A Spatially Explicit Model of Post-disaster Housing Recovery

Ali Nejat*
Dept. of Civil, Environmental & Construction Engineering, Texas Tech University, Lubbock,
TX, USA

Roxana J. Javid
Dept. of Engineering Technology, Savannah State University, Savannah, GA, USA

Souparno Ghosh
Dept. of Mathematics and Statistics, Texas Tech University, Lubbock, TX, USA

&

Saeed Moradi
Dept. of Civil, Environmental & Construction Engineering, Texas Tech University, Lubbock,
TX, USA

*To whom correspondence should be addressed. Dept. of Civil, Environmental, and Construction Engineering,
Texas Tech University, Box 41023, Lubbock, TX 79409-1023, E-mail: [email protected], Phone: 806.834.4065

Abstract: Although post-disaster housing recovery is an important player in community

recovery, its modeling is still in its infancy. This research aims to provide a spatial
regression model for predicting households’ recovery decisions based on publicly available
data. For this purpose, a hierarchical Bayesian geostatistical model with random spatial
effects was developed. To calibrate the model, households’ data that were collected from
Staten Island, New York in the aftermath of Hurricane Sandy were used. The model revealed
that on the scale of census tract, residents with higher income or larger household size were
significantly less likely to reconstruct. In contrast, odds of reconstruction rose with increase
of long-term residents. The model outputs were also employed to develop a reconstruction
propensity score for each census tract. The score predicts probability of
reconstruction/repair in each tract versus others. The model was validated through
comparison of the propensity scores with the distribution of Community Development Block
Grant Disaster Recovery (CDBG‐ DR) assistance and its resultant reconstruction. The
validation indicated capability of the model to predict the potential hotspots of
reconstruction. Accordingly, the propensity score can serve as a decision-support tool to
tailor recovery policies.

1 INTRODUCTION

Growth of population in hazard-prone areas, along with the increase of extreme events
(Pravettoni, 2009; Guha-Sapir et al., 2016) have increased disaster losses (Schwartz, 2006;

This is the author manuscript accepted for publication and has undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process, which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1111/mice.12487.

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 Nejat, Javid, Ghosh & Moradi

Cutter et al., 2008). This necessitates a better understanding of recovery dynamics. Within
community recovery, housing recovery is of vital importance. Housing is a primary element
of peoples’ lives that influences their well-being via providing a safe and secure place and a
positive sense of self-worth and power (Bratt, 2002). Further, this sector comprises the
majority of building stock in the United States (USCB, 2018) and shapes the built
environment through households’ preferences. These factors make the residential sector a
major part of the nation’s financial and social infrastructure (Comerio, 1998). Consequently,
post-disaster housing recovery affects the community recovery through ripple effects
(Peacock et al., 2007a, 2007b). Housing recovery is driven by a multitude of parameters and
requires a collective effort to be fully understood (Moradi et al., 2019). Several recovery
models have been proposed (Miles and Chang, 2003, 2011; Nejat and Damnjanovic, 2012;
Miles, 2017), however there are still gaps in the understanding of how individual decisions
lead to community recovery. This study seeks to contribute to the domain by developing a
spatial model capable of predicting households’ recovery decisions using publicly available
data. Such a model could help with enhancing recovery plans.
In the next section, recovery drivers and disaster models are outlined first. Then, the
research methodology, analysis, and results are discussed. Finally, conclusions, limitations,
and future lines of study are presented.

2 DRIVERS OF HOUSING RECOVERY

Various factors can affect households’ decisions and recovery progress. Herein, these factors
are classified into three categories as internal, interactive, and external, a selection of which
are summarized in Table 1 and described afterward.

Table 1
Drivers of recovery
Driver Instance
1. Internal
Demographic/socioeconomic
Race Bullard and Wright (2009); Peacock et al. (2014); Nejat et al. (2019)
Income Bolin and Bolton (1983); Peacock et al. (2014); Hamideh et al. (2018)
Education Burton (2015); Nejat et al. (2018); Nejat et al. (2019)
Marital status Nejat and Ghosh (2016); Nejat (2018)
Having children Nejat et al. (2018)
Age Sanders et al. (2004); Henderson et al. (2010); Nejat et al. (2018)
Gender Nejat et al. (2018)
Employment Bolin and Bolton (1983)
Household size Sadri et al. (2018)
Homeownership Peacock et al. (2014); Hamideh et al. (2018)
Disaster experience Kick et al. (2011); Binder et al. (2015); Sadri et al. (2018)
Disaster damage Bolin (1976); Peacock et al. (2014); Hamideh et al. (2018); Sadri et al.
(2018)
2. Interactive
Social capital Yandong (2007); Aldrich (2010, 2012); Burton (2015); Sadri et al. (2018)
Place attachment Chamlee-Wright and Storr (2009); Binder et al. (2015); McNeil et al.
(2015)
Neighbors’ recovery Dacy and Kunreuther (1969); Rust and Killinger (2006)
3. External
REDi™ Rating System The Arup Group (Almufti and Willford, 2013)
Financial assistance Kamel and Loukaitou-Sideris (2004); Nejat and Ghosh (2016)
Restoration of community assets Miles and Chang (2011); Comerio (2014); Burton (2015)
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 3

2.1 Internal drivers

Internal drivers are individual- and household-level factors that influence recovery, such as
demographic and socioeconomic attributes, disaster experience, and property damage.
Researchers have reported low income and unemployment as obstacles to recovery (Bolin
and Bolton, 1983). Peacock et al. (2014) found that income and tenure influenced recovery of
Miami-Dade, Florida after the 1992 Hurricane Andrew and Galveston, Texas following the
2008 Hurricane Ike, while race affected the former only. Further, education level (Nejat et
al., 2019), marital status (Nejat and Ghosh, 2016; Nejat, 2018), household size (Sadri et
al., 2018), age, gender, and having children (Nejat et al., 2018) can influence
households’ recovery decisions.
Moreover, recovery is influenced by households’ experience of previous disasters.
Residents who had underwent repetitive floods were more interested to relocate (Kick
et al., 2011). Similarly, Binder et al. (2015) associated relocation of Oakwood Beach,
New York households in the aftermath of the 2012 Hurricane Sandy with their
experience of recurring disasters. Conversely, Sadri et al. (2018) reported a positive
correlation between recovery rate and disaster experience.
Finally, level of damage affects households’ decisions and recovery duration. The
effect of damage is significant and long lasting. (Bolin, 1976; Peacock et al., 2014).
Researchers reported that this effect continued several years after Hurricanes Andrew
and Ike (Peacock et al., 2014; Hamideh et al., 2018). Similarly, greater damage led to
longer recovery of tornado-impacted regions in Indiana (Sadri et al., 2018).

