University of Southern California Dissertations and Theses

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

(USC Thesis Other)

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content PREPARING FOR IMMIGRATION REFORM:

A SPATIAL ANALYSIS OF UNAUTHORIZED IMMIGRANTS

by

Anna Jane Fischer

A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

December 2014

Copyright 2014 Anna Jane Fischer

ii
DEDICATION
I would like to dedicate this document to my parents and brother for always encouraging me to
pursue my academic goals and to Evan Colby, for putting up with my lack of availability on
seemingly endless weekends. Thank you, Evan, for being forever helpful and making the day-to-day
a little bit easier so that I could focus on accomplishing this goal.

iii
ACKNOWLEDGMENTS
I would like to acknowledge my thesis advisor and mentor, Dr. Karen Kemp, for guiding me through
the thesis process. Her feedback as well as her words of encouragement were a major guiding force
through the process of research and writing. I would also like to acknowledge Dr. Robert Vos for his
direction in the initial formulation of my thesis topic. Lastly, I would like to acknowledge Professor
Roberto Suro, Director of the Tomas Rivera Policy Institute, who has taught me much of what I
know about immigration policy and who was one of the initial sources of motivation for conducting
research on this topic. Thank you also to my family and friends, without whom I could not have
made it this far.

iv
TABLE OF CONTENTS
DEDICATION ii
ACKNOWLEDGMENTS iii
LIST OF TABLES vii
LIST OF FIGURES ix
LIST OF ABBREVIATIONS x
ABSTRACT xii
CHAPTER 1: INTRODUCTION 1
1.1 Approaches to Immigration Reform 1
1.2 Issues Addressed in Immigration Reform Legislation 2
1.3 Immigrant Processing Requirements 3
1.4 Research Objectives 5
1.5 Thesis Structure 5
CHAPTER 2: BACKGROUND 7
2.1 Methods for Estimating the Unauthorized Population 7
2.1.1 Residual Method for National and State Estimates 7
2.1.2 Residual Method Combined With Other Methods for Sub-state Estimates 9
2.1.3 Challenges and Weaknesses of Existing Estimation Methods 10
2.2 Results of Prior Research and Analysis 11
2.2.1 Estimates of the Unauthorized 11
2.2.2 Characteristics of the Unauthorized Population 12
2.3 Immigrant Settlement Patterns in the United States 15

v
CHAPTER 3: METHODOLOGY 17
3.1 Overview of Analysis Steps 18
3.2 Define Variables 20
3.2.1 The Dependent Variable 20
3.2.2 Corresponding Demographic Variables 24
3.3 State Level Analysis: Define Relationship Between Dependent and Independent Variables 26
3.3.1 Principal Components Analysis (PCA) 26
3.3.2 Exploratory Regression Analysis 33
3.3.3 Ordinary Least Squares 36
3.3.4 Geographically Weighted Regression 39
3.4 Census Tract Level Analysis 43
3.4.1 Calculating a Unique Component Score for each Census Tract 44
CHAPTER 4: RESULTS 48
4.1 Relative Densities and Distribution 48
4.2 Model Performance and Verification of Results 52
4.2.1 Comparison with State Level Estimates Generated by Warren and Warren 52
4.2.2 Comparing Results to Independent Sub-state Estimates for California 56
CHAPTER 5: CONCLUSIONS 60
5.1 Weaknesses, Challenges, Limitations and Next Steps 60
5.1.1 Missing Data and Data Uncertainty 60
5.1.2 Ecological Fallacy 61
5.1.3 Refinement of Independent Variables 61
5.1.4 Improved Method for Verifying the Results 62
5.1.5 Sensitivity and Reliability Analysis 63
5.1.6 Refine Display of Results 63

vi
5.2 Lessons Learned and Potential Impacts 64
REFERENCES 65
APPENDICES 71

vii
LIST OF TABLES

Residual Method: Common Data Sources 8
Table 1

Spearman’s Rank-order Correlation Between the Unauthorized and Foreign-born Table 2
Population by State 22

Demographic Variables Considered for Inclusion in the Analysis 25
Table 3

KMO Measures for Demographic Variables 28
Table 4

Eigenvalue-one Criterion: Total Variance Explained by Initial PCA 29
Table 5

KMO and Bartlett's Test 30
Table 6

Eigenvalue-one Criterion: Total Variance Explained by Final PCA 31
Table 7

Component Score Coefficient Matrix 32
Table 8

Independent Variables Included in the Exploratory Analysis 35
Table 9

Passing Model Variables and Direction 36
Table 10

Statistics of Passing Model 36
Table 11

Retained Variables in OLS Regression 37
Table 12

OLS Regression Results 38
Table 13

GWR Results 40
Table 14

Absolute % Difference Between Estimates 54
Table 15

Absolute % Difference Between Estimates of the Unauthorized by Region in CA 57
Table 16

Estimates of the Total Unauthorized Population in CA 58
Table 17

viii

Differences in the Distribution of Unauthorized Population by Region in CA 58
Table 18

Correlation Matrix: First PCA 71
Table 19

Anti-image Correlation, Final PCA 72
Table 20

Correlation Matrix: Final PCA 73
Table 21

Reproduced Correlations and Residuals: Final PCA 74
Table 22

Percent of Census Tracts with Missing Variables by State 76
Table 23

ix
LIST OF FIGURES

Scatterplot of the Unauthorized by the Total Foreign-born Population by State 22
Figure 1

Observed Dependent Variable: % Unauthorized out of Total Foreign-born Population Figure 2
(2006-2010) 24

Scree Plot 29
Figure 3

Component Scores by State 32
Figure 4

OLS Standard Residuals 39
Figure 5

GWR Standard Residuals 41
Figure 6

Strength of Independent Variable Coefficients as Predictors of the % of the Figure 7
Unauthorized Population 42

Unauthorized Population by Census Tract in the United States 50
Figure 8

Unauthorized Population by Census Tract in California 50
Figure 9

Unauthorized Population by Census Tract in Los Angeles County 51
Figure 10

Unauthorized Population by Census Tract in Los Angeles 52
Figure 11

Absolute % Difference from Warren Estimates 56
Figure 12

x
LIST OF ABBREVIATIONS
ACS American Community Survey
AICc Akaike's Information Criterion
CBO Congressional Budget Office
CIR Comprehensive Immigration Reform
CSII Center for the Study of Immigrant Integration (CSII)
CPS Current Population Survey
GDP Gross domestic product
GNP Gross national product
DACA Deferred Action for Childhood Arrivals
DHS Department of Homeland Security
DOJ Department of Justice (U.S.)
GWR Geographically Weighted Regression
INS Immigration and Naturalization Service
ITIN Individual Taxpayer Identification Number
KMO Kaiser-Meyer-Olkin
LAC-MILSS Los Angeles County Mexican Immigrant Legal Status survey
LPR Legal Permanent Resident
MPI Migration Policy Institute
OIS Office of Immigration Statistics
OLS Ordinary Least Squares
PCA Principal Component Analysis
PPIC Public Policy Institute of California

xi
PUMA Public Use Microdata Areas
RPI Registered Provisional Immigrant
TRPI Tomas Rivera Policy Institute
USC University of Southern California
USCIS United States Citizen and Immigration Services
VIF Variance Inflation Factor

xii
ABSTRACT
An estimated 11.7 million unauthorized immigrants resided in the United States in 2012
according to the Pew Hispanic Center (Passel, Cohn, and Gonzalez-Barrera 2013). Reforming
the U.S. immigration system is a clear policy priority for President Barack Obama, and an
agenda item for the 113
th
Congress (U.S. Congressional Research Service 2013). Based on prior
legislation, processing of immigrants for legalization is likely to be a complex and time
consuming task, necessitating the involvement of nonprofit and public infrastructure. The goal of
this study was to design a research methodology for estimating the unauthorized population at
the census tract level, as a means for visually representing the relative densities of the
unauthorized population in a way that would be useful for planning where to provide services for
the unauthorized populations within a community. Using statistical methods, the relationships
between the dependent and independent variables was defined at the state level. The state level
relationships were then applied to census tract level data in order to make census tract estimates.
The results of the analysis were displayed as relative densities using the dot density renderer in
ArcGIS Desktop. The performance of this model was verified by comparing the results generated
in this study to those of other studies. Based on this verification method, the performance of the
model varied by geography, with the western states, in particular, California seeming to have
performed the best. The states that appear to have performed the worst are primarily located in
northeastern United States and include six out of the eight states with the lowest number of
unauthorized persons (<3,000). Within California, between a 0.02 (Orange County) and 3.4 (Bay
Area) percentage point difference was found when comparing the regional distribution estimated
in this study with those of other studies.

1
CHAPTER 1: INTRODUCTION
An estimated 11.7 million unauthorized immigrants resided in the United States in 2012
according to the Pew Hispanic Center (Passel, Cohn, and Gonzalez-Barrera 2013). Reforming
the federal immigration system of the United States, a stated second-term policy priority for
President Barack Obama, and a clear agenda item for the 113
th
Congress, has garnered a great
deal of attention from all sides of the political spectrum (U.S. Congressional Research Service
2013). The 113
th
Congress has been marked by heated bipartisan debate around proposed
immigration related legislation. In the following section, three approaches to immigration reform
are introduced with examples describing how these approaches have played out over Obama’s
presidency. This chapter continues by outlining the research objectives, making a case for
utilizing spatial analysis methods in planning for immigration reform, and concludes with an
outline of the thesis structure
1.1 Approaches to Immigration Reform
Three leading approaches to reform have presented themselves during Obama’s presidency and
the 113
th
Congress, including: (1) comprehensive immigration reform (CIR), where wide-
ranging reforms are enacted in one “mega-bill,” (2) the piecemeal approach, where rather than
floating one bill, several immigration related bills are introduced, and (3) administrative or
executive action, unilateral action undertaken by Obama.
On June 27th 2013, the Senate passed a comprehensive immigration reform (CIR) bill:
Border Security, Economic Opportunity, and Immigration Modernization Act (S. 744). Although
this bill garnered a great deal of attention, as of August 2014, John Boehner, Speaker of the
House, has not brought S. 744 for a vote on the House floor. Additionally, reports have surfaced

2
claiming that Boehner does not plan to act on the Senate bill this year (Myers 2014). Although
the House has not gone to a vote on S. 744, they have continued to be active on the subject of
immigration, but in what could be described as a piecemeal approach. As of the end of March
2014, over one dozen immigration related bills, addressing facets of the immigration system
were pending in the House (U.S. Congressional Research Service 2013; What’s on the Menu?
2014).
As of June 2014, Obama has announced a plan to move forward on immigration reform
through unilateral action using his executive powers (Marshall and Garcia 2014). Although as of
the second week of August 2014, Obama has not announced a path to legalization, it is
speculated that a path to legalization may be announced before the fast approaching end of the
summer (Nakamura 2014). Obama employed executive action in 2012 with Deferred Action for
Childhood Arrivals (DACA), which offered young unauthorized immigrants that arrived in the
United States as children and met certain other criteria, reprieve from deportation and
authorization to work.
1.2 Issues Addressed in Immigration Reform Legislation
The bills acted on by the House in the 113
th
Congress have addressed a number of aspects of the
U.S. immigration system including: interior enforcement, employment eligibility verification,
worksite enforcement, border security, nonimmigrant visas, and immigrant visas (U.S.
Congressional Research Service 2013). Similarly, S.744 addressed many of the same facets
through various provisions in the bill. In contrast, S.744 also included provisions for the
legalization of unauthorized immigrants as well as humanitarian admissions (U.S. Congressional
Research Service 2013). The legalization of unauthorized immigrants was a controversial
element in S.744, which would have allowed for most unauthorized immigrants in the United

3
States to gain legal status. Legal status would first be granted through a new status, Registered
Provisional Immigrant Status (RPI). After a period of time, immigrants with RPI status would
have been able to apply to adjust to Legal Permanent Resident (LPR) status (U.S. Congressional
Research Service 2013).
Due to the provision that would have allowed most unauthorized immigrants to gain
legalization, S.744 was projected to have grown the U.S. labor force (U.S. Congressional Budget
Office 2013b). The U.S. Congressional Budget Office (CBO) projected that S.744 would boost
economic output and increase real gross domestic product (GDP). While per capita gross
national product (GNP) as well as average wages would initially fall slightly, they would
increase by 2033 (U.S. Congressional Budget Office 2013b). Although the average GNP and
wages were projected to have initially fallen, these averages would have included all those newly
authorized to live and work in the United States and would not have necessarily indicated a
decrease for those already legally present in the United States under current law (U.S.
Congressional Budget Office 2013b).
1.3 Immigrant Processing Requirements
Should a path to legalization for unauthorized immigrants be introduced, that targets anywhere
near the numbers of those that would have potentially been eligible under S.744, upwards of 8
million unauthorized immigrants may be in need of processing in the United States (U.S.
Congressional Budget Office 2013a). Based on past legislation, processing of immigrants for
legalization is likely to be a complex and time consuming task, necessitating the involvement of
nonprofit and public infrastructure, such as community groups, nonprofits, and legal service
providers. S.744 would have required unauthorized immigrants to supply proof of presence in
the United States on and after December 31, 2011, proof of immigration status, proof of identity,

4
as well as undergo a background check in order to obtain Registered Provisional Immigrant
(RPI) status (U.S. Senate 2013).
Similar detailed and thorough documentation was required to apply for DACA. Preliminary
findings from a study conducted by the Tomas Rivera Policy Institute (TRPI) in Los Angeles
County, estimates an average of 3 hours of assistance would be required per low-need applicant,
those who have a majority of required documents, to process applications for RPI status under S.
774 (Chan, Kabat, and Reyes 2013). Moderate need applicants, those missing required
documents, may require between 6–20 hours of assistance (Chan, Kabat, and Reyes 2013). High
need applicants, those with criminal records or previous interactions with U.S. Citizen and
Immigration Services (USCIS), are likely to require the greatest amount of resources and time.
However, no reliable estimate exists for this population because they are generally not served
within the network of non-profit service organizations but instead are referred out to attorneys
for legal advice (Chan, Kabat, and Reyes 2013). The estimates produced by TRPI only include
the time required to help applicants prepare their legalization application. They do not include
the time required to process the application once received by the Department of Homeland
Security (DHS).
Given these numbers, in Los Angeles County alone, the estimated 900,000 unauthorized
immigrants would require a minimum of 2.7 million hours of assistance (Chan, Kabat, and Reyes
2013). If the registration period is limited to one-year, a full-time workforce of 2,700 individuals
would be required to process RPI applications alone. This assumes 1,700 work hours per person
per year spending 100 percent of their time processing applications. This estimate does not
include time that would surely be needed for administrative duties such as set-up, supervision, or
training. Regardless of the final form that immigration reform may take, whether through S.774,

