University of Southern California Dissertations and Theses

The geographic connotations of reincarceration: a spatial analysis of recidivism in Washington State

(USC Thesis Other)

The geographic connotations of reincarceration: a spatial analysis of recidivism in Washington State

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THE GEOGRAPHIC CONNOTATIONS OF REINCARCERATION:
A SPATIAL ANALYSIS OF RECIDIVISM IN WASHINGTON STATE
by
Robert P. Woodmark
A Thesis Presented to the
FACULTY OF THE USC DORNSIFE COLLEGE OF LETTERS, ARTS AND SCIENCES
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY
December 2023
Copyright 2023 Robert P. Woodmark

ii
Dedicated to the incarcerated individual who aspires to make right, and the formerly incarcerated
individuals fulfilling the promises that we made to ourselves. Never forget to extend an arm to
the ones coming after you

iii
Acknowledgements
I am grateful to my advisor, Dr. Elisabeth Sedano, for pushing me to strive for my best. I would
also like to acknowledge the Thesis Committee which included Dr. Robert Vos and Dr. An-Min
Wu. Your expertise in your respective fields paid dividends to my success. Your assistance in
this process is much appreciated.

iv
Table of Contents
Dedication....................................................................................................................................... ii
Acknowledgements........................................................................................................................iii
List of Tables ................................................................................................................................. vi
List of Figures............................................................................................................................... vii
Abbreviations............................................................................................................................... viii
Abstract.......................................................................................................................................... ix
Chapter 1 Introduction .................................................................................................................... 1
1.1 Background......................................................................................................................... 3
1.2 Study Area .......................................................................................................................... 4
1.3 Project Overview ................................................................................................................ 7
1.3.1 Temporal Scope ......................................................................................................... 8
1.3.2 Scale of Analysis........................................................................................................ 8
1.3.3 Primary Dataset.......................................................................................................... 8
1.3.4 Limitations of the Primary WADOC Dataset............................................................ 9
1.4 Document Overview ......................................................................................................... 11
Chapter 2 Literature Review......................................................................................................... 12
2.1 Social Disorganization Theory ......................................................................................... 12
2.2 Social Disorganization Theory in Crime and Recidivism Research................................. 13
2.3 Mathematical Methodology.............................................................................................. 14
2.4 Areal Unit of Aggregation ................................................................................................ 16
2.5 Overview of Previous Analytical Research on Recidivism.............................................. 17
2.5.1 Washington State Specific Studies on Recidivism.................................................. 18
2.5.2 Relevant Studies in Spatial Criminology................................................................. 22
Chapter 3 Methods........................................................................................................................ 25
3.1 Variable Selection and Data.............................................................................................. 26
3.1.1 Description of the Data ............................................................................................ 27
3.1.2 Limitations of the Data ............................................................................................ 29
3.2 Data Preprocessing............................................................................................................ 30
3.2.1 Data Conversion and Standardization...................................................................... 30
3.2.2 Handling Missing Values......................................................................................... 31
3.2.3 Dependent Recidivism Outcome Variable and Date of Recidivism........................ 31
3.2.4 County of Release .................................................................................................... 33
3.2.5 Data Normalization.................................................................................................. 33
3.3 Exploratory Data Analysis................................................................................................ 34
3.3.1 Baseline Descriptive Statistics................................................................................. 36
3.3.2 Univariate Data Visualization.................................................................................. 37

v
3.3.3 Data Transformation ................................................................................................ 37
3.3.4 Multivariate Data Visualization............................................................................... 38
3.3.5 Correlation Analysis – Pearson’s Correlation Coefficients..................................... 38
3.3.6 Analysis of Variance (ANOVA).............................................................................. 39
3.3.7 One Sample T-Test .................................................................................................. 39
3.3.8 Bonferroni Correction.............................................................................................. 40
3.3.9 Cohens d – Calculating Effect Size ......................................................................... 40
3.4 Exploratory Spatial Data Analysis.................................................................................... 41
3.4.1 Spatial Weights Matrix (SWM)............................................................................... 42
3.4.2 Moran's I Index for Spatial Autocorrelation ............................................................ 42
3.4.3 Getis-Ord Hot Spot Analysis................................................................................... 43
3.4.4 Interpretation of Findings ........................................................................................ 44
Chapter 4 Results.......................................................................................................................... 45
4.1 Results of Exploratory Data Analysis............................................................................... 45
4.1.1 Baseline descriptive statistics .................................................................................. 45
4.1.2 Power analysis ......................................................................................................... 48
4.1.3 Eliminating the counties with insufficient sample sizes.......................................... 51
4.1.4 Summary statistics................................................................................................... 51
4.1.5 Univariate Data Visualization.................................................................................. 53
4.1.5.1 Rate of recidivism........................................................................................... 53
4.1.5.2 The rate enrolled in public assistance per 100k population............................ 54
4.1.5.3 Lacking basic prose literacy skills per 100k population................................. 54
4.1.6 Data Transformation ................................................................................................ 55
4.1.7 Reevaluating Data Distribution for Normality ........................................................ 56
4.1.8 Multivariate Data Visualization and Pearson’s Correlation Coefficients................ 57
4.1.9 Single-Factor Analysis of Variance (ANOVA)....................................................... 58
4.1.10 Two-tailed one sample T-test and Cohen’s d ........................................................ 59
4.1.11 Conclusion ............................................................................................................. 61
4.2 Results of Exploratory Spatial Data Analysis................................................................... 61
4.2.1 Spatial Weights Matrix ............................................................................................ 61
4.2.2 Moran’s I Statistic – Measuring Global Spatial Autocorrelation ............................ 62
4.2.3 Getis-Ord Statistics - Local Cluster Analysis.......................................................... 63
4.3 Conclusion ........................................................................................................................ 65
Chapter 5 Discussion .................................................................................................................... 67
5.1 Significant Findings.......................................................................................................... 67
5.2 Methodological Barriers ................................................................................................... 70
5.3 Future Research ................................................................................................................ 73
References..................................................................................................................................... 76

vi
List of Tables
Table 1 WADOC facilities.............................................................................................................. 7
Table 2 Variables.......................................................................................................................... 27
Table 3 Crosstab of recidivism vs. total releases with derivative rates of recidivism by county. 47
Table 4 Results of the weighted average pooled d for effect size. ............................................... 49
Table 5 Summary statistics. .......................................................................................................... 52
Table 6 Results of the Shapiro-Wilks tests................................................................................... 57
Table 7 Scatterplot matrix with density plots and Pearson's correlation coefficients. ................. 58
Table 8 Results of the single factor ANOVA. .............................................................................. 58
Table 9 Results of the T-test and Cohen's d.................................................................................. 59
Table 10 Results of the Moran's I index for spatial autocorrelation on the recidivism rates........ 63

vii
List of Figures
Figure 1. Rate of Recidivism (Reconfinement) 2012- 2022........................................................... 6
Figure 2. Workflow diagram......................................................................................................... 26
Figure 3. An image of the raw dataset as received from WADOC .............................................. 30
Figure 4. Workflow diagram of the data normalization process .................................................. 34
Figure 5 Results of a power analysis to determine sample size.................................................... 50
Figure 6 Map of the counties that were removed from further analysis....................................... 51
Figure 7 Histogram and boxplot of the rate of recidivism by county........................................... 53
Figure 8 Histogram and boxplot of the rate enrolled in public assistance per 100k..................... 54
Figure 9 Histogram and boxplot of rate lacking basic prose literacy skills per 100k................... 55
Figure 10 Histogram of the log transformed variable................................................................... 55
Figure 11 Results of the fixed distance Getis-ord Gl* Hot Spot Analysis.................................... 64
Figure 12 Results of the K Nearest Neighbors SWM Getis-ord Gl* Hot Spot Analysis……...…67
Figure 13 Pie chart of statewide WADOC recidivism rates (2012-2022). ................................... 70

viii
Abbreviations
ACS American Community Survey
AUA Areal unit of aggregation
BJS Bureau of Justice Statistics
CGP Criminal geographic profiling
df Degrees of freedom
DOC Department of Corrections
EDA Exploratory data analysis
ESDA Exploratory spatial data analysis
GIST Geographic information science and technology
GLM Generalized linear model
GWR Geographically weighted regression
HBLR Hierarchical binary logistic regression
ICE Influence on conditional expectation
LISA Local indicators of spatial association
MAUP Modifiable Areal Unit Problem
NIJ National Institute of Justice
OAA Offender Accountability Act
RMI Risk management identification
SPLOM Scatterplot matrix
SWM Spatial weights matrix
WDQ Weighted displacement quotient
WADOC Washington State Department of Corrections

ix
Abstract
Recidivism rates have important implications for public safety, the well-being of both those
reentering our communities, as well as the communities that the formerly incarcerated
individuals are being released back into. This study leverages a comprehensive prison
admission/release dataset from the Washington State Department of Corrections in a spatial
analysis looking at both individual level and county contextual variables with the intent to
identify whether the county of release of a formerly incarcerated individual is correlated to the
recidivism rates of the county. The analysis further considered social disorganization theory
aspects as potential contributors to recidivism patterns. By incorporating these variables into a
comprehensive exploratory data analysis, subsequent statistical analyses, and finally exploratory
spatial data analysis methodologies the study aimed to understand how the socioeconomic
context of the county of release may lead to a propensity to recidivate in Washington State. The
findings of this project show that there is no evidence of a correlation between the county of
release and propensity to recidivate in the State from 2012-2022, the conclusion drawn here is a
finding for the null hypothesis of this study, that while many counties display statistically
significant deviations from the sample mean, these deviations are not attributable to the county
of release itself. Furthermore, contrary to academic literature that has found significant
correlations between social disorganization and crime, no statistically significant correlations
with the socioeconomic contextual variables meant to reflect social disorganization were found
in this analysis. This suggests that while rates of public assistance enrollment and literacy rates
or educational attainment have been found to be correlated to crime and recidivism elsewhere,
they are not in Washington State at the county scale of analysis. The ultimate conclusion
underscores a critical concern regarding the selected scale of analysis, emphasizing that the

x
County, as an areal unit of aggregation, proves to be too broad to comprehensively capture the
nuances of recidivism. This assertion gains robust support from the evidence revealed by the
analysis' findings.

1
Chapter 1 Introduction
The primary intent of this study is to investigate the correlation between the county of release
and recidivism outcome in Washington State. Employing an evidence-based spatial analysis
approach at the county scale, the study engages in a comprehensive exploratory data analysis
(EDA), and exploratory spatial data analysis (ESDA) to assess recidivism likelihood in
Washington State. By leveraging and expanding upon existing research on recidivism in the
fields of Criminal Justice and Criminology within Washington State (Aos 2003; Miller, Drake,
and Nafziger 2013; Miller, Drake, and Aos 2006; Gagliardi et al. 2004; Beckett, Harris, and
Evans 2008; Lovell, Johnson, and Cain 2007; Manchak, Skeem, and Douglas 2008; Phipps et al.
1999; Camasso and Jagannathan 1995), this project seeks to inform decision-making processes,
targeted outreach efforts, and drive changes in public policy and discourse.
This research sheds light on the spatial patterns and potential contextual factors
influencing recidivism through the use of variables that reflect social disorganization and give
each county further context. To analyze these research goals, utilization of various statistical
methods, including descriptive statistics, a Power Analysis, Shapiro-Wilks test, T-tests, Cohen’s
d, ANOVA, Moran’s I Index for spatial autocorrelation, and Getis-Ord
∗
hot spot analysis
allowed this research to delve into the intricacies of recidivism trends and spatial patterns across
counties. Additionally, the identification of hotspots and spatial autocorrelation were analyzed in
a way that provides insights for informed decision-making regarding recidivism through targeted
interventions in areas exhibiting higher susceptibility.
This study also builds on prior research assessing the relationship between the
neighborhood context and the propensity to recidivate. Specifically, building on work by Hipp
(2007), this study aims to investigate whether the county of release is correlated with a

2
propensity to recidivate while recognizing the scale of analysis emerges as a pivotal
consideration. Drawing inspiration from this past work, which delved into the neighborhood
effects of crime at the block and tract levels, it is clear that selecting the most appropriate areal
unit of analysis is not merely a methodological choice but also a theoretical one. Prior studies
have primarily focused on more granular scales of analysis. However, some have examined the
neighborhood effects of crime using combinations ranging from two tracts together (e.g., Logan
and Stults 1999; Morenoff et al. 2001; Sampson et al. 1997) to aggregations of nine or ten tracts
(e.g., Almgren et al. 1998; Bursik 1988; Heitgerd and Bursik 1987). A consistent finding across
the referenced studies is the profound impact that the unit of analysis can have on our
understanding of crime dynamics.
While it is reasonable to assume that the existence of heterogeneity across blocks and
within tracts, as outlined in Hipp's work, may be masked at higher levels of aggregation for
certain measures and theoretical inquiries, by adopting the county as the unit of analysis, the
intent is to capture a broader spectrum of contextual variables that might shape an individual's
journey post-release. Beyond the mere correlation between county of release and reoffending,
this study seeks to understand how the diverse landscape of county-specific attributes—socioeconomic indicators, educational opportunities, or even community cohesion—might serve in the
future as predictors of recidivism. The county scale, thus, offers both a challenge and an
opportunity. A challenge in ensuring that critical nuances are not overlooked, and an opportunity
to understand recidivism in a broader socio-spatial framework (Hipp 2007).
Applying this understanding to the context of recidivism and county of release, it raises
pertinent questions: How do counties, as broader units compared to tracts or neighborhoods,
capture the myriad of factors influencing recidivism? Can the heterogeneities, often noted at

3
smaller spatial scales, be suitably represented at the county level? Another focus of this research
then, is to understand if the county scale of analysis accurately captures the spatial processes of
recidivism, with the ultimate goal of determining whether the county serves as an appropriate
areal unit of aggregation for the purposes of analyzing recidivism in Washington State.
1.1 Background
In 2019 alone, Washington State ranked in the top ten of all states for the number of total
admission and releases from incarceration (Bureau of Justice Statistics 2019). Between 2012-
2022 approximately 6,500 people were released from WADOC operated prisons per year. These
statistics highlight the importance of creating a more evidence-based data-driven approach to
reducing recidivism rates and promoting successful reentry in the state of Washington.
Reducing recidivism rates holds significant implications for public health, safety, crime
rates, tax burdens, and various measurable social benefits. It contributes to the creation of more
cohesive and crime-resilient neighborhoods (Drawve and McNeeley 2021; Hipp et al. 2010;
Petersilia 2009; Wang et al. 2013). However, it's crucial to acknowledge that recidivism
disproportionately impacts marginalized communities, often affecting them at higher rates than
their more prosperous and socially cohesive counterparts (Kubrin et al. 2007; Shaw and McKay
1942).
In the context of Washington State, the Offender Accountability Act (OAA), enacted in
1999, mandates the consideration of recidivism potential in risk assessments. The Level of
Service Inventory Revised (LSI-R) is a key tool used to determine sentencing guidelines for
judges (Washington State Legislature, 1999; Aos, Drake, and Miller 2006; Drake 2014). Given
that the majority of incarcerated individuals will eventually re-enter society, there is a
compelling argument for prioritizing rehabilitation alongside confinement as part of a holistic

4
approach to criminal justice. Reducing recidivism not only enhances equitable outcomes but also
aligns with the broader goals of public safety and community well-being.
Traditional statistical methods that are widely used in the analysis of recidivism assume
that the relationship between recidivism and the other variables (ecological and demographic
factors) remain constant over the entire State. This is typically not the case with dynamic
variables such as crime rates or recidivism that have geographic connotations and usually exhibit
spatial patterns (Cahill and Mulligan 2007). By conducting the analysis at the county level, the
study acknowledges that recidivism rates and their underlying correlates may vary spatially
across different counties. This approach allows the project to consider the unique social,
economic, and environmental factors specific to each county that may impact recidivism. Doing
so, makes the study better able to capture the localized influences on reoffending and understand
the potential importance of the neighborhood context to successful reentry. This approach can
help identify areas of high or low recidivism rates, detect spatial patterns or clusters, and inform
targeted interventions tailored to the specific needs of each county. Studies on the neighborhood
context of recidivism suggest that the social and environmental context to which the formerly
incarcerated reenter the community as well as the geographical accessibility to interventions and
mobility are a pivotal part of successful reentry (Hipp et al. 2010; Kubrin and Stewart 2006).
1.2 Study Area
In 2007 in Washington State, recidivism rates were recorded at 65.9% for men and 53.6%
for women (Caseload Forecast Council 2007). The Washington State Department of Corrections
maintains and operates 12 prisons across the State. As of June 2022, the WADOC had 12,972
person(s) incarcerated under their supervision (WADOC 2022). These statistics highlight the