2.2 Interactive drivers

Interactive drivers of recovery stem from individuals’ interfaces with community. This
category includes drivers such as social capital, place attachment, and neighbors’
recovery.
Social capital is referred to as the engine of recovery (Aldrich, 2010, 2012) and is defined
as the collaboration among community members to share resources and to achieve common
goals (Jamali and Nejat, 2016). As a contributor to social capital, trust enhanced recovery of
western China (Yandong, 2007) and Indiana towns (Sadri et al., 2018). Other players such as
social advocacy organizations, religious organizations, and recreation centers were also found
significant in recovery of Mississippi Gulf Coast following the 2005 Hurricane Katrina
(Burton, 2015).
Further, place attachment can affect recovery decisions. Place attachment is the
residents’ loyalty to their place of living since it accommodates their needs or brings
them a sense of identity (Jamali and Nejat, 2016). Chamlee-Wright and Storr (2009)
reported influence of sense of place on New Orleans post-Katrina recovery. Similarly, place
attachment was a major recovery driver of Rockaway Park, New York after Hurricane Sandy
(Binder et al., 2015).
Households’ recovery decisions are also influenced by their neighbors’ decisions (Rust
and Killinger, 2006) as families tend to rebuild or relocate together (Dacy and Kunreuther,
1969).

2.3 External drivers

External drivers are plans, supports and services that can help or enhance community
recovery, such as design initiatives, financial aids, and restoration of community assets and
infrastructure.

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 Nejat, Javid, Ghosh & Moradi

Among design initiatives is the Resilience-based Earthquake Design initiative (REDi™)

which has provided a platform to implement resilience-based earthquake design using a
rating system (Almufti and Willford, 2013). The REDi™ roadmap to resilience consists of
loss assessment and resilient design and planning which includes three categories of
organizational, building, and ambient resilience. While building resilience is focused on
minimizing the expected damages to buildings, organizational and ambient resilience target
utility/business continuity and site accessibility (Almufti and Willford, 2013).
Further, financial assistance affects recovery. Kamel and Loukaitou-Sideris (2004)
investigated distribution of federal assistance in the 1994 Northridge earthquake, California,
and observed a decline in residential population wherever less assistance had been received.
Similarly, Nejat and Ghosh (2016) reported significance of insurance reimbursements and
financial assistance from federal, state, and local organizations.
Moreover, housing recovery is affected by restoration of community assets and
infrastructure. Burton (2015) reported a positive correlation between post-Katrina
recovery of Mississippi Gulf Coast and availability of commercial establishments.
Business recovery is important since it provides reconstruction services and materials
as well as jobs, i.e. income resources (Comerio, 2014). In addition, restoration of
utilities, transportation systems, schools, and health-care facilities provide residents
with the crucial services required for recovery (Miles and Chang, 2011; Comerio,
2014).

3 DISASTER MODELING

Various models have been developed to simulate different aspects of a disaster such
as prediction, resilience, and recovery. Panakkat and Adeli (2008) reviewed earthquake
prediction models and classified them into models based on analysis of earthquake
precursor and historic earthquake data (Panakkat and Adeli, 2008). Panakkat and Adeli
(2007) proposed a method to evaluate short-term potential of a region to seismic
hazard. The method analyzes eight seismicity indicators using neural network approach
and predicts the largest earthquake magnitude in the following month (Panakkat and
Adeli, 2007). Panakkat and Adeli (2009) applied neural network modeling to historical
data, inputted as seismicity indicators, to predict time and location of future
earthquakes (Panakkat and Adeli, 2009). Rafiei and Adeli (2017) developed a model
for prediction of earthquake occurrence which applies machine learning classification
and neural dynamics optimization to predict location of earthquake with maximum
occurrence probability.
Further, several models have been proposed to assess resilience to disasters. Ouyang
and Fang (2017) developed a model to study infrastructure resilience to intentional
attacks that uses a decomposition algorithm to propose optimum pre-attack defense
strategies, worst attack scenario, and optimum post-attack recovery scenario. Wang et
al. (2017) proposed a method to identify the lifeline path optimized with respect to
construction cost and potential seismic failures. Bozza et al. (2017) proposed a
framework to quantify urban resilience using complex network theory for modeling
urban areas as social-physical networks and assessing resilience via time-independent
synthetic indices. Galbusera et al. (2018) proposed a Boolean network approach to
assess resilience of interdependent infrastructure embracing mitigation elements,
resource limitations, and allocation plans. Mahmoud and Chulahwat (2018) developed
a spatiotemporal model to quantify community resilience using a finite element
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 5

formulation to incorporate the components’ interactions. Batouli and Mostafavi (2018)

proposed a multiagent simulation method to assess resilience of road infrastructure to
sea-level rise by integrating sea-level rise, infrastructure conditions, and infrastructure
management.
Moreover, effective recovery planning is a main principle of recovery continuum
(FEMA, 2011) for which models capable of simulating the impacted community are
essential (Nejat and Damnjanovic, 2012). A few models have been developed for this
purpose. ResilUS (Miles and Chang, 2006, 2011), for example, simulates recovery
dynamics of households and businesses in the context of neighborhood and community
conditions. Nejat and Damnjanovic (2012) developed an agent-based model to capture
the collective nature of postdisaster recovery through simulation of homeowners’
interactions with their neighbors. Huling and Miles (2015) developed a discrete-event
simulation framework which models owner-occupied single-family wood-framed
houses as entities, recovery needs as events, and financial/human supplies as resources.
Miles (2017) expanded this model to add households’ eligibility for financial
assistance, renter and landlord households, sequential or parallel events, and
stochasticity in events’ durations (Miles, 2017; Miles et al., 2018). González et al.
(2017) proposed a data-driven process to tailor recovery strategies for interdependent
infrastructures. The model employs predefined recovery strategies to generate a linear
operator representing the system’s recovery dynamics and to customize the recovery
strategies for an assumed disaster scenario (González et al., 2017).

4 METHODOLOGY

This research proposes a spatial model to predict household’ recovery decisions based on
publicly available demographic and socioeconomic data. To develop the model, individual-
level data were first collected via interviewing households of Staten Island, New York affected
by Hurricane Sandy. Then, a spatial logistic regression model was developed to reveal the
relationship between the households’ aggregated recovery decisions and their attributes at
census tract level. The rationale behind selecting this unit of analysis is the fact that recovery
plans cannot be devised at a micro level. Finally, the model was applied to predict recovery
decisions. In the next subsections, data collection and analysis methodologies are explained.