5
a piecemeal approach, or executive action taken by Obama, preparing to process a substantial
number of applicants will not only require a large enough workforce, but outreach and services
in locations that are accessible to the eligible unauthorized population.
1.4 Research Objectives
The unauthorized population is neither limited to discrete locations nor evenly spread out.
Additionally, there is no large-scale survey that directly asks about legal status, no reliable
estimates at the sub-state level for a majority of the nation, not to mention the lack of estimates at
the neighborhood level. In fact, no existing estimates of the unauthorized population at the
census tract level were uncovered during the course of this research.
That being said, the goal of this analysis is to design a research methodology for estimating
the unauthorized population at the census tract level, as a means for visually representing the
relative densities of the unauthorized population in a way that would be useful for planning
where to provide services for the unauthorized population within a community.
1.5 Thesis Structure
Chapter two contains a thorough investigation of the current state of the field of research around
estimating the unauthorized population, examining several leading estimation methods and then
presenting the results of previous research and analysis, including estimates of the number and
likely characteristics of unauthorized population. Chapter three follows with a detailed account
of the study design and methodology utilized in this study. Chapter three begins with a section
on determining the variable inputs for the analysis and continues with defining the relationships
between the independent and dependent variables at the state level. The state level relationships
are then applied to the census tract level data in order to make census tract estimates of the

6
unauthorized population. Chapter three concludes with an overview of the rendering scheme for
mapping the results.
Chapter four presents the results of the analysis through maps of various scales and extents
that visualize the relative density of the unauthorized population using dot density renderer in
ArcGIS Desktop. Although only four maps are presented, a map could be produced for virtually
any location of interest within the study area (forty-eight contiguous U.S. states and Washington,
DC). Conclusions on the implication of the analysis and the viability and performance of the
analysis method are presented in chapter five. This report concludes with an overview of the
challenges, weaknesses, and limitations of the analysis and suggests next steps to carry the
research and methodology forward.

7
CHAPTER 2: BACKGROUND
This chapter presents an overview of the state of the field of research on estimating the
unauthorized population by presenting the leading estimation methods as well as the findings of
recent studies, which includes both existing estimates of the total numbers as well as
characteristics of the unauthorized population. This chapter concludes with an overview of
research on immigrant settlement patterns in the United States.
The material presented is the basis for many of the methodological decisions made
throughout this analysis. Specifically, the characteristics of the unauthorized population and their
settlement patterns in the United States, as determined from prior research and analysis, guided
the decisions on what independent variables to include in the analyses. The data generated from
previous estimates of the unauthorized were used as the dependent variable as well as the
primary method of verifying the results. Not to mention, knowledge of the existing estimation
methods influenced the overall study design.
2.1 Methods for Estimating the Unauthorized Population
The following section covers the residual method, community-based probability method, and
other statistical methods that have been used to calculate estimates of the unauthorized
population.
2.1.1 Residual Method for National and State Estimates
The “residual method” is the leading method for estimating the unauthorized population, used to
produce the estimates released by the Department of Homeland Security (DHS) Office of
Immigration Statistics (OIS) and the Pew Hispanic Center (henceforth referred to as Pew)
(Passel 2013; Baker and Rytina 2013). Simply put, the residual method subtracts the legal

8
foreign-born (legal nonimmigrants, refugees, asylees, and legal permanent residents) from the
total number of foreign-born residing in the United States. What remains, after making certain
adjustments for factors such as undercounting and mortality, is an estimate of U.S. foreign-born
that are not legally present in the United States, the unauthorized population, as they are referred
to in this report (Hill and Johnson 2011; Judson and Swanson 2011; Passel 2013; Pastor and
Marcelli 2013; Warren and Warren 2013;). A simplified equation for estimating the unauthorized
population using the residual method follows. In addition to the equation below, adjustments are
made to account for mortality and emigration rates.
total unauthorized population
equals (=)
Total foreign-born population
minus (-)
legal permanent residents (LPRs)
nonimmigrant resident population
refugees admitted
removals of unauthorized population
plus (+)
the undercount
The following data sources are commonly incorporated into the residual method to estimate the
number of unauthorized immigrants in the United States:
Residual Method: Common Data Sources Table 1
ORGANIZATION ESTIMATE/COUNT
ACS Total foreign-born
U.S. Census Bureau Total foreign-born
CPS Total foreign-born
Department of Homeland Security (DHS) Authorized immigrant population
Department of State Refugee characteristics
DHS and U.S. Citizenship and
Immigration Services (USCIS)
Legal permanent residents (LPR)
characteristics
USCIS Asylums granted affirmatively
Executive Office for Immigration Review
of the Department of Justice (DOJ)
Asylums granted defensively in
removal proceedings
U.S. Customs and Border Protection
*
Nonimmigrant admissions
National Center for Health Statistics Life expectancy tables
*
TECS system capturing I-94 arrival-departure records

9
The residual method alone is restricted to estimating the unauthorized population at the
national or state level because of the lack of granularity of required data. Using estimates made
from the residual method as a baseline, combined with additional methods, such as survey and
statistical methods as well as the use of administrative data, have been employed to estimate the
distribution and demographic characteristics of the unauthorized population at the sub-state level.
2.1.2 Residual Method Combined With Other Methods for Sub-state Estimates
Two examples of studies that produced sub-state estimates for California include Pastor and
Marcelli (2013) and Hill and Johnson (2011). Pastor and Marcelli (2013) use a “community-
based probability method,” a combined survey and statistical method, to generate estimates of
the unauthorized by sub-counties, or PUMAs. Using this method, of the individuals captured in
the ACS as non-citizen foreign-born (excluding those born in Cuba), the probability of being
unauthorized is calculated by using legal status predictors generated from Marcelli’s 2001 Los
Angeles County Mexican Immigrant Legal Status survey (LAC-MILSS). Those with the highest
calculated probabilities of being unauthorized are flagged until the total number of those flagged
equals the OIS estimates (derived from the residual method) of the total number of unauthorized
adults for the top ten countries of origin (Pastor and Marcelli 2013). The characteristics of those
flagged as unauthorized are then analyzed and presented as the characteristics of the
unauthorized (Pastor and Marcelli 2013).
Hill and Johnson (2011) use statistical methods that include administrative data, Individual
Taxpayer Identification Number (ITIN) filer counts, to estimate the total number of unauthorized
by zip codes and counties in California. The final zip code estimates are ultimately scaled so that
when summed, they equal the total number of unauthorized in California, as derived from the
residual method. Hill and Johnson use ITIN filers, excluding those that file from abroad, as a

10
proxy for the unauthorized because they have found that, “the vast majority of ITIN filers do
appear to be unauthorized (2011, 11).” Although not all of the unauthorized pay taxes or pay
taxes using an ITIN—some do not pay taxes at all or pay taxes using other methods like a false
or fraudulent Social Security number— it is unlikely that persons legally in the United States
would use an ITIN because they would use a Social Security number or other federal tax ID
number instead (Hill and Johnson 2011).
2.1.3 Challenges and Weaknesses of Existing Estimation Methods
Estimating the unauthorized is not an exact science, and there are several aspects of the leading
methodologies that are subjective. One such aspect is the undercount of the unauthorized
population. It is generally understood that a portion of the unauthorized population is missed in
the census and other surveys; what is debated is the percentage of the unauthorized population
that is not surveyed. OIS uses an undercount of 10 percent, and Pew uses an undercount in the
“range of 10-15 percent” (Baker and Rytina 2013; Passel 2013). Warren and Warren (henceforth
referred to as Warren), on the other hand, use an undercount of 20 percent (2013). The resulting
estimates of the unauthorized are sensitive to the estimated undercount used in the analysis, as
shown through sensitivity analysis conducted by OIS (Baker and Rytina 2013).
Another seemingly subjective area is determining “legal status indicators,” or characteristics
that may indicate an individual as likely to be unauthorized (Pastor and Marcelli 2013).
Although, statistical and multivariate regression analysis has been employed to determine these
indicators based on the results of smaller scale surveys, the results may be compromised for a
variety of factors, including the small numbers of those being surveyed and the known
difficulties in eliciting truthful responses when directly inquiring about legal status (Hill and
Johnson 2011; Pastor and Marcelli 2013).

11
2.2 Results of Prior Research and Analysis
2.2.1 Estimates of the Unauthorized
The unauthorized population residing in the United States in 2012 has been estimated at both
11.4 and 11.7 million by the OIS and Pew, respectively (Baker and Rytina 2013; Passel, Cohn,
and Gonzalez-Barrera 2013). Pew and OIS offer ongoing yearly reports estimating the
unauthorized population in total and by select demographic characteristics (Passel and Cohn
2011; Baker and Rytina 2013; Passel, Cohn, and Gonzalez-Barrera 2013). A third leading source
for estimating the number of unauthorized is Robert Warren, Statistics Division, U.S.
Immigration and Naturalization Service (INS), and John Robert Warren, Minnesota Population
Center, University of Minnesota, who as of January 2010, estimated 11.7 million unauthorized
persons were residing in the United States (Warren and Warren 2013).
In California, several studies have attempted to estimate the unauthorized population at a
sub-state level, including: county, Public Use Microdata Areas PUMAs (or “sub-county”), and
zip code level (Fortuny, Capps, and Passel 2007; Hill and Johnson 2011; Hill and Hayes 2013;
Pastor and Marcelli 2013). The finest geographic scale that estimates the unauthorized
population for all fifty states is at the congressional district level (Rob Paral and Associates
2006). At the county, sub-county, and zip code level (or for any smaller geography) estimates are
only available for select geographic regions.
Estimates of immigrant sub-populations have also been conducted, including estimates of the
number of legal immigrants eligible to naturalize and the unauthorized youth eligible for DACA.
These analyses have been conducted at various scales and geographies. Estimates of the eligible
DACA population have been conducted for the entire United States by metro area and
congressional district, and for the state of Illinois by cities/towns, House districts, and Senate

12
districts. For the city of Chicago, these estimates have been made down to the community level.
For the state of California, Rob Paral and Associates in collaboration with the USC Center for
the Study of Immigrant Integration (CSII) have estimated the number of legal immigrants
eligible to naturalize at the California Assembly, Senate and Congressional Districts as well as
the census tract level for Napa. (See Rob Paral and Associates “Map Gallery,”
http://www.robparal.com/gallery/index.html).
While existing studies have increased the overall knowledge of the location of the
unauthorized, because the current sub-state estimates are limited to certain regions and there is
an overall lack of estimates at a fine geographic scale, the existing estimates are not suitable for
planning the outreach and physical infrastructure at the community level for a national initiative.
2.2.2 Characteristics of the Unauthorized Population
Demographic characteristics of the unauthorized population at the national level have been
estimated by the OIS and Pew (Passel and Cohn 2009; Baker and Rytina, 2013). Demographic
characteristics presented by the OIS include: period of entry, state of residence in the United
States, region of birth, country of birth, age range, and sex (Hoefer, Rytina, and Baker 2011;
Hoefer, Rytina, and Baker 2012; Baker and Rytina, 2013). A 2009 study from Pew made
estimates of the number of unauthorized population by educational attainment, income, and
health insurance coverage for the unauthorized population in the U.S (Passel and Cohn 2009). A
2013 study from CSII presents estimates of the characteristics of the unauthorized population in
California at the regional (multi-county) level, including race/ethnicity, child population, child
poverty, speaks English well, industry, occupation, and labor force participation (Pastor and
Marcelli 2013). Based on the findings of these existing analyses, the unauthorized population in

13
the United States, including how they differ from the overall foreign-born and legal immigrant
population, can be characterized as follows:
• Country and region of birth. Fifty-nine percent of the unauthorized population is
from Mexico (Hoefer, Rytina and Baker 2012). And in California the percentage is
much higher, with 72 percent of the unauthorized population from Mexico, followed
by Central America at 12 percent (Pastor and Marcelli 2013). Of all the immigrants
from Mexico (an estimated 11.4 million) residing in the United States in 2008, more
than half were unauthorized (Terrazas 2010).
• Ethnicity. 76 percent of the unauthorized immigrant population is Hispanic (Passel
and Cohn 2009).
• Age and sex. The majority of unauthorized immigrants are between 25 and 44 (59
percent). Unauthorized immigrants are less likely to be 65 and older compared to
authorized foreign-born and U.S.-born population. Only 1.2 percent of unauthorized
immigrants are 65 and older, compared to 16 percent of authorized immigrants, and
12 percent of the U.S.-born (Passel and Cohn 2009). In California, the median age for
the unauthorized population is thirty-one compared to forty-four and fifty for
authorized and citizen foreign-born population respectively (2009-2011 data) (Pastor
and Marcelli 2013). More than half of the total unauthorized population is male (53
percent) (Hoefer, Rytina and Baker 2012).
• Period of entry. The vast majority (99 percent) of the unauthorized population
currently residing in the United States arrived after 1980 (based on author’s
calculation of total unauthorized population by year of entry in Hoefer, Rytina, and
Baker, 2011). In part, this is likely due to the Immigration Reform and Control Act of