5
importance of gaining a better understanding of the factors contributing to recidivism to inform
effective public policy in the State.
The geographic scope of the analysis focused on the state of Washington in the United
States. Washington State has shown a commitment to evidence-based practices and innovations
in correctional programs, making it a compelling case study for assessing the effectiveness of
interventions aimed at reducing recidivism rates.
Washington State is located in the Pacific Northwest region of the United States. It spans
an area of approximately 184,661 square kilometers and is bordered by the Pacific Ocean to the
west, Oregon to the south, Idaho to the east, and British Columbia, Canada to the north. The
U.S. Census Bureau’s 2020 Census indicates that the population in Washington State grew by
.58% year over year from 2010-2020 to roughly 7.71 million residents. Of those residents, a
majority reside around the Puget Sound Region on the Western Side of the Cascade Mountain
Range. Population hot spots are also centered in the southernmost Clark County which is near
Portland Oregon. In addition, the eastern half of the state’s population is largely centered in
Spokane and Benton Counties.
In Washington State, the socioeconomic landscape varies across counties, influencing
recidivism rates in distinct ways. For instance, urban areas like King County, which encompasses
Seattle, have numerous employment opportunities, vocational training centers, and support
services that could positively impact individuals' chances of finding stable employment postrelease. On the other hand, counties with essential healthcare infrastructure and support systems,
such as Thurston County, where the state capital Olympia is located, may potentially witness
better outcomes in terms of reducing recidivism (Figure 1). These urban areas across the state
tend to have larger populations and a more diverse economy, offering a broader range of reentry

6
opportunities. In contrast, rural areas, such as Adams County or Franklin County in Eastern
Washington, often face unique challenges associated with fewer economic opportunities, higher
poverty rates, and limited access to quality education and healthcare services.
Figure 1. Rate of Recidivism (Reconfinement) 2012- 2022
The Washington State Department of Corrections operates twelve prisons across the state
(Table 1). These facilities are located in 11 different counties (Figure 1). Of those twelve
facilities four, Clallam Bay Corrections Center (CBCC), Coyote Ridge Corrections Center
(CRCC), Monroe Correctional Complex (MCC), and Washington State Penitentiary (WSP) are
designed to house incarcerated individuals with higher security classifications, including
maximum security. Three of those, CRCC, MCC, and WSP are capable of housing roughly
2,400 individuals each making them the largest of the twelve facilities. The total capacity of all
WADOC operated prisons as of June, 2022 is 14,698 incarcerated individuals. The average daily
population across all WADOC facilities during that same time was 12,233 incarcerated

7
individuals (WADOC, 2022). The majority of WADOC-operated prisons within the state are
strategically situated around the Puget Sound Region. Upon comparing this distribution with the
population density map provided above, noticeable patterns emerges between prison locations
and population centers.
Table 1. WADOC facilities
Facility Name Acronym County Capacity Custody Level
Airway Heights Corrections
Center
AHCC Spokane 2258 Minimum and Medium
Cedar Creek Correctional
Complex
CCCC Thurston 480 Minimum
Clallam Bay Corrections Center CBCC Clallam 858 Medium, Close, and Maximum
Coyote Ridge Corrections Center CRCC Franklin 2468 Minimum
Larch Corrections Center LCC Clark 240 Minimum
Mission Creek Corrections Center
for Women
MCCCW Mason 321 Minimum
Monroe Correctional Complex MCCCW Snohomish 2400 Minimum, Medium, Close, and
Maximum
Olympic Corrections Center OCC Jefferson 272 Minimum
Stafford Creek Corrections Center SCCC Grays
Harbor
1936 Minimum, Medium, and
Maximum
Washington Corrections Center WCC Mason 1268 Medium, Close, and Maximum
Washington Corrections Center
for Women
WCCW Pierce 738 Minimum, Medium, and Close
Washington State Penitentiary WSP Walla
Walla
2439 Minimum, Medium, Close, and
Maximum
1.3 Project Overview
The primary objective of this study was to investigate the potential correlation between
the county of release and the propensity to recidivate. This exploration was based on an
extensive WADOC prison admission/release dataset spanning from 2012 to 2022. As a
secondary aim, the study aimed to contribute a small collection of variables linked to Social
Disorganization Theory. This contextual framework aims to offer insights into county dynamics

8
and assess if relationships found in studies from diverse jurisdictions hold true for the State of
Washington.
1.3.1 Temporal Scope
The temporal scope of the study covers a period of ten years, from July 1, 2012 to June
30, 2022. The spatiotemporal granularity allows for an in-depth examination with a sufficient
follow-up period to determine whether the propensity to recidivate is driven by spatial factors at
the County level. The study aims to capture the geographic connotations and trends in recidivism
rates over this period, being that it is the most recent data available.
1.3.2 Scale of Analysis
This analysis is conducted at the county scale to examine the variations in recidivism
rates and factors across different areal units within Washington State. Washington State
comprises 39 counties, each with its own unique characteristics, socioeconomic factors, and
criminal justice practices. Analyzing this data at the county level enables a spatial examination of
recidivism patterns and provides insights into county-specific factors that may contribute to
reoffending. While the County scale may be effective at evaluating the overall effectiveness of
WADOC operated prisons, it may fail to highlight the nuanced nature of recidivism. The choice
of the scale of analysis was limited to data availability.
1.3.3 Primary Dataset
The study utilized a comprehensive prison admission/release dataset obtained from
WADOC through a public records request. This dataset provides detailed information on the
admission and release of individuals under WADOC custody or released between July 1, 2012,
and June 30, 2022. The dataset comprises a total of 64,320 individuals, as received from the

9
WADOC prior to any data preprocessing. It also encompasses various variables, including an
individual's unique identifier number (DOCNUM), admission and release dates, admission
reasons, and their respective county of admission and release.
1.3.4 Limitations of the Primary WADOC Dataset
When analyzing this dataset, it became apparent that it poses certain limitations for
analysis. The dataset received from WADOC, included data entry that was not standardized
across the institution’s database. Each facility seemingly entered data in different ways which
made preparing the data for analyses a major obstacle. The WADOC only maintains records at
the county scale of analysis due to privacy concerns with finer scale data collection and
dissemination. This limitation is related to the modifiable areal unit problem (MAUP). The
MAUP refers to the potential distortion of results when aggregating data into different spatial
units (e.g., counties) due to the arbitrary nature of the boundaries (Openshaw and Taylor, 1979;
Openshaw, 1984). Aggregating data at the county level may mask finer-scale spatial patterns and
heterogeneity within counties, potentially leading to unreliable conclusions.
Another major limitation was posed when attempting to extrapolate independent
variables associated with individual-level characteristics. The efficiency of searching each
individual DOC Number within a government database was severely limited by the volume of
the data, thereby preventing the retrieval of demographic information and many other variables
initially considered valuable for the analysis. It should also be noted that the dataset excludes
records from county jails which typically hold incarcerated individuals until sentencing at which
point WADOC takes custody of an incarcerated individual, Federal prisons, Immigration
Detention facilities, and private prisons operating within the State. Consequently, the analysis is
focused solely on individuals under custody or released by WADOC.

10
Furthermore, the primary dataset encompasses a temporal span from 2012 to 2022. While
this ten-year period provides valuable insights into recidivism trends, it's important to consider
the temporal limitations inherent in the data. Notably, individuals released towards the latter part
of the 2012-2022 timeframe inherently have less opportunity post-incarceration to recidivate
when compared to those released earlier in the time span. This temporal asymmetry could
influence the observed recidivism rates and patterns, potentially skewing the analysis towards
individuals who had more time to reoffend. Therefore, the findings of the study should be
interpreted while keeping this temporal bias in mind.
Using the county as the scale of analysis has certain limitations that should be
highlighted. The county scale might not capture the actual scale at which relevant processes are
occurring. It is useful for analyzing general trends across a study area but typically does not
adequately represent the nuances that are typical of social phenomenon such as crime and
recidivism which are generally fluid across space. Another limitation is that it may not
adequately account for variations in population density across different regions. When using
counties, areas with higher populations will naturally have more incidents, which could lead to
biased results if not appropriately adjusted for population differences. The areal unit of
aggregation posed a major methodological limitation to the framework of this study.
Furthermore, while the study focused on Washington State, the findings may not be directly
generalizable to other states or jurisdictions. It is also essential to consider the unique
characteristics and local context of each jurisdiction (county) when applying the study's findings
to inform policy in different settings.

11
1.4 Document Overview
This thesis navigates through a structured exploration, beginning with a literature review
in Chapter 2 that examines social disorganization theory, assesses critically relevant research
conducted in Washington state, as well as studies employing spatial analysis on recidivism.
Subsequently, Chapter 3 outlines the methodological approach, detailing the steps taken to
achieve the research goals. Chapter 4 presents the results derived from these methods, offering
insights into potential correlations between the county of release and recidivism. Finally, Chapter
5 delves into a comprehensive discussion that contextualizes the findings within real-world
implications, thereby connecting research outcomes to practical applications.

12
Chapter 2 Literature Review
This chapter critically evaluates conventional statistical approaches to assessing recidivism,
conducts an in-depth analysis of existing studies focused on recidivism within Washington State,
and extends its scope to include a review of spatial analysis methodologies used in relevant
literature beyond the state. Furthermore, this section explores the foundational underpinnings of
social disorganization theory, which is considered a fundamental theoretical framework within
the criminology and criminal justice disciplines.
2.1 Social Disorganization Theory
Two predominant theoretical dichotomies, social disorganization theory (SDT) and
routine activities theory, are dominant in the academic literature on the neighborhood structural
characteristics of crime and recidivism in the broader fields of criminology and criminal justice.
In spatial analyses of recidivism, the predominant theoretical framework often employed is based
on SDT (Drawve and McNeeley 2021). This study delves into the significance of the location—
specifically the county—where an individual is released, a variable inherently tied to the social
disorganization theory. Rooted in the premise that community structures influence crime
tendencies, social disorganization theory aligns with our assessment.
SDT explains that communities characterized by poverty, residential instability, and
racial diversity suffer from higher rates of crime (Shaw & McKay 1942). It was a theoretical
shift away from the tendency to use only individual-level characteristics to describe crime and
recidivism, to a recognition that place, and geographical location also contribute to these
phenomena (Shaw and McKay 1972). The guiding principle of the theory is that crime and
recidivism are directly correlated to ecological context and that social and economic

13
disorganization which are typically characterized by poverty, residential instability, higher levels
of gentrification or housing instability, ethnic diversity, and low literacy rates or educational
attainment, lead to higher rates of crime and recidivism (Shaw and McKay 1942). Another major
component of the theory is characterized by the social cohesion of a neighborhood. The
neighborhoods that share common values and maintain strong social connections are more
resilient to crime according to this theory (Shaw and McKay 1942). The primary criticism of this
theory in criminology is that it ignores the individual-level factors that are often involved in
crime and recidivism (Bursik 1988). Individual-level factors include the personal characteristics
and environmental circumstances of an individual that may contribute to the likelihood of
committing a crime or recidivating. Some of these factors include things such as genetics,
poverty status, educational attainment, literacy, social/peer network, childhood experiences,
health, and substance use or abuse.
2.2 Social Disorganization Theory in Crime and Recidivism Research
SDT is often utilized in studies on crime and recidivism by guiding the research design,
influencing the selection of variables, and shaping the interpretation of data. A common
approach to utilizing the SDT is building a subset of variables into an index meant to reflect
social disorganization, and then testing for areas of concentrated disadvantage using the index
(Kirk and Laub 2010). Studies might also start with hypotheses that connect recidivism rates to
indicators of social disorganization such as residential mobility, socioeconomic status, family
structure, concentrated disadvantage, and neighborhood resources. When utilized in studies on
crime and recidivism a common approach is acquiring qualitative survey data from residents on
the context of a specific neighborhood (Sampson et al. 1997). More recent methodologies
include incorporating spatial analysis and multilevel models of census data and crime rates

14
across neighborhoods to see how key elements of social disorganization theory predict crime
rates (Kubrin and Weitzer 2003; Hipp 2007).
Many academic studies are grounded in the SDT framework. A recent study exemplifies
this approach by examining how neighborhood characteristics link to gun violence using data
from 2014 to 2019 (Maher et al. 2022). In another study grounded in the Social Disorganization
Theory framework, researchers employed specific variables to gauge concentrated disadvantage
within neighborhoods. This included assessing low income, which involved quantifying the
percentage of households in the community with incomes falling below the lowest quintile of the
Netherlands' income distribution in 2008. Additionally, the study considered the proportion of
households receiving welfare benefits in 2006 (Gerben et al. 2013). These chosen variables are
indicative of economic hardship and align with social disorganization theory's core premise that
areas marked by economic disadvantage are more likely to experience heightened crime rates
(Shaw and Mckay 1942).
2.3 Mathematical Methodology
Exploratory data analysis (EDA) is a set of methods that evaluate variations among
different groups, identify data outliers, detect clustering, and highlight non-linear relationships
between variables (David and Tukey 1977). The purpose of EDA is to visually interpret data in a
way that could potentially unveil underlying patterns, leading to novel and often unexpected
insights (David and Tukey 1977). Recent advances emphasize the application of multivariate
EDA techniques, specifically the utility of boxplots and scatterplots for large data sets
(Nicodemo and Satorra 2022).
Correlative analysis is a pivotal tool in unveiling potential relationships and associations
within complex datasets. This analytical approach enables researchers to explore the degree of

15
linear dependence between two or more variables, unveiling insights into how changes in one
variable might correspond to changes in another. By calculating correlation coefficients, such as
Pearson's correlation coefficient, researchers can quantify the strength and direction of
relationships, shedding light on patterns that might not be immediately apparent (Ostertagova,
Ostertag, and Kováč 2014). Correlation analyses then, are useful in preliminary hypothesis
testing used to guide further research endeavors. It should always be cautioned that correlation
does not imply causation.
Exploratory spatial data analysis (ESDA) is an extension of the traditional EDA
techniques to spatial data. Spatial data has geographical or spatial information (x,y coordinates)
associated with it, and ESDA techniques are designed to uncover spatial patterns, anomalies, or
other interesting characteristics in such data. One common aspect of ESDA is the assessment of
spatial autocorrelation, which is the tendency of nearby geographical entities to exhibit similar
attributes is a principle often attributed to Waldo Tobler and cited as Tobler’s First Law of
Geography (Tobler 1997). At its core, the concept of spatial autocorrelation assesses how
observations of the same variable are related to each other across a study area, offering insights
into the spatial patterns in the data distribution. This concept becomes especially pertinent when
considering crime or recidivism patterns, where geographic proximity can often translate to
similar characteristics and behaviors (Cahill and Mulligan 2007). When assessing dynamic
spatial patterns such as recidivism, it is uncommon that homogeneity exists across an entire,
global, study area. The spatial weights matrix (SWM) represents the spatial relationship amongst
all neighbors or observations in a dataset in a mathematical matrix format which makes it
possible to measure autocorrelation (O’Sullivan and Unwin 2010).