4.1 Data collection

A survey was conducted in Staten Island, New York in December 2012, about 35 days after
Hurricane Sandy. The data collection was supported by a RAPID grant (CMMI 1313946)
from the Civil, Mechanical and Manufacturing Innovation (CMMI) Division of the National
Science Foundation (NSF). The Office of Human Research Protection Program at Texas Tech
University approved the survey. The survey team included two faculty members and three
trained students from Texas Tech University who conducted sample recruitment and face-to-
face interviews. Recruitment included two sampling methods: door-to-door survey of in-
place residents, and in-shelter survey of individuals whose residence were not habitable. The
team randomly selected and approached addresses within predefined areas. The addresses
then served as benchmarks to approach the neighbors. Unapproachable addresses and those
without answer were marked for next-day follow-up and were deleted after two unsuccessful
follow-ups. Snow-ball sampling technique (Goodman, 1961; Berg, 2006) supplemented the

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 Nejat, Javid, Ghosh & Moradi

survey to inquiry households residing in shelters. Participants were required to be at least 18

years old and were compensated for their contribution.
This sampling method was applied due to the limited number of households in the study
area. A randomization procedure would have reduced the sample size further and thus,
defeated randomization in favor of generalizability of the results and inferences. Still, the
authors acknowledge the dependence induced by the survey technique. Therefore, the sample
is envisioned as a single realization of the underlying stochastic data generation process.
Particularly, a spatial data generation process is imposed due to the samples’ spatial
proximity.
The data collection lasted for 5 days (150 hours of face-to-face interviews) and resulted in
126 completed surveys of single-family households. The questionnaire included five sections.
First, the participants were asked about their demographic and socioeconomic attributes such
as age, race, education, employment, income, and religion. Second, pre-disaster status of
residence was enquired including type, tenure, residence length, insurance, and value. Third,
the households were asked about disaster damage and their decision whether to rebuild/repair,
wait/stay in temporary housing, or relocate. Fourth, the decision was more investigated by
questioning its drivers such as financial assistance, infrastructure restoration, and neighbors’
reconstruction. Participants who had decided to relocate were also asked about their new
location. Fifth, information on households’ financial losses and assistance received from
various entities were collected.
Power analysis with STATA© indicated that sample size provided adequate power (> 0.8)
for multiple regression assuming a medium effect size (Cohen, 1992). Table 2 shows how the
sample attributes compare to those reported by the 2012 census data for the sampled tracts. As
mentioned, there were 126 completed surveys (sample size). Meanwhile, according to the
2012 census data, there were 111,235 individuals and 40,386 households in the sampled tracts.
While race ratios are close for the dominant group (Caucasian), the sample is older, has lower
income, and features more males and less African Americans and Asians than the census
data. Since this research was aimed to develop a model for predicting recovery decisions on a
scale above individual-level, the survey data on recovery decisions were only utilized, and for
the others, publicly available data were employed. The utilized data are more described in the
next section.

Table 2
Characteristics of the sample versus census data
Variables Sample 2012 Census
Age (median) 45.52 35.9
Gender (male) 0.64 0.48
Gender (female) 0.36 0.52
Race (Caucasian) 0.81 0.78
Race (African American) 0.07 0.12
Race (Asian) 0.02 0.08
Race (other) 0.1 0.02
Income (median $) 52,278 73,496

4.2 Spatial logistic regression

The idea of regression is to obtain an empirical relationship between predictors and observed
responses. The estimates of the regression coefficients provide insight into the relative
strength of the predictors in explaining the variations in the responses. However, when
samples are dependent, it is required to explicitly account for that dependence in order to
remove the confounding effect of the predictive power of the samples themselves from that
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 7

of the predictors. In the current research, assumption of independence among sampling units
(tracts) is untenable due to the nature of the sampling. Figure 1 shows spatial clusters of the
sampled tracts that potentially induce local spatial dependence among the responses.

Figure 1. Spatial clusters of sampled tracts

In this paper, this dependence is handled by envisioning the samples to be a realization of a
random spatial field. This specification essentially suggests that observed responses in close
spatial proximity are more likely to be similar as compared to the responses that are far apart.
Once this within-response dependence is accounted for, a regression model is specified to
estimate the marginal effects of the predictors on the responses. This research applies spatial
logistic regression to impose spatial dependence among responses, and to model recovery
decisions of the households.
Since recovery policies cannot be devised at individual-level (Nejat et al., 2019), census
tract was selected as the unit of analysis. The spatial domain was partitioned into disjoint
regions corresponding to the tracts. Data on average household income, average household
size, percent of long-term residents, dominant race, and percent of native-born residents in
each tract were obtained from the 2012 census data. Ratio of long-term residents is defined as
the number of households who lived in the same place a year ago divided by the total number
of households. Regarding the race, Caucasian is dominant in 99 tracts while African
American prevails in the other. Moreover, the average percent of native-born residents was
79.5%. Since there was not much variation in dominant race and percent native-born among
the tracts, the other three variables, i.e. average household income, average household size,
and ratio of long-term residents were used as the model covariates. Additionally, the response
variable for each tract was calculated as the ratio of respondents who had decided to
repair/rebuild.
A Bayesian approach was employed in the analysis. Bayesian framework has been applied
to various research. In the field of structural health monitoring, Yuen and Huang (2018)
proposed a Bayesian frequency-domain method to improve substructure identification by
modeling the boundary force as filtered white noise. By adding to the constraints, this led to
enhanced identification of the inverse problem (Yuen and Huang, 2018). In the same filed,
Huang and Beck (2018) developed a Gibbs sampling procedure for sparse Bayesian learning
to specify the full posterior uncertainty and resulting in better identification of spatially
sparse stiffness reduction of substructure. Wang et al. (2018) applied Bayesian inference to
assess rail health using data from acoustic emission monitoring. Bayesian network approach
has also been applied to assess road safety and to identify common causes of incidents,
probabilities of incidents, critical elements of a system (Castillo et al., 2017a; Castillo et al.,
2017b; Grande et al., 2017), and to model degradation of bridge networks (Kosgodagan‐

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 Nejat, Javid, Ghosh & Moradi

Dalla Torre et al., 2017). Chen et al. (2017) also developed a vision-based method for
autonomous crack detection that employs local binary patterns, support vector machine, and
Bayesian decision theory to detect cracks on underwater metallic surfaces.
This research applies a hierarchical Bayesian geostatistical model with random spatial
effects. Let ( ) denote number of respondents sampled from the census
tract and denote number of respondents who decided to rebuild/repair. The basic model
can be written as

(1)

where is the probability of reconstruction in the grid cell. A spatial model for is
described as

(2)

where ’s are spatial grid-level predictors with regression coefficients and ’s are
random spatial effects from a Gaussian distribution