14
1986 (IRCA), which, allowed immigrants who arrived prior to and had been
continually present in the United States since 01 January 1982, to legalize. Of the
immigrants that qualified under the “pre-1982” provision, 1.6 million had legalized as
of 2009 (Baker 2010).
• Educational Attainment. Unauthorized immigrants are less likely to have completed
high school or to have attended college than authorized foreign-born. Nearly half (47
percent) of unauthorized immigrants between 25 and 64 did not complete high school
compared to around 23.5 percent of legal immigrants. Similarly, 25 percent of the
unauthorized population have attended or completed college compared to 54 percent
of legal immigrants (Passel and Cohn 2009).
• Income. An analysis conducted by Pew found that the 2007 median household
income was $14,000 less for the unauthorized than the U.S.-born ($36,000 versus
50,000) (Passel and Cohn 2009). A similar study found even greater income
disparities in California, where the median annual income for full time workers was
found to be $30,000 less for the unauthorized than the U.S.-born ($20,000 versus
$50,000) (Pastor and Marcelli 2013). Additionally, unlike other immigrant groups,
unauthorized immigrants do not “make notable gains” corresponding with longer time
in the United States (Passel and Cohn 2009).
• Health Insurance. Fifty-nine percent of the unauthorized adults did not have health
insurance for the entire year of 2007 (Passel and Cohn 2009).
• Household and home ownership. Unauthorized immigrants are more likely to live
in households with a partner and children (47 percent) than authorized immigrants (35

15
percent). Unauthorized immigrants are less likely to be homeowners than authorized
immigrants (Passel and Cohn 2009).
• Residency. In California, the median number of years in the country for unauthorized
is 9 compared to 19 for authorized noncitizen immigrants, and 27 for immigrant
citizens (2009-2011 data) (Pastor and Marcelli 2013).
• Language proficiency. A study conducted using data from 2009-2011 found that of
immigrants in California, 42 percent of unauthorized speak English well compared to
61 percent of authorized noncitizen (Pastor and Marcelli 2013).
2.3 Immigrant Settlement Patterns in the United States
Immigrant settlement patterns, defined as trends in where immigrant groups choose to reside in
the United States, are affected by a variety of factors, including existing family/social ties,
demographic make-up of a community, as well as economy and industry (Bohn 2009). One
major change in immigrant settlement patterns that started to occur in the 1990s is the dispersal
of immigrants from settling primarily in just a few states (or metro areas within these states) to
settling across the wider United States. In 1990, nearly 75 percent of immigrants of working age
in the United States resided in just six states, with over 30 percent residing in California (Bohn
2009). In the 1990s the proportion of immigrants residing in California began to fall for the first
time since the early 1900s and by the late 1990s, the combined proportion of immigrants living
in these six traditional immigrant-receiving states began to fall as well (Bohn 2009). In terms of
population growth, the states with the highest ratio of immigrants to nonimmigrants saw some of
the lowest immigrant growth rates from 2000-2007 (Bohn 2009). A similar analysis of settlement
patterns of Mexican immigrants, found that Mexicans had also begun to settle in non-traditional
states in the south and Midwest of the country, such as Georgia, North Carolina, Nebraska and

16
Ohio (Terrazas 2010). Furthermore, the growth rate of Mexican immigrants did not necessarily
coincide with the state’s overall growth rate. In Louisiana and North Dakota the Mexican
immigrant growth rate grew despite the total population shrinking from 2000 to 2008. And in
many states, the growth in Mexican immigrants contributed considerably to the overall
population growth of the state; In Rhode Island, Mexican immigrants accounted for nearly 60
percent of the total population growth (Terrazas 2010).
Due to lack of data on the unauthorized population, it is difficult to tell how these patterns
may have differed, if at all, between the unauthorized and the foreign-born population as a
whole. In the case of California, the change in the proportion of immigrants residing in the state,
in major part has been due to fewer newly arrived immigrants choosing to settle in California
versus established immigrants migrating out of California (Bohn 2009). A similar study of
immigrant settlement patterns conducted by the Brookings Institution, found that recently arrived
immigrants that are choosing to settle in non-traditional states are likely to be from Asia or
Mexico and have lower rates of U.S. citizenship (Singer 2004).

17
CHAPTER 3: METHODOLOGY
Given that existing methods for estimating the unauthorized are not suitable for making
estimates at the census tract level, the goal of this analysis was to design a methodology that may
be suitable for estimating the unauthorized population at the census tract level. That being said,
this analysis draws on existing methods and their findings as a basis for the methodology
outlined in this chapter. Specifically, known characteristics of the unauthorized population and
their settlement patterns, established in prior research and analysis, are a basis for determining
what variables to include in this analysis. One of the main data sources for this analysis and the
source of all of the demographic data (aside from the estimates of the unauthorized population at
the state level) is the Census Bureau’s American Community Survey (ACS).
To oversimplify the analysis method in an attempt to explain the methodology designed in
this study: suppose that majority of the unauthorized population in the United States is from
Mexico and speaks English less than “very well.” This method would bring those demographic
variables into the analysis as independent variables, define their relationship with the dependent
variable (the unauthorized population) using regression analysis and then use the resulting
equation to make estimates of the unauthorized at the census tract level by “plugging in” census
tract level data. While the method used in this study is fundamentally based on the
straightforward approach outlined above, there are several crucial ways that this analysis differs:
• The dependent variable is the percent of the unauthorized out of the total foreign-born
population. In fact, all demographic variables are transformed to be percentages of the
total foreign-born.
• All demographic characteristics are incorporated into one variable using Principal
Component Analysis (PCA). Many of the demographic characteristics of the

18
unauthorized population used in this analysis are highly correlated. In order to avoid
the multicollinearity problem that would arise from including all of the variables into
a regression analysis, the variables are reduced to one artificial variable, or
component score, using PCA.
• The relationship between the dependent and independent variables were defined using
a state level regression equation and then “brought down” or applied to the census
tract level data in order to make estimates for each census tract. While there are many
challenges (including ecological fallacy) with scaling down state level equations to a
smaller geography, this method was chosen because the state level estimates of the
unauthorized population are the only available and widely accepted as reliable
estimates of the unauthorized population.
3.1 Overview of Analysis Steps
An overview of the analysis steps is shown below. Details of the analysis follow in the next
sections:
I. Determine input variables:
1. Define what is being estimated (the dependent variable)
2. Identify the demographic variables (the independent variables) that correspond to the
characteristics of the unauthorized population, more specifically variables with the
potential to differentiate the unauthorized from the larger foreign-born population
II. State level analysis: Define relationship between dependent and independent variables
1. Derive the first principal component to account for joint variation in correlated
independent variables using Principal Component Analysis (PCA)
2. Compute a component score for each state

19
3. Conduct exploratory regression analysis; Independent variables include the component
score (defined in PCA) as well as other state level variables to identify best regression
model
4. Run Ordinary Least Squares (OLS) regression analysis based on results of exploratory
regression analysis to calculate the percent of unauthorized out of total foreign-born
(dependent variable)
5. Run Geographically Weighted Regression (GWR) analysis in order to determine a unique
equation for each state included in the analysis
III. Census tract-level analysis: Estimate unauthorized population at the census tract level using
previously defined state level equations
1. Compute component scores for each census tract using the coefficient scores defined in
the state level analysis
2. Based on GWR equation for each state, substitute the state level component score with
each individual census tract’s component score in order to calculate an estimated percent
of the unauthorized population out of total foreign-born (dependent variable) for each
census tract
3. Multiply the estimated percent of the unauthorized out of the total foreign-born
(dependent variable) by the total foreign-born population for each census tract in order to
come up with an estimate of the total number of the unauthorized for each census tract
IV. Visualize the results of the analysis
V. Verify results of the analysis and draw conclusions of the viability of the method

20
3.2 Define Variables
This section reviews the variables included in the analysis and the reasoning behind including (or
excluding) certain variables. Which variables to include were determined by consulting previous
research findings and methodological approaches. Because the goal of this analysis is to make
estimates at the census tract level, only demographic data that is available at the census tract
level could be incorporated into the equation. Alaska, Hawaii, and Puerto Rico were not included
in the analysis because of lack of geographically near neighbors, a requirement for running
Geographically Weighted Regression (GWR) analysis. The time period for the analysis is 2006-
2010.
3.2.1 The Dependent Variable
The dependent variable in this analysis is the percent of the unauthorized out of total foreign
born. The dependent variable was calculated by dividing the number of the unauthorized from
the number of foreign born by state. The estimates of the unauthorized by state (the numerator)
were generated from Warren using the residual method. The source for the foreign born
population estimates (the denominator) is the ACS.
The percent of the unauthorized was estimated out of a base population, rather than estimating
the total number of the unauthorized directly. A base population was used as a method for
standardizing all of the demographic data. Standardizing the data not only helps to minimize
outliers but also ensures that the patterns or correlations are due to underlying demographic
differences not differences in the total population numbers between each state. Two variables
were considered as the base population: (1) total foreign-born, and (2) total noncitizen foreign-
born. For the reasons outlined below, the base population chosen for the analysis was total

21
foreign-born, resulting in the dependent variable being the percent of the unauthorized out of the
total foreign-born population in the United States:
• There is precedence for using the foreign-born population as the base of the
estimation. One of the leading methods for estimating the unauthorized population, the
residual method, uses the foreign-born population as the base for estimating the number
of unauthorized at the state level. Similarly, PPIC’s estimates of the total number of
unauthorized by zip code uses the foreign-born population as a base (Hill and Johnson
2011).
• The ACS estimates for the foreign-born population have a smaller margin of error
than those for the noncitizen population. The noncitizen population is a subset of the
foreign-born population, meaning that the total number of noncitizens is smaller than or
equal to the total number of foreign-born in any given geography. Because the ACS
estimates are derived from surveying a sample of the population, estimates of smaller
populations or within small geographies tend to have lower levels of accuracy due to
larger margins of error (U.S. Department of Commerce 2008).
• There is a strong positive correlation between the foreign-born population and
unauthorized population. This is logical because the foreign-born population, as
captured by the ACS, invariably includes a portion of the unauthorized population
although the exact proportion is unknown. Visual inspection of the scatterplot (Figure 1)
shows a strong positive linear relationship between the foreign-born and unauthorized
population (as estimated by Warren), meaning that as the total number of foreign-born
increases, so does the number of unauthorized. The strength and direction of the
relationship is further corroborated by the results of the Spearman’s rank-order

22
correlation (Table 2). The Spearman’s correlation found that an increase in the foreign-
born population was strongly correlated with an increase in the unauthorized population
in the United States at the state level, r
s
(47)= .973, p < .0005.

Scatterplot of the Unauthorized by the Total Foreign-born Population by State Figure 1

Spearman’s Rank-order Correlation Between the Unauthorized and Foreign-born Table 2
Population by State
Unauthorized Foreign-born
Spearman's rho
Unauthorized
Correlation
Coefficient
1.000 .973
**

Sig. (2-tailed) . .000
N 49 49
Foreign-born
Correlation
Coefficient
.973
**
1.000
Sig. (2-tailed) .000 .
N 49 49
**Correlation is significant at the 0.01 level (2-tailed).

Once the dependent variable was determined, there was still the question of time period as
well as which source would be used to supply the data for the dependent variable. When
considering data options, special consideration was paid to accuracy and recentness of data. The

23
ACS data was the logical choice for the base population (denominator), offering the most
authoritative source for demographic data that is updated regularly and available nationwide at
the census tract level. The ACS releases data in 1-year, 3-year, and 5-year estimates. The 5-year
estimates were chosen because they are the most reliable and have the largest sample size,
particularly important when working with small geographies or when analyzing small
populations (U.S. Department of Commerce 2008).
While several state level estimates of the unauthorized population exist, the Warren estimates
were chosen as the numerator, because they have been released yearly and for all fifty states
(Warren and Warren 2013). A five-year (2006-2010) average of the unauthorized population was
taken for the numerator in order to correspond with the 5-year ACS data. This 5-year average
was then divided by the 2006-2010 estimates of the foreign-born population released by the ACS
in order to come up with the percent of the unauthorized out of the total foreign-born population,
the dependent variable.

24

Observed Dependent Variable: % Unauthorized out of Total Foreign-born Figure 2
Population (2006-2010)
3.2.2 Corresponding Demographic Variables
Based on findings on the characteristics of the unauthorized population, as determined by prior
research and analysis, the demographic variables in Table 3 (as shares of the total foreign-born)
were considered for inclusion in the analysis. Several variables were considered in many of the
categories (year of entry, language proficiency, educational attainment, income and country of
origin) including some “nested variables” where one or more variables make-up another
variable. For example, the variable “Income less than 50,000 or no income” includes the variable
“no income,” which was also initially considered in the study.

25
Demographic Variables Considered for Inclusion in the Analysis Table 3
Variable Characteristic Universe
Spearman’s rho:
correlation with
dependent variable
Entered the U.S. after 2000

Year of entry
Total population born
outside of the U.S.
.374
**

Entered the U.S. before 1980 Year of entry Foreign-born -.708
**

Speak a language other than English:
Speak English 'very well'
Language
proficiency
Foreign-born population
5 years and over
-.402
**

Speak a language other than English:
Speak English 'not at all'
Language
proficiency
Foreign-born population
5 years and over
.765
**

Speak a language other than English:
Speak English 'less than very well'
Language
proficiency
Foreign-born population
5 years and over
.776
**

Speak a language other than English:
Speak English 'less than well'
Language
proficiency
Foreign-born population
5 years and over
.807
**

65 years and over Age
Total foreign-born
population
-.777
**

Not a U.S. citizen
Citizenship
status
Total foreign-born
population
.874
**

Less than high school graduate
Educational
attainment
Foreign-born population
25 years and over
.760
**

Graduate or professional degree
Educational
attainment
Foreign-born population
25 years and over
-.570
**

Bachelor's degree or higher
Educational
attainment
Foreign-born population
25 years and over
.609
**

No income Income
Foreign-born population
15 years and over
.673
**

Income less than 50,000 or no income Income
Foreign-born population
15 years and over
.605
**

Median income in the last 12 months Income
Population 15 years and
over in the United States
with income
.507
**

Americas: Latin America: Central
America: Mexico
Country or
region of
origin
Foreign-born population
excluding population
born at sea
.773
**

Americas: Latin America: Other Central
America
Country or
region of
origin
Foreign-born population
excluding population
born at sea
.375
**

Americas: Latin America: Caribbean: Cuba
Country or
region of
origin
Foreign-born population
excluding population
born at sea
No clear correlation
Americas: Other Latin America
Country or
region of
origin
Foreign-born population
excluding population
born at sea
-.279
**

Median Age of Foreign-born Age Total population -.630
**

Income in the past 12 months below
poverty level: Foreign-born
Income
Total population for
which poverty status is
determined
.654
**

Total Hispanic or Latino foreign-born Ethnicity
Hispanic or Latino
Population
.799
**

**
Correlation is significant at the 0.01 level (2-tailed).
Data source for all demographic variables is the ACS 2006-10
Red columns indicate demographic variables that were ultimately not retained in the analysis.