16
Local statistics are a method for taking this heterogeneity that exists from one
observation to the next across a study area and using the descriptive statistics of each to derive a
more nuanced understanding of the spatial patterns that make up the global study area
(O’Sullivan and Unwin 2010). Since Unwin (1996) and Fotheringham (1997) highlighted the
importance of local statistics in their seminal works on the issue, the use of local statistics in
spatial analysis research has been limited to the technological capabilities – especially the
computing power of home computers. Within the last 15 to 20 years, access to extensive and
detailed spatial as well as spatiotemporal datasets has offered the chance to acquire new insights
and enhance our comprehension of intricate geographic phenomenon (Mennis and Guo 2009).
Spatial statistical methods that were previously computationally burdensome have become
achievable with the advances in computer science (Goodchild 2007). Common statistical
measures of spatial autocorrelation include the Moran’s I test (Moran 1948), Getis and Ord’s G
statistics (Getis and Ord 1992), Geary’s C (Geary 1954), join count statistics (Besag 1974), and
local indicators of spatial association (LISA) (Anselin 1995).This study has chosen to implement
the Getis-Ord
∗Statistic to assess localized autocorrelation.
2.4 Areal Unit of Aggregation
While a wide range of interdisciplinary research approaches have centered on the
significance of neighborhoods, a recurring aspect in many of these studies is the tendency to
overlook the essential component of determining the appropriate level of aggregation for
capturing the neighborhood effects (Hipp 2007). Choosing the areal unit of aggregation (AUA)
at an appropriate scale to define the neighborhood structure and underlying spatial processes
associated with such is critical to the results of any analysis. Too large of an aggregation has the
potential to obscure more localized spatial patterns and effects by combining the heterogeneity of

17
multiple neighborhoods into one aggregate (O’Sullivan and Unwin 2010). This makes the
analysis susceptible to the ecological fallacy or ecological inference, which occurs when it is
assumed that all observations in a dataset are homogenous across the whole of a study area, or in
other words, it is a failure to recognize that the data aggregate is not representative of each
observation within that aggregation (Robinson 1950).
An ideal approach to finding the most appropriate areal unit of aggregation to represent
the neighborhood contextual aspects of crime and recidivism is to flexibly aggregate the data to
varying geographic sizes and compare the results (Hipp 2007). The ideal aggregation would
mimic the scale at which the phenomenon of recidivism occurs, for instance, one study found
that Census Block Groups most adequately represented the actual geographic areal
measurements of the city neighborhoods in both Los Angeles and San Francisco, California
(Grannis 1998). Further research related to the proper geographical units of aggregation for
neighborhood level crime and recidivism have assessed the Census tract and Census block units
of aggregation (Hipp 2007). This same study also called for future research to assess the larger
geographical units of aggregation such as combinations of block groups, or the County scale of
analysis. That is one place where this study intends to contribute to existing literature; By
assessing whether the county of release is associated with a propensity to recidivate this study
seeks to contribute to research on the appropriate scale of aggregation for assessing the
neighborhood contextual elements of crime and recidivism.
2.5 Overview of Previous Analytical Research on Recidivism
Understanding the underlying spatial processes of recidivism is an important component
of studies in the fields of criminology and criminal justice. It not only sheds light on the
effectiveness of rehabilitation programs and the challenges of reintegration but also offers

18
insights into the broader social dynamics that contribute to criminal behavior. As the rates of
recidivism continue to pose significant challenges to the criminal justice system, researchers
have sought diverse statistical and modeling methodologies to unravel the complexity of factors
influencing the propensity to recidivate. Prior research in Washington State has utilized various
methods for assessing recidivism, a Level of Service Inventory-Revised (LSI-R) score analysis,
meta-analytical approaches that calculate effects sizes, and multivariate and bivariate analysis
methodologies such as correlative analyses. The LSI-R is a widely used assessment tool in the
field of criminology and criminal justice. It is designed to evaluate the risk and needs of
offenders, particularly those who are involved in the criminal justice system (Phipps et al. 1999).
When expanding the scope to encompass studies from outside jurisdictions, it is common to
observe the utilization of a wide range of statistical methods to assess recidivism. Binary logistic
regression is frequently employed, offering insights into the relationship between predictor
variables and the likelihood of recidivism. Geographically weighted regression and area under
the receiver operator characteristic (AUC) analysis are also popular choices, allowing for the
examination of spatial variations and predictive accuracy. In addition, Bayesian estimation
techniques provide a probabilistic framework to assess the complex interactions contributing to
recidivism. Correlative analyses further enhance understanding by investigating associations
among variables. These methodologies collectively contribute to a comprehensive assessment of
the multifaceted phenomenon of recidivism.
2.5.1 Washington State Specific Studies on Recidivism
The passing of the Offender Accountability Act in 1999 by the Washington State
legislature mandated the use of a risk assessment for post-release sentencing guidelines and
allocating resources to the highest-risk formerly incarcerated individuals, for this purpose the

19
DOC adopted the use of the LSI-R (Phipps et al. 1999). This assessment is used as the risk for
re-offense component in DOC’s Risk Management Identification (RMI) system (Phipps et al.
1999).
A common approach to assessing recidivism is to assess the effectiveness and accuracy
of the LSI-R risk assessment questionnaire as a predictor of recidivism (Manchak, Skeem, and
Douglas 2008; Aos 2003; Camasso and Jagannathan 1995). One study conducted by the
Washington State Institute for Public Policy (Aos 2003) analyzes rates of recidivism among a
sample of 22,533 offenders using the LSI-R scores as the predictor variable. The authors used
three statistical measures to assess the efficacy of the LSI-R. First, they employed correlation
coefficients. They then conducted an area under the receiver operator characteristic (AUC) to
assess the strength of the correlation. Finally, they obtained the odds ratio from multivariate
analyses and used them to assess the contribution of the variables that make up the LSI-R
questionnaire for predicting recidivism.
The research reveals a gradual increase in recidivism rates with higher LSI-R scores,
indicating a positive correlation between scores and re-offending likelihood. However, the
absence of distinct cut-off scores suggests a lack of naturally occurring low- and high-risk
categories (Aos 2003). This upward trend is supported by equations showing that each one-point
increase in the LSI-R score corresponds to incremental percentage point increases in
misdemeanor and felony recidivism rates, felony recidivism rates, and violent felony recidivism
rates (Aos 2003). Despite low correlation coefficients, the moderate AUC values suggest a
significant but not overwhelmingly strong association between LSI-R scores and recidivism,
emphasizing its predictive capability (Aos 2003). The distribution of LSI-R scores, as

20
demonstrated by a bell-shaped curve, highlights a clustering of scores around the mean, with
fewer offenders having notably high scores (Aos 2003).
The study further dissects the associations between LSI-R domain scores and recidivism.
It underscores that certain variables, particularly criminal history, exhibit stronger correlations
with recidivism, contributing significantly to the overall predictive power of the LSI-R (Aos
2003). Criminal history emerges as a robust predictor of recidivism in this analysis (Aos 2003).
The findings collectively highlight the dynamic nature of predicting recidivism and the varying
strengths of association between LSI-R variables and different types of recidivism. This offers
valuable insights for risk assessment and intervention strategies within the criminal justice
system (Aos 2003).
Another study conducted in Washington State assesses evidence-based interventions and
programs for adult incarcerated individuals to determine the efficacy of those interventions (Aos,
Miller, and Drake 2006). The intent is to determine which programs are working and which are
not to ultimately aid in lowering recidivism rates in the State of Washington. This study is a costbenefit analysis of sorts. Notably, only a limited subset of the evaluated program assessments
originated from within Washington State; instead, the study encompassed evaluations spanning a
40-year timeframe commencing from 1970, drawn from a wide range of English-speaking
nations (Aos, Drake, and Miller 2006).
Aos, Drake, and Millers’ study used a meta-analytic approach to calculate effect sizes.
They employed both fixed effects and random effects modeling in their analysis. These methods
involve pooling the results of multiple rigorous evaluations of different programs to calculate an
overall estimate of the effect size, which indicates the magnitude of the impact of each program
on recidivism rates.

21
The results of their study reveal that certain programs are effective in reducing recidivism
rates, while others are ineffective. Among the effective programs are drug courts, in-prison
therapy communities, cognitive-behavioral drug treatment while incarcerated, drug treatment in
the community, and cognitive-behavioral treatment for domestic violence offenders (Aos, Drake,
and Miller 2006). However, some programs, such as jail diversion for mentally ill individuals
and some intensive supervision programs, did not demonstrate significant recidivism reduction
(Aos, Drake, and Miller 2006). Their study emphasizes the need to focus resources on evidencebased programs to avoid ineffective approaches to successful reentry.
Other studies analyzing recidivism assess the impacts of individual-level variables on the
propensity to recidivate (Miller, Drake, and Nafziger 2013; Gagliardi et al. 2004; Beckett, Harris,
and Evans 2008; Lovell, Johnson, and Cain 2007). In a study examining recidivism among
mentally ill formerly incarcerated individuals, researchers analyzed a sample of 333 individuals
released from Washington State Prisons in 1996 and 1997 with a post-release follow-up period
ranging from 27 to 55 months. The study found that approximately 77% of mentally ill offenders
(MIOs) were re-arrested or charged with new crimes, with 41% convicted of another felony
offense and 23% convicted of violent crimes. Comparatively, the general prison population had a
38% conviction rate for new felonies. Several independent variables were assessed in this study,
including sentence length, past felonies, and past drug felonies, which showed statistically
significant correlations to recidivism. Surprisingly, the study concluded that MIOs were no more
likely to recidivate than the general population. The authors argued against including mental
health status as a predictor in recidivism assessment tools.

22
2.5.2 Relevant Studies in Spatial Criminology
In the context of spatial criminology, which is often used to assess recidivism,
researchers examine how crime and criminal behavior vary across specific localized
environments. Typically, research investigates whether certain factors that contribute to crime
remain consistent across different areas or if they are influenced by the characteristics of each
location. Understanding these variations is crucial for developing effective crime prevention
strategies and policies that consider the specific conditions of different places. (LeBeau and
Leitner 2011).
Within the realm of spatial criminology, researchers have conducted studies that focus on
diverse aspects of crime. For instance, a National Institute of Justice (NIJ) funded project
explores the connection between neighborhood characteristics and juvenile
delinquency/recidivism in Philadelphia. Their findings reveal that delinquency is concentrated in
impoverished, violent crime-prone neighborhoods (Mennis et al. 2011). In another study, the
concept of the "ambient population" was introduced to refine crime rate calculations, offering
empirical evidence for its application (Malleson and Andresen 2015). In the realm of criminal
geographic profiling (CGP), one study identified the center of minimum distance as a reliable
model to predict recidivism. This model calculates the geographical point or area where
individuals at risk of recidivism are most likely to gravitate towards, shedding light on the
direction and magnitude of the relationship between spatial factors and recidivism outcomes. In
that study they also explore Bayesian estimation techniques to enhance CGP accuracy (Levine
and Block 2011). Another methodology that has recently surfaced is to evaluate the weighted
displacement quotient (WDQ) for crime suppression operations, which are strategies aimed at
reducing criminal activities in specific areas (Ratcliffe 2004). Another study examined the
impact of population displacements after Hurricanes Katrina and Rita on crime rates in

23
Louisiana. Their study suggests that crime remains stable or declines in regions receiving
emergency evacuees from the disaster zones (Leitner et al. 2011).
When assessing recidivism using spatial analysis, it is far more common to apply global
statistics and regression models to the phenomenon than it is local statistics (Kim et al. 2013;
Leymon et al. 2022; Wang et al. 2013; Stahler et al. 2013; Hipp, Petersilia, and Turner 2010).
Only in the last decade and a half have studies begun to leverage spatial modeling to assess the
spatial relationship between the community contextual environment and recidivism empirically
(Gottfredson and Taylor 1988; Stahler et al. 2013; Kubrin and Stewart 2006). Those studies that
do assess recidivism using local statistics tend to rely on sampling units or surveys to derive their
results (Drawve and McNeeley 2021; Cahill and Mulligan 2007).
In one such study the relationship between community context and recidivism in
Minnesota is assessed. The study investigated the impact of neighborhood-level factors on
recidivism, extending beyond social disorganization. The analysis considered the influence of
prosocial-local institutions and criminogenic establishments on recidivism outcomes. While the
presence of criminogenic establishments showed no significant relationship with recidivism, an
increase in the number of prosocial institutions within neighborhoods was associated with a
decrease in arrest likelihood (Drawve and McNeeley 2021). This finding deviated from previous
research. Moreover, the study revealed that the effect of prosocial establishments was more
pronounced in less-disadvantaged neighborhoods (Drawve and McNeeley 2021). The research
suggests that the structural nature of disadvantaged neighborhoods might outweigh the impact of
prosocial institutions, emphasizing the need for tailored approaches in service provision. Despite
certain limitations, this study contributes to the understanding of neighborhood influences on
recidivism and advocates for localized research approaches. The results of this multilevel

24
Bernoulli model suggest that the greater number of prosocial establishments per neighborhood
leads to a decrease in recidivism rates (Drawve and McNeeley 2021).
In a study published in 2007, researchers aimed to understand violence levels in Portland,
Oregon's Multnomah County from 1998 to 2002. Initially using a traditional Ordinary Least
Squares (OLS) regression model, they found it inadequate, explaining less than 40% of violence
variance and yielding counterintuitive results (Cahill and Mulligan 2007). This underlined the
limitation of global regression models for spatial data, missing local variations.

25
Chapter 3 Methods
This chapter provides a detailed description of the methodological approach taken in this study to
assess recidivism. It begins with an introduction to the essential data required for the analysis,
followed by the steps taken to prepare and clean the data for spatial analysis. Subsequently, the
proposed methodological approach is outlined, highlighting how it will facilitate the project's
primary objective of determining the potential correlation between the county of release and the
likelihood of recidivism.
Figure 2 provides a visual representation of the project workflow. The initial phase
involved detailing the data cleaning and preparation process for the primary WADOC
admission/release dataset. Subsequent steps encompass dataset normalization, the presentation of
baseline and summary statistics, and a power analysis to determine a minimum sample size
(number of total releases for each county). Univariate and bivariate data visualization methods,
such as histograms, boxplots, and a scatterplot matrix (SPLOMs), assessed data distribution and
Pearson’s correlation coefficients. Both an ANOVA test and subsequent T-test and Cohen’s d
and their corresponding p-values served to explore the correlation between the county of release
and recidivism propensity in Washington State from 2012-2022.

26
Figure 2. Workflow diagram
3.1 Variable Selection and Data
The foundation of this analysis lies in a collection of data sourced from reputable
institutions, including the Washington State Department of Corrections (WADOC), the U.S.
Census Bureau, the American Community Survey, and relevant datasets from the Washington
State Open Data Portal. These diverse datasets were utilized to provide a comprehensive basis
for the ensuing workflow.
This study’s selection of contextual variables is aligned with the key tenets of social
disorganization theory (SDT), with each choice grounded in the scholarly discourse on the topic.
For instance, the inclusion of mass shootings and gun violence as a variable draws upon analyses
that demonstrate its significant impact on the social fabric and crime rates within communities,
thereby echoing SDT’s focus on the influence of community context on criminal behavior,

27
especially recidivism (Maher et al. 2022). Furthermore, the study’s focus on public assistance
enrollment rates draws on insights from a recent study which contextualized welfare benefits
within the SDT paradigm, highlighting economic disadvantage as a pivotal factor in community
social dynamics and crime occurrence. (Gerben et al. 2013). Literacy rates are also considered,
given their established role in indicating community cohesion, social control, and educational
attainment, factors that are fundamental to SDT’s explanation of how lower literacy levels can
undermine social stability and control, potentially leading to higher crime rates (Shaw and
McKay 1942).
The deliberate inclusion of these variables is designed to capture the complex interplay of
factors that contribute to social disorganization and its relationship with recidivism, consistent
with the empirical evidence presented in SDT-related research.
Table 2. Variables
Variable Source Temporal Extent Definition
Rate of Recidivism WADOC July 1, 2012 to June
30, 2022.
Recidivism rates by county of
WADOC operated prisons
from 2012-2022
County of Release WADOC July 1, 2012 to June
30, 2022.
The county that an
incarcerated individual
(DOCNUM) was released to.
Enrolled in Public
Assistance (SNAP/SSI)
American
Community Survey
2021 (five-year
estimate)
The population enrolled in
SNAP food assistance or
Supplemental Income public
assistance.
Lacking Basic Prose
Literacy Skills
National Center for
Education Statistics 2003 The percentage lacking basic
prose literacy skills by county.
3.1.1 Description of the Data
In this study, a comprehensive dataset spanning from 2012 to 2022 was utilized,
consisting of detailed prison release and admission records obtained from the Washington State
Department of Corrections (WADOC), as discussed in Chapter 1. The WADOC dataset provided
valuable information on the admission and release of individuals under WADOC custody during

28
the specified ten-year period. While the concept of recidivism in criminal justice and
criminology is often used to broadly describe a formerly incarcerated individual who was
sentenced to corrections, released, and then displayed further criminal behavior (Maltz 2001), for
the purpose of this analysis, it is defined instead as a formerly incarcerated individual who was
released from confinement and returned to incarceration at a later date. It is operationalized in
this manner to capture only the most serious relapses of criminal behavior that result in further
confinement. This dataset was complemented by several county contextual variables, which were
incorporated as additional social disorganization predictors to enhance the analysis.
The county contextual variables used as control variables included in this analysis were
carefully selected based on academic research on Social Disorganization Theory. They are meant
to capture various dimensions of the socio-economic and demographic context within each
county as a measure of social disorganization. These variables comprised the percentage of the
population lacking basic prose literacy skills, and the number of individuals enrolled in public
assistance programs, specifically SNAP food assistance, and SSI supplemental income. These
datasets were then normalized for population per 100k.
By integrating these diverse variables, the study aimed to explore potential associations
and shed light on the complex interplay of factors contributing to recidivism at the county level
in Washington State. The inclusion of county contextual variables enables a deeper
understanding of the social, economic, and demographic factors that may influence recidivism
rates across different counties. This approach allows for a more nuanced analysis, considering
the unique characteristics and challenges within each county and their potential impact on
reoffending rates. Furthermore, the temporal scope of the dataset (2012-2022) allows for an indepth examination of recidivism trends over that time, providing valuable insights into what

29
implications they might have for policy and intervention strategies. Overall, this dataset,
encompassing both prison release records and county contextual variables, forms the foundation
for the evidence-based spatial analysis approach used in this study to address the research
objective effectively.
3.1.2 Limitations of the Data
The primary dataset used in this study presents limitations that need to be considered
when interpreting the results (Figure 3). One major limitation is the use of the county scale of
analysis, which may not adequately represent the nuances of the neighborhood context. At this
scale, the spatial variation is constrained to the large areal measurements of the counties,
masking potential spatial heterogeneity that could exist at finer, more local scales of analysis.
Moreover, counties with larger populations will naturally have greater numbers of incidents for
count data variables, potentially biasing the results. Additionally, the dataset obtained from the
Washington State Department of Corrections (WADOC) lacks demographic-related information
and specifics regarding the prison from which individuals were released. The absence of these
crucial details limits a deeper exploration of individual-level characteristics and their potential
influence on recidivism rates. It is essential to acknowledge these limitations to ensure a
comprehensive interpretation of the findings. Despite these constraints, the dataset serves as a
valuable starting point for investigating the correlation between the county of release and the
likelihood of recidivism within Washington State.