{ } (3)

where is a spatial covariance matrix with a predefined covariance structure with the
parameter determining the range of spatial dependence. This paper assumes a stationary
distance-based exponential covariance structure defined as

( ) (4)
where ( ) is the Euclidean distance between centroids of the and census tract.
Under the model specified in equations (1) and (2), the joint distribution
is given by Banerjee et al. (2014)

∫ ∏ (5)

Although conditional independence is assumed in (5), the random effects induce marginal
spatial dependence between and . The random effects also allow for the possibility that
and to be different while their mean may be the same. Hence, this method of two-stage
spatial modeling is appropriate when spatial explanation in mean is of interest. Since this
study intends to explain the average decision-making behavior of residents at census-tract
level, hierarchical spatial modeling is resorted by which spatial dependence is incorporated in
the second stage via random effects. It is worth noting that while the partitioned spatial
domain may suggest areal models, the census tracts’ irregular sizes and the fact that most of
the cells were not sampled cause standard conditional autoregressive models to be unstable.
The formal specification of the hierarchical model is completed by assuming independent
Normal priors for regression parameters , Uniform prior for , and Inverse
Gamma prior for .
Although this model specification is mathematically straightforward, there are a few subtle
issues that made fitting the model a challenge. First, the marginal data model in (5) involves
n-dimensional integral and does not have a closed-form solution. Therefore, during the
fitting, the random spatial effects need to be continually updated. However, due to the
dimensionality of W, standard Markov Chain Monte Carlo (MCMC) updates have very low
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 9

acceptance rates. The Gibbs sampler can be used to speed up the computation. Nevertheless,
unlike data augmented probit models (Albert and Chib, 1993), the logit model will not
provide closed-form expressions for full conditionals of the posterior densities. Hence, the
sampling procedure requires Metropolis-Hastings steps embedded within the Gibbs sampler.
Indeed, adaptive Metropolis sampling needs to be resorted to update the random spatial
effects. For the remaining parameters, posterior samples are drawn from their full conditionals
using Metropolis-Hastings algorithm. The entire model was fitted using spBayes package
(Finley et al., 2007) available in R programming language (R Core Team, 2013).
Nevertheless, questions remain of why standard frequentist mixed effect logistic
regression was not fitted or why the authors chose to incur such computational cost. There are
three predominant frequentist approaches to fitting mixed effect logistic regression. First, a
standard Generalized Linear Mixed Model can be applied if one is interested only in the
probabilities conditional on the random effects and wishes to interpret the odds given the
random effects. On the other hand, if it is aimed to extract the marginal probabilities, by
integrating out the random Gaussian effects, the odds interpretation of the regression
parameters is lost during marginalization. Second, if it is wished to work explicitly with
marginally specified logistic model, marginalized random effects models can be used
(Heagerty, 1999), but then conditional interpretation is sacrificed. The third avenue is to use a
marginal analysis that does not need a complete specification of the likelihood (Lipsitz et al.,
1991; Carey et al., 1993). Thus, for the frequentist models, one needs to decide a-priori
whether the marginal or the conditional probabilities are of interest and then choose the
procedure. Parzen et al. (2011) is suggested to be reviewed for more discussion on challenges
of fitting a GLMM on correlated binary data in frequentist set up.
For a Bayesian model, that a-priori should be chosen. The random spatial effects are
continually updated during the iterative fitting procedure. Thus, the posterior distribution of
the conditional probabilities is always available. If study of the marginal probabilities post-
hoc is of interest, the marginal posteriors of can be generated by averaging the posterior
draws of , essentially examining the posterior predictive distribution of It is precisely
for this flexibility that the authors resorted to Bayesian modeling.

5 ANALYSIS AND RESULTS

The model has the following form:

(6)

Socio-demographic data for predictors were based on the 2012 census data at tract level. Out
of the 110 census tracts of Staten Island, 108 tracts are residential from which 24 tracts were
sampled in the survey. Statistics of the model covariates are summarized in Table 3.

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 Nejat, Javid, Ghosh & Moradi

Table 3
Statistics of the model covariates
Model Covariates N Mean SD Min Max
Average income ($) 108 86,467.31 24,426.71 18,648 241,964
Average household size 108 2.834259 0.345914 1.6 3.9
Ratio of long-term residents 108 0.931204 0.047315 0.75 1

To fit the model, socio-demographic data were extracted for the sampled tracts. Then,
posterior samples of the model parameters were obtained using an adaptive MCMC
algorithm with 200 sequential batches of length 200 iterations each. To assess the chain
convergence, three parallel chains, each with over-dispersed initial values, were run and the
Gelman-Rubin diagnostic measure was evaluated (Gelman, 1996). The maximum value of ̂
was , which is below the recommended threshold of (Gelman, 1996), and hence,
indicates convergence of the MCMC scheme. For each chain, the first 20,000 samples were
discarded as burn-in samples and the last 20,000 samples were retained with a unit thinning
interval to obtain the final posterior samples of the parameters. An overall acceptance rate of
approximately 40% was realized in the post burn-in period. Kernel density estimate of the
variables’ posterior distributions are shown in Figure 2, followed by the posterior summary
of the model parameters in Table 4.

Density of intercept Density of percent of long-term residents

Density of income Density of 𝝈𝟐

Density of household size Density of 𝝋

Figure 2. Kernel density estimates of the posterior distributions

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 11

Table 4
Posterior summary of the model parameters
Parameter Posterior mean 95% credible interval MCMC 50% quintile

-5.4519 (-8.39, -3.39) -5.2

-1.9181 (-3.50, -0.59) -1.85
-2.0642 (-3.81, -0.71) -1.96
2.6888 (0.82, 5.20) 2.54
0.2623 (0.11, 0.60) 0.24
0.1748 (0.04, 0.29) 0.18
As Figure 2 and Table 4 suggest, income and household size had decreasing effect while
ratio of long-term residents had increasing effect on the probability of reconstruction within a
tract. This means that tracts with higher average income and bigger household size were
significantly less likely to reconstruct. In contrast, odds of reconstruction were significantly
higher in tracts with more long-term residents. Despite imposing vague priors, with supports
spanning the real line, none of the 95% posterior credible intervals on the regression
coefficients included 0 that indicates significance of all predictors at 5% level.
Once the parameters were estimated, the study proceeded to predict probability of
reconstruction for all the 108 tracts using composition sampling. For each tract, the posterior
realization of the regression parameters ( ), random effect ( ), and predictor
values were plugged into the regression model (6) and a posterior predictive probability ,
set of sampled and unsampled tracts, was obtained. Since these estimates were not
adjusted for the spatially unbalanced nature of the sampling strategy, notion of propensity
score was introduced. The raw propensity score for tract , set of sampled tracts, was
defined as