26
Each potential demographic variable’s relationship with the dependent variable (percent of
the unauthorized out of total foreign-born), and therefore their viability as analysis variables, was
examined through visual inspection of scatterplots as well as Spearman’s rank-order correlation
to test the strength and direction of their relationships with the dependent variable, as shown in
Table 3. Where several variables were considered in a particular category, the variable(s) with
the strongest relationships with the dependent variable as well as existing theory were considered
in determining which variables would be retained for inclusion in the analysis. Ultimately,
thirteen demographic variables were retained.
3.3 State Level Analysis: Define Relationship Between Dependent and Independent
Variables
3.3.1 Principal Components Analysis (PCA)
Principal component analysis is a statistical method for reducing the number variables in an
analysis into a subset of linearly uncorrelated “artificial” variables, called principal components.
PCA is a data reduction technique, often utilized as a way of eliminating the redundancy between
variables that may be measuring the same or similar construct (O'Rourke and Hatcher 2013). In
the case of this analysis, a number of highly correlated demographic variables are reduced to one
principal component that represents the maximum variance between the original variables.
The PCA results in a set of actual scores, in this case, one score for each geography (forty-
eight contiguous U.S. states and Washington, DC) included in the analysis. These scores were
then used in subsequent regression analysis in place of the original variables. So instead of
entering all thirteen demographic variables into the regression analysis, only one composite
variable (the principal component) was entered into the analysis. PCA was chosen as an analysis
method because it eliminates the multicollinearity problems that would have arisen, should all

27
correlated variables have been entered into a regression analysis, without having to eliminate
variables altogether.
An initial PCA was run on all retained demographic variables (see Table 3), chosen because
of their strength of relationship with the dependent variable (the percent of the unauthorized out
of the total foreign-born population). To confirm PCA as an appropriate analysis method, the
correlation matrix as well as Bartlett’s test of Sphericity and Kaiser-Meyer-Olkin (KMO)
Measure of Sampling Adequacy were examined and are explained in the following section.
3.3.1.1 Variable correlation and sampling adequacy
From examining the correlation matrix (See Appendices, Table 19), it is clear that all variables
are strongly correlated with at least three other variables at a level of r ≥ 0.3. The only variable
with relatively low level of correlation with other variables is percent of the foreign-born
population born in Central America (other than Mexico), which is correlated with three variables
right around 0.3, and with all other variables at <0.3. Additionally, Bartlett's test of Sphericity is
statistically significant with a p-value <.0005, indicating that overall there are correlations in the
variables, suggesting that principal components analysis is an appropriate method for reducing
the number of variables in the analysis (Laerd Statistics 2013). While the correlation matrix and
Bartlett's test of Sphericity show that there is correlation between variables, there may in fact be
too high of correlation between variables. When examining the correlation matrix (Table 19,
Appendices), there are three variables with r ≥ 0.9 which may indicate multicollinearity or
singularity with the data.
Kaiser-Meyer-Olkin (KMO) analysis was used to test for sampling adequacy and linear
relationship between variables. Sampling adequacy was assessed for the overall equation as well
as for the individual variables using KMO analysis. The sampling adequacy for this PCA was

28
found to be .771, which is satisfactory or “middling” on Kaiser's (1974) classification of measure
values (Laerd Statistics 2013). This indicates linear relationships between variables and that PCA
may be an appropriate analysis method. When assessing KMO measures for individual variables,
all variables have strong linear relationships with other variables (KMO >= .5) except for born in
Central America (other than Mexico) (KMO = .285).
KMO Measures for Demographic Variables Table 4
Entered 2000 or later
*
0.7353
Entered before 1980
*
0.7958
Speak a language other than English:
Speak English 'less than well'
*

0.8598
65 years and over
*
0.8335
Not a U.S. citizen
*
0.905
Less than high school graduate
*
0.777
Bachelor's degree or higher
*
0.7348
Median income in the past 12 months 0.6382
Born in Mexico
*
0.7704
Born in Central America
*
0.2851
Median Age 0.7940
Income in the past 12 months below
poverty level
*

0.7194
Hispanic or Latino foreign-born
*
0.805
*
percent of total foreign-born
3.3.1.2 Retaining principal components
From examining the scree plot (Figure 3) and the eigenvalue-one criterion (Table 5) from the
initial PCA, it appears that three components could potentially be retained. The first three
components have eigenvalues greater than one and each account for over 10 percent of the total
variance. That being said, it logically does not make sense to have greater than one component
for the purpose of this study, because all demographic variables were included based on their
relationship and potential to estimate one variable, the percent of the unauthorized out of the total
foreign-born population. The decision to retain only one component is strengthened by
examining the component matrix. The component matrix shows that all variables, except “born
in Central America,” load on the first component at .3 or greater.

29

Scree Plot Figure 3

Eigenvalue-one Criterion: Total Variance Explained by Initial PCA Table 5
Component
Initial Eigenvalues
Extraction Sums of Squared
Loadings
Rotation Sums of Squared Loadings
Total
% of
Variance
Cumulative
%
Total
% of
Variance
Cumulative
%
Total
% of
Variance
Cumulative
%
1 7.515 57.805 57.805 7.515 57.805 57.805 5.301 40.775 40.775
2 2.568 19.755 77.560 2.568 19.755 77.560 4.263 32.794 73.569
3 1.527 11.750 89.310 1.527 11.750 89.310 2.046 15.741 89.310
4 .591 4.544 93.855

5 .270 2.081 95.935

6 0.188 1.447 97.382

7 0.134 1.031 98.413

8 0.068 0.523 98.936

9 0.051 0.394 99.330

10 0.028 0.218 99.548

11 0.027 0.207 99.755

12 0.017 0.131 99.886

13 0.015 0.114 100.000

Note: Extraction Method: Principal Component Analysis
The variable “born in Central America” was ultimately removed from the analysis because of
the lack of sampling adequacy as measured in the KMO test as well as the relatively low levels

30
of correlation as measured in the correlation matrix. A final PCA was rerun, omitting the “born
in Central America” variable and retaining only one component score.
3.3.1.3 PCA Results
Ultimately, one component score was retained with an eigenvalue of 7.5 and which accounts for
62.5 percent of the total variance (Table 7). Twelve demographic variables were incorporated
into the principal component, with only one variable having been dropped: “born in Central
America.” The output of the final PCA showed improvement from the initial PCA, as reflected
in a higher overall KMO measure of .813, which according to Kaiser's (1974) classifications is
“meritorious" sampling adequacy (Table 6) (Laerd Statistics 2013). Additionally, the KMO
measures for individual variables are now all above .65 (Table 20, Appendices).
KMO and Bartlett's Test Table 6
Kaiser-Meyer-Olkin Measure of Sampling Adequacy .813
Bartlett's Test of Sphericity
Approx. Chi-Square 936.359
df 66
Sig. 0.000

The correlation matrix and Bartlett's test of Sphericity (statistically significant with a p-
value <.0005), indicate that overall there are correlations between the variables. While these
indicators suggest that principal components analysis may be an appropriate method for reducing
the number of variables in the analysis, on the other hand, there are indicators that
multicollinearity may be a problem. Similarly to the original analysis, three variables continue to
be correlated with other variables at r ≥ 0.9 (Table 21, Appendices). Additionally, the
determinant of the correlation matrix is 3.797E-010. A determinant <.00001 indicates that their
may be a multicollinearity problem with the data although “strictly speaking,” when conducting
PCA, this is not a concern (Field 2013, 21). Although there is concern about model fit with 81

31
percent (54) of the residuals computed between observed and reproduced correlations are
nonredundant residuals with absolute values >0.05 (Table 22, Appendices) (Field 2013).
Eigenvalue-one Criterion: Total Variance Explained by Final PCA Table 7
Component
Initial Eigenvalues
Extraction Sums of Squared
Loadings
Total
% of
Variance
Cumulative
%
Total
% of
Variance
Cumulative
%
1 7.504 62.536 62.536 7.504 62.536 62.536
2 2.376 19.797 82.333

3 1.236 10.297 92.630

4 0.271 2.255 94.885

5 0.197 1.643 96.527

6 0.161 1.346 97.873

7 0.094 0.782 98.655

8 0.061 0.510 99.165

9 0.033 0.272 99.437

10 0.028 0.230 99.667

11 0.023 0.196 99.863

12 0.016 0.137 100.000

Note: Extraction Method: Principal Component Analysis
The final output of the PCA, visualized in Figure 4, a unique component score generated for
each state, is used in the regression in place of the twelve variables from which it was calculated.
The component scores for each state (Table 8) are calculated by multiplying a weight, generated
in the course of the PCA, by the original variable and summing the results (Laerd Statistics
2013).
The resulting component scores generated in the final PCA range from -2.05 (Vermont) to
1.54 (Arkansas). By comparing Figure 2 to Figure 4, it appears that in general (and with a few
exceptions) those states with high component scores also have high ratios of the unauthorized out
of total foreign-born and vise versa. This is one positive indicator of the suitability of using the
principal component moving forward in the analysis.

32

Component Scores by State Figure 4

Component Score Coefficient Matrix Table 8
Entered 2000 or later
*
0.062
Entered 1980 or before
*
-0.099
Speak a language other than English: Speak English 'less
than well'
*

0.114
65 years and over
*
-0.112
Not a U.S. citizen
*
0.125
Less than high school graduate
*
0.118
Bachelor's degree or higher
*
-0.099
Median income in the past 12 months -0.080
Born in Mexico
*
0.118
Median Age -0.104
Income in the past 12 months below poverty level
*
0.102
Hispanic or Latino
*
0.115
*
as a percent of the total foreign-born
Note: Extraction Method: Principal Component Analysis;
Rotation Method: Varimax with Kaiser Normalization

33
3.3.2 Exploratory Regression Analysis
Exploratory regression analysis looks at all possible combinations of independent or explanatory
variables and outputs a list of passing models that meet the specified model parameters.
Regression analysis was chosen as a method of analysis because of the complexity involved with
estimating the percent of the unauthorized out of the total foreign-born population, in particular
the challenge of estimating a population (the unauthorized) in which there is an overall lack of
reliable data. Therefore, exploratory regression was used as a method to investigate all potential
explanatory variables that may be important contributing factors for estimating the unauthorized
population. Aside from the principal component, generated in the PCA, several independent
variables related to immigrant settlement patterns and changes in settlement patterns were
considered for inclusion in the analysis.
3.3.2.1 Independent variables
There is great variance in the percent of the unauthorized out of the total foreign-born population
(the dependent variable) by state. In Vermont, the unauthorized make up just 1.2 percent of the
total foreign-born compared to 53 percent in Alabama. Because of this wide variance in the
percent of the unauthorized out of total foreign-born, and as supported by the literature, it is
hypothesized that various state factors may affect immigrant settlement patterns and therefore the
make-up of the immigrant population residing in a particular state. The settlement pattern
variables are introduced below:
• Immigrant growth rates. Rather than focus on the underlying causes of the changes in
immigrant settlement patterns, this analysis looks at changes in settlement patterns as
reflected by state growth rates of the unauthorized as well as the foreign-born as a whole
during three different time periods: (1) 1990-2000, (2) 2000-2010, and (3) the long-term

34
growth rate: 1990-2010. All growth rates were calculated using either the Warren or ACS
state level data.
• Low unauthorized population. In addition to the component score and growth rate
variables, a dummy variable was created for the eight states that had an average of less
than 3,000 unauthorized immigrants during the 2006-2010 analysis period. Of these eight
states, five were estimated to have fewer than 1,000 persons. These eight states have
significantly lower numbers of the total unauthorized, with all other U.S. states included
in this study having of greater than 20,000 unauthorized persons (when taking the
average of the analysis period). Because these eight states are outliers in many ways, a
dummy variable was used as an attempt to account for some of the distinctive
characteristics of these states rather than remove the states from the model. Removing
these states was not preferable due to the already low number of cases (forty-eight
contiguous U.S. states and Washington, DC) included in the analysis. Given the
information and theory previously outlined, the following variables were chosen as
potential explanatory variables in the exploratory regression analysis:

35
Independent Variables Included in the Exploratory Analysis Table 9
Description Universe Time period Type
Spearman’s rho:
correlation with
dependent variable
Principal component
(generated from PCA)
(see Table 3 for
universe of input
variables)
2006-2010 ordinal .832
**

States with growth rate
>100 percent unauthorized
Unauthorized 2000-2010
nominal
(dummy)
.361
*

States with a decline in
number of unauthorized
Unauthorized 2000-2010
nominal
(dummy)
-.559
**

Growth rate Unauthorized 2000-2010 ratio .542
**

Growth rate Foreign-born 2000-2010 ratio .571
**

Growth rate Foreign-born 1990-2000 ratio .825
**

Growth rate Foreign-born 1990-2010 ratio .795
**

States with more than
double the nations mean
immigrant growth rate
Foreign-born 1990-2000
nominal
(dummy)
.559
**

States with less than 3,000
unauthorized immigrants
Unauthorized 2006-2010
nominal
(dummy)
-.636
**

**
Correlation is significant at the 0.01 level (2-tailed).
*
Correlation is significant at the 0.05 level (2-tailed).

Additionally, the strength and direction of the relationships between the dependent variable
(the percent of the unauthorized out of total foreign-born) and the potential independent variables
(Table 9) were examined through visual inspection of scatterplots as well as Spearman’s rank-
order correlation before being introduced into the exploratory regression. Although the strength
of the relationship varied, Spearman’s rank order found that all potential explanatory variables
were significantly correlated with the dependent variable and all variables were retained for
inclusion in the exploratory regression.
3.3.2.2 Exploratory Regression Analysis Results
After careful considerations of the theory and examination of the data using the exploratory
regression method, one model presented itself as most suitable for estimating the rate of the
unauthorized population out of the total foreign-born. In order for a model to be considered
“passing,” it had to meet all of the following criteria:
• Minimum Adjusted R-Squared > 0.50

36
• Maximum Coefficient p-value < 0.05
• Maximum Variance Inflation Factor (VIF) Value < 7.50
• Minimum Jarque-Bera p-value > 0.10
• Minimum Spatial Autocorrelation p-value > 0.10
After careful consideration of the exploratory regression results, a four variable model met all
of the model criteria and all variables were found to be statistically significant at the 0.01 level
(see Table 10 and 11).
Passing Model Variables and Direction Table 10
Variable
Time
period
Description Type direction
Component score 2006-2010 Generated from PCA interval positive
Low unauthorized
population
2006-2010
States with less than
3,000 unauthorized
immigrants
nominal
(dummy)
negative
Unauthorized growth rate 2000-2010 Unauthorized growth rate ratio positive
Immigrant growth rate 1990-2000 Immigrant growth rate ratio positive

Statistics of Passing Model Table 11
Adjusted
R-Squared
AICc
Jarque-Bera
p-value
Koenker (BP)
Statistic p-value
Max VIF
Factor Global
Moran's I p-value
0.912280 -162.763921 0.440282 0.738423 2.540353 0.871584
3.3.3 Ordinary Least Squares
Once a suitable model was found using the exploratory regression method, Ordinary Least
Squares (OLS) linear regression analysis was performed in order to model the relationship
between key variables and the dependent variable. OLS regression analysis results in one set of
coefficients that can be multiplied by each state’s explanatory variables in order to produce an
estimate of the percent of the unauthorized population out of the total foreign-born (dependent
variable) for each state.

