Resilience Index Methodology

0704-0610_OSIE_Resilience Index Methodology_August2024.pdf

On-Site Installation Evaluations

Resilience Index Methodology

OMB: 0704-0610

Document [pdf]
Download: pdf | pdf
CUI
Chief Digital & Artificial Intelligence Office

OSIE Resilience Index Methodology
Contents

Points of Contact........................................................................................................................... 1
Background ................................................................................................................................... 1
Data Sources ................................................................................................................................ 2
Data Ingestion and Merging .......................................................................................................... 4
OSIE Resilience Index: Domain Calculations ............................................................................... 4
OSIE Resilience Index: Weighting Schemes and Stability Analysis ........................................... 12

Points of Contact
Office of Command Climate and Wellbeing Integration (OCCWI) Project Team:
Andra Tharp, PhD HQE | OCCWI Director | Andra.L.Tharp.civ@mail.mil
Travis Bartholomew | OCCWI Deputy Director | Travis.W.Bartholomew.civ@mail.mil
Andrew Moon, PhD | Decision Support & Performance Evaluation Director |
Andrew.M.Moon4.civ@mail.mil
Rachel Clare, PhD | Evaluation Specialist | Rachel.C.Clare.civ@mail.mil
Advana Project Team:
Brittney Davis | CDAO People & Health Portfolio Lead | Brittney.H.Cates.civ@mail.mil
Stephen Axelrad, PhD | BAH People & Health Portfolio Lead CDAO | Stephen.H.Axelrad.ctr@mail.mil
Melissa Macasieb, PhD | CDAO Product Lead | Melissa.L.Macasieb.ctr@mail.mil
Daniel Dockterman, PhD | Lead Dashboard Developer & Methodologist
Daniel.M.Dockterman.ctr@mail.mil
Zachary Alerte | Developer/Methodologist | Zachary.W.Alerte.ctr@mail.mil
Jessica Bianchi | Developer/Methodologist | Jessica.E.Bianchi.ctr@mail.mil
Sarah Leffingwell, PhD | Developer/Methodologist | Sarak.K.Leffingwell.ctr@mail.mil
Jennifer Phung | Developer/Methodologist | Jennifer.Phung2.ctr@mail.mil
Lauren Walker | Developer/Methodologist | Lauren.M.Walker19.ctr@mail.mil

Background
On February 26, 2021, Secretary of Defense Lloyd Austin issued the Memorandum, “Immediate Actions
to Counter Sexual Assault and Harassment and the Establishment of a 90-Day Independent Review
Commission on Sexual Assault in the Military,” which directed immediate actions to address sexual
assault and harassment. Immediate Action 2 directed the Office of the Secretary of Defense (OSD) to
conduct on-site installation evaluations (OSIEs) and to provide quarterly command climate updates.

1

CUI

CUI
Chief Digital & Artificial Intelligence Office

To support identification of installations for the 2021 evaluations, the Under Secretary of Defense for
Personnel and Readiness (USD(P&R)) directed the completion of a force-wide Defense Organizational
Climate Survey (DEOCS). The DEOCS was selected as the primary data source for the 2021 installation
evaluations because it serves as the most timely and sensitive Defense-wide measure of command
climate and because other relevant data were delayed due to COVID. In 2022, command climate
updates employed a multi-measure approach to better capture the many facets of installation risk.
Specifically, the OSIE team developed a Risk Index by leveraging several data sources, in addition to the
DEOCS, across five organizational levels: Individual, Workplace, Leadership, Installation, Community.
In 2024, the Risk Index was reframed as a resilience index, and incorporated updated and additional data
sources. The Resilience Index was calculated in March 2024 and August 2024. The data sources used in
index versions from 2022 and 2024, and the subsequent methodology used to identify outlier installations
in terms of resilience, are described in detail in the sections below.

