28 ISE Magazine | www.iise.org/ISEmagazine
Every year, Puerto Rico’s population enacts
a set of measures to get ready for the
hurricane season. The U.S. Federal Emergency
Management Agency (FEMA) assists in
accelerating survivor recovery in the United
States and its territories after a disaster.
In September 2017, hurricanes Irma and María made
landfall in Puerto Rico, which divided the island’s history
in two: “before and after María.” A recent news article
suggested that more than 60.5% of Puerto Ricans’ aid
applications were denied by FEMA related to those
events.
E
Measuring disaster preparation
in Puerto Rico
Data analysis shows key inuencers lead more people to take action
By Yesenia Cruz Cantillo, Carlos J. González Oquendo and Cristian Solano Payares
November 2023 | ISE Magazine 29
Five years later in September 2022, Hurricane Fiona
hit the island. If FEMA had a prole of vulnerable people,
would there be a way of predicting which population
segment is likely to request aid in an emergency – and
in turn, the agency could proactively intervene to better
handle their process?
Previous research has shown machine learning
techniques to help FEMA forecast Transitional Shelter
Assistance (TSA) eligibility. This research aims to
determine how predictive models can assist FEMA by
helping build population proles that support decision-
makers. Predictive analytics will extract information from
datasets retrieved from fema.gov to perform prediction
analysis about vulnerable population proles. Data
mining techniques will be applied to discover hidden
knowledge and unexpected patterns in databases,
analyze them from dierent perspectives, categorize
them, summarize the identied relationships and
establish key predictors. This will allow ocials to
know in advance the attributes of most likely people to
request aid in an emergency.
Hurricane impacts
FEMA reported in its mitigation assessment team
report, Hurricanes Irma and Maria in Puerto Rico: Building
Performance, Observations, Recommendations and
Technical Guidance, that in September 2017, Hurricane
Irma, a Category 5 storm, passed near Puerto Rico,
causing $s1 million in damage. Thirteen days later,
Category 4 Hurricane María landed on the island,
causing $100 billion and more than 4,000 fatalities.
An article in theconversation.com and govexec.com in
October 2022 reported that more than 60% of Puerto
Ricans seeking FEMA aid after Hurricane Maria had their
applications for housing assistance denied, mainly due
to the homeowners’ lack of legal property titles. When
Hurricane Fiona, Category 1 storm, impacted Puerto Rico
in September 2022, blue tarps still covered hundreds of
homes because new roofs still had not been built after
María.
Independent of FEMA’s reasons for declining aid, it is
relevant that the agency and Puerto Rican government
could discover useful patterns and trends in disaster
community preparedness actions, attitudes and
motivations. This would show that advanced resources
are needed and make them better prepared to aid
vulnerable populations. Using data extraction of
individual disaster preparedness from the 2021 FEMA
National Household Survey (NHS), a nationwide survey
that included Puerto Rico, predictive analytics revealed
important information. Therefore, nding signicant
predictor variables could help estimate future patterns
and trends.
A 2019 paper by C.E., et al., “Quantifying inequities in
U.S. federal response to hurricane disasters in Texas and
Florida compared with Puerto Rico,” reected this, as
did A CBS News study, “Puerto Rico received slower, less
‘generous’ federal disaster aid than Texas, Florida.” In
both, the information focused on how nancial support
to vulnerable populations in Puerto Rico impacted by
hurricanes has been relatively slow during the past ve
years. According to the studies, survivors in Florida and
Texas received approximately $100 million in FEMA
funds across the nine-day period when hurricanes
Harvey and Irma made landfall in the mainland U.S.
Meanwhile, Puerto Rico obtained almost $6 million
in FEMA aid for María survivors in a similar period. Also
worrying was the deployment of emergency workers:
31,000 were deployed for the storms’ peak in Texas
versus 19,000 in Puerto Rico a month after the impact
of María. The study also noted that Puerto Rico received
fewer helicopters, tarps, water and food than the
mainland states. The reports also indicated that FEMA
acknowledged its response to Hurricane María was
30 ISE Magazine | www.iise.org/ISEmagazine
Measuring disaster preparation in Puerto Rico
the result of being unprepared and understaed. As a
result, resources such as food supplies needed were
underestimated in the storm’s aftermath.
Preparedness influencers
The 2021 FEMA NHS dataset employed for this research
was retrieved from fema.gov. FEMA conducted an 11-week
survey from Feb. 24 to May 14, 2021, to assess American
disaster preparedness behaviors, attitudes and motives,
including those in Puerto Rico. They published their
ndings in the report, “2021 National Household Survey,
Key Findings: Individual and Community Preparedness
Division.”
It showed that in Puerto Rico, 73.5% of adult respondents
say they are pursuing three or more of the 12 preparedness
actions listed. According to 64.6% of respondents, they
saved for a rainy day, assembled or updated provisions
and prepared for emergencies or disasters in the previous
year.
The Boston University School of Public Health’s
“Behavioral Change Models-The Transtheoretical Model
(Stages of Change)” states that the Transtheoretical
Model, also known as the Stages of Change Model (SCM),
concentrates on individual decision-making. It operates
under the assumption that people do not make sudden,
drastic behavioral changes. Instead, a cyclical process
results in continuous behavior change, especially in the
case of habitual behavior. The SCM is a model, not a
theory, and can be used to test various behavioral ideas
and structures.
According to this model approach, 2% of individuals in
Puerto Rico are in “precontemplation” and have no plans
to prepare in the upcoming year; 16.3% have switched to
“contemplation” and plan to get ready during the coming
year; 20.4% have moved to “preparation,” where they are
ready to take action within the next six months; 28.6%
are in “action,” and have recently changed their behavior;
and 32.7% are in “maintenance,” sustaining their behavior
change (See Figure 1).
The FEMA report identied four critical inuencers
on disaster preparedness, which are standard behavior
change actions that are deeply associated with higher
preparedness actions: awareness, disaster experience,
preparedness ecacy and risk perception. Preparedness
actions are more likely to be taken for those who have one
of these inuencers, as applied to Puerto Rico:
• Awareness. To be more prepared for a crisis, 79.6% of
Puerto Ricans have read, seen or heard information
about assembling or updating supplies.
• Disaster experience. 95.9% of respondents possess
disaster-related personal or familial experience.
• Preparedness ecacy. Taking precautions to assist
them in surviving a calamity, in the opinion of 46.9%, was
a wise decision.
• Risk perception. 59.2% think that it is very likely a
disaster will impact them in the area they live.
Data mining, key influencers
to disaster preparedness
Employing data mining techniques allowed the discovery
of hidden knowledge and unexpected patterns in the NHS
datasets. Data analysis was performed from many dierent
perspectives; it was categorized, and the identied
relationships were summarized.
In particular, the classication tree method was applied
to generate rules for data classication. Dependent and
independent variables were carefully chosen, considering
aspects of Puerto Ricans’ disaster preparedness actions,
attitudes and motivations. Classication tree generation
Figure 1
Stages of change model
Puerto Rican residents’ behavior in preparation for a possible disaster.
November 2023 | ISE Magazine 31
helped to predict responses on a categorical dependent
variable. Exhaustive CHAID (chi-squared automatic
interaction detection) was the growing method chosen
for classication tree building. It chose independent
predictors with the strongest interaction with dependent
variables selected from the FEMA NHS data. Once the
trees were built, “if-then” statements were employed to
describe the likelihood of the selected variables.
The main ndings from applying classication trees with
exhaustive CHAID describe key inuencers to disaster
preparedness for Puerto Ricans.
Awareness. The main predictor of “information read,
seen, or heard about assemble or update supplies to
be better prepared for a disaster” is “supplies lasting
assembled.” People whose supplies lasted less
than one week had a likelihood of 67.7% of reading,
seeing or hearing information about assembling or
updating supplies. Those whose supplies lasted less
than a week and who lived in a single-unit home or a
multiunit apartment complex or condo, with or without
a basement, had a likelihood of 80.8% of being aware of
such information. This model has a risk estimate of 0.102,
indicating that in 10.2% of cases, the model’s predicted
category – yes or no – is incorrect. Therefore, the “risk” of
misclassifying a case is approximately 10%. The model
classies about 90% of Puerto Ricans correctly.
Disaster experience. The main predictor of “personal/
family experience of disaster impacts” is “type of disaster
experienced.” People or families that had experienced
hurricane disasters had a likelihood of 100% of ever
experiencing the impacts of a disaster. The “risk” of
misclassifying a case is approximately 2%. Therefore, the
model classies about 98% of Puerto Ricans correctly.
Preparedness ecacy. The main predictor of the
“impact of taking preparation steps that help people
get through a disaster” is “considering preparing for a
disaster” (see Figure 2). Puerto Ricans who considered
preparing for a disaster had a 72.7% likelihood of believing
it is a good idea to take preparation steps to help them
through a disaster. Those who considered preparing
for a disaster and knew evacuation routes as part of
disaster or emergency preparation in the last year had
a 100% likelihood of believing it was a good idea to take
preparation steps to get through a disaster. The “risk” of
misclassifying a case is approximately 12%. The model
classies about 88% of the Puerto Ricans correctly.
Risk perception. The main predictor of the “likelihood
perception of a disaster impacting the living area” is
“duration of living without power.” Puerto Ricans who lived
fewer than three months without power had a likelihood
of 60.9% of thinking that it is very likely they would be
impacted by a disaster in their area. Those who lived
The agency (FEMA) and the Puerto
Rican government could discover
useful patterns and trends in disaster
community preparedness actions,
attitudes and motivations. This
would show that advanced resources
are needed and make them better
prepared to aid vulnerable populations.
Figure 2
Level of preparation
The extent Puerto Ricans made themselves ready for a possible disaster.
32 ISE Magazine | www.iise.org/ISEmagazine
Measuring disaster preparation in Puerto Rico
fewer than three months without power and documented
property via photo or video as actions taken to prepare
had a 75.9% likelihood of thinking they would be impacted.
And 95.5% believed their area would be aected by a
disaster if they had experienced fewer than three months
without power, documented their property with photos
or videos and expressed great concern about not having
access to health and medical care (medical care, public
health, patient movement, medical supply chain, fatality
management and dispatch). The “risk” of misclassifying a
case is approximately 14%. Therefore, the model classies
about 86% of Puerto Ricans correctly.
Community preparedness proles: nancial assistance
aid. The main predictor of “types of aid expected: nancial
assistance” is total household annual income before taxes
(see Figure 3). Total household annual income before taxes
had an impact on the type of aid people expected from
nancial assistance. For example, Puerto Ricans whose
income was between $20,000 to $49,999 had a likelihood
of 88.9% of expected nancial assistance. For those whose
incomes were between $20,000 to $49,999 and do not
have a disability or a health condition that might aect their
capacity to respond to an emergency (a mobility, hearing,
vision, cognitive, or intellectual disability or physical,
mental or health condition) had a likelihood of 100% of
expected aid in the form of nancial assistance.
Following the theoretical stages of the change
model approach, 2% of people in Puerto Rico are in
precontemplation; 16.3% have changed to contemplation;
20.4% have moved to preparation; 28.6% are in action; and
32.7% are in maintenance. Decision trees were built based
on the 2021 FEMA NHS dataset that covers demographic
information and disaster preparedness actions, attitudes
and motivations.
Puerto Ricans who were in the contemplation stage had
a likelihood of 31.2% in believing the supplies assembled
would last more than one week; those in the maintenance
stage had a likelihood of 45.5% that supplies would last
more than three months.
In addition, Puerto Ricans whose income was between
$20,000 to $49,999 had a likelihood of 88.9% of expected
aid in the form of nancial assistance. This is the beginning
point of developing a scoring model that can be applied
to other datasets. The rst step is to estimate the number
of Puerto Ricans who will likely want the same kind of
assistance when their demographics are similar. Next, the
model is applied to compare datasets that lack information
on past disaster preparedness but do have demographic
information accessible.
Yesenia Cruz Cantillo, Ph.D., MSIE, MECE, is an industrial
engineer with a master’s degree in industrial engineering
focused on Quality Systems and expertise in Quality
Control Systems. She has six years of experience
developing, evaluating and implementing quality
The FEMA report identified four critical
influencers on disaster preparedness,
which are standard behavior change
actions that are deeply associated
with higher preparedness actions:
awareness, disaster experience,
preparedness ecacy and risk
perception.
Figure 3
Aid expectation
Percentage of residents who expect nancial assistance based on household income.
November 2023 | ISE Magazine 33
management systems using ISO 9000 standards and
statistical analysis applied to several production and
service processes. She also holds a Ph.D. in civil/
transportation engineering and a master’s degree
in transportation engineering and strong abilities
in simulation. She has over 17 years of experience
in research and teaching and almost eight years of
experience in transportation consulting. She has a strong
background in production engineering and expertise in
improving information systems using informatics and
production planning and control.
Carlos J. González Oquendo is a Ph.D. student who
earned a master’s degree in electrical engineering with
a specialization in telecommunications, signal & image
processing. He has a minor in advanced mathematics
and possesses knowledge and experience in statistics,
stochastic and random processes, estimation theory,
math modeling, scientic programming and simulation.
He is pursuing a Ph.D. in electrical and computer
engineering at Mississippi State University.
Cristian Solano Payares, a Ph.D. student, is an industrial
engineer with 21 years of experience in teaching and
research and 2½ years in logistic consulting. He is a
business logistic specialist with a master’s degree in
management engineering focused on logistics. He is
a doctoral student at Universidad Popular Autónoma
del Estado de Puebla in the Logistics and Supply Chain
Management Program. He is a professor in the Industrial
Engineering program at Universidad del Atlántico in
logistics and a researcher in hospital logistics and
process improvement projects for logistics process
of health services companies. He is a member of the
research engineering group, Research and Innovation for
Development (3i+d).
Note: For a full list of references used by the authors
for this article, see the ISE reference page, iise.org/
isemagazine/references.
Puerto Rico data sheet
Puerto Rico is a territory of the United States with a population of 3,221,789 (U.S. Census Population Estimates,
July 1, 2022) with 44% of the population ages 16 and over are employed.
The National Council on Disability, in its 2022 report “Disparate Treatment of Puerto Rico Residents with
Disabilities in Federal Programs and Benets,” established that the poverty rate in Puerto Rico is 44.1%, which is
double that of Mississippi, the poorest state in the U.S. It also indicated that the prevalence of disabilities is 22%,
slightly lower than the national prevalence rate of 26%.
The highest prevalence of disability among Puerto Rican residents is in mobility, with a 12.6% of prevalence
rate, almost double that of the U.S. The incidence of low vision or blindness in Puerto Rico is four times the rate of
those in the U.S.
In addition, 54.7% of Puerto Ricans have a household income of less than $24,999. The median household
income in Puerto Rico is $21,967, in contrast with $57,406 (2023) in the U.S.