37
Regression equation:
Y
=
β
0
+
β
1
X
1
+
β
2
X
2
+…
β
n
X
n
+
ε
OR percent of the unauthorized population out of total foreign-born =

β
0
+
β
1
(population <3,000) +
β
2
(component score) +

β
3
(unauthorized growth rate) +
β
4
(immigrant growth rate)
+
ε
Where,
Dependent variable (Y)
Explanatory variables (X)
Intercept (β
0
)
Coefficients (β
1…
β
n
)
Residuals (ε)
Retained Variables in OLS Regression Table 12
Explanatory
variable (x)
Coefficient
(β)
StdError
t-
Statistic
Probability
Robust_
SE
Robust_t Robust_Pr VIF
Intercept 0.2446 0.0182 13.4763 0.000000* 0.0172 14.2366 0.000000* --------
Low
unauthorized
population
-0.1511 0.0201 -7.5172 0.000000* 0.0239 -6.3332 0.000000* 1.5292
Unauthorized
growth rate
0.0457 0.0133 3.4434 0.001273* 0.0158 2.8974 0.005844* 1.5158
Component
Score
0.0527 0.0097 5.4466 0.000002* 0.0077 6.8600 0.000000* 2.5404
Immigrant
growth rate
0.0586 0.0147 3.9862 0.000250* 0.0111 5.2668 0.000004* 2.1818
* An asterisk next to a number indicates a statistically significant p-value (p < 0.01).

All signs are expected. All variables are statistically significant. No major aspatial
autocorrelation as indicated by the low VIF scores.

38

OLS Regression Results Table 13
Dependent Variable
% unauthorized out of
total foreign-born
Input Features Contiguous U.S. states
Number of Observations 49
Multiple R-Squared 0.919590
Adjusted R-Squared 0.912280
AICc -162.7639

Joint F-Statistic 125.799207 Prob(>F), (4, 44) degrees of freedom 0.000000
*

Joint Wald Statistic 592.529532 Prob(>chi-squared), (4) degrees of freedom 0.000000
*

Koenker (BP) Statistic 1.985517 Prob(>chi-squared), (4) degrees of freedom 0.738423
Jarque-Bera Statistic 1.640680 Prob(>chi-squared), (2) degreesof freedom 0.440282
* An asterisk next to a number indicates a statistically significant p-value (p < 0.01).
Adjusted r-squared is .91, indicating that 91 percent of the variance in the dependent variable
is explained by the model. The Jarque-Bera Statistic was not statistically significant, indicating
that the residuals are normally distributed; a second test, Moran’s I, was performed to test
whether the residuals exhibit spatial randomness. The results of Moran’s I test of spatial
autocorrelation, as indicated by a z-score between -1.65 and 1.65 (z = .678) that is not
statistically significant (p = .498), implies that the residuals are randomly spatially distributed
(see Figure 5 for visual inspection of standard residuals). The default neighborhood search
threshold for testing spatial autocorrelation was around 315 miles. To put this in perspective, it is
roughly the driving distance from San Diego to Las Vegas or Boston to Philadelphia.

39

OLS Standard Residuals Figure 5

The Koenker test is not statistically significant, signifying that the relationships between the
explanatory variables and the dependent variable are non-stationary, and that the strength of the
relationships is likely to stay relatively constant across geographies. Although a non statistically
significant Koenker test indicates that the model may not be greatly improved by using
geographically weighted regression (GWR), GWR was chosen to be performed regardless due to
the known differences in the distribution and characteristics of the unauthorized population
across the nation.
3.3.4 Geographically Weighted Regression
Once the OLS regression equation is properly specified, the same dependent and explanatory
variables were included in a Geographically Weighted Regression (GWR) analysis. GWR

40
analysis is a type of linear regression that allows for the strength and direction of relationships of
variables to vary across space. Similar to the OLS regression, one of the outputs of GWR
analysis is coefficients (β) to be multiplied by the explanatory variables (x) and summed to come
up with an estimate of the percent of the unauthorized population out of the total foreign-born
(dependent variable) for each state. The primary difference between the two methods is unlike
OLS, which outputs one set of coefficient scores for all geographies, GWR outputs unique sets
of coefficient scores (β) for each geography. In the case of this analysis, a unique set of
coefficient scores (β) is specified for each state, resulting in forty-nine unique regression
equations, one for each contiguous states and Washington, DC.
GWR Results Table 14
Input Features
Contiguous U.S.
states and D.C.
Number of Observations 49
Dependent Variable
% of unauthorized
population out of
total foreign-born
Multiple R-Squared 0.9314
Adjusted R-Squared 0.9158
Residual Squares 0.0664
Sigma 0.0412
AICc -161.2483
Effective number 9.8813

Comparing the results of the GWR to the OLS analysis, the Akaike's Information Criterion
(AICc) went up slightly, from -162.76 to -161.25 in the GWR, but the adjusted R-squared also
went up slightly, from 0.9123 to 0.9158. Results of Moran’s I, test of spatial autocorrelation on
StdResiduals of the GWR, show no spatial autocorrelation:
• Moran’s index: .021
• p-value: .729
• z-score: .346

41

GWR Standard Residuals Figure 6

In addition to outputting a unique regression equation and coefficient scores for each state in
the analysis, GWR outputs an adjusted r-squared for each state. The results of the GWR show an
adjusted r-squared value between .90-.92, indicating that, depending on the state, between 90 and
92 percent of the variance in the dependent variable is explained by the model.
Additionally, a measure of the influence of each independent variable as a predictor of the
percent of the unauthorized immigrant population by state can be explored through each
independent variable’s coefficients. Figure 7 shows a clear spatial relationship between each
variable’s strength as a predictor of the dependent variable. In the case of the component score,
the strength of this variable as a predictor of the dependent variable is strongest in Texas,
Oklahoma, and Kansas and appears to diminish radially from these states. The two growth rate
variables appear to have strongest influence on the East coast, with the strength diminishing from

42
East to West. While the GWR output variable coefficients for all states included in this study for
the “<3,000 unauthorized immigrants” variable, because this is a dummy variable, only the eight
states that meet this criteria vary in influence on the dependent variables. Additionally, because
“<3,000 unauthorized immigrants” variable has a negative relationship with the dependent
variable, a lower standard deviation of the variable coefficient indicates a stronger influence on
the dependent variable.

Component Score

Foreign-born Growth Rate (1990–2000)

Unauthorized Population <3,000 Persons
(highlighted states)

Unauthorized Population Growth Rate
(2000–2010)

Standard Deviation

Strength of Independent Variable Coefficients as Predictors of the % of the Figure 7
Unauthorized Population

43

3.4 Census Tract Level Analysis
Once a regression equation was specified for each state using GWR, the next step of this analysis
was to apply the state level equations to the corresponding census tract data in each state in order
to generate an estimate of the percent of the unauthorized out of the total foreign-born population
(dependent variable) for each individual census tract within the geography of the analysis. The
key to making an estimate for each census tract was to calculate a unique component score for
each census tract.
Although the variables remain the same, the data has changed from state level to census tract
level data. Even though it is likely that the relationship between each variable is somewhat
different at the census tract level than at the state level, the PCA is not rerun using the census
tract data, but the component scores for the census tracts are computed using the coefficient
scores generated in the state level analysis (Table 8). The reasoning for not conducting a new
PCA using the census tract data is that in order to estimate the unauthorized population, the
relationship between the dependent variable and the principal component (generated in the state
level PCA) as well as the rest of the independent variables must be defined. This was done using
GWR at the state level and is explained in the previous section.
If the PCA were to be rerun using census tract data, the relationship between the component
score and the dependent variable would have to be redefined. This is simply not possible because
no estimates of the dependent variable exist at the census tract level. Therefore, acknowledging
the flaws in this method, the relationships between the independent and the dependent variables
were defined at the state level and then applied to the census tract level. The process of using the

44
state level equations to generate census tract level estimates is explained in the following
sections.
3.4.1 Calculating a Unique Component Score for each Census Tract
The first step to calculating individual estimates for each census tract is to calculate a unique
component score for each census tract. In fact, when applying each state level equation
(generated in the GWR) to the census tracts within the state, the only input that changes is the
component score variable. Therefore, the key to making unique estimates for each census tract is
the component score.
As previously mentioned, the PCA was not rerun using the census tract level data. Instead,
component scores were generated for each census tract using the coefficient scores previously
generated in the PCA. The component scores for the census tracts were computed manually by
multiplying the twelve demographic variable data specific to each census tract by the coefficient
scores previously generated in the PCA (Table 8) and then summing the results.
The same coefficient scores were used to calculate every component score generated in this
study (for every states and census tract). While the coefficient scores stay the same, the
demographic variable inputs are specific to the geography for which the component score is
being calculated. The equation, with the coefficient scores, follows:
component score equals (=):
0.062 (Entered 2000 or later)
+ -0.099 (Entered before 1980)
+ 0.114 (Speak English 'less than well')
+ -0.112 (65 years and over)
+ 0.125 (Not a U.S. citizen)
+ 0.118 (Less than high school graduate)
+ -0.099 (Bachelor's degree or higher)
+ -0.080 (Median income in the past 12 months)
+ 0.118 (Born in Mexico)
+ -0.104 (Median Age)

45
+ 0.102 (Income in the past 12 months below poverty level)
+ 0.115 (Hispanic or Latino)
After calculating a unique component score for each census tract in the United States, the scores
were inserted into their respective state level regression equations (depending on which state the
census tract was located) in order to come up with an estimate of the rate of the unauthorized out
of the total foreign-born for each census tract in the United States.
For example, given the regression equation,
Y
=
β
0
+
β
1
X
1
+
β
2
X
2
+…
β
n
X
n

Where,
Estimate of the dependent variable (Y)
Explanatory variables (X)
Intercept (β
0
)
Coefficients (β
1…
β
n
)
From the results of the GWR, the equation for California was found to be:
Y
=.2697 + -0.1875 x
1
+ 0.0575 x
2
+ 0.0393 x
3
+ 0.0412 x
4

Where,
Y = percent of unauthorized out of total foreign born (dependent variable)
x
1
= dummy variable for unauthorized population <3,000 (where “1” indicates <3,000
persons and all other states are “0”.)
x
2
= component score
x
3
= unauthorized growth rate (2000-2010)
x
4
= immigrant growth rate (1990-2000)
In order to come up with estimates for each census-tract level (in this case for the state of
California) the only variable that would change in the equation would be (X
2
), the component
score. After calculating the percent of the unauthorized out of total foreign-born, the final step is
to multiply the result by the total number of foreign-born per census tract as released by the
2006-2010 ACS.
For example, take two census tracts (A and B):
Census tract Component score Total foreign-born
A 0.0554 1,000
B 1.13 200

46

Census tract A
• Calculate the percent of the unauthorized out of total foreign-born using the regression
equation for California (generated in GWR):
0.2697 + -0.1875(0)+ 0.0575(0.0554) + 0.0393(0.1283) + 0.0412(0.3724) = 0.2933,
• Calculate the total unauthorized (percent of the unauthorized out of total foreign-born
multiplied by total foreign-born):
0.2933 * 1,000 = 293
Results: 29.33 percent of the foreign-born population is unauthorized, an estimated 293
unauthorized out of the 1,000 foreign-born persons.

Census tract B
• Calculate the percent of the unauthorized out of total foreign-born using the regression
equation for California (generated in GWR):
0.2697 + -0.1875(0) + 0.0575(1.13) + 0.0393(0.1283) + 0.0412(0.3724) = 0.3551
• Calculate the total unauthorized (percent of the unauthorized out of total foreign-born
multiplied by total foreign-born):
0.3551 * 200 = 71
Results: 35.51 percent of the foreign-born population is unauthorized, an estimated 71
unauthorized out of the 200 foreign-born persons.

47
If census tracts A and B are the exact same size, hypothetically, in a neighborhood that
consisted of only these two census tracts, with all other factors the same, more services and/or
greater outreach should be provided in census tract A than B, given that there is a higher density
unauthorized persons in census tract A.

