2.683751 14.01249 L1. | 13.23777 7.114203 1.86 0.063 -.7058146 27.18135 |

CBSA Unemployment Rate| 1.069682 .0841072 12.72 0.000 .9048347 1.234529 State Unemp. Rate| .0195549 .0929135 0.21 0.833 -.1625523 .2016621 | year | 2005 | -.200318 .079006 -2.54 0.011 -.355167 -.0454691 2006 | -.1788229 .0850651 -2.10 0.036 -.3455474 -.0120984 2007 | -.2317344 .0950773 -2.44 0.015 -.4180824 -.0453863 2008 | -.1787427 .0937816 -1.91 0.057 -.3625513 .0050658 2009 | -.1243112 .2674095 -0.46 0.642 -.6484242 .3998017 2010 | .0242011 .289686 0.08 0.933 -.543573 .5919752 2011 | .1518596 .2365022 0.64 0.521 -.3116762 .6153954 2012 | .1299393 .2140392 0.61 0.544 -.2895699 .5494484 2013 | .0455897 .1794884 0.25 0.799 -.3062011 .3973805 2014 | -.0727036 .1128966 -0.64 0.520 -.2939768 .1485697 2015 | -.2030906 .0966791 -2.10 0.036 -.3925782 -.013603 2016 | -.2632216 .1136974 -2.32 0.021 -.4860643 -.0403789 2017 | -.2346484 .1250963 -1.88 0.061 -.4798325 .0105358 | Constant | 2.80934 .4196327 6.69 0.000 1.986875 3.631805 ---+--- sigma_u | 0 sigma_e | .54635193

rho | 0 (fraction of variance due to u_i)