(7)

where is the unadjusted predictive probability of reconstruction for tract obtained from
the spatially explicit model, and is the relative effort for tract defined as the number of
samples in tract divided by the total sample size

∑ (8)

This adjustment allowed to put more weight on the tracts that were more intensely
sampled. Since sampling intensity was directly proportional to the severity of damage in each
tract, this weighing scheme put more weight on decisions of the households who resided in
more impacted areas. Given that relative efforts do not exist for the unsampled tracts, they
were simulated by the overall sampling rate defined as

∑ ∑ (9)

where set of sampled tracts and denotes the total population in the tract. For the
unsampled tract, was approximated by

(10)

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 Nejat, Javid, Ghosh & Moradi

A normalized relative effort was then calculated across all the census tracts as

̅̅̅ ∑ (11)

Finally, the normalized propensity score for tract was given by

̅̅̅̅ ̅̅̅ ∑ ̅̅̅ (12)

Figure 3 depicts the calculated reconstruction observed probabilities and reconstruction

propensity scores. While magnitude of the propensity scores cannot be individually
interpreted, the normalization allows to interpret propensity score of a tract in comparison to
the others. For example, if tract A has a greater normalized propensity score than tract B,
residents of tract A on average are more likely to decide to reconstruct/repair than tract B.
The propensity score therefore provides an index for housing recovery in each tract. Figure 3
shows a cluster of high propensity scores in the eastern sector, indicating a potential hotspot
for housing reconstruction/repair. Overlaps between the two maps are clearly observable,
especially in areas with the highest observed probabilities.
To validate the model, the reconstruction propensity scores were compared to distribution
of Community Development Block Grant Disaster Recovery (CDBG‐ DR) assistance and its
resultant reconstruction. This assistance is provided by the U.S. Department of Housing and
Urban Development to help with long-term recovery of infrastructure, housing, and economy
in the aftermath of presidentially-declared disasters (HUD, 2017a). CDBG‐ DR is based on
the recovery needs that have not been met by other recovery programs and supplements
programs of the Federal Emergency Management Agency, the Small Business
Administration, and the U.S. Army Corps of Engineers (HUD, 2017b). Further, 70% of the
total allocations should benefit households with low and moderate income, a requirement that
may be revised to 50% through a waiver (HUD, 2001, 2018). Members of the public cannot
apply for CDBG-DR funds directly. Funds are awarded to state and local governments as
grantees (HUD, 2017b).
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 13

Figure 3. Reconstruction observed probabilities and reconstruction propensity scores

In post-Sandy recovery of New York city, Build It Back program is the local CDBG-DR
authority that helps with recovery of homeowners, landlords, renters, and tenants (NYC,
2018a). Figure 4 illustrates the progress of the program from April 2014 to October 2016
(HRO, 2016). The figure shows a positive relationship between number of started/completed
constructions and number of checks issued (i.e. amount of assistance paid) to the
homeowners. In other words, progress of housing recovery can be approximated by the
amount of allocated funds. Further, the CDBG-DR serves homeowners after all other disaster
assistances are exhausted (NYC, 2018c). The reimbursements associated with the Build It
Back program started in early 2014, more than one year after the hurricane (HRO, 2016).
Thus, the recovery progress depicted in Figure 4 can be attributed to the properties which
sustained damage such severe that other forms of assistance could not completely cover them.

Figure 4. Progress of Build It Back program, April 2014 – October 2016 (HRO, 2016)

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14 Nejat, Javid, Ghosh & Moradi

Figure 5. Distribution of single-family projects funded by the Build It Back program by

September 30, 2018 (NYC, 2018b)

Figure 5 illustrates the distribution of Build It Back assistance to single-family homes by

September 30, 2018 (NYC, 2018b) at zip-code level. As shown, the east and south shores of
Staten Island had the major share of applicants and CDBG-DR reimbursements. Moreover,
since reconstruction advanced with the growth of reimbursements (Figure 4), these sectors
observed more reconstruction as well. This conclusion is consistent with the model
predictions that showed a cluster of high propensity scores in the eastern sector of the island
(Figure 3). Therefore, the model was reasonable in predicting the potential reconstruction
hotspots.

6 CONCLUSION
While there is extensive literature on disaster recovery, few models have been proposed
for simulation of housing recovery. This research intends to develop a spatial model capable
of predicting households’ post-disaster recovery decisions. To calibrate the model, data on
recovery of Staten Island residents affected by Hurricane Sandy was collected via a face-to-
face interview. Further, the model was validated through comparison of the propensity scores
with the distribution of CDBG‐ DR assistance and its associated reconstruction. The model
indicated that households with higher average income and bigger household size were less
likely to reconstruct, while odds of reconstruction grew with increase of long-term residents.
These results conform to the literature as, for instance, Anton and Lawrence (2014) suggest
that lower-income residents have a stronger place attachment due to the lack of other
alternatives. Therefore, they unconsciously adopt the idea that their current place of living is
their best option (Anton and Lawrence, 2014). Also, the influence of household size on
recovery has been reported by various researchers (Nigg and Miller, 1993; Mock et al., 2018;
Sadri et al., 2018). Further, in an effort to explore recovery drivers, Nejat and Ghosh (2016)
observed a positive correlation between reconstruction and duration of residence.
The model has the potential to be applied for customizing recovery policies as it can
predict the areas where more residents are expected to reconstruct. Prioritizing restoration of
infrastructure, for example, may more accelerate progress of housing recovery in such
regions.
The methodology used in this research is generalizable and scalable to cases where a
strong spatial dependence exists. However, although the methodology is robust, limitations
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 15

were associated with the research. One limitation, resulting from financial and situational
constraints, was the limited sample size. The limitation also made it impossible to interview
periodically to check for changes of the households’ decisions. Therefore, conducting a survey
in the areas which had limited participants could be another line of study for finer calibration
of the model.
Further, more research is required to study the influence of other parameters such as
recovery and relocation incentives or disaster damage. Financial assistance offered to
residents for returning to their homes play an important role (Kim, 2015; Nejat and Ghosh,
2016). In contrast, post-disaster situation is sometimes so severe, or the area is so prone to
future extreme events that it is essential to relocate the residents (Binder, 2014; Binder et al.,
2015; Greer, 2015). Level of damage can also influence the progress of recovery (Peacock et
al., 2014) and may be included in a model. However, the idea of the current research was to
develop a model which can provide policymakers with a ballpark measure of recovery either
before disasters or immediately after. Therefore, inclusion of damage data would defeat the
main purpose of the model as it would entail waiting for a decent amount of time to collect
data.
In this research, distribution of CDBG‐ DR assistance and its accompanying
reconstruction were applied for model validation. Availability of more comprehensive
recovery reports at proper spatial resolutions may better help with this purpose. Extending the
work to a diverse range of disasters can also assist with evaluating the model applicability.
Finally, future research can explore the pros and cons of changing the unit of analysis to a
finer or coarser scale such as block/block group or county/state.

ACKNOWLEDGEMENTS

This research was supported in part by the National Science Foundation awards #1313946
and #1454650 for which the authors express their appreciation. Publication of this paper does
not necessarily indicate acceptance by the funding entities of its contents, either inferred or
specially expressed herein. The authors would like to thank Renee Hooper and Vance Pryor
who helped with the review of literature.

REFERENCES

Albert, J. H. & Chib, S. (1993), Bayesian Analysis of Binary and Polychotomous Response
Data, Journal of the American Statistical Association, 88(422), 669-679.
Aldrich, D. P. (2010), Fixing Recovery: Social Capital in Post-Crisis Resilience, Journal of
Homeland Security, 6, 1-10.
Aldrich, D. P. (2012), Building Resilience: Social Capital in Post-Disaster Recovery,
University of Chicago Press, Chicago and London.
Almufti, I. & Willford, M. (2013), Reditm Rating System: Resilience-Based Earthquake
Design Initiative for the Next Generation of Buildings, Arup Co.
Anton, C. E. & Lawrence, C. (2014), Home Is Where the Heart Is: The Effect of Place of
Residence on Place Attachment and Community Participation, Journal of environmental
psychology, 40, 451-461.
Banerjee, S., Carlin, B. P. & Gelfand, A. E. (2014), Hierarchical Modeling and Analysis for
Spatial Data, CRC press.

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
16 Nejat, Javid, Ghosh & Moradi

Batouli, M. & Mostafavi, A. (2018), Multiagent Simulation for Complex Adaptive Modeling
of Road Infrastructure Resilience to Sea‐ Level Rise, Computer‐ Aided Civil and
Infrastructure Engineering, 33(5), 393-410.
Berg, S. (2006), Snowball Sampling-I, Encyclopedia of Statistical Science, 12, 7817-7821,
Wiley & Sons, Hoboken, New Jersey.
Binder, S. B. (2014), Resilience and Postdisaster Relocation: A Study of New York's Home
Buyout Plan in the Wake of Hurricane Sandy, University of Hawaii at Manoa.
Binder, S. B., Baker, C. K. & Barile, J. P. (2015), Rebuild or Relocate? Resilience and
Postdisaster Decision-Making after Hurricane Sandy, American Journal of Community
Psychology, 56(1-2), 180-196.
Bolin, R. (1976), Family Recovery from Natural Disaster: A Preliminary Model, Mass
Emergencies, 1(4), 267-277.
Bolin, R. C. & Bolton, P. (1983), Recovery in Nicaragua and the USA, International Journal
of Mass Emergencies and Disasters, 1(1), 125-144.
Bozza, A., Asprone, D., Parisi, F. & Manfredi, G. (2017), Alternative Resilience Indices for
City Ecosystems Subjected to Natural Hazards, Computer‐ Aided Civil and Infrastructure
Engineering, 32(7), 527-545.
Bratt, R. G. (2002), Housing and Family Well-Being, Housing Studies, 17(1), 13-26.
Bullard, R. D. & Wright, B. (2009), Race, Place, and Environmental Justice after Hurricane
Katrina: Struggles to Reclaim, Rebuild, and Revitalize New Orleans and the Gulf Coast,
Westview Press, Boulder.
Burton, C. G. (2015), A Validation of Metrics for Community Resilience to Natural Hazards
and Disasters Using the Recovery from Hurricane Katrina as a Case Study, Ann. Am.
Assoc. Geogr., 105(1), 67-86.
Carey, V., Zeger, S. L. & Diggle, P. (1993), Modelling Multivariate Binary Data with
Alternating Logistic Regressions, Biometrika, 80(3), 517-526.
Castillo, E., Grande, Z., Mora, E., Lo, H. K. & Xu, X. (2017a), Complexity Reduction and
Sensitivity Analysis in Road Probabilistic Safety Assessment Bayesian Network Models,
Computer‐ Aided Civil and Infrastructure Engineering, 32(7), 546-561.
Castillo, E., Grande, Z., Mora, E., Xu, X. & Lo, H. K. (2017b), Proactive, Backward Analysis
and Learning in Road Probabilistic Bayesian Network Models, Computer‐ Aided Civil
and Infrastructure Engineering, 32(10), 820-835.
Chamlee-Wright, E. & Storr, V. H. (2009), “There's No Place Like New Orleans”: Sense of
Place and Community Recovery in the Ninth Ward after Hurricane Katrina, J. Urban Aff.,
31(5), 615-634.
Chen, F. C., Jahanshahi, M. R., Wu, R. T. & Joffe, C. (2017), A Texture‐ Based Video
Processing Methodology Using Bayesian Data Fusion for Autonomous Crack Detection
on Metallic Surfaces, Computer‐ Aided Civil and Infrastructure Engineering, 32(4), 271-
287.
Cohen, J. (1992), Quantitative Methods in Psychology: A Power Primer, Psychol. Bull., 112,
1155-1159.
Comerio, M. C. (1998), Disaster Hits Home: New Policy for Urban Housing Recovery, Univ
of California Press.
Comerio, M. C. (2014), Disaster Recovery and Community Renewal: Housing Approaches,
Cityscape, 16(2), 51-64.
Cutter, S. L., Barnes, L., Berry, M., Burton, C., Evans, E., Tate, E. & Webb, J. (2008), A
Place-Based Model for Understanding Community Resilience to Natural Disasters, Global
environmental change, 18(4), 598-606.
Dacy, D. C. & Kunreuther, H. (1969), Economics of Natural Disasters; Implications for
Federal Policy, Free Press, New York.
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 17

FEMA. (2011), National Disaster Recovery Framework - Strengthening Disaster Recovery

for the Nation, The Federal Emergency Management Agency (FEMA), US Department of
Homeland Security.
Finley, A. O., Banerjee, S. & Carlin, B. P. (2007), Spbayes: An R Package for Univariate and
Multivariate Hierarchical Point-Referenced Spatial Models, Journal of Statistical
Software, 19(4), 1.
Galbusera, L., Giannopoulos, G., Argyroudis, S. & Kakderi, K. (2018), A Boolean Networks
Approach to Modeling and Resilience Analysis of Interdependent Critical Infrastructures,
Computer-Aided Civil and Infrastructure Engineering, 33(12), 1041-1055.
Gelman, A. (1996), Inference and Monitoring Convergence in Markov Chain Monte Carlo in
Practice. Wr Gilks, S. Richardson, and Dj Spiegelhalter, Eds, Chapman & Hall, New
York.
González, A. D., Chapman, A., Dueñas‐ Osorio, L., Mesbahi, M. & D'Souza, R. M. (2017),
Efficient Infrastructure Restoration Strategies Using the Recovery Operator, Computer‐
Aided Civil and Infrastructure Engineering, 32(12), 991-1006.
Goodman, L. A. (1961), Snowball Sampling, The Annals of Mathematical Statistics, 148-170.
Grande, Z., Castillo, E., Mora, E. & Lo, H. K. (2017), Highway and Road Probabilistic
Safety Assessment Based on Bayesian Network Models, Computer‐ Aided Civil and
Infrastructure Engineering, 32(5), 379-396.
Greer, A. (2015), Household Residential Decision-Making in the Wake of Disaster: Cases
from Hurricane Sandy, Unpublished Doctoral dissertation). University of Delaware, USA.
Guha-Sapir, D., Hoyois, P. & Below, R. (2016), Annual Disaster Statistical Review 2015:
The Numbers and Trends, Centre for Research on the Epidemiology of Disasters (CRED),
Brussels.
Hamideh, S., Peacock, W. G. & Zandt, S. V. (2018), Housing Recovery after Disasters:
Primary Versus Seasonal and Vacation Housing Markets in Coastal Communities, Natural
Hazards Review, 19(2), 04018003.
Heagerty, P. J. (1999), Marginally Specified Logistic‐ Normal Models for Longitudinal
Binary Data, Biometrics, 55(3), 688-698.
Henderson, T. L., Roberto, K. A. & Kamo, Y. (2010), Older Adults' Responses to Hurricane
Katrina Daily Hassles and Coping Strategies, Journal of Applied Gerontology, 29(1), 48-
69.
HRO. (2016), Build It Back Progress Update, Mayor’s Office of Housing Recovery
Operations (HRO), New York City, New York, New York, U.S.
Huang, Y. & Beck, J. L. (2018), Full Gibbs Sampling Procedure for Bayesian System
Identification Incorporating Sparse Bayesian Learning with Automatic Relevance
Determination, Computer‐ Aided Civil and Infrastructure Engineering, 33(9), 712-730.
HUD. (2001), Community Development Block Grant Program: Guide to National Objectives
and Eligible Activities for Entitlement Communities.
HUD. (2017a), Fact Sheet: Community Development Block Grant Disaster Recovery (Cdbg-
Dr), The United States Department of Housing and Urban (HUD),
https://www.hudexchange.info/resources/documents/CDBG-DR-Fact-Sheet.pdf.
HUD. (2017b), Programs of Hud: Major Mortgage, Grant, Assistance, and Regulatory
Programs, The United States Department of Housing and Urban (HUD),
https://www.hud.gov/sites/dfiles/Main/documents/HUDPrograms2017.pdf.
HUD. (2018), Key Things to Do When You Receive Cdbg-Dr Funds, The United States
Department of Housing and Urban (HUD)
Huling, D. & Miles, S. B. (2015), Simulating Disaster Recovery as Discrete Event Processes
Using Python, Global Humanitarian Technology Conference (GHTC), 2015 IEEE, IEEE,
pp. 248-253.

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
18 Nejat, Javid, Ghosh & Moradi

Jamali, M. & Nejat, A. (2016), Place Attachment and Disasters: Knowns and Unknowns, J.
Emerg. Management, 14(5), 349-364.
Kamel, N. M. & Loukaitou-Sideris, A. (2004), Residential Assistance and Recovery
Following the Northridge Earthquake, Urban Studies, 41(3), 533-562.
Kick, E. L., Fraser, J. C., Fulkerson, G. M., McKinney, L. A. & De Vries, D. H. (2011),
Repetitive Flood Victims and Acceptance of Fema Mitigation Offers: An Analysis with
Community–System Policy Implications, Disasters, 35(3), 510-539.
Kim, K. (2015), Planning for Post-Disaster Recovery: Next Generation, Edited by James C.
Schwab, Journal of the American Planning Association, 81(2), 159-160.
Kosgodagan‐ Dalla Torre, A., Yeung, T. G., Morales‐ Nápoles, O., Castanier, B., Maljaars,
J. & Courage, W. (2017), A Two‐ Dimension Dynamic Bayesian Network for Large‐
Scale Degradation Modeling with an Application to a Bridges Network, Computer‐ Aided
Civil and Infrastructure Engineering, 32(8), 641-656.
Lipsitz, S. R., Laird, N. M. & Harrington, D. P. (1991), Generalized Estimating Equations for
Correlated Binary Data: Using the Odds Ratio as a Measure of Association, Biometrika,
78(1), 153-160.
Mahmoud, H. & Chulahwat, A. (2018), Spatial and Temporal Quantification of Community
Resilience: Gotham City under Attack, Computer‐ Aided Civil and Infrastructure
Engineering, 33(5), 353-372.
McNeil, S., Mininger, K., Greer, A. & Trainor, J. (2015), Household Residential Decision-
Making in the Wake of Disaster: Report of Results Prepared for Sea Bright Residents,
Disaster Research Center.
Miles, S. B. (2017), Desaster: A Discrete Event Disaster Recovery Simulation Built on Top
of the Simpy Discrete Event Simulation Python Library.
Miles, S. B., Burton, H. V. & Kang, H. (2018), Community of Practice for Modeling Disaster
Recovery, Natural Hazards Review, 20(1), 04018023.
Miles, S. B. & Chang, S. E. (2003), Urban Disaster Recovery: A Framework and Simulation
Model, Multidisciplinary Center for Earthquake Engineering Research (MCEER),
University at Buffalo, Buffalo, NY.
Miles, S. B. & Chang, S. E. (2006), Modeling Community Recovery from Earthquakes,
Earthquake Spectra, 22(2), 439-458.
Miles, S. B. & Chang, S. E. (2011), Resilus: A Community Based Disaster Resilience Model,
Cartogr. Geogr. Inf. Sci., 38(1), 36-51.
Mock, N. B., Kadetz, P., Papendieck, A. & Coates, J. (2018), Dynamics of Early Recovery in
Two Historically Low-Income New Orleans’ Neighborhoods: Tremé and Central City,
Creating Katrina, Rebuilding Resilience, Elsevier, pp. 305-328.
Moradi, S., Vasandani, V. & Nejat, A. (2019), A Review of Resilience Variables in the
Context of Disasters, Journal of Emergency Management, In press.
Nejat, A. (2018), Perceived Neighborhood Boundaries: A Missing Link in Modeling Post-
Disaster Housing Recovery, International Journal of Disaster Risk Reduction, 28, 225-
236.
Nejat, A., Binder, S. B., Greer, A. & Jamali, M. (2018), Demographics and the Dynamics of
Recovery: A Latent Class Analysis of Disaster Recovery Priorities after the 2013 Moore,
Oklahoma Tornado, International Journal of Mass Emergencies & Disasters, 36(1), 23-
51.
Nejat, A. & Damnjanovic, I. (2012), Agent‐ Based Modeling of Behavioral Housing
Recovery Following Disasters, Computer-Aided Civil and Infrastructure Engineering,
27(10), 748-763.
Nejat, A. & Ghosh, S. (2016), Lasso Model of Postdisaster Housing Recovery: Case Study of
Hurricane Sandy, Natural Hazards Review, 17(3), 04016007.
 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
A spatially explicit model of post-disaster housing recovery 19