48
CHAPTER 4: RESULTS
This chapter reviews the results of the analysis as visually represented using dot density renderer
in ArcGIS Desktop. Estimates generated in this analysis are then verified by comparing the
results of this analysis with estimates made in prior analyses.
4.1 Relative Densities and Distribution
The estimates of the number of unauthorized by census tract generated in this study were not
released. Rather, the estimates were visualized using dot density renderer in ArcGIS as a method
for communicating relative densities and concentrations of the unauthorized population. This
analysis concludes with relative density maps rather than releasing estimates for each census
tract, for two primary reasons:
1. Census tract boundaries do not have much meaning on their own in regards to this
analysis, as they are administrative boundaries that do not necessarily correspond with
neighborhood or community boundaries nor service areas for providing immigrant
services. In fact, in dense areas, several hundred or even thousands of census tracts could
be located in a particular service areas. When taken together, on the other hand, the total
number of unauthorized per census tract paints a picture of the landscape of the service
area.
2. Given that there is not even a consensus as to how many unauthorized people reside in
the entire United States, it is unreasonable to believe that the number of unauthorized
population can be estimated at as fine a geographic scale as the census tract level with
any real accuracy. Rather than try to present these frequencies, the total numbers are used
to present the relative density or distribution of the population.

49
4.1.1.1 Dot Density Renderer
The dot density renderer displays the number estimates of the unauthorized as a random dot
pattern within each census tract, where each dot represents a certain number of people. In order
to maintain density, as the zoom level increases, the number of people represented by each dot
diminishes, while the size of the dot stays the same. Using dot density renderer in Esri ArcGIS
Desktop is the preferred method for presenting the results for the following reasons:
• The optimal dot to person ratio can be manually adjusted to best communicate density
depending on the particular geography being displayed.
• By mapping the results, the distribution patterns and clusters of high numbers of
unauthorized become apparent. This would be difficult to determine looking at a table of
estimates alone.
• The results could in the future be combined with other potentially relevant infrastructure
information for planning purposes, such as accessibility by public transportation or
existing physical office locations of service providers.
4.1.1.2 Maps of Relative Density
In Figure 8, the density of the unauthorized population is displayed for the entire United States.
In Figure 8 one dot represents 1,500 people. Because the density of the unauthorized population
varies greatly, the dot density map is not particularly informative at this level.

50

Unauthorized Population by Census Tract in the United States. 1 dot = 1,500 people Figure 8

Figure 9 starts to show areas of density in California. In Figure 9, one dot represents 500
people. While this is potentially useful for state level planning and implementation, that is not
the goal of this analysis.

Unauthorized Population by Census Tract in California. 1 dot = 500 people Figure 9

51

The applicability of the analysis for local level planning starts to become apparent by looking
at Figure 10 (one dot represents one-hundred people) and even more so with Figure 11 (one dot
represents fifty people). Although only four maps are presented here, using this methodology and
the dot density renderer scheme, maps could be made for virtually any geography in the 48 states
and Washington, D.C.

Unauthorized Population by Census Tract in Los Angeles County. 1 dot = 100 people Figure 10

52

Unauthorized Population by Census Tract in Los Angeles. 1 dot = 50 people Figure 11

4.2 Model Performance and Verification of Results
The primary method for drawing conclusions about the accuracy of the estimates produced in
this analysis, was to compare the results of this analysis to those of other studies. Specifically,
the census tract level estimates produced in this analysis were summed up to various geographies
and compared to estimates made for those geographies by Warren and Warren, the Public Policy
Institute of California (PPIC) and the USC Center for the Study of Immigrant Integration (CSII).
In the following sections, the estimates generated in this study are referred to as “Fischer”
estimates.
4.2.1 Comparison with State Level Estimates Generated by Warren and Warren
The census tract level estimates generated in this analysis were summed by state and compared
to those estimates generated by Warren (2013) using the residual method. The state level

53
estimates generated by Warren are the very same estimates upon which this analysis was based.
The Warren estimates were the numerator in the dependent variable (percent of the unauthorized
out of total foreign born) for the state level regression analyses conducted in this study. The
absolute percent differences of estimates generated in this study were compared to those
generated by Warren were calculated by taking the absolute value of the following equation:
(Warren estimate - Fischer estimate) / Warren estimate. The results are presented below in Table
15 and Figure 12.

54
Absolute % Difference Between Estimates Table 15
Unauthorized population estimates by… absolute % difference
of Fischer estimate
from Warren estimate
Warren Fischer
California 3,059,069 3,074,782 1
Georgia 395,838 391,721 1
Louisiana 51,467 52,021 1
Mississippi 23,807 24,186 2
Idaho 34,183 33,529 2
Iowa 46,373 45,150 3
Indiana 103,268 100,294 3
Wisconsin 81,988 79,384 3
Nebraska 39,494 40,832 3
Arizona 343,887 327,291 5
Virginia 264,453 277,257 5
Oregon 134,817 127,897 5
Missouri 70,031 73,651 5
New Mexico 80,317 75,570 6
Nevada 184,848 202,035 9
North Carolina 354,355 320,780 9
South Carolina 99,470 88,615 11
Texas 1,612,281 1,429,830 11
Kansas 72,618 64,332 11
Washington 255,464 284,721 11
Tennessee 130,475 115,481 11
Florida 988,384 1,106,241 12
Maryland 215,259 242,902 13
Colorado 207,881 181,054 13
Wyoming
*
2,945 2,550 13
Illinois 598,574 518,163 13
Delaware 21,337 24,414 14
Arkansas 64,789 54,787 15
Montana
*
793 924 17
Oklahoma 87,584 72,490 17
Utah 102,534 84,840 17
Washington, DC 23,006 18,962 18
Connecticut 112,595 133,308 18
Minnesota 102,516 121,538 19
New Jersey 416,144 494,469 19
Alabama 84,291 67,223 20
Kentucky 43,809 52,973 21
Ohio 98,564 122,933 25
Rhode Island 27,985 35,355 26
Massachusetts 202,790 257,687 27
New York 756,996 1,052,310 39
Pennsylvania 148,215 208,818 41
Michigan 91,766 149,391 63
Maine
*
2,024 3,645 80
West Virginia
*
818 1,814 122
South Dakota
*
953 2,166 127
New Hampshire
*
2,047 5,181 153
North Dakota
*
425 1,209 185
Vermont
*
298 1,693 468
*
states with unauthorized population <3,000 persons as estimated by Warren (2013)

55

The states with the largest absolute percentage difference from the Warren estimates were the
six out of eight states with immigrant populations less than 3,000, indicating that this study’s
analysis method may not be suitable for states with such low unauthorized immigrant
populations. While the actual absolute difference between the estimates was between 784 and
3,134 persons for those states with <3,000 unauthorized persons, this equated to an 80 to 468
percent difference from the original Warren estimates. For example, the Warren estimate for
Vermont is 298 persons, while this study estimated 1,693 persons. This equates to a difference of
1,395 people or 468 percent of the total Warren estimate. Of these 8 states, Wyoming and
Montana performed moderately, with differences from the Warren estimates being 13 and 17
percent respectively. Excluding the six “low-population” states that performed very poorly, of
the remaining forty-three states and D.C. included in this analysis the results varied:
Very good. Sixteen states had less than 10 percent difference from the Warren estimates,
with California’s estimate being the best with a less than 1 percent difference, followed
closely by Georgia, Louisiana, Mississippi, and Idaho (all less than 2 percent).
Good. Nineteen states performed well with differences from 10-20 percent of the Warren
estimates.
Moderate: Five states performed moderately with 20-30 percent differences from the
Warren estimates.
Poor: Four states performed poorly with differences of over 39 percent from the Warren
estimates, with Michigan having the largest percent difference, 62.8 percent.
Each state’s absolute percent difference from the Warren estimates was mapped in order to
identify spatial patterns. Based on the absolute value measures, Figure 12, illustrates that the

56
model performed well in the western United States. While, seven out of nine states that
performed the worst (with an absolute difference of 31 percent or higher) were located in the
northeast of the United States.

Absolute % Difference from Warren Estimates Figure 12
4.2.2 Comparing Results to Independent Sub-state Estimates for California
The results were further verified at the county level in California, where independent estimates
of the unauthorized population have been released by the PPIC and CSII. In order to compare the
results of this study with those of PPIC and CSII, the census tract level estimates were summed
to correspond with the county areas for which PPIC and/or CSII estimates have been released.
Table 16 shows the results of the comparison, with the estimates generated in this study labeled
as “Fischer” estimates.

+

57
Absolute % Difference Between Estimates of the Unauthorized by Region in CA Table 16

Unauthorized population estimates
Absolute % difference of
Fischer estimate from…
Fischer
(2006-10)
CSII
(2009-11)
PPIC
(2008)
CSII PPIC
EAST BAY (Alameda & Contra Costa
Counties)
201,935 153,910 203,000 31 1
INLAND EMPIRE (San Bernardino
and Riverside Counties)
290,473 259,130 296,000 12 2
ORANGE COUNTY 274,677 236,569 289,000 16 5
SILICON VALLEY (Santa Clara and
San Mateo Counties)
249,168 173,815 235,000 43 6
CENTRAL VALLEY (Fresno, Kern,
Kings, Madera, Merced, San Joaquin,
Stanislaus, and Tulare Counties)
286,978 331,584 260,000 13 10
BAY AREA (Alameda, Contra Costa,
Marin, Napa, Santa Clara, San Mateo,
and San Francisco Counties)
552,499 386,947 498,000 43 11
LOS ANGELES COUNTY 1,081,991 892,081 916,000 21 18
SACRAMENTO METRO (El Dorado,
Placer, Sacramento, Sutter, Yolo, and
Yuba Counties)
118,398 83,480
not
available
42
not
available

Looking at the regional estimates in Table 16, it appears that the estimates produced in this
study are comparable to those produced by PPIC with a 1–18 percent absolute difference. On the
other hand, there is a 12–42 percent difference from the estimates of CSII. It is important to note
that similar to the methodology of this study, PPIC used the Warren estimates as the basis of
their estimates, while CSII does not (For more detailed information about their estimates, see
section: Residual Method Combined With Other Methods For Sub-state Estimates) (Pastor and
Marcelli 2013; Hill and Johnson 2011). Another reason that the estimates generated in this study
may differ from those of other studies is because the study period differs.
Because this analysis is focused on the relative densities or the distribution of the
unauthorized population, another test of the validity of the results was conducted by looking at
the estimated differences of the distribution of the unauthorized population between studies. The
first step was to calculate the distribution or percent, rather than the frequencies, of the

58
unauthorized population by region in the state of California. This was calculated by dividing
each regional estimate by the corresponding total state estimate. For example, in the case of
CSII, all regional estimates were divided by CSII’s estimate for the state of California,
2,654,752. The results being the distribution or percent of the total unauthorized population by
region across the state of California. The results are compared in Table 17 and 18 with the
estimates generated in this study labeled as “Fischer” estimates.
Estimates of the Total Unauthorized Population in CA Table 17
CSII PPIC Fischer
California 2,654,752 2,876,000 3,074,782

Differences in the Distribution of Unauthorized Population by Region in CA Table 18

% of unauthorized by region as
estimated by…
Percentage point difference
from Fischer results
COUNTY CSII PPIC Fischer CSII PPIC
ORANGE County 8.9 10.0 8.9 0.0 -1.1
INLAND EMPIRE (San
Bernardino and Riverside
Counties)
9.8 10.3 9.4 -0.3 -0.8
EAST BAY (Alameda &
Contra Costa Counties)
5.8 7.1 6.6 0.8 -0.5
SILICON VALLEY (Santa
Clara and San Mateo
Counties)
6.5 8.2 8.1 1.6 -0.1
CENTRAL VALLEY (Fresno,
Kern, Kings, Madera, Merced,
San Joaquin, Stanislaus, and
Tulare Counties)
12.5 9.0 9.3 -3.2 0.3
BAY AREA (Alameda, Contra
Costa, Marin, Napa, Santa
Clara, San Mateo, and San
Francisco Counties)
14.6 17.3 18.0 3.4 0.7
LOS ANGELES County 33.6 31.8 35.2 1.6 3.3
SACRAMENTO METRO
(El Dorado, Placer,
Sacramento, Sutter, Yolo, and
Yuba Counties)
3.1

not
available
3.9 0.7 not available

While the estimate of the unauthorized population made in this analysis varied by 12–42
percent from those of CSII when comparing the number estimates of the unauthorized, the

59
differences in distribution of the unauthorized varied by 0.02–3.4 percentage points. These
results are encouraging because while the estimates generated in this analysis vary from those of
PPIC and CSII, the differences between this study and the other two leading studies in the
percent of the unauthorized by regions in California is no more than 3.4 percentage points (Bay
Area).