Data Sources
Table 1 below outlines differences between the 2022 Risk, March 2024 Resilience, and August 2024
Resilience versions of the OSIE Index. The data sources are described in more detail following the table.
Table 1: Data Source Usage in 2022 Risk and 2024 Resilience Index Versions
Time period used in Time period used in Time period used in
Data Source
2022 Index
Mar 2024 Index
Aug 2024 Index
Defense Organizational Climate Survey (DEOCS)
Jan 2021-Jul 2022
Jan-Dec 2022
-5.0
DEOCS 5.1
--Aug 2023-Jan 2024
Workplace and Gender Relations Survey of Active2018
2018 & 2021
2018 & 2023
Duty Members (WGRA) - Contextual Analysis
Defense Sexual Assault Incident Database (DSAID)
FY 2018
CY 2021
FY 2023
Defense Suicide Prevention Office (DSPO) Suicide
2020-2022
2016-2022
2016-2024
Counts
Status of Forces Survey of Active-Duty Members [not used]
2020
2022
Contextual Analysis
Family Advocacy Program (FAP) Domestic and
FY 2021
FY 2022
FY 2022
Child Abuse Counts
U.S. County Health Rankings & Roadmaps
2022
2023
2024
(CHR&R)
Defense Organizational Climate Survey (DEOCS): Designed by the Office of People Analytics (OPA),
the DEOCS assesses 19 protective and risk factors that can impact a unit/organization’s climate and the
ability to achieve their mission.
Protective factors are attitudes, beliefs, and behaviors associated with positive outcomes for
organizations or units. Higher favorable scores on protective factors are linked to a higher likelihood of
positive outcomes, such as improved performance or readiness and higher retention, and are also linked
to a lower likelihood of negative outcomes, such as suicide, sexual harassment, and sexual assault. The
DEOCS identifies 10 Protective factors. However, for this analysis, transformational leadership ratings for
the unit/organization leader and the non-commissioned officer, where applicable, are treated as two
separate factors. Thus, the 11 Protective factors are as follows: Cohesion, Connectedness, Engagement
& Commitment, Fairness, Inclusion, Morale, Safe Storage for Lethal Means, Work-Life Balance,

2

CUI

CUI
Chief Digital & Artificial Intelligence Office

Leadership Support, Transformational Leadership (Commander), and Transformational Leadership (Noncommissioned Officer).
Risk factors are attitudes, beliefs, and behaviors associated with negative outcomes for organizations or
units. Higher unfavorable scores on risk factors are linked to a higher likelihood of negative outcomes,
such as suicide, sexual harassment, and sexual assault, and are also linked to a lower likelihood of
positive outcomes, such as higher performance, readiness, and retention. The DEOCS identifies nine
Risk factors. However, for this analysis, passive leadership ratings and toxic leadership ratings for the
unit/organization leader and the non-commissioned officer, where applicable, were treated as separate
factors. Thus, the 11 Risk factors are as follows: Alcohol Impairing Memory, Binge Drinking, Stress
Passive Leadership (Commander), Passive Leadership (Non-Commissioned Officer), Toxic Leadership
(Immediate Supervisor), Toxic Leadership (Non-Commissioned Officer), Racially Harassing Behaviors,
Sexually Harassing Behaviors, Sexist Behaviors, and Workplace Hostility. For more information on the
DEOCS, see Prevention | Home.
The 2022 OSIE Risk Index used data from DEOCS 5.0, spanning January 2021 through July 2022. The
March 2024 OSIE Resilience Index used data from DEOCS 5.0, spanning January 2022 through
December 2022. The August 2024 OSIE Resilience Index used data from DEOCS 5.1, spanning August
2023 through January 2024.
Workplace and Gender Relations Survey of Active-Duty Members (WGRA) – Contextual Analysis:
OPA’s 2018, 2021, and 2023 WGRA provide insights regarding the estimated prevalence and
characteristics of sexual assault, sexual harassment, and gender discrimination in the Active Component;
Service members’ experiences with reporting these types of incidents; and perceptions of unit culture and
climate. A follow-up contextual analysis using 2018 WGRA data was done by OPA to understand how
rates of sexual assault and sexual harassment vary across installations and ships. For more information,
see https://www.opa.mil/research-analysis/health-well-being/gender-relations/contextual-studiesworkplace-and-gender-relations-survey-of-active-duty-members/2018-contextual-risk-factors-associatedwith-sexual-assault-and-sexual-harassment-in-active-duty-overview-report. i For the 2021 and 2023
surveys, a limited contextual analysis was performed to replicate estimated prevalence rates. In addition,
installation level estimates of general reporting climate and trust in the military system were produced
from the 2018 survey, though these items were not produced for the 2021 or 2023 survey.
Defense Sexual Assault Incident Database (DSAID): DSAID is the Department’s authoritative,
centralized database used to collect and maintain information about sexual assault cases involving
members of the U.S. Armed Forces. The Sexual Assault Prevention and Response Office (SAPRO)
provided a record of every reported case of military sexual assault (restricted and unrestricted) in FY
2018, CY 2021, and FY 2023, by installation. These years were chosen to align with WGRA fielding
windows. For more information on DSAID, see https://www.sapr.mil/dsaid-overview.
Status of Forces Survey of Active-Duty Members (SOFA) - Contextual Analysis: The Office of
People Analytics (OPA)’s SOF enables the DoD to assess the attitudes and opinions of the DoD
community, such as retention, satisfaction, stress, and readiness. The survey includes questions relating
to suicide-related behaviors, such as suicide attempts and suicidal ideation. For the 2020 and 2022
SOFA, OPA performed a limited contextual analysis to assess installation level estimates of suicide
attempts and ideation. For more information on the SOF, see https://www.opa.mil/research-analysis/opasurveys/status-of-forces-surveys
Defense Suicide Prevention Office (DSPO) Suicide Counts: DSPO provided the OSIE team with a
record of every military suicide from 2016 through 2024 (as of June 2024), by UIC and installation. ii
DSPO is part of the Office of the Under Secretary of Defense for Personnel and Readiness and is the

3

CUI

CUI
Chief Digital & Artificial Intelligence Office

authoritative source for suicide data in the Department of Defense (DoD). For more information on
DSPO, see https://www.dspo.mil/.
Family Advocacy Program (FAP) Domestic and Child Abuse Counts: The Office of Military
Community and Family Policy provided the OSIE team with FAP records of every substantiated incident
of domestic abuse and child abuse and neglect in FY 2021 and FY 2022, by installation. iii FAP is the
DoD’s program designated to address domestic abuse, child abuse and neglect, and problematic sexual
behavior in children and youth. For more information on FAP, see
https://www.militaryonesource.mil/family-relationships/family-life/preventing-abuse-neglect/the-familyadvocacy-program/.
U.S. County Health Rankings & Roadmaps (CHR&R): CHR&R is a program of the University of
Wisconsin Population Health Institute that compiles local U.S. health data to help communities identify
opportunities to improve their health. The CHR&R spans several health focus areas: length of life, quality
of life, tobacco use, diet and exercise, alcohol and drug use, sexual activity, access to clinical care,
quality of clinical care, education, employment, income, family and social support, community safety, air
and water quality, housing and transit, and demographics. For more information on the CHR&R, see
https://www.countyhealthrankings.org.

Data Ingestion and Merging
Each data source informing the OSIE Resilience Index was ingested into Advana. Validation consisted of
confirming record counts match and comparing individual values to the original file for select rows and
registrations. We also verified all variables to ensure they were transferred properly and contained valid
values. 
We then merged each of the separate data sources into a single table using Databricks. Specifically, we
merged installation names from the DEOCS, WGRA, SOFA, DSAID, DSPO, and FAP data files using a
standardized list from the OSIE master database of installations and ships (n = 1,668 in the full
installation list). iv This external master list allowed us to match installations with different names/aliases
across data files (e.g., Eglin Air Force Base vs. Duke Field vs. Camp Bull Simons). Where applicable, we
also aggregated installations from the data files to match the OSIE master database. For instance,
McGuire Air Force Base, Fort Dix, and Naval Air Engineering Station Lakehurst—listed as separate bases
in the WGRA data—were collapsed into the OSIE installation Joint Base McGuire-Dix-Lakehurst. Finally,
we merged the CHR&R to the master database of installations by matching OSIE installations with U.S.
Counties using FIPS codes.

OSIE Resilience Index: Domain Calculations
We categorized data sources and DEOCS factors into five levels or domains based on a social ecological
model. A social ecological model is a public health framework used to understand the complex interaction
between the individual, interpersonal, organizational, and community factors that affect a person’s overall
health and well-being. This framework enables scholars to better understand the causal processes
behind incidents or harm or violence, including why and how individuals are at risk or protected from harm
or violence. To create environments free from harm and violence, it is necessary to enhance protective
factors and reduce risk factors at every level of the social ecological model. v
To produce a social ecological model appropriate for a military environment, we tailored the levels of the
model to better suit an installation setting in which a Service member is embedded in an existing chain of

4

CUI

CUI
Chief Digital & Artificial Intelligence Office

command or leadership structure. The social ecological model we used to produce the installation
resilience index reflects risk and protective factors at five different levels (see Figure 1):
1. Individual (e.g., individual behaviors, attitudes)
2. Workplace (e.g., work peers, interpersonal teams, etc.)
3. Leadership (e.g., organizational factors controlled by the command team or supervisor)
4. Installation (e.g., installation historical prevalence or incidence rates)
5. Community (e.g., health trends in the surrounding civilian community)
These five levels constitute a robust social ecological model tailored for the military environment. We
used a “best fit” approach and placed each risk and protective factor into a single level of the social
ecological framework.
Figure 1: OSIE Resilience Index, by Domain

Individual: The Individual domain is comprised of six factors from the DEOCS: Connectedness, Sexually
Harassing Behaviors, Racially Harassing Behaviors, Sexist Behaviors, Alcohol Impairing Memory, and
Binge Drinking. For each factor, we converted installation raw scores to percentiles by comparing each
installation’s factor score to the factor scores of all other installations with DEOCS data (n = 931). vi
Because percentiles are a measurement of relative resilience, we assigned percentiles for
Connectedness in ascending order and for the five risk factors in descending order. (i.e., A higher value
on a protective factor translates to a higher percentile, or a higher level of resilience, and vice versa.) We
then averaged the six factor percentiles to create a DEOCS Individual domain composite score for each
installation. Finally, depending on the weighting scheme employed (Original Weights, Data Coverage,
and Domains Equally Weighted) we assigned the DEOCS Individual domain a weight of either 15% or
20% in the OSIE Resilience Index (see Table 2). Note that we detail each weighing scheme in the final
section of the Methodology.

5

CUI

CUI
Chief Digital & Artificial Intelligence Office

Table 2: OSIE Resilience Index, Weighting Schemes by Domain
Domain

Items/Inputs

Data Source

Scoring Type

Original
Weights

Data Coverage

Equal Domain
Weights

DEOCS Individual

Average Factor
Percentile

15%

20%

20%

DEOCS Workplace

Average Factor
Percentile

15%

20%

15%

WGRA Climate

Average Scale Score
Percentile

5%

10%

5%

DEOCS Immediate
Supervisor

Average Factor
Percentile

5%

7%

7%

DEOCS Commander

Average Factor
Percentile

5%

7%

7%

DEOCS NonCommissioned Officer

Average Factor
Percentile

5%

7%

7%

WGRA Male

Rate Per Capita
(Percentile)

5%

10%

3%

WGRA Female

Rate Per Capita
(Percentile)

5%

--

3%

WGRA & DSAID

Rate Percentile

5%

--

3%

FAP

Rate Per Capita
(Percentile)

10%

--

6%

Suicide Risk Group
Average Percentile
Average Percentile

15%
5%
5%

10%
5%
10%

5%
10%
10%

100%

100%

100%

Connectedness
Sexually Harassing Behaviors
Individual

Racially Harassing Behaviors
Sexist Behaviors
Alcohol Impairing Memory
Binge Drinking
Stress
Work-life Balance
Engagement & Commitment
Morale

Workplace

Fairness
Inclusion
Cohesion
General Reporting Climate
Trust in the Military System
Toxic Leadership
Leadership Support
Transformational Leadership

Leadership Passive Leadership
Transformational Leadership
Passive Leadership
Toxic Leadership
Estimated Male Sexual Assault
Rate
Estimated Male Sexual
Harassment Rate
Estimated Female Sexual Assault
Rate
Installation Estimated Female Sexual
Harassment Rate
Estimated Sexual Assault NonReporting Rate
Domestic Abuse Counts
Child Abuse Counts
Suicide Counts
Health Outcomes
Community
Health Factors

DSPO & SOFA
CHR&R

Total
Notes: Percentages may not sum to 100% due to rounding. See Table 5 for full list of CHR&R measures.

Workplace: The Workplace domain utilized two data sources: DEOCS and WGRA. The DEOCS
component included seven factors: Stress, Work-life Balance, Engagement & Commitment, Morale,
Fairness, Inclusion, and Cohesion. Like the DEOCS factors in the Individual domain, we converted an
installation’s raw score to a percentile by comparing each installation’s factor score to the factor scores of
all other installations with DEOCS data (n = 931). We assigned percentiles for the six Protective factors
in ascending order and assigned percentiles for the Risk Factor Stress in descending order. (i.e., Higher

6

CUI

CUI
Chief Digital & Artificial Intelligence Office

percentiles were associated with more resilience and less risk). We then averaged the seven factor
percentiles to create a DEOCS Workplace domain composite score for each installation.
The second component of the Workplace domain was comprised of two survey items from the WGRA:
General Reporting Climate and Trust in the Military System. For each item, we converted an installation’s
raw score to a percentile by comparing each installation’s scale score to the scale scores of all other
installations with WGRA data (n = 398). We then averaged the two percentile scores to create a WGRA
climate composite score for each installation.
Leadership: The Leadership domain was comprised of seven DEOCS factors across three subdomains:
Immediate Supervisor (Toxic Leadership and Leadership Support, n = 931), Commander
(Transformational Leadership and Passive Leadership, n = 931), and Non-Commissioned Officer
(Transformational Leadership, Passive Leadership, Toxic Leadership, n = 829). vii Like the DEOCS factors
in the other domains, we converted each installation’s raw score to a percentile by comparing their factor
score to the factor scores of all other installations with DEOCS data. Again, we assigned percentiles for
the Protective Factors in ascending order and for the Risk Factors in descending order. We then
averaged the percentile scores in the Immediate Supervisor, Commander, and Non-Commissioned
Officer subdomains and averaged across subdomains to create a DEOCS Leadership domain composite
score for each installation. viii
Installation: The Installation domain was the most complex in terms of breadth and variety of data
sources. First, the domain included the estimated male sexual assault and sexual harassment rates from
the WGRA.9 For both sexual assault and sexual harassment, we converted each installation’s rate to a
percentile by comparing their rates to the rates of all other installations with WGRA data (n = 398). Given
that these rates measure negative constructs, we assigned risk percentiles for the two rates in ascending
order (i.e., higher rates were coded using lower percentile scores). We then averaged the two percentiles
to create a WGRA male sexual assault and harassment composite score for each installation.
Similarly, the domain includes the estimated female sexual assault and sexual harassment rates from the
WGRA (n = 201). ix Like male sexual assault and sexual harassment, we converted female rates for each
installation into risk percentiles in ascending order. We then averaged the two percentiles to create a
WGRA female sexual assault and harassment composite score for each installation.
Third, we estimated installation sexual assault reporting rate by comparing the total number of estimated
sexual assaults at an installation (from the 2018, 2021, and 2023 WGRA) to the total number of reported
sexual assaults from DSAID (FY 2018, CY 2021, and FY 2023, respectively). Specifically, we defined
reporting rate as DSAID reported sexual assaults divided by WGRA total estimated sexual assaults. x We
then converted each installation’s sexual assault reporting rate to a percentile by comparing their
reporting rate to the reporting rates of all other installations with both male and female WGRA sexual
assault data and DSAID data (n = 179).
Fourth, we ranked installations according to their FAP per capita rate of domestic abuse incidents and
child abuse incidents. To convert raw counts to per capita rates, we used installation size estimates
derived from the total rostered individuals at an installation per the DEOCS aligned with the FAP year of
interest. xi We then divided both the count of domestic abuse incidents and child abuse incidents by this
estimate of installation size and multiplied by 1,000 to standardize each of these rates per 1,000 Service
members. Lastly, we converted each installation’s per capita rate of domestic and child abuse incidents
to a percentile by comparing their rates to the rates of all other installations with FAP data (n = 181). We
then averaged the two percentiles to create a FAP domestic and child abuse composite score for each
installation.

7

CUI

CUI
Chief Digital & Artificial Intelligence Office

Lastly, we classified installations into suicide risk groups. The data sources and methodology used in this
classification differ between the Risk (2022) and Resilience (March 2024 and August 2023) versions of
the index. Both are described below.
OSIE Risk Index (2022) Suicide Risk Groups: (n = 1,165). xii Like FAP domestic and child abuse
incidents, we converted raw suicide counts into a per capita rate, by dividing suicides by the
estimated active-duty population at each installation and then multiplied the rate by 1,000 Service
members. However, rather than converting these rates to percentiles, we categorized
installations based on a predetermined matrix (see Table 3) given that suicides are low incidence
events. Essentially, we wanted to consider both the raw suicide count and per capita rate when
scoring installations on this risk measure.
Table 3: OSIE Risk Index Suicide Scoring Matrix

Based on data from 2020 Q1 through 2022 Q1

Suicide Rate per 1,000
< 0.25

.25 - .49

.50 - .74

0
1
Suicide Count

2
3
4-5
6 - 10
>10

0

.75 - 99

1 - 1.49

>1.5

20
20

40

60

20

40

40

40

60

60

20

40

40

60

80

80

20

40

60

80

80

100

20

40

60

80

100

100

OSIE Resilience Index Suicide Risk Groups: (March 2024, n=1,055; August 2024, n = 931)
The March 2024 version of the OSIE Resilience Index utilized both raw suicide counts spanning
2016 through 2022 and installation-level estimates of suicidal ideation from the 2020 Status of
Forces Survey of Active-Duty Members (SOFA). The August 2024 Resilience Index uses suicide
counts spanning 2016 through June of 2024 and 2022 SOFA estimates of suicidal ideation. Raw
suicide counts were converted to a suicide score in a similar method as the 2022 methodology,
incorporating both suicide counts and per capita rates. The matrix included more categories than
the previous version due to the larger pool of data (making both suicide counts and rates have a
wider range) and in order to introduce more variation across installations (with 11 categories
rather than 6) (see Table 4).

8

CUI

CUI
Chief Digital & Artificial Intelligence Office

Table 4: OSIE Resilience Index Suicide Scoring Matrix
Based on data from 2016 Q1 through 2022 Q4

Suicide
Count

0
1
2
3
4-5
6-7
8-10
11-14
15-19
20-30
31-70
71+ (max
104)

0
0
0
0
0
0
0
0
0
0
0
0

1-4
0
10
10
10
10
20
30
40
50
60
70

5-9
0
10
10
10
20
30
40
50
60
70
80

0

80

80

Suicide Rate per 100,000
10-16 17-24 25-34 35-49 50-69 70-99
0
0
0
0
0
0
10
10
20
30
40
50
10
20
30
40
50
60
20
30
40
50
60
70
30
40
50
60
70
80
40
50
60
70
80
80
50
60
70
80
80
90
60
70
80
80
90
90
70
80
80
90
90
100
80
80
90
90
100
100
80
90
90
100
100
100
90

90

100

100

100

100

100+
0
60
70
80
80
90
90
100
100
100
100
100

The Resilience Index method also incorporated suicidal ideation estimates at the installation
level. Installations were categorized into four groups of low, medium-low, medium-high, and high
levels of suicidal ideation, resulting in a score of 25, 50, 75, or 100, respectively.
To combine these two scores in the final suicide risk groups, we computed a weighted average of
the scores, with the suicide count-derived score holding 80% weight and the SOFA-derived score
a 20% weight. In cases where an installation had no SOFA data (n = 540), the suicide countderived score was used alone.
Community: Lastly, the Community domain consisted of U.S. health data compiled and maintained by
the CHR&R. The CHR&R classifies measures into Health Outcomes and Health Factors. Health
Outcomes are comprised of 5 measures: years of potential life lost before age 75 per 100,000 population,
the percentage of adults reporting fair or poor health, average number of physically unhealthy days,
average number of mentally unhealthy days, and percentage of live births with low birthweight. Health
Factors are comprised of 28 measures from four subdomains (health behaviors, clinical care, social and
economic factors, and physical environment), including: the percentage of adults who are current
smokers, the ratio of population to primary care physicians, and the percentage of the workforce that
drives alone to work. (For the full list of factors, see Table 5.)
We converted each CHR&R measure into a county-level percentile by comparing each U.S. County with
an associated installation to all U.S. Counties with installations. xiii This meant that if multiple installations
were in the same county, every installation was assigned the same county percentile for each CHR&R
measure. For example, both Creech and Nellis Air Force Bases are in Clark County, Nevada. Therefore,
both installations were linked to the same percentile scores across all the CHR&R measures. In total, we
linked 909 installations and ships to 463 unique U.S. counties. We assigned resilience percentiles for

9

CUI

CUI
Chief Digital & Artificial Intelligence Office

positive measures (e.g., percentage of adults ages 25 and over with a high school diploma or equivalent)
in ascending order and negative measures (e.g., percentage of adults reporting fair or poor health) in
descending order. xiv Once every CHR&R measure had been converted into a percentile, we computed a
weighted average of measures for each installation comprising both the Health Outcomes and Health
Factors domains in accordance with CHR&R’s original weighting scheme (see Table 5). We then
calculated a percentile of those weighted averages to create the two County Health composite scores.

10

CUI

CUI
Chief Digital & Artificial Intelligence Office

CUI

Table 5: County Health Measures and Weighting Scheme
Domain
Health
Outcomes
Total

Measure

Data Source (Year)

Years of potential life lost before age 75 per 100,000 population (age-adjusted).
Percentage of adults reporting fair or poor health (age-adjusted).
Average number of physically unhealthy days reported in past 30 days (age-adjusted).
Average number of mentally unhealthy days reported in past 30 days (age-adjusted).
Percentage of live births with low birthweight (< 2,500 grams).

National Center for Health Statistics - Mortality Files (2018-2020)
Behavioral Risk Factor Surveillance System (2019)
Behavioral Risk Factor Surveillance System (2019)
Behavioral Risk Factor Surveillance System (2019)
National Center for Health Statistics - Natality files (2014-2020)

Weight

Behavioral Risk Factor Surveillance System (2019)

50.0%
10.0%
10.0%
10.0%
20.0%
100.0%
10.5%

Behavioral Risk Factor Surveillance System (2019)

5.3%

USDA Food Environment Atlas, Map the Meal Gap from Feeding America (2019)
Behavioral Risk Factor Surveillance System (2019)
Business Analyst, ESRI, YMCA & US Census Tigerline Files (2010 & 2021)
Behavioral Risk Factor Surveillance System (2019)
Fatality Analysis Reporting System (2016-2020)
National Center for Health Statistics - Natality files (2014-2020)
Small Area Health Insurance Estimates (2019)
Area Health Resource File/American Medical Association (2019)
Area Health Resource File/National Provider Identification file (2020)
CMS, National Provider Identification (2021)
Mapping Medicare Disparities Tool (2019)
Mapping Medicare Disparities Tool (2019)
Mapping Medicare Disparities Tool (2019)
American Community Survey, 5-year estimates (2016-2020)
American Community Survey, 5-year estimates (2016-2020)
Bureau of Labor Statistics (2020)
Small Area Income and Poverty Estimates (2020)
American Community Survey, 5-year estimates (2016-2020)
American Community Survey, 5-year estimates (2016-2020)
County Business Patterns (2019)
National Center for Health Statistics - Mortality Files (2016-2020)

2.1%
2.1%
1.1%
2.6%
2.6%
2.6%
5.3%
3.2%
1.1%
1.1%
5.3%
2.6%
2.6%
5.3%
5.3%
10.5%
7.9%
2.6%
2.6%
2.6%
2.6%

Average daily density of fine particulate matter in micrograms per cubic meter (PM2.5).

Environmental Public Health Tracking Network (2018)

2.6%

Indicator of the presence of health-related drinking water violations.
Percentage of households with at least 1 of 4 housing problems: overcrowding, high housing costs, lack of
kitchen facilities, or lack of plumbing facilities.

Safe Drinking Water Information System (2020)

2.6%

Comprehensive Housing Affordability Strategy (CHAS) data (2014-2018)

2.1%

Percentage of the workforce that drives alone to work.

American Community Survey, 5-year estimates (2016-2020)

2.1%

Among workers who commute in their car alone, the percentage that commute more than 30 minutes.

American Community Survey, 5-year estimates (2016-2020)

Percentage of adults who are current smokers (age-adjusted).
Percentage of the adult population (age 18 and older) that reports a body mass index (BMI) greater than or
equal to 30 kg/m2 (age-adjusted).
Index of factors that contribute to a healthy food environment, from 0 (worst) to 10 (best).*
Percentage of adults age 18 and over reporting no leisure-time physical activity (age-adjusted).
Percentage of population with adequate access to locations for physical activity.*
Percentage of adults reporting binge or heavy drinking (age-adjusted).
Percentage of driving deaths with alcohol involvement.
Number of births per 1,000 female population ages 15-19.
Percentage of population under age 65 without health insurance.
Ratio of population to primary care physicians.
Ratio of population to dentists.
Ratio of population to mental health providers.
Rate of hospital stays for ambulatory-care sensitive conditions per 100,000 Medicare enrollees.*
Percentage of female Medicare enrollees ages 65-74 that received an annual mammography screening.*
Health Factors Percentage of fee-for-service (FFS) Medicare enrollees that had an annual flu vaccination.*
Percentage of adults ages 25 and over with a high school diploma or equivalent.*
Percentage of adults ages 25-44 with some post-secondary education.*
Percentage of population ages 16 and older unemployed but seeking work.
Percentage of people under age 18 in poverty.
Ratio of household income at the 80th percentile to income at the 20th percentile.
Percentage of children that live in a household headed by a single parent.
Number of membership associations per 10,000 population.*
Number of deaths due to injury per 100,000 population.

Total

1.1%
100.0%

Note: N = 463 U.S. counties; “Number of newly diagnosed chlamydia cases per 100,000 population” and “Number of reported violent crime offenses per 100,000 population” omitted from OSIE Resilience
Index given these measures are not comparable across state lines. The symbol * indicates resilience percentiles assigned in ascending order (i.e., higher levels on a measure indicate higher resilience).
Source: https://www.countyhealthrankings.org/explore-health-rankings/county-health-rankings-measures

11

CUI

CUI
Chief Digital & Artificial Intelligence Office

CUI

OSIE Resilience Index: Weighting Schemes and Stability Analysis
We employed three weighting schemes to rank installations in terms of resilience: Original Weights, Data
Coverage, and Domains Equally Weighted (see Table 2). Utilizing three separate weighting schemes
allowed us to be more confident in the reliability of the rankings. Importantly, for each weighting scheme,
if a data source (or sources) were missing, all coefficients were removed from the weighting
formula. Using this approach, installations were not penalized for having missing data. Instead, the
weights of the other data sources increased proportionately to compensate for the missingness of the
other data sources.
Original Weights. The Original Weights scheme assigned percentages to domains based largely on
socio-ecological theory. Moreover, Original Weights ensured every data source contributed to the overall
ranking as each subdomain was assigned a weight between 5% and 15%. In this weighting scheme,
DEOCS factors comprised 45% and suicide risk group made up 15% of the OSIE Resilience Index.
Data Coverage. The Data Coverage scheme prioritized data sources for which more installations had
data. Because there were only 386 installations with estimated female sexual assault rates, sexual
harassment rates, and sexual assault reporting rates, these measures were omitted from the OSIE
Resilience Index. Likewise, domestic abuse and child abuse rates were removed from the OSIE
Resilience Index given that only 181 installations had FAP data. As a result, in this weighting scheme
DEOCS factors comprised 60% of the OSIE Resilience Index to compensate for the absence of the
omitted data sources.
Domains Equally Weighted. The Domains Equally Weighted scheme followed an atheoretical approach
by assigning a weight of 20% to each of the five domains. Therefore, this domain-agnostic weighting
scheme de-emphasized data sources in the Installation domain (e.g., estimated male sexual assault rates
and sexual harassment rates comprised only 3% of the OSIE Resilience Index and suicide risk group
accounted for only 5%). Conversely, greater prominence was given to the Community domain by
increasing the overall weight of the CHR&R from 10% to 20%.
Stability Analysis, August 2024 OSIE Resilience Index. We performed a stability analysis to examine
the potential influence of the weighting schemes on the installation rankings for the August 2024 OSIE
Resilience Index. First, we categorized installations into quintiles in terms of resilience and then analyzed
the frequency with which installations’ resilience quintile changed depending on the weighted scheme
employed. This section details results of the stability analysis for the August 2024 Resilience Index. As
shown at the top of Table 6, 100% of the 931 total installations exhibited no change when their resilience
index quintiles were computed using Original Weights versus Data Coverage.
When resilience quintiles produced using Original Weights were compared against those using Data
Coverage, a smaller but still high percentage of installations exhibited no change (94%). Similarly, 806 of
the 847 total installations (87%) exhibited no change when their resilience index quintiles were computed
with Domains Equally Weighted compared to using Original Weights. Across all three weighting schemes,
almost 99% of installations were sorted into the same or an adjacent resilience quintile, lending credence
to the consistency of the OSIE Resilience Index.

12

CUI

CUI
Chief Digital & Artificial Intelligence Office

CUI

Table 6: Stability of August 2024 OSIE Resilience Index Rankings by Weighting Schemes
Original Weights vs.
Data Coverage

Original Weights vs.
Domains Equally Weighted

Data Coverage vs.
Domains Equally Weighted

931 (100%)

872 (94%)

806 (87%)

+/- 1 Quintile Change

-

55 (6%)

117 (13%)

+/- 2 Quintile Change

-

4 (0%)

8 (1%)

+/- 3 Quintile Change

-

-

-

931 (100%)

931 (100%)

931 (100%)

All Installations
No Change

Total

Notes: Percentages may not sum to 100% due to rounding.

If you do not currently have access to OPA.mil, there is a Request Access procedure accessible from the top-right
corner of the home page. If you need help accessing the site of establishing an account, please email
ContactOPA@mail.mil
ii Records of suicide do not include civilian population.
iii Records of domestic abuse and child abuse and neglect do not include military population on ships.
iv We define an installation as: A facility or municipality housing the primary quarters, correspondence, and body of
military service units at the lowest echelon available such that each location be geographically unique and reasonably
encompassing all its associated units.
v The Centers for Disease Control and Prevention. (2022) “The Social-Ecological Model: A Framework for
Prevention.” About Violence Prevention | Violence Prevention | CDC
vi DEOCS raw factor scores were originally computed for each installation in two steps. First, we converted the
proportion of responses in each category to an average unit score for each factor. Specifically, each negative
category for a protective factor was assigned a value of -1 (e.g., non-cohesive organization), each neutral category
was assigned a value of 0 (e.g., neutral), and each positive category was assigned a value of 1 (e.g., cohesive
organization). For risk factor scores, we use the opposite coding structure: each negative category was assigned a
value of 1 (e.g., frequent binge drinking), each neutral category was assigned a value of 0 (e.g., some binge
drinking), and each positive category was assigned a value of -1 (e.g., no binge drinking). For factors with only two
response categories, each positive category was assigned a value of 1 (e.g., no presence of racially harassing
behaviors) and each negative category was assigned a value of -1 (e.g., presence of racially harassing
behaviors). Second, we aggregated all unit-level individual factor scores to the installation-level according to the
number of individuals rostered for each unit. This process ensures that the responses of each survey respondent in
an installation (regardless of unit) were assigned equal weight in the overall factor score of the installation.
vii 101 installations were comprised only of units without non-commissioned officers, and therefore, this subdomain
was omitted from their Index.
viii Two DEOCS factors were excluded from the Risk Index. Safe Storage of Lethal Means was omitted because
descriptive and exploratory factor analysis revealed this factor to behave differently than all the other DEOCS factors.
Additionally, Workplace Hostility was excluded given that this factor was rescored part way through DEOCS
administration.
ix The number of installations with estimated female sexual assault and sexual harassment rates was significantly
lower than the number with male rates (n = 228 vs. n = 461 for 2023). This is because installations with fewer than
100 Service members of a given gender were excluded from the WGRA Contextual Analysis.
x Total sexual assaults at an installation were estimated by summing the sexual assault rate for men multiplied by the
total number of male service members and the sexual assault rate for women multiplied by the total number of female
i

13

CUI

CUI
Chief Digital & Artificial Intelligence Office

CUI

service members. Therefore, to ensure an accurate estimate of this total, sexual assault non-reporting rates were
only calculated for installations where both male and female sexual assault rates were available (n = 201). In
addition, any installation with a sexual assault non-reporting rate less than 0% (i.e., DSAID reported sexual assaults
greater than WGRA estimated total sexual assaults) were recoded as 0%.
xi For the 2022 OSIE Risk Index, installation size estimates were derived using a different method. We estimated the
active-duty population at each installation by averaging the total DEOCS roster count of non-civilians for all units
mapped to that installation and the 2018 ADMF count of installation size (a variable we obtained from the 2018
WGRA Contextual Analysis).
xii Unlike the other data sources, we assumed that installations without data were the result of no suicides and not
data missingness. Thus, all 1,165 installations were classified into a suicide risk group, ranging from 0 to 100.
xiii We assigned ships the U.S. County of their homeport.
xiv The CHR&R measure presence of health-related drinking water violations was binary. Therefore, we coded “Yes”
as 50 and “No” as 100.

14

CUI


File Typeapplication/pdf
AuthorDockterman, Daniel [USA]
File Modified2024-12-10
File Created2024-09-09

© 2025 OMB.report | Privacy Policy