Nejat, A., Moradi, S. & Ghosh, S. (2019), Anchors of Social Network Awareness Index: A
Key to Modeling Postdisaster Housing Recovery, Journal of Infrastructure Systems, 25(2),
04019004.
Nigg, J. M. & Miller, K. S. (1993), Event and Consequence Vulnerability: Effects on the
Disaster Recovery Process, University of Delaware, Disaster Research Center.
NYC. (2018a), About the Mayor's Office of Housing Recovery Operations, New York City
(NYC) Housing Recovery, https://www1.nyc.gov/site/housingrecovery/about/about.page.
NYC. (2018b), Sandy Funding Tracker, New York City (NYC),
https://www1.nyc.gov/assets/sandytracker/data/maps/sft-bib-business-master/index.html.
NYC. (2018c), Welcome to Nyc Housing Recovery, New York City (NYC) Housing
Recovery, https://www1.nyc.gov/site/housingrecovery/index.page.
Ouyang, M. & Fang, Y. (2017), A Mathematical Framework to Optimize Critical
Infrastructure Resilience against Intentional Attacks, Computer‐ Aided Civil and
Infrastructure Engineering, 32(11), 909-929.
Panakkat, A. & Adeli, H. (2007), Neural Network Models for Earthquake Magnitude
Prediction Using Multiple Seismicity Indicators, International journal of neural systems,
17(1), 13-33.
Panakkat, A. & Adeli, H. (2008), Recent Efforts in Earthquake Prediction (1990–2007),
Natural Hazards Review, 9(2), 70-80.
Panakkat, A. & Adeli, H. (2009), Recurrent Neural Network for Approximate Earthquake
Time and Location Prediction Using Multiple Seismicity Indicators, Computer‐ Aided
Civil and Infrastructure Engineering, 24(4), 280-292.
Parzen, M., Ghosh, S., Lipsitz, S., Sinha, D., Fitzmaurice, G. M., Mallick, B. K. & Ibrahim, J.
G. (2011), A Generalized Linear Mixed Model for Longitudinal Binary Data with a
Marginal Logit Link Function, The annals of applied statistics, 5(1), 449.
Peacock, W. G., Dash, N. & Zhang, Y. (2007a), Sheltering and Housing Recovery Following
Disaster, in Kaplan, H. (ed.), Handbook of Disaster Research, Springer, Newyork, pp.
258-274.
Peacock, W. G., Dash, N. & Zhang, Y. (2007b), T, in Kaplan, H. (ed.), Handbook of Disaster
Research, Springer, Newyork, pp. 258-274.
Peacock, W. G., Van Zandt, S., Zhang, Y. & Highfield, W. E. (2014), Inequities in Long-
Term Housing Recovery after Disasters, J. Am. Plann. Assoc., 80(4), 356-371.
Pravettoni, R. (2009), Number of Disasters Per Year, GRID-Arendal,
http://www.grida.no/resources/7324.
R Core Team. (2013), R: A Language and Environment for Statistical Computing, R
Foundation for Statistical Computing, Vienna, Austria.
Rafiei, M. H. & Adeli, H. (2017), Neews: A Novel Earthquake Early Warning Model Using
Neural Dynamic Classification and Neural Dynamic Optimization, Soil Dynamics and
Earthquake Engineering, 100(9), 417-427.
Rust, E. B. & Killinger, K. (2006), The Financial Services Roundtable Blue Ribbon
Commission on Mega-Catastrophes: A Call to Action, Financial Services Roundtable.
Sadri, A. M., Ukkusuri, S. V., Lee, S., Clawson, R., Aldrich, D., Nelson, M. S., Seipel, J. &
Kelly, D. (2018), The Role of Social Capital, Personal Networks, and Emergency
Responders in Post-Disaster Recovery and Resilience: A Study of Rural Communities in
Indiana, Natural Hazards, 90(3), 1377-1406.
Sanders, S., Bowie, S. L. & Bowie, Y. D. (2004), Chapter 2 Lessons Learned on Forced
Relocation of Older Adults: The Impact of Hurricane Andrew on Health, Mental Health,
and Social Support of Public Housing Residents, J. Gerontol. Soc. Work, 40(4), 23-35.
Schwartz, E. (2006), A Needless Toll of Natural Disasters, The Boston Globe.

This article is protected by copyright. All rights reserved.

 14678667, 2020, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mice.12487, Wiley Online Library on [28/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
20 Nejat, Javid, Ghosh & Moradi

USCB. (2018), Annual Estimates of Housing Units for the United States, Regions, Divisions,
States, and Counties: April 1, 2010 to July 1, 2017, The United States Census Bureau.
Wang, J., Liu, X. Z. & Ni, Y. Q. (2018), A Bayesian Probabilistic Approach for Acoustic
Emission‐ Based Rail Condition Assessment, Computer‐ Aided Civil and Infrastructure
Engineering, 33(1), 21-34.
Wang, Z., Wang, Q., Zukerman, M., Guo, J., Wang, Y., Wang, G., Yang, J. & Moran, B.
(2017), Multiobjective Path Optimization for Critical Infrastructure Links with
Consideration to Seismic Resilience, Computer‐ Aided Civil and Infrastructure
Engineering, 32(10), 836-855.
Yandong, Z. (2007), Social Capital and Post-Disaster Recovery: A Sociological Study of
Natural Disaster, Sociological Studies, 5, 164-187.
Yuen, K. V. & Huang, K. (2018), Identifiability‐ Enhanced Bayesian Frequency‐ Domain
Substructure Identification, Computer‐ Aided Civil and Infrastructure Engineering, 33(9),
800-812.