60
CHAPTER 5: CONCLUSIONS
While the validation method indicates that the methodology generated in this study may be an
appropriate analysis method for estimating the unauthorized population at the census tract level,
it is worthwhile to discuss some limitations of this analysis. This chapter begins with the
weaknesses, challenges and limitations of this analysis method, focusing on limitations around
data availability. The chapter continues with ideas for future research, including suggestions on
refining the methodology as well as the need for greater verification of results.
5.1 Weaknesses, Challenges, Limitations and Next Steps
There were a number of challenges in the analysis, including lack of available data, missing data
and data uncertainty, as well as concerns of model accuracy, and difficulty in verifying reliability
of the methodology and overall results of the analysis. Additionally, there are a number of ways
the research presented in this report could be continued in order to strengthen and further verify
the results. Lastly, the visual display of the results could be refined and presented in a way that
allows users to interact and query the results based on their area of interest.
5.1.1 Missing Data and Data Uncertainty
One weakness of the analysis is the number of census tracts with missing demographic data.
Estimates could not be calculated for census tracts with missing variables. Due to missing data,
no estimate were generated for 4,420 census tracts, roughly 6 percent of the 72,539 census tracts
within the geography of this analysis. Fourteen states had over 10 percent of census tracts where
no estimates were generated due to at least one missing demographic variable. The missing
predictions could greatly change the impression of the visual patterns in the analysis and
therefore changing the interpretation of the results. Additionally, the census tracts with missing

61
variables could have greatly changed the interpretation of the verification method, which
involved summing all of the census tract estimates. That being said, it is hypothesized that a
great many of census tracts with missing demographic variables are missing because of very low
numbers of foreign-born (and therefore likely low numbers or no unauthorized), but this may not
exclusively be the case. See Appendices, Table 23, for the percentage of census tracts with
missing variables by state.
Additionally, all of the variables used in the analysis were estimates rather than known
counts. This is unavoidable, given that no known counts exist of the unauthorized population. As
previously explained, the Warren estimates of the unauthorized population are made using the
residual method. The ACS data used to generate the component scores is also an estimate, albeit
statistically sound, based off a survey of a subset of the population (U.S. Department of
Commerce 2008).
5.1.2 Ecological Fallacy
The estimates of the rate of the unauthorized out of the total foreign-born for each census tract
were based on the relationship of the independent variables at the state level. The state level
relationships are assumed to be the same at the census tract level in order to make census tract
level estimates. The ecological fallacy being that inferences about the group at the census tract
level are deduced from correlations of the variables at the state level. While this is not ideal, the
assumptions about the census tract level relationships between variables was necessary given the
lack of data available at the census tract level.
5.1.3 Refinement of Independent Variables
While the choice of a dependent variable is quite limited by data available (or lack thereof), there
are numerous possibilities on independent input data, particularly demographic data input in the

62
PCA. In future analysis, it is recommended that greater exploration be conducted to determine
which variables to include in the analysis. There are two weaknesses to the PCA analysis that are
further discussed in the next sections: (1) a concern of the multicollinearity and singularity of the
data, and (2) not all characteristics of the unauthorized were captured in the input data.
5.1.3.1 Multicollinearity and Singularity of PCA Input Data
As previously mentioned, while there were a number of indicators that PCA was an appropriate
analysis method, there were other indicators that there may be a problem with the
multicollinearity of singularity of the data, namely that some variables were measuring
ostensibly the same thing. In a future analysis, including or omitting variables should be
considered, particularly those variables that were related to year of entry in the U.S. and age.
5.1.3.2 Differing Characteristics of the Unauthorized
There is also the issue that the unauthorized population is not uniform. While similar
demographic characteristics may be used to describe the majority of the unauthorized population,
in reality, every person has varying combinations of demographic characteristics. There are
certainly characteristics, other than the ones included in this analysis, which would better
differentiate different groups of unauthorized.
5.1.4 Improved Method for Verifying the Results
Further verification of the results is necessary to determine the reliability of the methodology
outlined in this report. As previously discussed, this report has only verified the results at the
sub-state level for the state of California and no verification has been conducted for the accuracy
of the estimates below the county level. A similar method of comparing the results of this report
to that of prior studies, could be conducted for other states where prior studies have produced

63
sub-state estimates. To verify the results at a finer geography than that of other studies (such as
the census tract level, where it is believed that no other estimates exist), a survey that asks about
legal status could be conducted.
5.1.5 Sensitivity and Reliability Analysis
Another method for verifying the robustness of the analysis method would be to conduct a
sensitivity analysis to see how sensitive the analysis is to changes in the analysis inputs. There
are countless ways that the analysis or the input data could be adjusted to conduct a sensitivity
analysis. One idea is to fill in a number in place of the missing variables to see how big an
influence the missing variables may have on the results. Similarly, reliability analysis for the
PCA could be conducted in SPSS.
5.1.6 Refine Display of Results
5.1.6.1 Use of Masking in the Dot Density Renderer
The results of this analysis is an estimate of total unauthorized population for each census tract in
the United States, visualized in ArcGIS Desktop using dot density renderer to display relative
densities. The dot density rendering method could be improved upon through the use of masking.
Through masking, the area for which dots can be rendered is restricted within the polygon
boundaries (census tracts in the case of this analysis) to those that may be inhabited. No dots
would be rendered in areas that are within the census tract boundaries but are clearly
uninhabited, such as bodies of water or national park land. By removing uninhabited lands from
the rendering area, the dot density renderer more accurately displays relative density.

64
5.1.6.2 Interactive Web Application
Creating an interactive web application could increase the accessibility of the results of this
research. A web application would allow users to explore the results of the analysis for their
geography of interest, without requiring the manual adjustment of density display properties.
Ideally, the dot density renderer would automatically adjust the dot size and density display
properties to best communicate distribution and relative densities in the selected area.
The optimal density display properties would need to be more nuanced than those that are
standard in ArcGIS Desktop that maintain density by making adjustments based solely on zoom
level, but would also require consideration of the average number of unauthorized in the
geography being viewed. While these settings can be manually adjusted in ArcGIS Desktop,
automating this process and making it available online may improve the access and therefore
usefulness of this tool as an applied research product.
5.2 Lessons Learned and Potential Impacts
Overall, the results of this analysis indicate that the method designed in this study may be a
viable means for estimating the unauthorized at the neighborhood level, at least in certain
geographies, such as the West Coast. That being said, this is a first attempt at an entirely new
methodology, which will undoubtedly require both refinement of the method and greater
verification of the results before being useful for planning purposes. Now is the time to start
investigating methods such as this one, so that if and when immigration reform occurs, those on
the ground providing services to the unauthorized will have the information needed to effectively
and efficiently process potentially upwards of 8 million people.

65
REFERENCES
Baker, Bryan. 2010. “Naturalization Rates among IRCA Immigrants: A 2009 Update.”
Washington, D.C.: Office of Immigration Statistics, Department of Homeland Security.
Retrieved http://www.dhs.gov/xlibrary/assets/statistics/publications/irca-natz-fs-2009.pdf
(last accessed 11 August 2014).
Baker, Bryan, and Nancy Rytina. 2013. “Estimates of the Unauthorized Immigrant Population
Residing in the United States: January 2012.” Washington, D.C.: Office of Immigration
Statistics, Department of Homeland Security. Retrieved
http://www.dhs.gov/sites/default/files/publications/ois_ill_pe_2012_2.pdf (last accessed 10
August 2014).
Bohn, Sarah. 2009. “New Patterns of Immigrant Settlement in California.” San Francisco, CA:
Public Policy Institute of California. Retrieved
http://www.ppic.org/content/pubs/report/R_709SBR.pdf (last accessed 10 August 2014).
Chan, Alex, Joanna Kabat, and Jesse Reyes. 2013. “Implementing Immigration Reform in Los
Angeles: Lessons from DACA” Los Angeles, CA: Tomas Rivera Policy Institute,
forthcoming.
Esri. Released 2012. ArcGIS Desktop for Windows, Version 10.1. Redlands, CA: Esri.
Field, Andy P. 2013. “Chapter 17: Exploratory Factor Analysis” In Discovering Statistics using
IBM SPSS Statistics. London; Thousand Oaks, CA: SAGE Publications. Retrieved
http://www.sagepub.com/field4e/study/resources.htm (last accessed 15 August 2014).

66
Fortuny, Karina, Randy Capps, and Jeffrey S. Passel. 2007. “The Characteristics of
Unauthorized Immigrants in California, Los Angeles County, and the United States.”
Washington, DC: The Urban Institute. Retrieved
http://www.urban.org/UploadedPDF/411425_Characteristics_Immigrants.pdf (last accessed
10 August 2014).
Hill, Laura, and Hans Johnson. 2011. “Unauthorized Immigrants in California: Estimates for
Counties.” San Francisco, CA: Public Policy Institute of California. Retrieved
http://www.ppic.org/main/publication.asp?i=986 (last accessed 10 August 2014).
Hill, Laura, and Joseph Hayes. 2013. “Unauthorized Immigrants.” San Francisco, CA: Public
Policy Institute of California. Retrieved
http://www.ppic.org/main/publication_show.asp?i=818 (last accessed 10 August 2014).
Hoefer, Michael, Nancy Rytina, and Bryan Baker. 2011. “Estimates of the Unauthorized
Immigrant Population Residing in the United States: January 2010.” Washington, D.C.:
Office of Immigration Statistics, Department of Homeland Security. Retrieved
http://www.dhs.gov/xlibrary/assets/statistics/publications/ois_ill_pe_2010.pdf (last accessed
10 August 2014).
———. 2012. “Estimates of the Unauthorized Immigrant Population Residing in the United
States: January 2011.” Washington, D.C.: Office of Immigration Statistics, Department of
Homeland Security. Retrieved
http://www.dhs.gov/xlibrary/assets/statistics/publications/ois_ill_pe_2011.pdf (last accessed
10 August 2014).

67
IBM Corp. Released 2012. IBM SPSS Statistics for Windows, Version 21.0. Armonk, NY: IBM
Corp.
Judson, Dean H., David A. Swanson. 2011. Estimating Characteristics of the Foreign-Born by
Legal Status: An Evaluation of Data and Methods. Vol. 2. Dordrecht; New York: Springer
Netherlands. doi:10.1007/978-94-007-1272-0.
Laerd Statistics. 2013. “Principal components analysis in SPSS.” https://statistics.laerd.com (last
accessed 11 August 2013).
Marshall, Serena, and Jon Garcia. 2014. “Obama Announces Unilateral Action on Immigration.”
abc News, June 30. Accessed 11 August 2014. http://abcnews.go.com/Politics/president-
obama-announces-unilateral-action-immigration/story?id=24368748
Minnesota Population Center. National Historical Geographic Information System: Version 2.0.
Minneapolis, MN: University of Minnesota 2011. Retrieved https://www.nhgis.org (last
accessed 10 August 2014).
Myers, Laura. 2014. “Reid: No chance of comprehensive immigration reform this summer.” Las
Vegas Review-Journal, July 2. Retrieved http://www.reviewjournal.com/politics/reid-no-
chance-comprehensive-immigration-reform-summer (last accessed 10 August 2014).
Nakamura, David. 2014. “Obama readies executive action on immigration.” The Washington
Post, August 1. Accessed 11 August 2014. http://www.washingtonpost.com/politics/obama-
readies-executive-action-to-legalize-millions-of-undocumented-
immigrants/2014/08/01/222ae2e8-18f8-11e4-85b6-c1451e622637_story.html

68
O'Rourke, Norm, and Larry Hatcher. 2013. A step-by-step approach to using SAS for factor
analysis and structural equation modeling, second edition. Cary, N.C: SAS Institute Inc.
Passel, Jeffrey S. 2013. “Unauthorized Immigrants: How Pew Research Counts Them and What
We Know About Them.” Washington, DC: Pew Hispanic Center. Retrieved
http://www.pewresearch.org/2013/04/17/unauthorized-immigrants-how-pew-research-
counts-them-and-what-we-know-about-them (last accessed 10 August 2014).
Passel, Jeffrey S., and D’Vera Cohn. 2009. “A Portrait of Unauthorized Immigrants in the United
States.” Washington, DC: Pew Hispanic Center. Retrieved
http://www.pewhispanic.org/2009/04/14/iii-demographic-and-family-characteristics/ (last
accessed 10 August 2014).
———. 2012. “Unauthorized Immigrants: 11.1 Million in 2011.” Washington, DC: Pew
Hispanic Center. Retrieved http://www.pewhispanic.org/2012/12/06/unauthorized-
immigrants-11-1-million-in-2011 (last accessed 10 August 2014).
Passel, J.S., D’Vera Cohn, and Ana Gonzalez-Barrera, A. 2013. “Population Decline of
Unauthorized Immigrants Stalls, May Have Reversed.” Washington, DC: Pew Hispanic
Center. Retrieved http://www.pewhispanic.org/2013/09/23/population-decline-of-
unauthorized-immigrants-stalls-may-have-reversed (last accessed 10 August 2014).
Pastor, Manuel, and Enrico A. Marcelli. 2013. “What's at Stake for the State: Unauthorized
Californians, Immigration Reform, and Our Future Together.” Los Angeles, CA: Center for
the Study of Immigrant Integration. Retrieved
http://csii.usc.edu/documents/whats_at_stake_for_the_state.pdf (last accessed 10 August

69
2014).
Rob Paral and Associates. 2006. “Unauthorized Immigrants in Congressional Districts.”
Retrieved www.robparal.com/MapPage.html?map=14&type=G. Accessed May 5, 2011 (last
accessed 10 August 2014) (Google Earth plug-in required).
Singer, Audrey. 2004. “The Rise of New Immigrant Gateways.” New Yoek, NY: The Brookings
Institution. Retrieved
http://www.brookings.edu/~/media/research/files/reports/2004/2/demographics%20singer/20
040301_gateways.pdf (last accessed 11 August 2014).
Terrazas, Aaron. 2010. “Mexican Immigrants in the United States,”
http://www.migrationpolicy.org, last modified 22 February 2010,
http://www.migrationpolicy.org/article/mexican-immigrants-united-states-0 (last accessed
11 August 2014).
U.S. Congress, Senate. 2013. Border Security, Economic Opportunity, and Immigration
Modernization Act. S.744. 113th Cong., 1st sess., June 27.
U.S. Congressional Budget Office. 2013a. Cost Estimate: S.744 Border Security, Economic
Opportunity, and Immigration Modernization Act. Washington, DC: Congressional Budget
Office, 2013. Retrieved http://www.cbo.gov/sites/default/files/cbofiles/attachments/s744.pdf
(last accessed 28 August 2014)
———. 2013b. The Economic Impact of S. 744, the Border Security, Economic Opportunity,
and Immigration Modernization Act. Washington, DC: Congressional Budget Office, 2013.
Retrieved http://www.cbo.gov/sites/default/files/cbofiles/attachments/44346-Immigration.pdf

70
(last accessed 28 August 2014)
U.S. Congressional Research Service. 2013. Immigration Legislation and Issues in the 113th
Congress. By Andorra Bruno, Michael John Garcia, William A. Kandel, Margaret Mikyung
Lee, Marc R. Rosenblum, Alison Siskin, and Ruth Ellen Wasem. CR R43320, Washington
DC: Library of Congress, Congressional Research Service, 2013. Retrieved
http://fas.org/sgp/crs/homesec/R43320.pdf (last accessed 10 August 2014)
U.S. Department of Commerce, Economics and Statistics Administration, U.S. Census Bureau.
2008. A Compass for Understanding and Using American Community Survey Data What
General Data Users Need to Know. Washington, D.C.: United States Government Printing
Office, 2008. Retrieved
http://www.census.gov/acs/www/Downloads/handbooks/ACSGeneralHandbook.pdf (last
accessed 10 August 2014)
Warren, Robert, and John Robert Warren. 2013. "Unauthorized Immigration to the United
States: Annual Estimates and Components of Change, by State, 1990 to 2010." International
Migration Review 47 (2): 296-329. doi:10.1111/imre.12022.
“What’s on the Menu? Immigration Bills Pending in the House of Representatives in 2014,”
immigrationpolicy.org, last modified 26 March 2014,
http://www.immigrationpolicy.org/sites/default/files/docs/summary_of_house_bills_final_4-
15-14.pdf (last accessed 10 August 2014)