30
Figure 3. An image of the raw dataset as received from WADOC
3.2 Data Preprocessing
The raw data received from the Washington Department of Corrections (WADOC) was initially
in PDF format and needed to be converted to CSV format for further processing. The data
provided by the WADOC was merged by the Public Records Unit to combine 10 years of data,
resulting in a large dataset. However, the merged dataset had inconsistent cell sizing and
formatting, requiring reformatting and standardization before proceeding with data exploration
and preparation.
3.2.1 Data Conversion and Standardization
The dataset contained many invalid date values which required transforming them into
valid dates. Many of the date values were in text format, contained leading spaces, or included
special characters as a result of an error in data entry at the point it was collected and entered.
Conversion to valid date values was achieved by using the TRIM and DATEVALUE functions
in Microsoft Excel. The dataset was then resized to extend the range horizontally to
accommodate all the new column headers associated with the admission and release dates. These
headers included the first, second, third, fourth, and fifth admission dates and counties, as well as

31
the first, second, third, fourth, and fifth release dates and counties. The objective was to
consolidate multiple entries into a single entry for each unique identifier (DOCNUM variable),
allowing for a more streamlined analysis.
3.2.2 Handling Missing Values
During the exploratory phase, it was observed that the dataset contained numerous
missing values across various variable columns. Each missing value was carefully assessed to
determine appropriate measures for handling them before proceeding with further analysis. In
some cases, missing values could be explained, such as multiple admission dates and only one
release date indicating a sentence extension while incarcerated. In these instances, the
corresponding cells were left blank. However, there were also cases where admission or release
counties were missing, indicating transfers to the custody of another jurisdiction or department,
such as Immigration and Customs Enforcement or the US Marshal. These instances were
removed from the analysis.
Additionally, approximately 9,000 cells had one admission date without a corresponding
release date, which could be attributed to life sentences, individuals in-custody but awaiting
sentencing, escaped prisoners who were never relocated, or individuals who died while
incarcerated. These entries were removed from the dataset as they did not contribute to the
overall analysis as there was no chance for recidivism.
3.2.3 Dependent Recidivism Outcome Variable and Date of Recidivism
A critical component of this analysis is the dependent outcome variable, which in this
case, indicates whether an individual has recidivated within the given spatiotemporal granularity
of the analysis. By leveraging the IF, AND, and MAX functions, individuals with admission
dates that came after their first release dates were identified in a helper column and labeled as

32
RECIDIVATED. Additionally, all DOCNUMs that exhibited a first release date that was prior to
the first admission date were identified as RECIDIVATED. These dates were stored in a helper
column and used later as the date to identify the correct county of release. The individuals
identified represent cases of recidivism, as they left prison and returned later between 2012 and
2022. Another column was inserted, and a formula was created to label all blank values, or
individuals who had not recidivated, as "0" (“NO RECIDIVISM”), while those labeled as
"RECIDIVATED" were assigned a value of "1". It's worth mentioning that out of the formerly
incarcerated individuals who experienced recidivism after their release, a total of 1,117
individuals faced multiple instances of recidivism. This discovery adds an intriguing layer to the
analysis, but it also brings about important considerations for the results. The decision to treat
these cases of multiple recidivism as separate entries, or to consolidate them as a single entry,
has significant implications on the interpretation and outcomes of the analysis. Treating multiple
instances of recidivism as separate entries could potentially provide a more comprehensive
understanding of the recidivism phenomenon, capturing the varied experiences of individuals.
On the other hand, aggregating these cases into a single entry will simplify the analysis but could
potentially mask nuances in the recidivism patterns.
Treating multiple recidivism cases as a single entry simplifies the dataset, making it
easier to manage and analyze however, given that a substantial portion of the entries involve
multiple recidivism instances and a significant proportion have different release counties for each
instance, consolidating these cases into a single entry might not fully capture the nuanced
variations present in your dataset. For this reason, the data was used with each instance treated
separately.

33
3.2.4 County of Release
After all of the 45,731 incarcerated individuals had been screened to determine
recidivism status from 2012-2022 and that status had been recorded, the next step required
determining which County each individual entry was released to post incarceration. This
required aligning the date of recidivism that we logged into a new helper column (admit date that
most closely follows the first release date) with the county of release for that corresponding
entry. The remaining admission/release entries for each individual DOCNUM (most typically
sentence extensions during incarceration) were removed from the analysis. This left the instances
of recidivism as a binary helper column 1,0 and the County of release next to that helper column.
The next step was to calculate the counts of total releases and incidents of recidivism for each
individual county. To achieve these results the IF and ANDIF functions were utilized to check
both the County name matches the newly created horizontal headers for the County of release
and the recidivism status was a 1 (RECIDIVATED) if that were true, then the formula added a 1
in a new helper column, otherwise it returned zeroes. This permitted the calculation of the total
number of releases and incidents of recidivism per county.
3.2.5 Data Normalization
Once the recidivism data was sorted into the 39 counties with totals for the number of
incidents of recidivism, and the total releases the data was normalized to the total rate of
recidivism per county (Figure 4). To calculate that rate the formula employed was:
= (#/ ) (1)
Opting against normalizing the recidivism rates per 100k as was done with the remaining
variables, was a deliberate choice. The unnormalized rates inherently account for population
sizes, considering that areas with larger populations naturally yield more incidents. Since densely

34
populated regions also have more releases, the existing rates inherently incorporate population
dimensions into the analysis, making normalization unnecessary.
Figure 4. Workflow diagram of the data normalization process
The County social disorganization contextual variables needed to be normalized for
population. The percentage of those lacking basic prose literacy skills in decimal form simply
needed to be multiplied by 100,000 to derive the rate lacking basic prose literacy skills per 100k.
The number enrolled in public assistance (SNAP/SSI) needed to be normalized by dividing the
number enrolled by the county population and then multiplying by 100,000 to get the rate
enrolled in public assistance per 100k.
3.3 Exploratory Data Analysis
In the subsequent section of this study, a transition occurred from data preparation and
normalization to the statistical analysis phase. This analysis followed a step-by-step approach
aimed at comprehensively examining the relationship between the county of release, social

35
disorganization contextual variables, and recidivism rates in Washington State. Initially, baseline
descriptive statistics were calculated, including the recidivism rate for each county, which was
used to determine the overall recidivism rate as well as county rates for WADOC-operated
facilities in the State of Washington during the period from 2012 to 2022. Subsequently, a power
analysis was conducted to ascertain the appropriate number of total releases for each county
(sample size) for this study. Following this, summary statistics were computed and reported to
provide a clear overview of the dataset's characteristics.
To gain deeper insights into the data, univariate data visualizations were employed to
assess distributional characteristics, and transformations were applied to variables that exhibited
significant deviations from a normal distribution. After transformation, the Shapiro-Wilks test
was utilized to reevaluate the normality of the data.
The analysis then shifted focus to exploring the linear relationships between the social
disorganization contextual variables and recidivism rates through bivariate data visualizations.
To quantify these relationships, Pearson’s correlation coefficients were used. Finally, attention
was turned to the crucial question of whether the county of release was correlated with
recidivism rates in Washington State. This was achieved through the application of both
ANOVA and T-tests, further supported by Cohen’s d for effect size calculations.
Descriptive statistics were employed to gain initial insights into the distribution and
characteristics of the data. Key summary statistics, including means, standard deviations,
medians, and ranges were calculated for the variables incorporated into this study within Excel
utilizing the Data Analysis Tool. This step provided a snapshot of the overall recidivism rates
and a glimpse into the context of the county of release of the incarcerated individuals across
Washington State, especially in terms of social disorganization.

36
3.3.1 Baseline Descriptive Statistics
To assess the relationship between the county of release and propensity to recidivate, a
contingency table was created in Excel. A contingency table, also known as a cross-tabulation or
crosstab, displays the frequency distribution of various categorical variables in tabular format
(Agresti 2002). Crosstabs aid in better understanding the relationship between the variables. The
creation of a crosstab table allowed for pairwise comparison between the counts of recidivism
and the identification of initial patterns present in the data. Furthermore, it allowed for the
calculation of the mean rate of recidivism per county which was used as the dependent variable
in this analysis. After creating a contingency table of the total number of releases per county and
the total number of recidivism incidents, the rate of recidivism was calculated for the 39 counties
across the State of Washington, notating the counties that exhibited low numbers of total
releases.
A power analysis was conducted in R.studio utilizing the ‘pwr’ package to derive a
sample size with a .80 statistical power (Travers et al. 2021). A weighted average of Cohen’s d
(“Pooled d”) was derived and input as the estimated effect size for the analysis. The alpha was
set to .05, with 39 groups. The results of the power analysis derive adequate sample sizes per
group for the selected statistical power. Once the number of total releases threshold was
established, counties that did not meet that threshold were removed from the analysis.
To delve into the data’s characteristics, summary statistics were calculated and analyzed
for each of the four variables within Excel, utilizing the Data Analysis Tool. This process aimed
to reveal central tendencies, comprehend the data distribution, and gauge variance. Furthermore,
attention was paid to any significant deviations from normal distributions, which could
potentially violate the assumptions of subsequent statistical tests. The results were organized in

37
separate tables for each variable, facilitating a thorough examination and interpretation of the
data's key attributes and overall distributional characteristics.
3.3.2 Univariate Data Visualization
Analyzing the data's distribution provides a foundational understanding of the dataset's
characteristics. It aids in formulating hypothesis testing, assessing the normality of the data, and
determining the most suitable statistical test for a given use case. This study leveraged both
histograms and boxplots to further assess the data’s distribution. To do so, the data was imported
into R.Studio and the ggplot2 package was utilized. Charts were created for all of the variables in
the analysis.
3.3.3 Data Transformation
Based on the findings of the summary statistics and the univariate data visualizations it
was determined that the rate of those lacking basic prose literacy skills per 100k needed to be
transformed using a natural logarithmic function, this was done in Excel using the LN() function.
In Excel, the LN() function calculates the natural logarithm of a given number. The natural
logarithm is the logarithm to the base "e," where "e" is the mathematical constant approximately
equal to 2.71828 (Agresti and Finley 2008). To reassess for normalcy, a Wilks-Shapiro test was
conducted on each of the variables, summary statistics were calculated and univariate data
visualizations were created for the log transformed variables in R.studio with the ggplot2
package (Kassambara 2019). The results of the Wilks-Shapiro tests were used to reassess the
data’s distribution prior to further analysis.

38
3.3.4 Multivariate Data Visualization
Scatter plot matrices (SPLOMs) efficiently visualize pairwise relationships among
multiple variables in a matrix format (David and Tukey 1977). They consolidate multivariate
information, allowing for simultaneous comparisons across different variable pairs (Chambers et
al. 1983). The increased interpretability of SPLOMs makes identifying linear and nonlinear
relationships, clusters, and outliers more feasible when working with large datasets (Agresti
2002). This analysis constructed a SPLOM using the ‘ggally’ package and the ggpairs command
in R.Studio. The resulting visualization showcases Pearson's correlation coefficients and
scatterplots, enhancing the ability to discern and understand linear relationships within the data.
3.3.5 Correlation Analysis – Pearson’s Correlation Coefficients
The Pearson’s correlation coefficient assesses how closely the data points in a scatterplot
align along a straight line, indicating whether one variable tends to increase as the other
increases, decrease as the other increases, or shows no relationship. The formula to calculate
Pearson's correlation coefficient is as follows:
=
∑(−̅)(−̅)
√∑(−̅)2(−̅)2
(2)
Where is Pearson’s correlation coefficient, and are individual observations from the two
variables, and ̅ and ̅ are the mean averages of the corresponding group (Agresti & Finley
2008). The formula calculates the strength and direction of linear relationships between two
variables (Pearson 1900). For the purpose of this analysis the Pearson’s correlation coefficients
are used to determine if a relationship exists between recidivism rates and the variables meant to
reflect elements of SDT.

39
3.3.6 Analysis of Variance (ANOVA)
A single-factor Analysis of Variance (ANOVA) tests whether the means of two or more
groups are significantly different from each other. The equation for a one-way ANOVA statistic
can be expressed as follows:
=

ℎ
(3)
For the purpose of this analysis, the ANOVA test is utilized to determine if there is a relationship
between the county of release and a propensity to recidivate in Washington State from 2012-
2022.
3.3.7 One Sample T-Test
After the initial ANOVA test was employed to assess whether there were significant
differences in recidivism rates among counties, a pairwise T-test was executed within RStudio on
the recidivism rates per county to both verify the findings of the ANOVA and identify potential
individual counties that exhibit significant deviations from the sample mean. The t-test, when
used to compare each county's recidivism rate to the sample mean rate, is a univariate test that
examines whether each county's rate of recidivism significantly deviates from the overall
population mean. The t-test approach treats each county's recidivism rate independently and
doesn't consider potential differences between groups. It is useful in identifying the statistically
significant rates of recidivism that fall above and below the sample mean. The results of this test
will be leveraged to verify the results of the ANOVA, and to identify individual counties that
may deviate from the population mean that could not be identified in an ANOVA.
Utilizing the Base R package and the t.test() function the T-test was used to compare each
county's recidivism rate to the average rate of all counties. The T-statistic measures how many
standard errors the county's rate deviates from the mean rate of all counties. This statistic

40
indicates the extent to which the county's mean varies from the average rate of all counties. The
sign (+/-) of the T-statistic denotes the direction of this variance. The P-value of the test reveals
whether the difference from the collective average rate holds statistical significance. A onesample T-test is specified as:
=
−
( ./√
(4)
3.3.8 Bonferroni Correction
A Bonferroni Correction is a statistical correction made to control the familywise error
rate when multiple comparisons or hypothesis tests are conducted simultaneously. It is a method
used to mitigate the problem of inflated Type I error rates that can occur when conducting
multiple statistical tests on the same dataset (Agresti & Finley 2008). To implement the
correction, the significance level (alpha) is divided by the number of tests being conducted.
3.3.9 Cohens d – Calculating Effect Size
After determining whether there were significant differences in the mean rates of
recidivism across the 30 counties in the analysis via ANOVA and pairwise T-tests, the effect size
of these differences was calculated to determine the extent to which those differences are
influenced by the County of release. This was achieved by utilizing Cohens d which is specified
as:
=
(̅1−̅2
)

(5)
Where d is the effect size, x̅1 is the mean of the first(dependent) group, x̅2 is the mean of the
second (independent) group, and s is the pooled standard deviation of both groups (Cohen 1988;
Rosenthal and Rosnow 1991). The pooled standard deviation is calculated as:

41
=
√(1−1)1
2+(2−1)2
2
1+2−2
(6)
Where is the pooled standard deviation, 1 is the sample size of the first group and 2is the
sample size of the second group, and 1 and 2 are the standard deviations of their respective
groups (Rosenthal & Rosnow 1991). Cohen's d provides a measure of the standardized effect
size, where larger values of d indicate a larger difference between the group means. The results
of the analysis are usually interpreted based on common threshold d values and significance is
interpreted based on corresponding p-values. The results of the analysis can be read based on
these threshold d values:
Small effect size: d ≈ 0.2
Medium effect size: d ≈ 0.5
Large effect size: d ≈ 0.8
3.4 Exploratory Spatial Data Analysis
To gain a deeper understanding of the underlying spatial patterns and dependencies in the
dataset, an Exploratory Spatial Data Analysis was performed. This study utilized global spatial
autocorrelation measures, followed by an assessment of local clusters and spatial autocorrelation.
All of this is meant to provide a more meaningful understanding of the data, and to help answer
the underlying research question of whether the county of release is related to recidivism rates
for those incarcerated in WADOC facilities from 2012-2022.
The Moran's I test is used to assess spatial autocorrelation, which measures the degree of
similarity or dissimilarity between the values of neighboring locations across a global study area.
In this case, the Moran's I test was applied to the Recidivism rates to assess the underlying
spatial patterns of the dataset. This study further explored the spatial patterns by employing local

42
spatial autocorrelation measures. These local measures provided a more detailed examination of
clustering or dispersion at the individual observation level, shedding light on specific areas with
notable high-value hotspots or low-value coldspots.
3.4.1 Spatial Weights Matrix (SWM)
When assessing spatial autocorrelation of recidivism rates at the county scale, it is crucial
to establish a Spatial Weights Matrix (SWM), which represents the neighborhood relationships
in a matrix format. The selection of an appropriate SWM should align with the underlying spatial
processes at play.
To identify the optimal distance threshold for this assessment, the study drew insights
from previous studies in academic literature that employed similar mathematical methodologies
for crime or recidivism analysis. These prior research efforts often incorporated a distance decay
function to model the movement patterns of formerly incarcerated individuals (Piquero,
Farrington, and Blumstein, 1999; Hipp, 2010). In addition, some studies have used the median
Census Tract size to determine a plausible distance threshold (Hipp, 2010). This analysis utilizes
a combination of those two elements. To further cross validate these findings a K-nearest
neighbors SWM was also utilized.
3.4.2 Moran's I Index for Spatial Autocorrelation
To examine the presence of spatial autocorrelation, the Moran's I index was utilized. The
Moran's I test for spatial autocorrelation measures the degree of spatial clustering or dispersion
in the data and assesses whether recidivism rates in neighboring counties were more or less
similar than expected under a random distribution (O’Sullivan and Unwin 2010). Positive values
of Moran's I indicate spatial clustering, while negative values suggest spatial dispersion. The
Global Moran’s I statistic is defined as:

43
=

∑ ( − ̅ )
2
=1
×
∑ ∑

=1

=1
( − ̅ )( − ̅ )
∑ ∑

=1

=1
(7)
where x is the attribute value at a given location, ̅ is the mean attribute value across the study
area, is the weight of the distance between location i and its neighbor j, and n is the number
of features within a specified distance from location i (O’Sullivan and Unwin 2010).
3.4.3 Getis-Ord Hot Spot Analysis
The Getis-Ord
∗
statistic, a local spatial autocorrelation technique, was employed to
identify spatial clusters (hot spots) of high and low recidivism rates across Washington State.
This analysis considered the county-level recidivism data and explored whether certain areas
exhibited significant clustering patterns. Getis and Ord have both alluded to the benefits of
leveraging both the Moran’s I and Getis-ord
∗
statistics in tandem when analyzing spatial
datasets (Getis and Ord 1992). Both global and local measures of spatial autocorrelation measure
different things and can point to different underlying spatial processes and patterns (O’Sullivan
and Unwin 2010). The global Moran’s I measure of spatial autocorrelation measures across an
entire study area without distinguishing between concentrations of high and low values, while
the Getis Ord
∗
statistic allows for that distinction to be analyzed. This gives a more nuanced
finer resolution look into the local clusters of hotspots and coldspots that may arise across the
entire study area. The Getis-Ord
∗
is specified as:
∗ () =
−1
(ⅆ)−̅−1
(ⅆ)
√∑
2 ()

=1
−(
=1
()
)
2
−1
(8)
Where (d) is the spatial weight between observation i and observation j at distance d. In our
case using a contiguity based SWM the distance would be substituted for a binary value based on
whether observations i and j are neighbors. Xj
is the attribute value for the
ℎ
location. ̅is the

44
mean of the attributes for all values. is the standard deviation of the attribute values from all
locations, and is the total number of observations (Getis and Ord 1992). The result of a Getisord
∗
hotspot analysis are
∗
values which are essentially z-scores. Positive values indicate
clustering of high values while negative values indicated clustering of low values. The
magnitude of the z-score indicates the intensity of clustering (Getis and Ord 1992).
3.4.4 Interpretation of Findings
Finally, the statistical findings were interpreted to draw meaningful conclusions
regarding the correlation between the county of release and recidivism likelihood in Washington
State. The results from the various statistical methods were synthesized to identify significant
associations, spatial patterns, and potential areas of intervention.

45
Chapter 4 Results
This chapter describes the results of the analytical workflow described in Chapter 3. It begins
with the results of the EDA and moves on to the results of the ESDA.
4.1 Results of Exploratory Data Analysis
This study conducted a comprehensive analysis of the summary statistics for the rate of
recidivism across counties. Additionally, it delved into key variables associated with social
disorganization, a factor that has been linked to recidivism rates in previous research. The
primary focus was on determining if the county of release is correlated with a propensity to
recidivate among WADOC incarcerated individuals from 2012 to 2022. By examining essential
variables such as ‘Public assistance enrollment per 100k’ and the rate ‘Lacking basic prose
literacy skills per 100k’, another objective is to extract meaningful insights into the intricacies of
recidivism and the prospective role of social disorganization as a predictor. Each summary
statistic reveals the central tendencies, variability, and distributional properties of these variables,
allowing us to draw more informed conclusions regarding the potential relationship between the
county of release and recidivism rates.
4.1.1 Baseline descriptive statistics
The total recidivism rate for all WADOC Formerly Incarcerated Individuals from 2012-
2022 is 41.0028%. Specifically, of 45,731 incarcerated individuals, 18,751 were reconfined in
WADOC operated facilities between 2012-2022.When broken down by County, the rate of
recidivism highlights unique patterns across the study area (Table 3). These rates highlight
unique patterns and trends that further enhance understanding of the phenomenon of recidivism
in Washington State. Notably, the range, approximately 35.71% to 58.06%, highlights significant

46
heterogeneity in reoffending behavior across counties. These rates underscore the nuanced nature
of recidivism and suggest that localized factors may be contributing to the observed variations.
Upon closer examination of the data, several counties stand out with small sample sizes and
outlier recidivism rates. San Juan County, for instance, records the highest rate of 58.06% with a
sample size of 31 total releases, and Wahkiakum County at 50.00% has a sample size of only 18
total releases. On the other end of the spectrum, Ferry County has the lowest rate at 35.71% and
a sample size of only 28 total releases. While these disparities may reflect distinct social,
economic, and environmental contexts within each county, to run a meaningful statistical
analysis to determine whether the county of release is related to recidivism outcome, adequate
sample sizes are necessary. To determine an adequate number of total releases the weighted
average of the Cohens d values was input into a power analysis as the estimated effect size.

47
Table 3 Crosstab of recidivism vs. total releases with derivative rates of recidivism by county.
COUNTY ADAMS
ASOTI
N
BENTO
N CHELAN
CLALLA
M CLARK
COLUMBI
A
RECIDIVATED
YES 39 79 616 254 172 1210 15
TOTAL RELEASES 84 184 1565 565 405 2828 39
PERCENT
RECIDIVATED 0.4643 0.4293 0.3936 0.4495 0.4246 0.4278 0.3846
COUNTY
GRAYS
HARBOR ISLAND
JEFFERS
ON KING KITSAP
KITTITA
S KLICKITAT
RECIDIVATED
YES 353 89 32 3579 832 90 46
TOTAL RELEASES 927 206 80 8941 1973 175 102
PERCENT
RECIDIVATED 0.3807 0.4320 0.4 0.4002 0.4216 0.5142 0.4509
COUNTY PIERCE
SAN
JUAN SKAGIT
SKAMANI
A
SNOHOMI
SH
SPOKAN
E STEVENS
RECIDIVATED
YES 2889 18 407 17 1504 1711 104
TOTAL RELEASES 7070 31 950 39 3625 4215 234
PERCENT
RECIDIVATED 0.4086 0.5806 0.4284 0.4358 0.4148 0.4059 0.4444
COUNTY FERRY
FRANK
LIN
GARFIE
LD GRANT MASON
OKANOG
AN PACIFIC
RECIDIVATED
YES 10 235 4 295 208 127 74
TOTAL RELEASES 28 559 8 681 540 300 165
PERCENT
RECIDIVATED 0.3571 0.4203 0.5 0.4331 0.3851 0.4233 0.4484
COUNTY LEWIS LINCOL
N
WHITM
AN
THURSTO
N
WAHKIA
KUM
WHATC
OM
WALLA
WALLA
RECIDIVATED
YES 413 9 16 739 8 594 158
TOTAL RELEASES 1026 25 44 1850 16 1416 382
PERCENT
RECIDIVATED 0.4025 0.36 0.3636 0.3994 0.5 0.4194 0.4136
COUNTY COWLITZ DOUGL
AS
YAKIM
A
PEND
OREILLE
RECIDIVATED
YES 654 76 1057 18
TOTAL RELEASES 1676 192 2538 47

48
4.1.2 Power analysis
To ascertain a suitable sample size of total releases for subsequent statistical analyses, a
power analysis was performed. One of the decisions that needs to be made when conducting a
power analysis is on the chosen statistical power of the analysis. Statistical power is the
probability that a study will correctly detect a true effect or relationship if it exists in the
population. To settle on a statistical power of 80% for the power analysis in this study, a
literature review within the field of recidivism and crime studies was conducted. While the
literature revealed limited studies that explicitly detailed power analysis parameters, a common
practice emerged in human and behavioral science studies, as well as related research within this
field, where a statistical power of 80% was employed (Travers et al. 2021; Nesset et al. 2020).
This choice also aligns with the conventional approach adopted in most research domains,
ensuring a reasonable balance between the probability of detecting true effects and the risk of
making a Type I error.
The next decision when conducting a power analysis involves estimating the effect size—
a crucial parameter that gauges the magnitude of the phenomenon under investigation. In this
study, we opted for a robust approach by utilizing Cohen's d to derive a weighted average effect
size, often referred to as the "Pooled d," across all 39 counties.
The calculation of the Pooled d involves summing the weighted Cohen's d values and
subsequently dividing this sum by the total of the weights. The results revealed that the Pooled d
equates to approximately -0.0291039964 (Table 4). In more concrete terms, this signifies that the
estimated effect size is characterized as a small negative effect, providing a nuanced
understanding of the impact being examined in our research.
PERCENT
RECIDIVATED 0.3902 0.3958 0.4164 0.3829

49
Table 4 Results of the weighted average pooled d for effect size.
COUNTY COHENS D VARIANCE WEIGHTS
WEIGHTED
COHENS D STD DEV
ADAMS 0.9357 0.0171 58.4230 54.6678 0.0544
ASOTIN 0.1389 0.0054 182.2412 25.3184 0.0364
BENTON -0.6738 0.0007 1275.4633 -859.4108 0.0123
CHELAN 0.6010 0.0020 478.5530 287.6437 0.0209
CLALLAM 0.0342 0.0024 404.7631 13.8456 0.0245
CLARK 0.1070 0.0003 2811.8863 301.0310 0.0093
COLUMBIA -0.8786 0.0355 28.137497 -24.7242 0.1153
COWLITZ -0.7512 0.0007 1307.1741 -981.9568 0.0119
DOUGLAS -0.6237 0.0062 160.7350 -100.2534 0.0352
FERRY -1.5047 0.0761 13.1324 -19.7609 0.1428
FRANKLIN -0.0636 0.0017 557.8686 -35.5280 0.0208
GARFIELD 1.7484 0.3160 3.1638 5.5318 0
GRANT 0.2277 0.0015 663.79 151.1544 0.0189
GRAYS HARBOR -0.9652 0.0015 632.4158 -610.4092 0.0159
ISLAND 0.2003 0.0049 201.9451 40.4688 0.0345
JEFFERSON -0.5281 0.0142 70.2095 -37.0778 0.0547
KING -0.5212 0.0001 7871.5478 -4103.2290
0.0051
KITSAP -0.0340 0.0005 1971.8542 -67.2205
0.0111
KITTITAS 2.0740 0.01800 55.5424 115.1951 0.0377
KLICKITAT 0.6329 0.0117 84.9782 53.7862 0.0492
LEWIS -0.4711 0.0010 923.4843 -435.1358 0.0153
LINCOLN -1.4387 0.0813 12.2852 -17.6750 0.14
MASON -0.8650 0.0025 392.9725 -339.9345 0.0209
OKANOGAN 0.0023 0.0033 299.9991 0.7004 0.0285
PACIFIC 0.5760 0.0070 141.5211 81.5200 0.0387
PEND OREILLE -0.9151 0.0301 33.1284 -30.3164 0.1170
PIERCE -0.3323 0.0001 6700.0373 -2226.5506 0.0058
SAN JUAN 3.5833 0.2393 4.1777 14.9704 0.0806
SKAGIT 0.1184 0.0010 943.3831 111.7337
0.0160
SKAMANIA 0.2891 0.0267 37.4347 10.8254 0.0641
SNOHOMISH -0.1888 0.0002 3561.4605 -672.7455 0.0081

50
In crafting a rigorous power analysis for this study, essential parameters were carefully
considered. The significance level was set at 0.05 and aimed for a statistical power of 0.80,
ensuring a strong likelihood of detecting true effects. Our analysis included 39 distinct counties,
each with unique characteristics. The estimated effect size, Pooled d ≈ -0.0291, indicates a small
negative effect. These parameters establish a robust foundation aligned with our study's
objectives. The analysis determined that a total number of releases sample size of n=2403 yields
a power of 0.80. This implies that for a balanced one-way ANOVA, each group should have a
minimum of 62 observations or total releases, providing a power of 0.8035 (Figure 5).
Figure 5 Results of a power analysis to determine sample size.
SPOKANE -0.3937 0.0002 3911.7103 -1540.3775 0.0075
STEVENS 0.4826 0.0047 209.5846 101.1641
0.0324
THURSTON -0.5394 0.0006 1614.9835 -871.2608 0.0113
WAHKIAKUM 1.7484 0.15803 6.3277 11.0637 0
WALLA WALLA -0.2184 0.0026 373.0945 -81.5178 0.0251
WHATCOM -0.0841 0.0007 1411.0013 -118.7700 0.0131
WHITMAN -1.3567 0.0436 22.9117 -31.0859 0.1363
YAKIMA -0.1524 0.0003 2508.838 -382.5251 0.0097

51
4.1.3 Eliminating the counties with insufficient sample sizes
After determining the sample size threshold of 62 total releases, the counties that
exhibited less than that were eliminated from further analysis. Those counties include Columbia
with 39 total releases, Ferry with 28, Garfield with 8, Lincoln with 25, Pend Oreille with 47, San
Juan with 31, Skamania with 39, Wahkiakum with 16, and Whitman with 44.
Figure 6 Map of the counties that were removed from further analysis.
4.1.4 Summary statistics
To delve deeper into the characteristics of the data, the summary statistics of each
variable were calculated and examined (Table 5). The analysis of recidivism rates reveals a
considerable degree of variability. With a standard deviation of approximately 2.72%, the data's
dispersion is significant. A positively skewed distribution (skewness ≈ 1.364) suggests a
tendency toward higher rates of recidivism, while a peaked distribution (kurtosis ≈ 3.448)
indicates the presence of extreme values. The 95% confidence interval (±0.1335) provides a
COUNTIES ELIMINATED
FROM ANALYSIS
Columbia San Juan
Ferry Skamania
Garfield Wahkiakum
Lincoln Whitman
Pend Oreille

52
reliable estimate of the mean recidivism rate, showing a diverse range of post-release
experiences.
The average enrollment in public assistance programs is approximately 36,485.06 people
per 100k population, indicating a moderate level of enrollment. Substantial variability is evident,
with a standard deviation of approximately 3,721.79, reflecting differing levels of need and
economic circumstances across regions. A mildly negatively skewed distribution (skewness ≈ -
0.57) suggests regions with higher enrollment rates, indicative of reliance on public assistance.
The distribution's relatively low kurtosis (≈ 0.84) indicates a moderate presence of extreme
values or outliers.
The rate of individuals lacking basic prose literacy skills is approximately 12,466.67
people per 100k, indicating a moderate level of individuals with this challenge. The distribution
exhibits a positively skewed pattern (skewness ≈ 2.07), implying areas with higher rates of those
lacking literacy skills. A wide range of rates (from 6,000 to 34,000) underscores significant
disparities in literacy levels across regions. A moderately high kurtosis value (≈ 3.91) indicates a
distribution with a notable presence of extreme values or outliers.
In summary, these variables' distributions demonstrate significant variability, skewed patterns,
and the presence of extreme values. These irregularities in the distributions prompt consideration
of data transformation. Specifically, the highly positively skewed distribution and peaked
kurtosis observed in literacy rates underscore the potential presence of extreme values.
Addressing these irregularities through transformation may allow for more accurate analysis,
interpretation, and identification of potential underlying factors.
Table 5 Summary statistics.
Variab
le Mean
Stand
ard
Error
Medi
an
Stand
ard
Deviat
ion
Sampl
e
Varia
nce
Kurt
osis
Skew
ness
Ran
ge
Minim
um
Maxim
um Sum
Confid
ence
Lvl.
95.0%

53
Recidiv
ism
Rates
0.421
3 0.0049
0.419
9 0.0272 0.0007
3.447
7 1.3644
0.133
5 0.3808 0.5143
12.63
99 0.0102
Enrolle
d in
Public
Assista
nce per
100k
36485
.06
679.50
24
36854
.47
3721.7
88
13851
706
0.836
2
-
0.5725
1669
8.2
27081.
01
43779.
21
1094
552 1389.72
Lackin
g Basic
Prose
Literac
y Skills
per
100k
12466
.67
1264.6
69 10000
6926.8
76
47981
609
3.907
9 2.0717
2800
0 6000 34000
3740
00 2586.52
4.1.5 Univariate Data Visualization
Both histograms and box plots were leveraged to assess the data for normality and to
identify outliers.
4.1.5.1 Rate of recidivism
The analysis of the histograms and boxplots revealed interesting patterns in the
distributions of various variables. The rate of recidivism data reveals that the majority of
counties have recidivism rates falling within the range of approximately 0.39 to 0.43 (Figure 7).
This central cluster suggests consistency in these counties' rates of recidivism. There is a single
county outlier with a high recidivism rate greater than 0.50. The outlier emphasizes the diversity
in recidivism rates among the counties in Washington State.
Figure 7 Histogram and boxplot of the rate of recidivism by county.

54
4.1.5.2 The rate enrolled in public assistance per 100k population
The distribution of the rate enrolled in public assistance per 100k population across the
30 counties in Washington State reveals a range of frequencies within distinct enrollment rate
intervals. Notably, most counties exhibited enrollment rates primarily falling within two intervals
of approximately 27,000 to 30,500 and 37,000 to 40,500. This suggests consistency in
enrollment patterns among these counties (Figure 8). Moreover, two counties stand out with
substantially lower rates, approximately 27,000 and 30,500, highlighting potential
socioeconomic disparities among counties in the State.
Figure 8 Histogram and boxplot of the rate enrolled in public assistance per 100k.
4.1.5.3 Lacking basic prose literacy skills per 100k population
The rate lacking basic prose literacy skills per 100k population variable exhibited positive
skewness in the distribution of the histogram as well. This indicates that most counties have a
rate lacking literacy between 5,000 and 15,000 people per 100k. Furthermore, there are
substantial outliers present in the dataset that tail off to the right side of the distribution. This
means that the tail of the distribution is elongated towards the higher values more than that of a
normal distribution. This is further confirmed by visually inspecting the boxplot and histogram
(Figure 9). The boxplot appears to show that there are four substantial outliers in the data above
the sample mean. Based on the summary statistics and the univariate visual analysis on this
variable, transformation is required.

55
Figure 9 Histogram and boxplot of the rate lacking basic prose literacy skills per 100k.
4.1.6 Data Transformation
Based on the skewness values and the results of the univariate data visualizations, the rate
lacking basic prose literacy skills per 100k exhibited a moderate to highly skewed distribution
which required transformation. The symmetry of the variable’s distribution improved
significantly after a natural logarithmic transformation (Figure 10).
Figure 10 Histogram of the log transformed variable - highlighting the improved symmetry.
Prior to logarithmic transformation, the individuals lacking basic prose literacy skills per
100k exhibited a positively skewed distribution with a kurtosis value of 3.908 and a skewness of
2.072. However, after applying the logarithmic transformation, these distribution characteristics
improved, resulting in reduced kurtosis (1.082) and skewness (1.227) values. Although it still
displayed positive skewness, the distribution became more balanced. The improvement in
distributional symmetry was confirmed through visual inspection of the histograms and the
results of the proceeding Shapiro-Wilks tests.

56
4.1.7 Reevaluating Data Distribution for Normality
The Shapiro-Wilks test was employed to assess the normality of the variables’
distributions, a critical consideration for the proceeding statistical analyses. While the test
resulted in a p-value of 0.01623, which is below the typical alpha threshold of 0.05, suggesting
statistical significance, it's crucial to look beyond this. The test's W-statistic for the rate of
recidivism is 0.91149, indicating a distribution that is relatively close to normal (Table 6). There
is a pronounced outlier above a 50% recidivism rate that could be impacting the p-value. This
nuanced perspective suggests that while there may be some departure from perfect normality, the
distribution is within a range that is compatible with the assumptions of ANOVA. Therefore,
despite the statistical significance, the rate of recidivism distribution exhibits characteristics that
align reasonably well with the requirements for ANOVA, supporting its applicability for
subsequent analyses.
The public assistance enrollment per 100k population variable displayed a higher Wstatistic of 0.96346 with a p-value of 0.3786. In this case, the p-value is not statistically
significant, implying that the distribution is more consistent with a normal distribution. It is
important to note that some departure from normality may still exist.
The logarithm of the rate lacking basic prose literacy skills per 100k exhibited a Wstatistic of 0.8794, accompanied by a small p-value of 0.002732. While this p-value suggests
statistical significance and some departure from perfect normality, it's essential to note that the
W-statistic of 0.8794 is relatively high. Compared to the W-statistics of other variables in the
study, which are mostly at or above 0.90, this value indicates that the deviation from normality,
while statistically significant, is not as pronounced as it may seem at first glance. This means that
while there is evidence of non-normality, the distribution of the logarithm of those lacking basic

57
prose literacy skills per 100k is not dramatically far from a normal distribution, and it remains
within a range that is suitable for statistical analyses.
Table 6 Results of the Shapiro-Wilks tests.
Variable Rate of Recidivism Public Assistance Log Lacks Literacy
W-statistic 0.9115 0.9635 0.8794
P-value 0.0162 0.3786 0.0027
4.1.8 Multivariate Data Visualization and Pearson’s Correlation Coefficients
This study conducted a thorough scatterplot analysis to examine the connections between
recidivism and county contextual variables that align with prior research grounded in Social
Disorganization Theory. The key contextual variables included the public assistance enrollment
per 100k, and the rate lacking basic literacy prose per 100k (log transformed). The results
yielded intriguing insights. Notably, the Pearson’s Correlation Coefficients, which found no
significant correlations between the rate of recidivism and the other variables, indicating an
absence of linear relationships. Based on these results it appears that while these variables have
contributed to a positive correlation to crime and recidivism in other studies conducted
elsewhere, in Washington State, these variables do not contribute to a propensity to recidivate at
the County scale of analysis.
The only noteworthy findings that surfaced in the form of robust and statistically
significant correlations were seen in the variables linked to social disorganization. Notably, a
strong and statistically significant negative correlation was evident between public assistance
enrollment per 100k and the rate lacking basic prose literacy skills per 100k with a Pearson’s
correlation coefficient of -0.726 (p < 0.001). This correlation suggests that as levels of public
assistance enrollment increase, the rate lacking literacy tends to decrease. This is a

58
counterintuitive finding that would require further analysis to dissect, which is outside the realm
of this study. While these findings are interesting, they show no relevance to our analysis.
Table 7 Scatterplot matrix with density plots and Pearson's correlation coefficients.
4.1.9 Single-Factor Analysis of Variance (ANOVA)
The one-sample ANOVA test compared the distribution of recidivism rates among
different counties to determine if there are statistically significant differences in these rates based
on the county of release. The test statistic is calculated to be nearly zero for the between group
variation. For a one-sample ANOVA, the test statistic is used to determine whether the observed
differences between the sample mean and the population mean are statistically significant across
groups. The p-value associated with the F-statistic is 1. This p-value indicates the likelihood of
observing these differences due to random chance alone.
Table 8 Results of the single factor ANOVA.
Source of
Variation SS df MS F P-value F crit
Between
Groups 5.55E-17 29 1.91E-18 1.39E-16 1 1.8474
Within
Groups 0.4142 30 0.0138
Total 0.4142 59

59
Since the p-value is very high (1), it suggests that there isn't enough evidence to conclude
that the differences in recidivism rates among the counties are statistically significant. In other
words, based on this test, there is insufficient evidence to reject the null hypothesis, which states
that there are no differences in recidivism rates among the counties. In this case, the high p-value
suggests that the differences observed are likely due to random variation rather than the effect of
the county of release.
4.1.10 Two-tailed one sample T-test and Cohen’s d
The t-test, when used to compare each county's recidivism rate to the sample mean rate,
is a univariate test that examines whether each county's rate of recidivism significantly deviates
from the overall population mean. The t-test approach treats each county's recidivism rate
independently and doesn't consider potential differences between groups. It is useful in
identifying the statistically significant rates of recidivism that fall above and below the sample
mean. The results of this test were leveraged to verify the results of the ANOVA, and to identify
individual counties that may deviate from the population mean that could not be identified in an
ANOVA (Table 9).
Table 9 Results of the T-test and Cohen's d.
COUNTY
RATE
RECID
STD
DEVIATION
SQRT
SAMPLESIZE
T
STATISTIC P_VALUE BONFERRONICOHENS_D
COHENS
P_VALUE
ADAMS 0.4643 0.0544 9.1652 7.237454 0.0000 0.0017 0.7897 0.5680
ASOTIN 0.4293 0.0365 13.5647 2.96269 0.0060 0.0017 0.2184 0.1726
BENTON 0.3936 0.0124 39.5601 -88.8291 0.0000 0.0017 -2.2454 0.9751
CHELAN 0.4496 0.0209 23.7697 32.1089 0.0000 0.0017 1.3508 0.8227
CLALLAM 0.4247 0.0246 20.1246 2.7612 0.0098 0.0017 0.1372 0.1090
CLARK 0.4279 0.0093 53.1789 37.5528 0.0000 0.0017 0.7062 0.5198
COWLITZ 0.3902 0.0119 40.9390 -106.958 0.0000 0.0017 -2.6126 0.9909
DOUGLAS 0.3958 0.0353 13.8564 -10.0235 0.0000 0.0017 -0.7234 0.5296
FRANKLIN 0.4204 0.0209 23.6432 -1.0532 0.3009 0.0017 -0.0445 0.0355
GRANT 0.4332 0.0190 26.0960 16.3132 0.0000 0.0017 0.6251 0.4679

60
GRAYS
HARBOR 0.3808 0.0159 30.4467 -77.3736 0.0000 0.0017 -2.5413 0.9887
ISLAND 0.432 0.0345 14.3527 4.43722 0.0001 0.0017 0.3092 0.2424
JEFFERSON 0.4 0.0548 8.9443 -3.48317 0.0015 0.0017 -0.3894 0.3019
KING 0.4003 0.0052 94.5569 -383.8 0.0000 0.0017 -4.0586 0.9999
KITSAP 0.4217 0.0111 44.4185 1.47826 0.1501 0.0017 0.0333 0.0265
KITTITAS 0.5143 0.0378 13.2288 32.5528 0.0000 0.0017 2.4608 0.9851
KLICKITAT 0.451 0.0493 10.0995 6.08198 0.0000 0.0017 0.6022 0.4516
LEWIS 0.4025 0.0153 32.0312 -39.3949 0.0000 0.0017 -1.2299 0.7809
MASON 0.3852 0.0209 23.2379 -40.0917 0.0000 0.0017 -1.7253 0.9149
OKANOGAN 0.4233 0.0285 17.3205 1.19614 0.2413 0.0017 0.0691 0.0550
PACIFIC 0.4485 0.0387 12.8452 9.01407 0.0000 0.0017 0.7017 0.516
PIERCE 0.4086 0.0058 84.0833 -183.085 0.0000 0.0017 -2.1774 0.9705
SKAGIT 0.4284 0.0161 30.8221 13.5728 0.0000 0.0017 0.4404 0.3402
SNOHOMISH 0.4149 0.0082 60.2080 -47.3078 0.0000 0.0017 -0.7857 0.5679
SPOKANE 0.4059 0.0076 64.9230 -132.4399 0.0000 0.0017 -2.0400 0.9585
STEVENS 0.4444 0.0325 15.2971 10.8640 0.0000 0.0017 0.7102 0.5217
THURSTON 0.3995 0.0114 43.0116 -82.4552 0.0000 0.0017 -1.9170 0.944
WALLA
WALLA 0.4136 0.0252 19.5448 -5.9959 0.0000 0.0017 -0.3068 0.2408
WHATCOM 0.4195 0.0131 37.6298 -5.2511 0.0000 0.0017 -0.1395 0.1109
YAKIMA 0.4165 0.0098 50.3786 -24.8665 0.0000 0.0017 -0.4936 0.3783
Initially, the pairwise T-tests revealed deviations in 25 out of the 30 counties under
examination, which raised both questions and interest. To gain a better understanding of the
significance of these deviations, this study calculated effect sizes (Cohen's d) and the associated
p-values. To address the challenge of multiple comparisons associated with the analysis of 30
counties, a Bonferroni correction was applied to the alpha level. Following this correction, none
of the observed Cohen's d values remained statistically significant at the conventional alpha level
of 0.05 (Only two did prior to correction). In essence, the fact that the deviations were not
significant after the Bonferroni correction supports the null hypothesis – that the differences in
rates across counties are not substantial enough to be attributed to the counties themselves.
Instead, it is more likely that these observed differences are a result of omitted variables or
random variation.

61
4.1.11 Conclusion
This study utilized both ANOVA and t-tests, along with effect size calculations using
Cohen's d. The initial ANOVA test was conducted to explore whether significant differences
exist in recidivism rates among various counties. Initial results were not statistically significant.
The analysis suggests that, based on the available data, the county of release alone does not exert
a statistically significant impact on recidivism rates. This finding indicates that other factors,
such as individual characteristics, socioeconomic conditions, or more localized environmental
factors, may exert more substantial influences on recidivism rates within Washington State.
Understanding the spatial distribution of these rates could still reveal spatial clusters or
patterns of high and low values, highlighting counties with unique characteristics influencing
recidivism rates. Incorporating further spatial visualization techniques, such as spatial
autocorrelation analysis and hotspot mapping, enables a deeper understanding of the geographic
aspects of the phenomenon. By exploring the spatial dimension, we can uncover spatial
dependencies, spatial clusters of similar values, and potential spatial patterns related to
recidivism and its contributing factors across counties. Such analyses can serve as a basis for
targeted interventions and policy decisions to effectively address recidivism and improve public
safety in different regions.
4.2 Results of Exploratory Spatial Data Analysis
This sections describes the results of the ESDA.
4.2.1 Spatial Weights Matrix
The selection of the Spatial Weights Matrix (SWM) for this analysis is underpinned by a
rationale grounded in prior research findings. Existing literature has suggested the presence of a
distance decay function for formerly incarcerated individuals (Piquero, Farrington and

62
Blumstein, 1999; Pyle, 1974). Other studies have leveraged the use of the median area of the
scale of analysis (Hipp, 2010).
This study utilized this same methodology to derive a distance based SWM. First the
median square mileage of the counties in the analysis was found to be 1,719 square miles. To
estimate the approximate distance across a square or rectangular area with an area of 1,719
square miles, a straightforward method involves calculating the square root of the given square
mileage. In this case, taking the square root of 1,719 yields a value of approximately 41.49 miles.
To further refine this measurement into meters, a simple conversion can be applied. By
multiplying the 41.49 miles by the conversion of approximately 1,609.34 meters per mile, the
approximate distance is transformed into meters, resulting in a rounded value of 66,734 meters
across. Thus, the choice of a SWM is a fixed distance band threshold set to 66,734 meters.
To cross validate the findings of the distance based SWM when utilized in the Getis Ord
Hot Spot Analysis a K-nearest neighbors algorithm was also utilized with the neighbor parameter
set to 3. It is also important to note that alternative contiguity based SWMs were initially
considered. However, they were eliminated from further consideration due to a specific
challenge posed by the dataset. The removal of counties with small numbers of total releases
resulted in the creation of isolated island counties, rendering the contiguity based SWMs useless
for this analysis (one county, Asotin, would have no neighbor).
4.2.2 Moran’s I Statistic – Measuring Global Spatial Autocorrelation
The assessment of Moran's Index for recidivism rates for each county revealed insightful
spatial patterns. The calculated Moran's Index value of 0.029836 suggests a slight positive spatial
autocorrelation, indicating that, to a small extent, areas with higher rates of recidivism tend to be
located near similar values. The Expected Index, which was found to be -0.034483, provides a

63
baseline against which the actual Moran's Index can be compared. It represents the anticipated
value of Moran's Index under the assumption of no spatial autocorrelation. The p-value of
0.754924, exceeding the significance level of 0.05, does not provide enough evidence to reject
the null hypothesis of complete spatial randomness. Thus, the analysis suggests that there is no
evidence of spatial autocorrelation in the distribution of recidivism cases among counties.
Further investigation and consideration of the underlying spatial patterns are warranted to
comprehend the complex dynamics influencing recidivism rates across Washington State.
Table 10 Results of the Moran's I index for spatial autocorrelation on the recidivism rates.
Global Moran’s I Summary
Moran’s Index 0.0298
Expected Index -0.0345
Variance 0.0425
z-score 0.3122
p-value 0.7549
4.2.3 Getis-Ord Statistics - Local Cluster Analysis
Overall, the local spatial autocorrelation analysis using the Getis-Ord
∗
statistic
provided deeper insights into the distribution and spatial patterns of recidivism rates at the
county level. These results offer a more nuanced understanding of the spatial dynamics of
recidivism in Washington State at the County level. The initial hot spot analysis conducted using
the Getis-Ord
∗
statistic in ArcGIS Pro with a distance band parameter of 66,734 meters,
observed a statistically significant hotspot at the 99% confidence interval of recidivism rates in
Kittitas County (Figure 11). The calculated Z-score of 3.48 reflects a substantial clustering of
high recidivism rates compared to its neighboring areas. This strong positive Z-score indicates
that the observed high rates are not just a product of random chance but represent a statistically
significant hot spot. Additionally, the low p-value of 0.0005 underscores the significance of this

64
spatial clustering, which is well below the conventional significance threshold of 0.05. While
these findings highlight Kittitas County as a possible hotspot for recidivism rates in the broader
context of the study area, it also invites further examination into the underlying factors
contributing to this localized phenomenon. It is abnormal, but not uncommon to find no global
spatial autocorrelation, yet local clusters at such a high level of significance.
Figure 11 Results of the 66,734-meter fixed distance Getis-ord
∗ Hot Spot Analysis
To cross-validate whether the finding in Kittitas County is an artifact, or indeed a hotspot
of recidivism rates, a further Getis-ord
∗ Hot Spot Analysis was conducted using the K-nearest
neighbors algorithm utilizing 3 as the number of neighbors parameter (Figure 12). The results of
this analysis resulted in a Z-score of 2.53 and a P-value of .0114, which cross-validates the
findings of the distance based SWM. The only notable change detected was the hot spot in
Kittitas County which was initially detected at the 99% confidence threshold, but when using the
K-nearest neighbors SWM changed to the 95% confidence threshold.

65
Figure 12 Outcome of Getis-ord
∗ Hot Spot Analysis utilizing the K-nearest neighbors
parameter set to 3
4.3 Conclusion
Throughout this comprehensive exploratory data analysis (EDA) and exploratory spatial
data analysis (ESDA), we have gained valuable insights into the dynamics of recidivism and its
potential underlying correlates with the county of release as the predictive factor. Our findings
have provided a deeper understanding of the spatial distribution and patterns of various variables
that have been found to be related to recidivism in other studies conducted elsewhere, in
Washington State, laying the groundwork for further research and testing.
In the EDA, we analyzed multiple variables, including the rate of recidivism across
counties, the rate of enrollment in public assistance programs per 100k, and the rate of those
lacking basic prose literacy skills per 100k (log transformed). Notably, no statistically significant
correlation was found between the rate of recidivism per county and either of the county
contextual variables meant to reflect social disorganization. Additionally, the outcomes of the

66
ANOVA test which were then verified with a T-test and Cohen’s d do not provide sufficient
evidence to assert that the variations in recidivism rates among the different counties are
statistically significant. In simpler terms, the county of release does not exhibit a significant
correlation with the propensity to recidivate in Washington State from 2012-2022.
The ESDA delved into spatial autocorrelation, highlighting noteworthy spatial patterns in
the data. While the Moran’s I index for spatial autocorrelation revealed no global spatial
autocorrelation, further measures of local spatial autocorrelation revealed a statistically
significant cluster of recidivism in Kittitas County. Upon closer examination utilizing an
alternative SWM it may be reasonably inferred that the clustering present is due to the low
variance in the recidivism data. Thus, what the Getis-ord
∗ Hot Spot Analysis might be
detecting as hotspots are not traditional hotspots or cold spots of high or low values, but rather
clusters of areas with similar rates. This could be because, with low variance, even small
deviations from the mean could be statistically significant (as was found in the T-tests), leading
to the identification of clusters that are not necessarily extreme but are consistently different
from the overall mean.
In conclusion, this extensive EDA and ESDA has tested whether the county of release is
associated with a propensity to recidivate, whether county social disorganization contextual
factors influence recidivism rates in Washington State, and finally, assessed the spatial patterns
of the phenomenon across Washington State. The findings of this study suggest that the County
scale of analysis is too broad an areal unit of aggregation to provide the detailed nuances of the
neighborhood context for crime or recidivism. Although the study offered valuable insights into
the general effectiveness of WADOC-operated facilities between 2012 and 2022, it fell short in

67
capturing the diverse nature of recidivism patterns across different microenvironmental
locations.
Chapter 5 Discussion
The preceding chapters have explored the nuances of our research goals, the methodology
utilized, and the outcomes achieved. As we transition into this chapter, our focus shifts from the
analytical framework to the theoretical, where the findings of the study are summarized in terms
of real-world implications. It should be noted that the results of this analysis are specific to
Washington State and should not be applied to other jurisdictions.
5.1 Significant Findings
The primary findings of this analysis are threefold. First, the county of release of
WADOC formerly incarcerated individuals from 2012-2022 showed no relationship with
recidivism status. This was verified by the outcomes of the ANOVA test in combination with the
subsequent T-test and Cohens d analyses. Second, the absence of a statistically significant
correlation between recidivism and the social disorganization contextual variables implies that
these factors, as measured in this study, do not appear to be strong drivers of recidivism rates in
Washington State counties. This finding challenges the idea that concentrated disorganization,
which has been found to correlate with crime and recidivism in some other jurisdictions, plays a
similar role in explaining recidivism patterns in Washington State at this scale of analysis.
Lastly, the findings of this study suggest that the County scale of analysis is too broad an areal
unit of aggregation to provide the detailed nuances of the neighborhood context for crime or
recidivism. Although the study offered valuable insights into the general effectiveness of

68
WADOC-operated facilities between 2012 and 2022, it fell short in capturing the diverse nature
of recidivism patterns across different microenvironmental locations.
Translating our findings into real-world terms, the analysis suggests that recidivism is not
substantially impacted by the geographical factor of the county of release. This outcome
underscores the importance of considering the spatial scale of analysis in studies of this nature.
It's noteworthy to acknowledge that existing academic literature has revealed significant
correlations between social disorganization indicators of crime and recidivism and increased
crime and recidivism rates in other jurisdictions. This contrasts with our findings and implies that
the county scale, while useful for certain analyses, might be too broad of an aggregate to capture
the nuanced nature of crime, which frequently occurs within subsections of cities or regions that
could be more accurately expressed through finer-grained spatial units like census tracts, blocks,
zip codes, or neighborhoods.
Revisiting the literature discussed in sections 1 and 2, Hipp (2007) emphasizes the
importance of considering scale when investigating crime and other neighborhood effects such as
recidivism. Drawing on data from a specific subset of the American Housing Survey, his study
delved into how various variables impact crime and disorder at both the block and census tract
levels. Notably, his research uncovers that the influence of these variables varies according to the
level of aggregation. For example, racial/ethnic diversity consistently influenced perceptions of
disorder, regardless of the aggregation level in this study, while economic resources exhibit only
localized effects at the block level, demonstrating disparities in their impact on crime and
disorder. This underscores the significance of researchers contemplating aggregation schemes
when examining crime and recidivism especially from a social disorganization theory
perspective.

69
Within the broader context, numerous studies exploring the influence of neighborhood
structural factors on various outcomes, such as recidivism, often struggle to define the
"neighborhood." Often, researchers poll household responses within a specific geographic unit or
employ statistical estimates to create such measures, without considering whether this unit aligns
with the outcome or predictor variables. Studies frequently use diverse geographic units, such as
blocks, block groups, tracts, zip codes, or amalgamations of multiple tracts, as proxies for the
neighborhood, without a determination of the appropriate level of aggregation. Hipp’s study
addressed exactly that, determining the correct geographic level of aggregation for use in studies
measuring or assessing crime and recidivism.
The author warns that in cases where the unit of analysis is too large, researchers risk
inadvertently including geographic units that contain multiple neighborhoods. For instance,
constructs like social disorder and crime result from the aggregation of individual instances—
each dilapidated building or piece of litter contributes to physical disorder, and each additional
crime event contributes to the overall crime rate. Therefore, the question then arises regarding
which is the “most” suitable geographic unit for aggregating these instances when constructing a
neighborhood measure of crime or disorder. This dilemma is pertinent to all studies, regardless
of their methodology for measuring such. Employing too high a level of aggregation carries the
risk of merging crime and disorder rates from different neighborhoods into a larger unit, often
masking meaningful relationships.
The study that we conducted at the county scale was a prime example of what Hipp was
warning readers about when aggregating at too high of a level, the phenomenon of masking most
of the heterogeneity of recidivism rates within the larger areal unit of analysis impacted the
results of the analysis. However, by solely focusing on the WADOC operated facilities for

70
purposes of this analysis, the results do speak to the efficacy of the Department of Correction’s
ran facilities at rehabilitating the population of incarcerated individuals that they are responsible
for.
This study found that the average rate of recidivism for all WADOC incarcerated
individuals released from prison between 2012-2022 was 41% (Figure 13).
Figure 5 Pie chart of statewide WADOC recidivism rates (2012-2022).
When comparing the findings of this study to the national average, the most often-cited
statistic is that within three years of release, roughly 67.8% of released prisoners were rearrested
(U.S. Bureau of Justice Statistics). Considering this study specifically gauged reconfinement
rather than rearrest as that of the BJS, the findings present a consistent narrative with broader
national trends.
5.2 Methodological Barriers
Throughout the course of this study, methodological challenges emerged from the initial
state of the raw dataset as it was received. The process of overcoming these obstacles was a timeintensive endeavor, involving months dedicated to meticulously preparing the data for analysis.
Yet, the significance of these barriers extended beyond data transformation, significantly shaping
the study's overall framework.

71
The data received from the WADOC was limited to the county scale of analysis due to
privacy concerns in the public records process. This limitation caused by the way the data is
reported imposed certain parameters that significantly constrained this project’s ability to
perform various spatial statistical analyses that could have potentially unveiled the
microenvironmental nuances typically associated with crime and recidivism. These limitations
are commonly known as the modifiable areal unit problem (MAUP).
In essence, the MAUP acknowledges the role of the chosen scale when aggregating data,
as it can significantly impact the results of any analysis. Eliminating much of the MAUP as a
constraint to our spatial analysis would require individual level point data, which, unfortunately,
is not feasible due to the previously alluded to data availability limitations and restrictions
imposed by the public records request process. Thus, in this project, we have effectively
highlighted the impact of the MAUP; as arbitrary data aggregates become larger, many of the
finer and more nuanced spatial relationships become obscured as they are aggregated within the
confines of the larger unit while they were collected at a much finer scale. This limitation
underscores the importance of recognizing and accounting for the MAUP in spatial analyses.
Another barrier imposed by the data was that limited sample sizes (total releases from
WADOC prison facilities) prevented the inclusion of all 39 counties in Washington State, which
reduced the level of detail attainable from the analysis. Additionally, the data received from the
WADOC did not include any demographic information that could have been leveraged for a
more comprehensive analysis of their impact on recidivism in Washington State. This limitation
restricted our ability to employ many multivariate data analysis methods, especially regression
analysis.

72
The absence of demographic information ultimately forced a shift in the project's focus
from exploring contributing contextual factors to recidivism (especially the context of the
geography to which the formerly incarcerated individual is released) to a more specific
investigation into whether the county of release is correlated with recidivism rates. Thus, the data
limitations imposed required an adaptation of research questions.
Certainly, if the dataset had contained additional information, such as the specific prison
locations upon release, gender, age, race, and crime type, would have undeniably enriched the
dataset, offering a more robust and comprehensive analysis of recidivism factors.
Beyond these variables, numerous other demographic and contextual factors could have
significantly enhanced the depth of the analysis. These might encompass aspects such as an
individual's educational background, including educational attainment, which could have
facilitated an examination of the impact of education on recidivism rates rather than utilizing
literacy as a measure of education. Likewise, exploring their employment history, both before
and after release, would have provided valuable insights into the role of employment in reducing
recidivism which is commonly found to be a substantial contributor to recidivism in other
geographies.
Moreover, considering marital and family status, such as whether individuals were
married, had dependents, or maintain family support systems, could have allowed for a more
intricate understanding of the influence of family dynamics and peer-support networks on
recidivism. Information related to substance abuse history, including previous treatment and
relapse rates while incarcerated, could have been instrumental in studying the connection
between substance abuse and recidivism, as well as the efficacy of programs offered while
incarcerated. Additionally, data on mental health history, such as diagnoses, treatment, and

73
access to mental health services, could have been invaluable in assessing the impact of mental
health on recidivism. Integrating these additional demographic and contextual variables into the
analysis would have fostered a more comprehensive understanding of the multifaceted factors
contributing to recidivism rates in Washington State.
If this dataset had been available at a more nuanced scale such as the census block or
census tract level, the study would have been far better positioned to gain a deeper understanding
of the spatial components of recidivism. The intricacies that remained obscured when analyzed at
the county scale could have potentially been unveiled had the data been accessible at a more
fine-grained, microenvironmental level. It's worth noting that some states, such as California,
offer access to higher precision geographical data upon public request. Assessing whether there
have been any adverse consequences from the release of such data in other jurisdictions could
help evaluate the risks associated with making finer-scale data more accessible in Washington
State.
5.3 Future Research
In essence, the outcomes of this study emphasize the significance of context and
granularity in the examination of recidivism. They underscore the necessity for future research to
delve into the effects of more localized factors and spatial scales, ultimately leading to a deeper
comprehension of the intricate dynamics influencing recidivism patterns within Washington
State. Furthermore, the finding of a statistically significant hotspot of recidivism rates in Kittitas
County suggests that further research could focus on this county specifically when assessing
recidivism.
Future studies aiming to delve into more nuanced patterns of recidivism would greatly
benefit from accessing higher-resolution data, which would enable a more detailed analysis at a

74
smaller geographic scale, such as neighborhoods or streets. This finer granularity could reveal
critical insights into localized environmental or social factors influencing recidivism rates, which
are not discernible at broader geographic levels. However, this approach presents a challenge, as
currently the most detailed data available through public records requests is at the county level.
Gaining access to more granular data would require navigating privacy concerns and data
sharing policies, as well as potentially collaborating with government agencies or criminal
justice organizations that have the capacity to provide such detailed information. This access is
crucial for future research to effectively explore and understand the complex spatial dynamics
underpinning recidivism patterns, thereby informing more targeted and effective intervention
strategies.
Additionally, future research could delve into the geographical histories of individuals
who have been incarcerated. This exploration would encompass not only their residential
locations prior to incarceration but also the specific prison locations and the sites of their release.
Analyzing how these geographical factors correlate with recidivism outcomes could be
particularly enlightening. For example, the study might investigate whether the proximity of the
prison to the individual's home community or the characteristics of the release site have any
bearing on the likelihood of reoffending. Such research would offer a deeper understanding of
the complex spatial dynamics that underlie recidivism patterns, potentially leading to more
effective and targeted intervention strategies.
Even when the relationships observed in EDA and ESDA do not show strong or
statistically significant correlations as was the case in this study, further multivariate analyses
including regression modeling could help to strengthen the findings of this study. Regression can
unveil hidden and nonlinear relationships that might not be evident through basic correlation

75
analysis. It can help uncover subtle and complex associations between variables that EDA and
ESDA might not capture. Furthermore, regression allows for hypothesis testing regarding the
significance of individual predictor variables. This can provide evidence for or against the
importance of specific factors, regardless of the strength of correlations. Given these
considerations, it is recommended that future research extends the findings of this study by
employing advanced modeling techniques to evaluate the interplay of variables of social
disorganization more comprehensively. In this same capacity, future research may incorporate
more readily available measures of social disorganization into an analysis, or perhaps create a
composite index of social disorganization to test against the recidivism rates derived from this
analysis.

76
References
Agresti, A. 2002. Categorical Data Analysis. Hoboken, NJ: John Wiley & Sons.
Agresti, A., Finlay, B. 2008. Statistical methods for the Social Sciences. 4th ed. Pearson Prentice
Hall.
Almgren, Gunnar, A. Guest, G. Immerwahr, and M. Spittel. 1998. “Joblessness, Family
Disruption, and Violent Death in Chicago, 1970-90.” Social Forces 76, no. 4: 1465.
Anselin, L. 1995. "Local Indicators of Spatial Association-LISA." Geographical Analysis 27, no.
2: 93-115.
Aos, S., Miller, Mama, and Drake, E. 2006. “Evidence Based Adult Corrections Programs: What
Works and What Does Not.” Olympia: Washington State Institute for Public Policy.
Aos, S. 2003. “Washington’s Offender Accountability Act: Update and Progress Report on the
Act’s Evaluation”. Washington State Institute for Public Policy Publications. 1-12.
https://www.wsipp.wa.gov/ReportFile/822/Wsipp_Washingtons-OffenderAccountability-Act-Update-and-Progress-Report-on-the-Acts-Evaluation_FullReport.pdf
Average Daily Population of Incarcerated Individuals Fiscal Year 2022, 2023. Washington State
Department of Corrections Publications https://doc.wa.gov/docs/publications/reports/400-
RE002-2206.pdf
Beckett, K.A., A.M. Harris, and H. Evans. 2008. The Assessment and Consequences of Legal
Financial Obligations in Washington State: Research Report. Olympia, WA: Washington
State Minority and Justice Commission.
Besag, J. “Spatial Interaction and the Statistical Analysis of Lattice Systems.” Journal of the
Royal Statistical Society: Series B (Methodological) 36, no. 2 (1974): 192–225.
Bursik, R.J. 1988. “Social Disorganization and Theories of Crime and Delinquency: Problems
and Prospects.” Criminology, 26(4), 519-551.
Cahill, M., and Mulligan, G. 2007. "Using Geographically Weighted Regression to Explore
Local Crime Patterns." Social Science Computer Review 25, no. 2: 174-193.
Camasso, M.J., Jagannathan, and Radha. 1995. “Prediction accuracy of the Washington and
Illinois risk assessment instruments: An application of receiver operating characteristic
curve analysis.” Social Work Research, 19, no. 3: 174–183,
https://doi.org/10.1093/swr/19.3.174
Cohen, J. 1988. Statistical power analysis for the behavioral sciences. Hillsdale, NJ: Lawrence
Earlbaum Associates.

77
David, F.N., and Tukey, J.W. 1977. "Exploratory Data Analysis." Biometrics 33, no. 4 (1977):
768-769. https://doi.org/10.2307/2529486.
Drake, E. 2014. "Predicting Criminal Recidivism: A Systematic Review of Offender Risk
Assessments in Washington State." Olympia, WA: Washington State Institute for Public
Policy.
Drawve, G., and McNeeley, S. 2021. "Recidivism and Community Context: Integrating the
Environmental Backcloth." Journal of Criminal Justice 73: 101786.
https://doi.org/10.1016/j.jcrimjus.2021.101786.
Fotheringham, A.S. 1997. "Local Statistics for Spatial Analysis." In Quantitative Geography:
Perspectives on Spatial Data Analysis, edited by Fotheringham, A.S., Jordan, N., and
Rogerson, P., 175-197. London, UK: Sage Publications Ltd.
Gagliardi, GJ., D. Lovell, P.D. Peterson, and R. Jemelka. 2004. “Forecasting Recidivism in
Mentally Ill Offenders Released from Prison.” Law and Human Behavior 28, no. 2: 133–
155.
Geary, R.C. 1954. “The Contiguity Ratio and Statistical Mapping.” The Incorporated Statistician
5, no. 3: 115.
Gerben J.N. Bruinsma, Lieven J.R. Pauwels, Weerman, F.M., Bernasco, W. 2013. “Social
Disorganization, Social Capital, Collective Efficacy and the Spatial Distribution of Crime
and Offenders: An Empirical Test of Six Neighbourhood Models for a Dutch City.” The
British Journal of Criminology 53:5, 942–963. https://doi.org/10.1093/bjc/azt030
Getis, A., and Ord, J.K. 1992. "The Analysis of Spatial Association by Use of Distance
Statistics." Geographical Analysis 24, no. 3 (1992): 189-206.
Goodchild, M.F. 2007. “Citizens as sensors: The world of volunteered geography.”
Gottfredson, S.D., and Taylor, R.B. 1988. "Community Contexts and Criminal Offenders." In
Communities and Crime Reduction, edited by Hope, T., and Shaw, M., 62-82. London:
Her Majesty’s Stationary Office.
Heitgerd, J.L., and R.J. Bursik. 1987. “Extracommunity Dynamics and the Ecology of
Delinquency.” American Journal of Sociology 92, no. 4: 775–787.
Hipp, J.R. 2007. "Block, Tract, and Levels of Aggregation: Neighborhood Structure and Crime
and Disorder as a Case in Point." American Sociological Review 72, no. 5: 659-680.
https://doi.org/10.1177/000312240707200501.
Kassambara, A. 2019. Practical statistics in R II - comparing groups: Numerical variables.
Datanovia.

78
Kim, R.H., and Clark, D. 2013. "The Effect of Prison-Based College Education Programs on
Recidivism: Propensity Score Matching Approach." Journal of Criminal Justice 41, no. 3:
196-204. https://doi.org/10.1016/j.jcrimjus.2013.03.001.
Kirk, D.S., and J.H. Laub. 2010. “Neighborhood Change and Crime in the Modern Metropolis.”
Crime and Justice 39, no. 1: 441–502.
Kubrin, C.E., and R. Weitzer. 2003. “Retaliatory Homicide: Concentrated Disadvantage and
Neighborhood Culture.” Social Problems. 50: 157–180.
Kubrin, C.E. and Stewart, E.A. 2006. “Predicting Who Reoffends: The Neglected Role of
Neighborhood Context in Recidivism Studies.” Criminology, 44: 165-197.
https://doi.org/10.1111/j.1745-9125.2006.00046.x
Kubrin, C.E., Squires, G.D., and Stewart, E.A. 2007. "Neighborhoods, Race, and Recidivism:
The Community-Offender Nexus and Its Implications for African-Americans." Race
Relations Abstracts 32: 7-37.
LeBeau, J.L. & M. Leitner. 2011. “Introduction: Progress in Research on the Geography of
Crime.” The Professional Geographer, 63:2, 161-173, DOI:
10.1080/00330124.2010.547147
Leitner, M., Barnett, M., Kent, J., and Barnett, T. 2011. “The Impact of Hurricane Katrina on
Reported Crimes in Louisiana: A Spatial and Temporal Analysis.” The Professional
geographer 63, no. 2: 244–261.
Levine, N. and Block, R. 2011. “Bayesian Journey-to-Crime Estimation: An Improvement in
Geographic Profiling Methodology.” The Professional geographer 63, no. 2: 213–229.
Leymon, M.G., Campbell, C.M., Henning, K., and Renauer, B.C. 2022. "The Impacts of Length
of Prison Stay on Recidivism of Non-Violent Offenders in Oregon." Journal of Criminal
Justice 82: 102002. https://doi.org/10.1016/j.jcrimjus.2022.102002.
Logan, J.R., and Brian J. Stults. 1999. “Racial Differences in Exposure to Crime: The City and
Suburbs of Cleveland in 1990*.” Criminology 37, no. 2: 251–276.
Lovell, D., Johnson, L.C., & Cain, K.C. 2007. “Recidivism of Supermax Prisoners in
Washington State.” Crime & Delinquency. 53 no. 4, 633–656.
https://doi.org/10.1177/0011128706296466
Maher, E.J., Gerlinger, J., Wood, A.D. et al. 2022. “Won’t You be My Neighbor? Neighborhood
Characteristics Associated with Mass Shootings in the USA.” Race Soc Probl 14, 223–
237. https://doi.org/10.1007/s12552-021-09350-3
Manchak, S. M., J. L.Skeem, and K.S. Douglas. 2008. “Utility of the Revised Level of Service
Inventory (LSI-R) in Predicting Recidivism after Long-Term Incarceration.” Law and
Human Behavior 32, no. 6: 477–488.

79
Malleson, N. and Andresen, M.A. 2015. “Spatio-Temporal Crime Hotspots and the Ambient
Population.” Crime Science 4, no. 1.
Maltz, M. 2001. Recidivism. Orlando, FL: Academic Press Incorporated.
Mennis, J., and Diansheng G. 2009. “Geographic Data Mining and Knowledge Discovery - An
Introduction.” Computers, Environment and Urban Systems: 403–408.
Mennis, J, Harris, P.W.,Obradovic,Z., Izenman,A.J., Grunwald, H.E. & Lockwood, B. 2011.
The Effect of Neighborhood Characteristics and Spatial Spillover on Urban Juvenile
Delinquency and Recidivism, The Professional Geographer, 63:2,174-192, DOI:
10.1080/00330124.2010.547149
Miller, M., Drake, E., & Nafziger, M. 2013. What Works to Reduce Recidivism by Domestic
Violence Offenders? (Document No. 13-01-1201). Olympia: Washington State Institute
for Public Policy.
Moran, P.A.P. 1948. “The Interpretation of Statistical Maps.” Journal of the Royal Statistical
Society. Series B (Methodological) 10, no. 2: 243-251.
Nesset, M., Berg, M., Loret, L.C., Bjørngaard, J., Whittington, R. and Palmstierna, T. 2020.
“Cognitive Behavioural Group Therapy versus Mindfulness-Based Stress Reduction
Group Therapy for Intimate Partner Violence: A Randomized Controlled Trial.” BMC
Psychiatry 20, no. 1:5.
Nicodemo, C., & Satorra, A. 2022. “Exploratory data analysis on large data sets: The example of
salary variation in Spanish Social Security Data.” BRQ Business Research Quarterly,
25(3), 283–294. https://doi.org/10.1177/2340944420957335
Openshaw, S. 1984. “Ecological fallacies and the analysis of areal census data.” Environment
and Planning A 16, 17–31.
Openshaw, S. and Taylor, P. 1979. A million or so correlation coefficients: three experiments on
the modifiable areal unit problem. Statistical Applications in the Spatial Sciences. Edited
by N. Wrigley. Pion, London. 127–144.
Ostertagova, E. & Ostertag, Oskar & Kováč, Jozef. 2014. “Methodology and Application of the
Kruskal-Wallis Test.” Applied Mechanics and Materials. 611. 115-120.
O'Sullivan, D., and Unwin, D. 2014. Geographic Information Analysis. 2nd ed. Hoboken, NJ:
Wiley.
Pearson, K. 1900. "On the Criterion That a Given System of Deviations from the Probable in the
Case of a Correlated System of Variables is Such That it can be Reasonably Supposed to
Have Arisen From Random Sampling." Philosophical Magazine Series 5, vol. 50, no.
302, pp. 157-175.

80
Petersilia, J. 2009. When Prisoners Come Home: Parole and Prisoner Reentry. Oxford, UK:
Oxford University Press.
Phipps, P.A., K. Korinek, S. Aos, and R. Lieb. Research Findings on Adult Corrections
Programs: A Review. Olympia, WA: Washington State Institute for Public Policy, 1999.
Piquero, A.R., Farrington, D.P., and Blumstein, A. 2003. "The Criminal Career Paradigm."
Crime and Justice: A Review of Research 30, edited by Tonry, M., 359-506. Chicago, IL:
The University of Chicago Press.
Ratcliffe, J. 2004. “The hotspot matrix: A framework For the spatio-temporal targeting of crime
reduction.” Police Practice and Research 5:5–23.
Recidivism of Adult Felons 2007, 2008, Washington State Caseload Forecast Council.
https://cfc.wa.gov/sites/default/files/Publications/Adult_Recidivism_FY2007.pdf
Robinson, W.S. 1950. “Ecological Correlations and the Behavior of Individuals.” International
Journal of Epidemiology 38, no. 2: 337-341.
Rosenthal, R. & Rosnow, R.L. 1991. Essentials of behavioral research: Methods and data
analysis. New York: McGraw Hill.
Sampson, R.J., S. W. Raudenbush, and Felton Earls. 1997. “Neighborhoods and Violent Crime:
A Multilevel Study of Collective Efficacy.” Science 277, no. 5328: 918–924.
Shaw, C.R., and H. D. McKay. Juvenile Delinquency and Urban Areas. Chicago, IL: Univ of
Chicago Press, 1942.
Stahler, G.J., Mennis, J., Belenko, S., Welsh, W.N., Hiller, M.L., and Zajac, G. 2013. "Predicting
Recidivism for Released State Prison Offenders." Criminal Justice and Behavior. 40, no.
6 pg. 690–711. https://doi.org/10.1177/0093854812469609.
Tobler, W. 1997. “The First Law of Geography.” Economic Geography 47(1). Pgs. 1-7.
Travers, Áine, T. McDonagh, T. Cunningham, C. Armour, and M. Hansen. 2021. “The
Effectiveness of Interventions to Prevent Recidivism in Perpetrators of Intimate Partner
Violence: A Systematic Review and Meta-Analysis.” Clinical Psychology Review 84: 5.
Unwin, D.J. 1996. "GIS, Spatial Analysis, and Spatial Statistics." Progress in Human Geography
20(4): 540-551. doi:10.1177/030913259602000408.
Wang, X., Hay, C., Todak, N.E., and Bales, W. 2013. "Criminal Propensity, Social Context, and
Recidivism." Criminal Justice and Behavior 41(3): 300–317.
https://doi.org/10.1177/0093854813507055.
Washington State Department of Corrections. 2019. Improve Lives | Decrease First-Year Rate of
Return to Institutions. https://doc.wa.gov/about/agency/strategic-plan-g1o1.htm

81
Washington Sentencing Guidelines Commission. 2007. Recidivism of Adult Felons. By the
Caseload Forecast Council.
cfc.wa.gov/sites/default/files/Publications/Adult_Recidivism_FY2007.pdf
Washington State Institute for Public Policy. 1999. The Community Juvenile Accountability Act:
Research-Proven Interventions for the Juvenile Courts, by Robert Barnoski. Document
No. 99051201. Olympia, WA. https://www.wsipp.wa.gov/ReportFile/1311/Wsipp_TheCommunity-Juvenile-Accountability-Act-Research-Proven-Interventions-for-theJuvenile-Courts_Full-Report.pdf
Washington State Legislature. 1999. Offender Accountability Act (OAA). Engrossed 2nd
Substitute Senate Bill 5421, Chapter 196, Laws of 1999 56th Legislature Regular Session.
July, 25.

Abstract (if available)

Abstract Recidivism rates have important implications for public safety, the well-being of both those reentering our communities, as well as the communities that the formerly incarcerated individuals are being released back into. This study leverages a comprehensive prison admission/release dataset from the Washington State Department of Corrections in a spatial analysis looking at both individual level and county contextual variables with the intent to identify whether the county of release of a formerly incarcerated individual is correlated to the recidivism rates of the county. The analysis further considered social disorganization theory aspects as potential contributors to recidivism patterns. By incorporating these variables into a comprehensive exploratory data analysis, subsequent statistical analyses, and finally exploratory spatial data analysis methodologies the study aimed to understand how the socioeconomic context of the county of release may lead to a propensity to recidivate in Washington State. The findings of this project show that there is no evidence of a correlation between the county of release and propensity to recidivate in the State from 2012-2022, the conclusion drawn here is a finding for the null hypothesis of this study, that while many counties display statistically significant deviations from the sample mean, these deviations are not attributable to the county of release itself. Furthermore, contrary to academic literature that has found significant correlations between social disorganization and crime, no statistically significant correlations with the socioeconomic contextual variables meant to reflect social disorganization were found in this analysis. This suggests that while rates of public assistance enrollment and literacy rates or educational attainment have been found to be correlated to crime and recidivism elsewhere, they are not in Washington State at the county scale of analysis. The ultimate conclusion underscores a critical concern regarding the selected scale of analysis, emphasizing that the County, as an areal unit of aggregation, proves to be too broad to comprehensively capture the nuances of recidivism. This assertion gains robust support from the evidence revealed by the analysis' findings.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A field study of juvenile recidivism in Los Angeles

PDF

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

PDF

A qualitative examination of postsecondary education as a remedy to recidivism among formerly incarcerated African-American men

PDF

A spatial analysis of beef production and its environmental and health impacts in Texas

PDF

Jobs, not jail: Homeboy Industries, a radical approach to redirecting adolescents away from the criminal justice system

PDF

A spatiotemporal analysis of environmental risk factors of Lyme disease in the Northeastern United States

PDF

That sinking feeling: predicting land subsidence in California’s San Joaquin Valley with a spatial regression model

PDF

Education based incarceration: educate to change the organizational culture of corrections in the Los Angeles County Sheriff’s Department

PDF

Assessing correctional educator capacity in adult correctional education programs: a gap analysis

PDF

Network accessibility and population change: historical analysis of transportation in Tennessee, 1830-2010

PDF

Water quality in the Los Angeles River: a remote sensing based analysis

PDF

Bogoliubov quasiparticles in Andreev bound states of aluminum nanobridge Josephson junctions

PDF

Temporal analysis of soil degradation in San Joaquin County, California: a close examination of soil erosion using RUSLE

PDF

The interplay between tobacco exposure and polygenic risk score for growth on birthweight and childhood acute lymphoblastic leukemia

PDF

Projecting vulnerability: a combined analysis of sea-level rise, hurricane inundation, and social vulnerability in Houston-Galveston, Texas

PDF

Association between the depth of implant cover screw and the marginal bone loss with the use of a removable provisional restoration: a retrospective two-dimensional radiographic evaluation

PDF

Geospatial analysis of the Round Fire: a replication of burn severity analyses in the Sierra Nevada

PDF

Providing a new low-cost primary care facility for under-served communities: a site suitability analysis for Service Planning Area 6 in Los Angeles County, California

PDF

Evaluating the utility of a geographic information systems-based mobility model in search and rescue operations

PDF

Understanding acute lymphoblastic leukemia in different ethnic groups in the United States

Asset Metadata

Creator Woodmark, Robert Patrick (author)

Core Title The geographic connotations of reincarceration: a spatial analysis of recidivism in Washington State

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Degree Conferral Date 2023-12

Publication Date 12/15/2023

Defense Date 11/13/2023

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

Tag 2012-2022,correctional policy reforms,correctional program effectiveness,criminal justice system in Washington State,criminal relapse,data analysis,formerly incarcerated,incarceration and reoffending,inmate rehabilitation,longitudinal study on incarcerated and formerly incarcerated individuals,OAI-PMH Harvest,post-release support,prison reintegration,recidivism rates,reentry,WADOC facilities,Washington State Department of Corrections

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Sedano, Elisabeth (committee chair), Vos, Robert (committee member), Wu, An-Min (committee member)

Creator Email robbywoodmark@gmail.com,woodmark@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113797221

Unique identifier UC113797221

Identifier etd-WoodmarkRo-12564.pdf (filename)

Legacy Identifier etd-WoodmarkRo-12564

Document Type Thesis

Format theses (aat)

Rights Woodmark, Robert Patrick

Internet Media Type application/pdf

Type texts

Source 20231218-usctheses-batch-1116 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

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

Repository Email cisadmin@lib.usc.edu