71
APPENDICES
Correlation Matrix: First PCA Table 19

Entered
2000 or
later
Entered
1980 or
before
Speak
English
'less
than
well'
Age
65+
Not a
U.S.
Citize
n
Less than
high
school
Bachelor's
degree+
Median
income
Born in
Mexico
Born in
Central
America
Median
Age
Income
in the
past 12
months
below
poverty
level
Hispanic
or Latino
Correlation Entered
2000 or
later
1.000 -.666 .094 -.650 .631 .072 .063 -.265 .149 .223 -.824 .368 .120
Entered
1980 or
before
-.666 1.000 -.594 .948 -.772 -.473 .239 .106 -.456 -.339 .889 -.340 -.524
Speak
English
'less than
well'
.094 -.594 1.000 -.646 .701 .889 -.790 -.419 .794 .138 -.469 .539 .888
Age 65+
-.650 .948 -.646 1.000 -.866 -.605 .374 .210 -.611 -.300 .925 -.448 -.622
Not a U.S.
Citizen
.631 -.772 .701 -.866 1.000 .748 -.555 -.477 .792 .209 -.848 .687 .785
Less than
high
school
.072 -.473 .889 -.605 .748 1.000 -.861 -.532 .883 .075 -.478 .676 .887
Bachelor's
degree+
.063 .239 -.790 .374 -.555 -.861 1.000 .642 -.774 .113 .267 -.606 -.781
Median
income
-.265 .106 -.419 .210 -.477 -.532 .642 1.000 -.568 .360 .323 -.886 -.403
Born in
Mexico
.149 -.456 .794 -.611 .792 .883 -.774 -.568 1.000 -.139 -.523 .707 .873
Born in
Central
America
.223 -.339 .138 -.300 .209 .075 .113 .360 -.139 1.000 -.165 -.241 .229
Median
Age
-.824 .889 -.469 .925 -.848 -.478 .267 .323 -.523 -.165 1.000 -.514 -.452

72
Income in
the past 12
months
below
poverty
level
.368 -.340 .539 -.448 .687 .676 -.606 -.886 .707 -.241 -.514 1.000
.549

Hispanic
or Latino
.120 -.524 .888 -.622 .785 .887 -.781 -.403 .873 .229 -.452 .549 1.000

Anti-image Correlation, Final PCA Table 20

Entered
2000 or
later
Entered
1980 or
before
Speak
English
'less than
well'
Age 65+
Not a
U.S.
Citize
n
Less
than high
school
Bachelor's
degree+
Median
income
Born in
Mexico
Median
Age
Income in the
past 12
months below
poverty level
Hispanic
or Latino
Entered 2000
or later
.682
a

Entered 1980
or before
.775
a

Speak English
'less than well'
.870
a

Age 65+ .902
a

Not a U.S.
Citizen
.887
a

Less than high
school
.831
a

Bachelor's
degree+
.819
a

Median
income
.645
a

Born in
Mexico
.851
a

Median Age .777
a

Income in the
past 12
months below
poverty level
.745
a

Hispanic or
Latino
.845
a

Measures of Sampling Adequacy (MSA)

73

Correlation Matrix: Final PCA Table 21

Born in
Mexico
Entered
2000 or
later
Entered
1980 or
before Age 65+
Less than
high school
Bachelor’s
degree +
Median
income
Speak
English 'less
than well'
Not a U.S.
citizen Median age
Income in
the past 12
months
below
poverty
level
Hispanic or
Latino
Correlation
a
Born in
Mexico
1.000 .149 -.456 -.611 .883 -.774 -.568 .794 .792 -.523 .707 .873
Entered
2000 or
later
.149 1.000 -.666 -.650 .072 .063 -.265 .094 .631 -.824 .368 .120
Entered
1980 or
before
-.456 -.666 1.000 .948 -.473 .239 .106 -.594 -.772 .889 -.340 -.524
Age 65+ -.611 -.650 .948 1.000 -.605 .374 .210 -.646 -.866 .925 -.448 -.622
Less than
high school
.883 .072 -.473 -.605 1.000 -.861 -.532 .889 .748 -.478 .676 .887
Bachelor’s
degree +
-.774 .063 .239 .374 -.861 1.000 .642 -.790 -.555 .267 -.606 -.781
Median
income
-.568 -.265 .106 .210 -.532 .642 1.000 -.419 -.477 .323 -.886 -.403
Speak
English
'less than
well'
.794 .094 -.594 -.646 .889 -.790 -.419 1.000 .701 -.469 .539 .888
Not a U.S.
citizen
.792 .631 -.772 -.866 .748 -.555 -.477 .701 1.000 -.848 .687 .785
Median
age
-.523 -.824 .889 .925 -.478 .267 .323 -.469 -.848 1.000 -.514 -.452
Income in
the past 12
months
below
poverty
level
.707 .368 -.340 -.448 .676 -.606 -.886 .539 .687 -.514 1.000 .549
Hispanic or
Latino
.873 .120 -.524 -.622 .887 -.781 -.403 .888 .785 -.452 .549 1.000
a. Determinant = 3.797E-010

74
Reproduced Correlations and Residuals: Final PCA Table 22

Born in
Mexico
Entered
2000 or
later
Entered
1980 or
before
Age
65+
Less than
high
school
Bachelor’
s degree
+
Median
income
Speak
English
'less than
well'
Not a
U.S.
citizen
Median
age
Income
in the
past 12
months
below
poverty
level
Hispanic
or Latino
Reproduced
Correlation
Born in
Mexico
.780
a
.414 -.653 -.741 .781 -.657 -.532 .754 .831 -.690 .677 .762
Entered 2000
or later
.414 .220
a
-.347 -.393 .414 -.348 -.282 .400 .441 -.366 .359 .404
Entered 1980
or before
-.653 -.347 .547
a
.620 -.654 .550 .445 -.632 -.696 .578 -.567 -.638
Age 65+ -.741 -.393 .620 .703
a
-.741 .623 .505 -.716 -.789 .655 -.643 -.724
Less than
high school
.781 .414 -.654 -.741 .782
a
-.657 -.532 .755 .832 -.691 .678 .763
Bachelor’s
degree +
-.657 -.348 .550 .623 -.657 .553
a
.448 -.635 -.699 .581 -.570 -.642
Median
income
-.532 -.282 .445 .505 -.532 .448 .362
a
-.514 -.566 .470 -.462 -.520
Speak
English 'less
than well'
.754 .400 -.632 -.716 .755 -.635 -.514 .730
a
.804 -.667 .655 .737
Not a U.S.
citizen
.831 .441 -.696 -.789 .832 -.699 -.566 .804 .885
a
-.735 .721 .812
Median age -.690 -.366 .578 .655 -.691 .581 .470 -.667 -.735 .610
a
-.599 -.674
Income in the
past 12
months
below
poverty level
.677 .359 -.567 -.643 .678 -.570 -.462 .655 .721 -.599 .588
a
.662
Hispanic or
Latino
.762 .404 -.638 -.724 .763 -.642 -.520 .737 .812 -.674 .662 .745
a

Residual
b

Born in
Mexico
-.265 .198 .129 .102 -.117 -.036 .039 -.038 .167 .030 .110
Entered 2000
or later
-.265 -.319 -.257 -.342 .412 .017 -.307 .190 -.458 .009 -.284
Entered 1980
or before
.198 -.319 .328 .181 -.311 -.339 .037 -.077 .312 .227 .114
Age 65+ .129 -.257 .328 .136 -.249 -.295 .070 -.077 .270 .195 .102

75
Less than
high school
.102 -.342 .181 .136 -.204
-6.273E-
005
.133 -.084 .213 -.002 .124
Bachelor’s
degree +
-.117 .412 -.311 -.249 -.204 .195 -.155 .145 -.313 -.035 -.139
Median
income
-.036 .017 -.339 -.295
-6.273E-
005
.195 .095 .089 -.147 -.425 .117
Speak
English 'less
than well'
.039 -.307 .037 .070 .133 -.155 .095 -.103 .198 -.116 .150
Not a U.S.
citizen
-.038 .190 -.077 -.077 -.084 .145 .089 -.103 -.114 -.034 -.027
Median age .167 -.458 .312 .270 .213 -.313 -.147 .198 -.114 .085 .222
Income in the
past 12
months
below
poverty level
.030 .009 .227 .195 -.002 -.035 -.425 -.116 -.034 .085 -.113
Hispanic or
Latino
.110 -.284 .114 .102 .124 -.139 .117 .150 -.027 .222 -.113
Extraction Method: Principal Component Analysis.
a. Reproduced communalities
b. Residuals are computed between observed and reproduced correlations. There are 54 (81.0%) nonredundant residuals with absolute values greater than 0.05.

76
Percent of Census Tracts with Missing Variables by State Table 23
State
%
missing
variables
California 0
Oregon 1
Nevada 1
Washington 1
Florida 1
Connecticut 1
New Jersey 1
Rhode Island 1
Massachusetts 1
New Hampshire
*
1
Arizona 2
Texas 2
Utah 2
New York 2
Vermont
*
2
Idaho 3
New Mexico 3
Maryland 3
Colorado 3
Delaware 3
District of Columbia 3
Minnesota 3
Wisconsin 4
Maine
*
4
Virginia 7
North Carolina 7
Illinois 7
Georgia 8
Kansas 8
Wyoming
*
8
Michigan 8
Nebraska 9
Pennsylvania 9
Oklahoma 11
Iowa 12
Missouri 12
South Carolina 12
Indiana 13
Montana
*
13
Tennessee 14
Arkansas 15
Ohio 15
North Dakota
*
16
Louisiana 18
Alabama 18
Kentucky 18
Mississippi 20
South Dakota
*
20
West Virginia
*
29
* unauthorized population <3,000 (Warren and Warren 2013)

Abstract (if available)

Abstract An estimated 11.7 million unauthorized immigrants resided in the United States in 2012 according to the Pew Hispanic Center (Passel, Cohn, and Gonzalez-Barrera 2013). Reforming the U.S. immigration system is a clear policy priority for President Barack Obama, and an agenda item for the 113th Congress (U.S. Congressional Research Service 2013). Based on prior legislation, processing of immigrants for legalization is likely to be a complex and time consuming task, necessitating the involvement of nonprofit and public infrastructure. The goal of this study was to design a research methodology for estimating the unauthorized population at the census tract level, as a means for visually representing the relative densities of the unauthorized population in a way that would be useful for planning where to provide services for the unauthorized populations within a community. Using statistical methods, the relationships between the dependent and independent variables was defined at the state level. The state level relationships were then applied to census tract level data in order to make census tract estimates. The results of the analysis were displayed as relative densities using the dot density renderer in ArcGIS Desktop. The performance of this model was verified by comparing the results generated in this study to those of other studies. Based on this verification method, the performance of the model varied by geography, with the western states, in particular, California seeming to have performed the best. The states that appear to have performed the worst are primarily located in northeastern United States and include six out of the eight states with the lowest number of unauthorized persons (<3,000). Within California, between a 0.02 (Orange County) and 3.4 (Bay Area) percentage point difference was found when comparing the regional distribution estimated in this study with those of other studies.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The spatial effect of AB 109 (Public Safety Realignment) on crime rates in San Diego County

PDF

Effect of spatial patterns on sampling design performance in a vegetation map accuracy assessment

PDF

Analysis of future land use conflict with volcanic hazard zones: Mount Rainier, Washington

PDF

Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand

PDF

Selection of bridge location over the Merrimack River in southern New Hampshire: a comparison of site suitability assessments

PDF

Building a geodatabase design for American Pika presence and absence data

PDF

Estimating populations at risk in data-poor environments: a geographically disaggregated analysis of Boko Haram terrorism 2009-2014

PDF

Surface representations of rainfall at small extents: a study of rainfall mapping based on volunteered geographic information in Kona, Hawaii

PDF

Social media to locate urban displacement: assessing the risk of displacement using volunteered geographic information in the city of Los Angeles

PDF

Analyzing earthquake casualty risk at census block level: a case study in the Lexington Central Business District, Kentucky

PDF

Modeling burn probability: a Maxent approach to estimating California's wildfire potential

PDF

A comparison of urban land cover change: a study of Pasadena and Inglewood, California, 1992‐2011

PDF

Spatial distribution of the Nile crocodile (Crocodylus niloticus) in the Mariarano River system, Northwestern Madagascar

PDF

Calculating solar photovoltaic potential on residential rooftops in Kailua Kona, Hawaii

PDF

A model for placement of modular pump storage hydroelectricity systems

PDF

Increase in surface temperature and deep layer nitrate in the California Current: a spatiotemporal analysis of four-dimensional hydrographic data

PDF

Demonstrating GIS spatial analysis techniques in a prehistoric mortuary analysis: a case study in the Napa Valley, California

PDF

Modeling nitrate contamination of groundwater in Mountain Home, Idaho using the DRASTIC method

PDF

Integration of topographic and bathymetric digital elevation model using ArcGIS interpolation methods: a case study of the Klamath River Estuary

PDF

Bringing GIS to a small community water system

Asset Metadata

Creator Fischer, Anna Jane (author)

Core Title Preparing for immigration reform: a spatial analysis of unauthorized immigrants

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 09/10/2014

Defense Date 08/26/2014

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag ArcGIS,foreign born,immigrant,immigration,immigration reform,OAI-PMH Harvest,spatial analysis,unauthorized immigrant

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kemp, Karen K. (committee chair), Lee, Su Jin (committee member), Ruddell, Darren M. (committee member)

Creator Email ajfische@usc.edu,fischer.ajf@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-471706

Unique identifier UC11286972

Identifier etd-FischerAnn-2909.pdf (filename),usctheses-c3-471706 (legacy record id)

Legacy Identifier etd-FischerAnn-2909.pdf

Dmrecord 471706

Document Type Thesis

Format application/pdf (imt)

Rights Fischer, Anna Jane

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA