Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Reliability Engineering and System Safety

journal homepage: www.elsevier.com/locate/ress

Static and dynamic metrics of economic resilience for interdependent

infrastructure and industry sectors
Raghav Pant a, Kash Barker b,n, Christopher W. Zobel c
a
Environmental Change Institute, Centre for the Environment, University of Oxford, Oxford, UK
b
School of Industrial and Systems Engineering, University of Oklahoma, Norman, OK 73019, USA
c
Department of Business Information Technology, Virginia Polytechnic Institute and State University, VA, USA

art ic l e i nf o a b s t r a c t

Article history: Infrastructures are needed for maintaining functionality and stability of society, while being put under
Received 21 January 2013 substantial stresses from natural or man-made shocks. Since avoiding shock is impossible, increased
Received in revised form focus is given to infrastructure resilience, which denotes the ability to recover and operate under new
21 July 2013
stable regimes. This paper addresses the problem of estimating, quantifying and planning for economic
Accepted 18 September 2013
resilience of interdependent infrastructures, where interconnectedness adds to problem complexity. The
risk-based economic input–output model enterprise, a useful tool for measuring the cascading effects of
Keywords: interdependent failures, is employed to introduce a framework for economic resilience estimation. We
Resilience propose static and dynamic measures for resilience that confirm to well-known resilience concepts of
Interdependent infrastructures
robustness, rapidity, redundancy, and resourcefulness. The quantitative metrics proposed here (static
Inoperability
resilience metric, time averaged level of operability, maximum loss of functionality, time to recovery)
guide a preparedness decision making framework to promote interdependent economic resilience
estimation. Using the metrics we introduce new multi-dimensional resilience functions that allow
multiple resource allocation scenarios. Through an example problem we demonstrate the usefulness of
these functions in guiding resource planning for building resilience.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction widespread social and economic impacts. Recent adverse impacts of

natural disasters (e.g., Hurricanes Katrina and Rita in 2005, the
Modern technological developments have been aimed at increas- Japanese earthquake and tsunami in 2011, Hurricane Sandy in
ing and improving the interactions among multiple infrastructure 2012) and man-made incidents (e.g., the Deepwater Horizon oil spill
systems. This has led to realizations that infrastructures no longer in 2010) underscore the vulnerabilities associated with infrastruc-
behave as independent and isolated systems, but are rather part of a tures, industries, and their associated workforce. Increased connec-
larger system comprising interdependent industrial sectors. In the tivity of such systems has meant that direct impacts resulting from
United States (U.S.) this is reflected in several recent policy state- these kinds of disruptive events often lead to broader indirect
ments that highlight the “open, technologically sophisticated, highly impacts [36].
interconnected, and complex” nature of infrastructures and industry Owing to the large-scale and complex nature of extreme disrup-
sectors spanning “important aspects of the U.S. government, econ- tive events and infrastructure responses, there is an increased under-
omy, and society” [9]. Many countries around the world share similar standing that it is not possible to protect every potential target against
views. The European Union (E.U.), for example, identifies the impor- every conceivable attack and eliminate all vulnerabilities [12]. From a
tance of the system of infrastructures and industry sectors as security point of view there is a paradigm shift from emphasis on
“physical and information technology facilities, networks, services protection and prevention towards preparedness and response. Accept-
and assets… that impact on the health, safety, security and economic ing that some widespread hazards cannot be prevented, the over-
well-being of citizens and effective functioning of governments in arching concept of system resilience includes preparedness measures
E.U. countries” [26]. to reduce their resulting direct and indirect impacts and to recover in
As large-scale systems are ubiquitous, they can be rendered a timely manner.
inoperable by external shocks and disruptive events, resulting in The ultimate usefulness of understanding large-scale interdepen-
dent system impacts and recovery, particularly from the standpoint of
a preparedness-focused decision maker, is as a descriptor not just of
n
Corresponding author. Tel.: þ 1 405 325 3721; fax: þ 405 325 7555. physical damage, but also of business economic interruption [38,34].
E-mail address: kashbarker@ou.edu (K. Barker). That is, at the heart of understanding and planning for disruptive

0951-8320/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ress.2013.09.007

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
2 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎

events is the quantification of (i) dollars of losses, and (ii) extent of and measures of system performance and recovery speeds as available
duration of system inoperability. Infrastructure recovery planning can investment options might be limited. In light of the above
proceed from such understanding. This is the essence of under- observations, the current literature lacks a quantitative resilience
standing and modeling economic resilience of interdependent infra- metric that accounts for interdependence among multiple infra-
structures and industries, which is the primary focus of this paper. structures and industries. With this work, we address those
In its broadest definition, as specified by The Infrastructure concerns by developing metrics for effectively quantifying inter-
Security Partnership (TISP) [37], a resilient infrastructure or dependent economic resilience. Our unique contributions in this
industry sector would “prepare for, prevent, protect against, paper include: (i) the integration of a mathematical construct of
respond or mitigate any anticipated or unexpected significant resilience with a data-driven interdependent damage assessment
threat or event” and “rapidly recover and reconstitute critical model, the inoperability input–output model (IIM) [36,18], (ii)
assets, operations, and services with minimum damage and resilience metrics derived from static and dynamic representations
disruption.” While different viewpoints about resilience exist of the interdependency model, (iii) analytical approaches to
within the research community [32,10,1], there is consensus that represent the resilience metrics as multi-dimensional resilience
resilience estimation is vital for risk-based decision making [16]. functions, and (iv) the use of resilience functions for fitting
For our purposes, we define economic resilience here as an ability different planning scenarios to yield similar resilience measures,
exhibited by a system that allows it to recover economic produc- thereby providing multiple perspectives (supported by graphical
tivity after a disruptive event, in a desired time and with an depictions) of resilience for decision support.
acceptable cost, noting that resilience is planned for in advance of The methodological background for the static and dynamic
a disruptive event through preparedness investments and inoperability input–output models is provided in Section 2, where
activities. we also provide a sample of the type of interdependent economic
In existing literature, economic resilience has been defined as systems being considered. Measures of static and dynamic resi-
the “inherent ability and adaptive response that enables firms and lience, extended from Pant and Barker [29] are introduced in
regions to avoid maximum potential losses” [34]. Economic Section 3. Section 4 illustrates, with a small data-driven case study,
resilience has primarily been studied in the context of seismic how these measures can be interpreted and used for constructing
response and recovery [38,7], community behavior [8], and dis- multi-dimensional contour plots that provide decision-making
aster hazard analysis [31], among other contexts [33]. Generally, guidelines to inform recovery options for interdependent sector
resilience is measured in terms of (i) the amount by which a resilience. Concluding remarks and future work are provided in
system is able to avoid maximum impact (i.e., static resilience [31], Section 5.
robustness [22], the opposite of vulnerability [5,6]), and (ii) the
speed at which the system recovers from a disruption (i.e.,
dynamic resilience [31], rapidity [43], recoverability [5,6]). A report 2. Models of interdependent inoperability
by the Joint Committee on Structural Safety [15] suggests that
vulnerability describes direct consequences of a disruption, while 2.1. Macroeconomic interdependency model
robustness describes broader indirect consequences, remaining
closely tied to a structural risk context. In recent work, Vugrin Interdependent economic resilience estimation first requires an
et al. [40] have developed an economic resilience framework for interdependent disruption evaluation model. Here we use models
measuring the targeted economic response of infrastructures. derived from the economic input–output model [17], which has been
To measure resilience of infrastructures and industry sectors, widely accepted as a useful tool for analyzing the interdependent
the approach adopted in this study centers around the integration connections among industry sectors. The economic input–output
of engineering resilience concepts (TISP approach) with the model explains the supply and demand balance at equilibrium for
properties interdependent macroeconomic systems. In line with an economy consisting of interacting infrastructures or sectors. For an
the TISP definition, for resilience quantification a model should n sector economy, this equilibrium is expressed through the equation
have parameters that depict (i) the ability to resist the initial x¼Axþc, where x is the n  1 vector of industry outputs, c is the
adverse effects of a disruptive event, and (ii) the ability to return to n  1 vector of exogenous demands, and A is the n  n interdepen-
an appropriate operability following the disruption. The resilience dency matrix that quantifies the proportions of inter-industry com-
triangle model suggested by Bruneau et al. [7] confirms to the TISP merce and interdependency. The wide use and popularity of
resilience approach, and has been widely used in engineering Leontief's model is evidenced from the availability of nation-wide
resilience modeling. Zobel [43,44] has addressed some of the economic input–output tables across several countries [42,13], includ-
limitations of the original resilience triangle approach and pro- ing annual data collection efforts in the U.S. by the Bureau of
vided a more quantitative treatment of the model for individual Economic Analysis (BEA).
system resilience quantification. Network-based models have been Alluded to previously, the input–output model is built for and
used to quantify resilience and other characteristics for electric applied to large-scale macroeconomic systems operating at regio-
power systems [27], communications systems [39], and rail nal, national, or international levels. In the U.S. the BEA releases
systems [14], but they tend to be infrastructure or network different input–output (I–O) data including annual I–O tables
specific. Reed et al. [30] have combined, for power delivery aggregated to 15 sectors or disaggregated to 65 sectors and five-
systems, the infrastructure characteristics with interdependent year benchmark I–O tables covering over 500 sectors that can be
economic modeling concepts to measure resilience of networked aggregated to the 65 or 15 sector levels. Table 1 shows a sample of
infrastructures. the A matrix for 2011 national level input–output accounts,
Even though the above approaches are suitable for measuring representing both infrastructure systems and industry sectors in
resilience of multiple systems, there are certain aspects of inter- the U.S. From this data we can derive the mapping of strength of
dependent system behaviors that are not addressed in current interdependent coupling of a particular infrastructure (e.g., Trans-
literature. First, in most studies, resilience is measured by a single portation) or industry (e.g., Agriculture) sector with other such
global metric based on the Bruneau et al. [7] resilience triangle, sectors. This is also illustrated in Fig. 1, where the Transportation
which does not capture the multi-dimensional aspect of resilience sector's interdependence with other sectors is highlighted in terms
across multiple interdependent systems. Second, for interdepen- of its demand towards (aTj ) and the supply it needs (aiT ) from
dent economic systems, we need to consider tradeoffs between other sectors. From the values we can infer, for example, that

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

Table 1
Sample A matrix from the BEA 2011 input–output accounts.

IOCode Commodities/industries 11 22 23 48TW

Name Agriculture Utilities Construction Transportation

11 Agriculture 0.1949842 0.0000011 0.0010711 0.0000444

21 Mining 0.0027765 0.1331650 0.0083818 0.0014131
22 Utilities 0.0154424 0.0002903 0.0032619 0.0028235
23 Construction 0.0044906 0.0107309 0.0007315 0.0075727
31G Manufacturing 0.2005193 0.0080531 0.2442287 0.1873453
42 Wholesale trade 0.0475103 0.0010087 0.0232085 0.0135453
44RT Retail trade 0.0018242 0.0000471 0.0359828 0.0047366
48TW Transportation and warehousing 0.0215651 0.0341889 0.0134875 0.0862129
51 Information 0.0006798 0.0008374 0.0060602 0.0049617
FIRE Finance, real estate 0.0782274 0.0066511 0.0300241 0.0548634
PROF Professional services 0.0103202 0.0129371 0.0812154 0.0614684
6 Educational services 0.0021031 0.0000422 0.0000136 0.0000766
7 Arts, entertainment 0.0010167 0.0029302 0.0026978 0.0082843
81 Other services 0.0019078 0.0004941 0.0130497 0.0053673
G Government 0.0002524 0.0001314 0.0000283 0.0105593

Inoperability for sector i, qi, refers to the inability of the

industry to perform its intended functions. The inoperability of a
system lies between 0 and 1, where qi ¼0 (x~ i ¼ x^ i ) is a measure of a
perfectly operable industry i, and qi ¼ 1 (x~ i ¼0) is a measure of
complete failure of industry i. Demand perturbation for sector i, ci ,
which refers to the change in final demand for the output of sector
i due to disruptive events, also lies between 0 and 1. Such demand
perturbation can occur due to the inability of the producing sector
to meet the demands of the final consumers when there is a
failure in the system.
From the metrics developed in Eqs. (1) and (2), the IIM in
Eq. (3) is derived from the economic input–output model by
subtracting the perturbed economic equilibrium from the as-
planned equilibrium and normalizing with the as-planned output
^ thus essentially maintaining a form similar to the Leontief
x,
economic input–output model. The normalized interdependency
matrix A of size n  n is a modified version of the original Leontief
A matrix describing the extent of economic interdependence
among a set of infrastructure and industry sectors. Central to this
development is the assumption that the interdependency struc-
Fig. 1. Sample relationship among the Transportation sector and others.
ture of the economy, measured by the A matrix, does not change
when demands and resulting outputs change. The A calculated
from the sample A matrix shown in Table 1 can suggest how a
Transportation is more dependent upon supply from Manufactur- disruption in the transportation infrastructure affecting its pro-
ing compared to the demand Manufacturing places on Transporta- ductivity will economically impact other sectors.
tion. Hence any change in Manufacturing output will have a higher        
impact on Transportation compared to the other-way around. ^  1 ðx^  xÞ
diagðxÞ ~ ¼ diagðxÞ^  1 A diagðxÞ
^ diagðxÞ ^  1 ðx^  xÞ
~
 

þ diagðxÞ^ c^  c~ ) q ¼ A q þ c ð3Þ
2.2. Risk-based interdependency model
A natural result from Eq. (3) is that widespread interdependent
inoperability is driven by perturbations in demand in one or more
An extension of the economic input–output model is the  1 
sectors, as q ¼ I  A c .
inoperability input–output model (IIM) [36], which models how
Certain economic impact analyses require modeling the
inoperability propagates throughout a set of interconnected infra-
dynamic interdependnent effects between an engineering and
structure and industry sectors. Inoperability quantifies the propor-
economic perspective, and representing the different time scales
tional extent to which an infrastructure or industry sector is not
over which the actual disruptive events take place and direct and
functioning in an as-planned manner. For n interdependent
indirect impacts are felt [24]. Modeling dynamic interdependent
sectors, x^ and c^ represent equilibrium as-planned output and final
inoperabilities allows us to examine initial impact effects and
demand levels, respectively, while x~ and c~ represent disrupted/
subsequent trajectories of multiple sectors as they progress
perturbed equilibrium levels for output and final demands,
towards recovery or further decline. This helps a preparedness
respectively. As such, the metrics of inoperability q and demand
planner to model baseline recovery and also estimate how
perturbation c are defined in Eqs. (1) and (2).
recovery trajectories can be improved through efficient prepared-

  1

q ¼ diag x^ x^  x~ ð1Þ ness strategies.
An extension to the IIM, the dynamic inoperability input–

  1
 output model (DIIM) [18] is a first-order differential equation that
c ¼ diag x^ c^  c~ ð2Þ
incorporates a rate constant into the static IIM structure, thereby

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
4 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎

modeling the propagation, then the inoperability trajectory, of 3.1. Static economic resilience
interdependent infrastructures and industries over time. The
n-sector DIIM is provided in Eq. (4). The general notion of static resilience, or “the ability of system

to maintain functionality when shocked”, Rose [32] is depicted in
_ ¼  KðI A ÞqðtÞ þ Kc ðtÞ; 8 t 40
qðtÞ ð4Þ Fig. 2 and given in Eq. (8). %ΔDY represents the actual percentage
change in the performance of the system following a disruptive
The analytic solution of the differential Eq. (4) is given in
event and %ΔDYmax represents the maximum percentage change
Eq. (5).
given the worst-case level of performance [33].
Z t

qðtÞ ¼ e  KðI  A Þt qð0Þ þ e  KðI  A

Þðt  zÞ
Kc ðzÞdz ð5Þ %ΔDY max %ΔDY
static economic resilience ¼ ð8Þ
0 %ΔDY max
Eq. (5) shows that the temporal evolution of sector inoper- Our definition of static resilience is adapted from the conven-
ability q(t) depends on: (i) the initial inoperability q(0) experi- tion defined by Rose [33]. If we define qmax as the maximum
enced after the disruption, (ii) additional demand perturbation possible inoperability that could be experienced after a disruptive
c ðt Þ at all times subsequent to the disruption, (iii) the interde- event, a measure of interdependent static resilience, similar in
pendent behavior of industries A , and (iv) the rate term in the concept to Eq. (8), is provided in Eq. (9). Static resilience for sector
i is referred to with convention Я i , adopting the Я notation of
S
DIIM represented with matrix K. When K is a diagonal matrix its
elements are given by Eq. (6). Whitson and Ramirez-Marquez [41] as R often refers to reliability.
 qi; max  qi
K ¼ diagðkii Þ; 0 rkii r 1 ð 8 i ¼ f1; 2; …; ngÞ ð6Þ Я Si ¼ ð9Þ
qi; max
Elements of K represent the speed at which sectors attain Calculated for sector i, Eq. (9) is an interdependent economic
particular responses to changes in production outputs and resilience metric, with qi capturing interdependence with Eq. (3).
demands. The bounds between 0 and 1 that are established for This demand-driven approach for inoperability, and hence resi-
the elements of K are determined by the consideration that the lience, measurement is consistent with the idea that static
matrix KðI  A Þ has positive eigenvalues. This guarantees that the economic resilience is a consequence of efficient utilization of
solution of Eq. (5) does not diverge, an important consideration for resources and not system repair [32]. Using the convention
modeling a trajectory that converges towards a stable state. Most h i  1

D ¼ dij ¼ I A , we have q ¼ D c , which means that inop-
applications of the DIIM (e.g., [18]) assume strictly diagonal
elements in K, the structure of which follows Eq. (7), or the erability in sector i can be represented by Eq. (10).
solution to Eq. (5) when no c ðtÞ is assumed. Value qi(0) is the n

initial inoperability experienced in sector i following a disruptive qi ¼ ∑ dij cj ð10Þ
j¼1
event, qi(Ti) is the desired inoperability state after recovery
(assumed to be small but nonzero), which requires Ti time periods As shown in Eq. (11), using the demand-driven model, the
to achieve (with T as the vector of all Ti), and aii is the diagonal static resilience for sector i can be written as a function of
entry in the interdependency matrix. maximum and expected demand perturbation levels, cj; max and
cj , respectively. Я i measures the proportional “savings” in inoper-
S
lnðqi ð0Þ=qi ðT i ÞÞ
kii ¼ T i ð1  aii Þ
ability when a priori planning can stave off the worst-case
inoperability outcome in favor of reduced inoperability.
    1
  1  
) K ¼ diagðln½qð0ÞÞ  diagðln½qðTÞÞ diagðTÞ diag I  A    
∑nj¼ 1 dij cj; max  ∑nj¼ 1 dij cj ∑j ¼ 1 dij cj; max  cj
n 

ð7Þ Яi ¼S
 ¼  ð11Þ
∑nj¼ 1 dij cj; max ∑nj¼ 1 dij cj; max
The IIM and its extensions have been deployed in several While calculating static resilience is mathematically straightfor-
domains, including interdependent impact analysis due to disrup- ward, the nuances lie in estimating the maximum and expected
tion in transportation systems [28,19], inventory decision making demand perturbation values and the influence of the interdependency
[20,3,4], and electric power recovery [21], among others [35]. The matrix. As we will examine later in Section 4, Я i can provide us with
S

aforementioned studies have used such models as demonstrative decision making insights for the functional relationship between cj
tools for recovery assessment, providing evidence that the DIIM is and cj; max .
suitable for use in a resilience estimation framework.

No disruption
level
3. Characterizing interdependent resilience
% DY
System output

Demonstrated in several of the works previously cited is the

need to address resilience from two aspects: (i) reducing the Expected degraded
impact of such disruptions, and (ii) improving the speed at which performance level
recovery occurs. As discussed, resilience is characterized by static
resilience for time-independent impact and damage avoidance % DY max
measures, and by dynamic resilience for impact and recovery
capability measures. Having established the IIM capabilities for Worst-case degraded
interdependent impact and recovery analysis, we concentrate on performance level
building such models for resilience estimation. In the subsequent
subsections we introduce metrics for static and dynamic resilience
quantification for interdependent infrastructures and industry
sectors. Fig. 2. Static economic resilience quantification (Rose [33]).

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

3.2. Dynamic economic resilience between 0 and 1, higher resilience is indicated by closeness of
elements toward 1. However, treating the K matrix as the only
Dynamic resilience adds the temporal perspective to recovery indicator of resilience results in an incomplete analysis. A illus-
properties desired from a system that is rebounding from a disruptive trative example shown in Fig. 4 illustrates that we can easily find
event. For an economic system, the “speed of the system towards examples where an industry sector having a higher value for the
achieving a desired state” Rose [32] becomes a factor in determining diagonal values of K leads to faster time to recovery but comes at a
how dynamically resilient the system is. In the dynamic interdepen- cost of a more significant initial impact. Fig. 4 shows that plot for
dency model construct, system behavior is generally indicative of a the time dependent operability of the sector having no initial
progression towards a stable state, and also the system inoperability perturbation (qi ð0Þ ¼ 0) for low (kii ¼ 0:2) and high (kii ¼ 1) design
responses are bounded. Hence, the dynamic model is a suitable K matrices. The resulting behaviors are seen because K also
candidate for analyzing and characterizing dynamic resilience in contributes towards interdependent coupling, which initially
interdependent infrastructure and industry sectors. magnifies the cascading impact of the disruption before recovery
The much used resilience triangle [7] framework, explained in kicks in. As such, there exists a trade-off between achieving faster
Fig. 3, captures technical, organizational, social, and economic aspects recovery at the cost of higher initial impact. Due to the inter-
of community behavior. It is suggested that resilience has four dependent nature of how inoperability propagates among inter-
dimensions, applicable to any general system behavior [7]: (i) connected infrastructures and industries, the choice of higher K
Robustness, the measure of the sector's ability to resist the initial can lead to adverse undesirable effects.
impact, (ii) Rapidity, the measure of the time required for the sector to To effectively use the dynamic inoperability model for resi-
attain recovery, (iii) Redundancy, the ability of the sector to substitute lience quantification multiple metrics are required in addition to
for lost production through inventories and other means, and (iv) the K matrix. As such, we propose three metrics to explain
Resourcefulness, the capacity enhancing capabilities of the sector to resilience through the dynamic interdependency model. It is
improve its performance. Henry and Ramirez-Marquez [11] also important that these metrics should be able to explain the
provide a similar discussion of the components of resilience. Quanti- standard properties expected in resilience behavior. Our metrics
tatively the resilience triangle measures system resilience in terms of conform to the principles outlined in previous resilience estima-
robustness and rapidity, which are achieved through incorporating tion modeling frameworks [7,11].
redundancy and resourcefulness into the system.
Although resilience, as quantified with the above approach, is 3.2.1. Time averaged level of operability
based on the four system dimensions, it ultimately measures one The overall level of functionality maintained by a system,
quantity encapsulating entire system behavior. Zobel [44] con- starting from the time of initiation of a disruptive event until
tends that a multi-dimensional representation of resilience is any time horizon, provides a good indication of its resilience to the
necessary, as a single metric could be misleading. Mathematically, disruptive event. From the dynamic interdependency model, the
the areas within two triangles can be the same indicating the same measure 1  qi(t) is the level of operability of industry i at each
amount of resilience, but one behavior could be due to high instance of time and can be used to quantify the overall sector
robustness and low rapidity while the other might have the functionality level. Starting at t ¼0, if the system is being mon-
opposite. Such considerations require multiple metrics in the itored for a significant amount of time t¼T (T denotes an upper
resilience estimation framework. limit of the time for which the system is monitored and effectively
By its very definition and structure, the dynamic interdepen- is chosen such that full system recovery is expected at time T),
dency model in Eq. (5) can be used to measure and quantify then overall system performance can be quantified in terms of the
resilience. The four dimensions of infrastructure resilience can be time averaged level of operability over the analysis period. This
explained with the model to develop a framework of multiple metric, referred to as Fi (or vector F for all sectors), is given in
metrics to measure, and ultimately plan for, economic resilience. Eq. (12).
In previous DIIM analyses, recovery is understood only in terms of Z Z
the K matrix. Lian and Haimes [18] refer to the coefficients of K as 1 T 1 T
Fi ¼ 1  qi ðt Þdt 3 F ¼ 1  qðt Þdt ð12Þ
industry resilience coefficients that influence the speed of sector T 0 T 0
recovery. For the diagonal-only K, with diagonal elements Zobel [44] used a similar approach to define a normalized
resilience metric from the area beneath the resilience triangle,
referred to as the predicted resilience of a system.
100
System Quality(X)

Loss of Resilience:
t1
X0 R = [100 X ( t )] dt
t0
Robustness

Rapidity

0
t0 t1
Time(t) Fig. 4. The dynamic response profiles of operability versus time for the sector with
no initial inoperability (q(0) ¼0) corresponding to two different K matrices (K1
Fig. 3. The resilience triangle depicting initial impact and recovery, adapted from indicates slower recovery and K2 indicates faster recovery). This helps in visualizing
Bruneau et al. [7]. the trade-off between initial impact and recovery time.

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
6 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎

We observe that 0 rFi r1, with Fi ¼0 occurring for a comple- performance of individual sectors as well as the overall intercon-
tely inoperable system (qi ðt Þ ¼ 1; 8 t A ½0; T ) and Fi ¼1 correspond- nected system.
ing to a fully functional system qi ðt Þ ¼ 0; 8 t A ½0; T ) for the duration
of the analysis period. Consideration should be given in deciding 3.3. Time to recovery
the time T over which the system functionality is being measured.
By its nature, the DIIM provides an exponentially decaying solu- Dynamic resilience is best understood in the context of the
tion for sector inoperability. After a certain time the model values speed of system recovery: the faster a system is able to recover
converge towards almost zero values. If T is large, the integral from a disruptive event, the more resilient it is defined to be. For
RT
solution of Fi in Eq. (12) will converge towards zero ( lim T1 0 qi an ideal resilient system, recovery implies a return to pre-
T-1
disruption levels of productivity and, thereafter, the capability to
ðt Þdt ¼ 0) and metric F i ðT Þ-1.
maintain functionality at such levels. In reality, systems might not
Fi can also be related to a metric of sector economic losses
RT be able to reorganize and recover to pre-disruption output levels
Q i ¼ 0 xi qi ðtÞdt, which is the integral of the industry sector i
due to the existence of permanent losses or different evaluation
inoperability over time T multiplied by xi which is the pre-
standards. Hence, recovery is best understood in terms of the
disruption output level for the sector. We can easily deduce that
capability of the system to achieve a stable condition, where
if qi ðtÞ ¼ 1 for all T then Q max ¼ xi T represents the maximum loss
productivity levels are higher than they were immediately after
the sector could have incurred. The functional relationships
the disruptive event. The assumptions of the dynamic interde-
among F i , Q i , and Q max is expressed in Eq. (13).
pendency model suggest that infrastructure and industry sectors
Q i Q max  Q i approach equilibrium conditions from different initial inoperabil-
Fi ¼ 1  ¼ ð13Þ
xi T Q max ity and demand perturbation conditions. As such, we can define a
Eq. (13) shows that Fi is a measure of the amount of loss the suitable resilience metric, called the time to recovery, to indicate
sector avoided as a fraction of potential maximum damage. This is sector recovery through the dynamic model. As the dynamic
consistent with the conventional static economic resilience defini- interdependency analysis suggests, the equilibrium levels of sector
tion [32] and the Я i metric proposed in Eq. (9), which is often inoperability are known, depending upon different dynamic con-
S

used as an overall performance measure even for dynamic system ditions. Thus, time to recovery is measured in terms of said
analysis. equilibrium level inoperabilities: if for industry i the equilibrium
inoperability is given as qei , then its time to recovery τi is defined
3.2.2. Maximum loss of sector functionality as the time when its inoperability qi ðt Þ is within an ε ( 5 1)
The immediate impact of a disruptive event on an infrastruc- neighborhood of qei , as shown in Eq. (15).

ture system is felt through the degradation of its capacity to τi ¼ t : t 4 0; qi ðt Þ  qei ðt Þ r ε 3 τ ¼ diag½τ1 ; τ2 ; …; τn  ð15Þ
produce output. If external demands for output are perturbed,
system operability is further eroded. Within the context of the Time to recovery is synonymous with the notion of rapidity
inoperability model, a resilient system is one that is capable of [32], with a slight difference in its role here. Rapidity is typically
maintaining a high level of initial degraded operability (1  qi ð0Þ) associated with improved dynamic resilience, because it is under-
following disruption, and continues maintaining high operabilities stood that recovery brings the system from its worst initial
(1  qi ðtÞ) at all times following the disruption. In reality, however, condition to a desired state, and that increased speed implies
there are events that are capable of causing systems to lose most more resilience. As shown in Fig. 4, however, there is a tradeoff
of their functionality, making it difficult for them to operate at between the time to recover and maximum inoperability, which
high levels of productivity. There is interest in understanding the shows that increased speed does not always lead to a better
worst effect a disruption can have on the infrastructures before performance throughout. Again, interdependence brings about
they can recover to better levels of performance. Since resilience is such behaviors.
associated with the capability of a system to bounce back or recoil The graphical representation of the three resilience metrics
from disruptions, getting a perspective of the lowest productivity generated from a general sector response to disruption is shown in
levels during recovery is needed to develop an understanding for Fig. 5, which is comparable to other engineering resilience
system resilience. In the dynamic interdependency model, it is research work [7,44,11]. Here the representation of a sector's
assumed that an industry sector is capable of recovering from any resilience is not just due to its own properties but also due to its
level of inoperability (o1) to an equilibrium condition. Hence, in interdependence with other sectors.
the model, inoperability reaches a maximum value before the
sector rebounds towards recovery. This maximum industry i
4. Decision support for interdependent static and dynamic
inoperability, called maximum loss of functionality, is quantified
resilience
in Eq. (14). It is clear from Eq. (14) that 0r qm i r1 since sector
inoperability lies on [0, 1].
Revisiting the criteria for measuring resilience, we have
 
qm m
i ¼ max qi ðt Þ 3 q ¼ max½qðt Þ ð14Þ demonstrated that the suggested static and dynamic resilience
tZ0 tZ0
metrics describe: (i) the overall impact effects (Я , F, qm ), and
S

1  qm
i is referred to as the robustness measure in the resilience (ii) recovery speeds (τ), of interdependent industry sectors. Since
triangle framework, which is slightly different than its interpreta- resilience is a pre-disruption measure (as defined in Section 1), the
tion in the result of Eq. (14). In the engineering-based interpreta- metrics are calculated from anticipated disruption and planning
tion, robustness implies the ability to resist the direct initial scenarios. Therefore, these metrics are more useful if they are used
impact and to avoid immediate damages [7,22], which would be as feedback measures to quantify the effectiveness of prepared-
measured (in DIIM terms) as 1 qi ð0Þ. However, due to the ness investments and activities. A sector decision maker would
interdependent nature of infrastructure and industry sectors, like to estimate appropriate resource allocation by setting targets
for acceptable levels of functionality (informed by Я i , F i , qm
S
some of which are initially impacted and some of which experi- i ) and
ence subsequent indirect impacts, qm i can manifest itself at a later the time to recovery (informed by τi ). In this Section we introduce
time as seen in Fig. 4. Ultimately, strengthening robustness, or multi-dimensional analytical functions of the resilience metrics
reducing qm i , requires similar planning strategies that enhance the that guide such decision making. These resilience functions not

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

here, βi is a measure of resource allocation and function f ðβi Þ A ½0; 1

represents the proportional extent to which a preparedness strategy
adopted by sector i reduces the maximum demand perturbation
impact. The effectiveness of f ðβi Þ is measured in terms of the new
resulting demand perturbation. If β i indicates the magnitude of effort
put into a planning strategy (investment, resource allocation), then
mathematically f ðβi Þ is considered to be a monotonically non-
increasing function of βi , as increased resources should result in
greater reduction of maximum impact ci; max . Examples of resources
allocated with βi could include activities that reduce the initial impact
of a disruption: system hardening, sensor technology to detect issues
early, and redundancy in sector-sustaining components. Substituting
Eq. (16) into Eq. (11), we calculate static resilience for sector i in
Eq. (17).
∑nj ¼ 1 dij cj; ð1  f ðβj ÞÞ
Я Si ¼ max

∑nj ¼ 1 dij cj;
!
max

n

dij cj; max   ð17Þ
¼ 1 ∑  f βj
Fig. 5. The three metrics of dynamic resilience for industry i, accounting for the j¼1 ∑nj¼ 1 dij cj; max
interdependent impact on industry i from other interconnected infrastructures and
industry inoperability. 
In the above equation, dij cj; max is a measure of the strength of
direct and indirect impact on sector i due to demand perturbation
only estimate one planning option but can also generate the same in sector j, which influences interdependent resilience. It has been
resilience measures over different planning perspectives. As such, normalized by the total impact felt across all sectors. The formula-
we are able to generate several resilience scenarios that can be tion of Eq. (17) can be modified as shown in Eq. (18), where
mapped to produce planning decision spaces. We demonstrate  
 
wij ¼ dij cj; max =∑nj¼ 1 dij cj; max is indicative of a weight exerted by
quantitative (and graphical) decision support schema for utilizing
the static and dynamic resilience metrics in informing resilience one sector's interdependency and demand perturbation on the
planning. resilience of another.
The subsequent subsections provide representations of the n  
decision spaces created by varying the parameters of the static Я Si ¼ 1  ∑ wij f βj ð18Þ
j¼1
and dynamic metrics of interdependent resilience defined in
Section 3. The decision space for static resilience is generated by Eq. (18) indicates that the higher the value of wij the lesser the
varying different investment levels in resilience-building resource desired value for f ðβj Þ; since a low value of wij f ðβj Þ indicates higher
allocation, while the decision space for dynamic resilience is resilience. Comparing the different wij 8 j is useful as an indicator in
generated by varying the values of the three resilience metrics in determining possible values for f ðβj Þ 8 j because it tells us where to
concert with the K matrix which represents investment options. put more effort in planning. As such the interdependent impacts of
A simple two-sector example adapted from Miller and Blair [23] sectors represented through the weights can be useful in indicating
illustrates the decision spaces. possible resource allocation strategies.
Another metric derived from static resilience planning, and a
4.1. Static resilience decision space useful indicator for planning, is the sensitivity of the sector
resilience Я i with respect to investment βj .
S

Through the static resilience formulation of Eq. (11), we have ∂Я Si ∂f ðβ Þ

¼  wij ∂β j ð19Þ
shown that the sector resilience metric Я i is influenced by the
S ∂β j j

amount by which the expected demand perturbations differ from

here we note that ∂f ðβ j Þ=∂β j is negative because f ðβ j Þ is a
the maximum demand perturbation. This is achieved through the
allocation of resources that require developing strategies to reduce decreasing function of βj . Sensitivity indicates the effectiveness
expected consequences of demand perturbations effectively, thus of the planning effort in reducing the resilience. If the magnitude
of ∂Я i =∂βj is large, then even a small planning effort for sector j
leading to economic resilience. While ideally we would like to S

reduce the demand perturbation for every sector as much as will have a greater impact in increasing the resilience of sector i.
possible, in reality such planning is limited by the availability of
The duple ðЯ i ; ∂Я i =∂β j Þ can be used as a multi-dimensional
S S
resources and investments. The next best option is to explore
measure of the value of different resilience and planning scenarios
where to invest more so that the best resilience outcome is
for strategic decision making. A numerical example explains this
achieved. We show here that Я i can be used to inform such
S
further.
decisions.
One of the ways for the implementation of preparedness, including
resilience building, activities deals with avoiding the maximum 4.1.1. Numerical example
impact, which is achievable if the expected demand perturbation is A two sector numerical example (adapted from Miller and Blair
a reduced proportion of the maximum impact that could have [23] and Barker and Haimes [2]) illustrates the usefulness of the
affected the sector. Proportionally reducing ci; max (i.e., avoiding the static resilience metric in deriving a decision-space that is useful
maximum impact) for sector i through the efficient allocation of for guiding investment into planning. Table 2 shows the commod-
resources to sector i. We assume that the expected demand ity flow matrix for the two sample industry sectors with all values
perturbation, ci , is the product of the
maximum demand perturba- in million USD. In this illustrative example we assume that the

tion, ci; max , and a planning function, f βi , as given in Eq. (16). industries represent a manufacturing sector and a transport sector

 that are significant at a regional scale. Though this is a theoretical
ci ¼ f βi ci; max 8 i ð16Þ illustration, the disruptions described here could be due to a

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
8 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎

disruption in raw manufacturing material resulting from a closure From the expression in Eq. (21), it can be inferred that the
of major roads, rails, or ports in the transport infrastructure, or a manufacturing sector resilience depends substantially on prepa-
major disaster to manufacturing plants that stops the transport of redness measures to protect the transportation sector. Also the
goods. These disruptions are then translated into supply or Transportation sector resilience is dependent largely on its own
demand loss metrics being used in the IIM (or DIIM) framework. preparedness measures due to the large multiplier (0.95) on f 2 ðβ 2 Þ
in calculating Я 2 . Eq. (22) confirms these observations because the
S
Our aim here is to show how the economic resilience planning
∂Я 2 =∂β 1 value shows that resilience in the Transportation sector is
S
proceeds once we get quantifiable estimates for economic losses
emerging from physical disruptions. much less sensitive to investments in Manufacturing, while the
From the values provided in the table we get the following other sensitivity values validate that the Transportation sector
matrices of interest. cmax corresponds to the total loss of oper- investment is critical for both sectors. This information indicates
ability of the two sectors. In a real world scenario, demand losses that there may be a preference for investing in Transportation
to the Transportation sector occur due to loss of passenger and sector planning. In physical terms, this implies that it is beneficial
freight travels along the transportation infrastructures, which to invest more in safeguarding the transportation infrastructure
affect economic earnings made for using the transportation infra- because it is critical for both sectors in the economy.
structure. For the Manufacturing sector, demand losses occur Through multiple resilience contour curves, shown in Fig. 6, we
when final consumers are not using the manufactured products. can graphically represent both sector resilience values and resilience
sensitivities with respect to the investment options. Such representa-
0:25 0:50 1:25 0:66 0:35
A ¼ ; D ¼ ; cmax ¼ ð20Þ tions provide an intuitive decision space for understanding the value
0:10 0:05 0:13 1:12 0:85 of investment to be made and the expected returns. Using the
For our case study we assume that the type of the planning contour plots three planning strategies are suggested and their

 investments outcomes are evaluated as represented in Table 3 and
function is same for both sectors and is given as f β i ¼ e  βi ,
where βi Z 0 indicates the amount of investment in USD million to in Fig. 6. Here investment decisions are informed by setting resilience
be made for resilience planning. The Eq. (18) static resilience targets corresponding to the contour values. Comparison between
metrics for the two sectors are as given in Eq. (21) by inputting the strategies A and C shows that for the same resilience for manufactur-
appropriate values from Eq. (20) into the calculation of wij . ing sector, improving Transportation sector resilience requires sub-
2 3 " stantial investment (1.25 million USD more investment into Sector
#
ЯS  β1
 0:56e  β2 1 and 1.15 million USD more investment overall). Between strategies
4 1 5 ¼ 1  0:44e ð21Þ B and C, opting for strategy B requires a total investment across the
Я S2 1  0:05e  β1  0:95e  β2
two sectors of an additional 0.1 million USD compared to C but results
in a significant increase in resilience outcomes for the Transportation
Similarly the sensitivity measures of Eq. (19) are calculated in
Eq. (22). sector.
2 S 3
∂Я 1 ∂Я S1 " #
6 ∂β 1 ∂β 2 7 0:44e  β1 0:56e  β2 4.2. Dynamic resilience decision space
4 ∂Я S ∂ Я S 5 ¼ ð22Þ
2 2 0:05e  β1 0:95e  β2
∂β 1 ∂β 2
As explained previously the rationale behind generating the
three resilience metrics is to present a comprehensive resilience
Table 2
Two-sector commodity flow table for input–output analysis (Miller and Blair [23]) Table 3
(in million USD). Outcomes of different investment strategies based on planned resilience decisions.

Industry Mfg Trans External demand (c) Total output (x) Planning Sector Sector investment
strategy resilience (Я S1 ; Я S2 ) (million USD) (β1 ; β2 )
Manufacturing 150 500 350 1000
Transportation 200 100 1700 2000 A (0.9, 0.95) (1.75, 3.1)
Value added 650 1400 B (0.95, 0.97) (2.5, 3.6)
Total output (x) 1000 2000 C (0.95, 0.95) (3.0, 3.0)

Fig. 6. Evolution of sector resilience based on amount of investments made into sector preparedness in the two-sector case study. Here Sector 1 is the Manufacturing sector
while Sector 2 represents the Transportation sector.

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

estimation through the dynamic inoperability model. This Eq. (26) is a useful visual representation of a decision space that
improves on the previous notion that the parameter K can alone reflects overall sector operability behavior in terms of maximum
signify resilience. To demonstrate the usability of the dynamic impact, recovery time and speed of recovery. Similar analysis is
resilience metrics for decision making we show that these metrics also presented in generating a multi-dimensional decision space
can be related to each other to form a single analytic function. using the resilience triangle construct [44].
Hence, we propose a multi-dimensional approach that fits all In essence the contour plot generates a decision space that can
resilience characteristics into a single function. For dynamic be used to estimate the outcome of system performance when
resilience planning we show that using this function we can choosing recovery strategies reflected through choices made for
estimate different K matrix options. the K matrix elements. Combining the constant functional rela-
As the three resilience metrics are measured with a dynamic tionship αi τi ¼ Li with Eq. (25), we get the expression in Eq. (27)
interdependency model, there exists a functional relationship for kii , which is similar in principle to the interpretation of kii
between the three metrics. Investigating such a functional rela- through Eq. (7). Li here can be treated as a functional measure of
tionship that relates the three metrics provides a sense of bounds the fraction of recovery from initial (or maximum) impact. Because
 
associated with each metric based on its dependence on the other we know that kii r 1, we can infer that Li r τi 1 ∑nj¼ 1 aij . Unlike
two. Moreover the values of two of the metrics can be specified,
thereby allowing for an educated guess to be made about the the expression in Eq. (7) in which kii for sector i was derived in
possible value of the third metric. The K is still a parameter in terms of its own interdependency coefficient (anii ), here we have kii
resilience planning because it influences all the three metrics. modeled in terms of a sector's interdependence on all others as
From the definitions of the resilience metrics, we seek to represent well.
the time averaged level of operability in terms of the maximum Li

 kii ¼   ð27Þ
i ; τi ; K .
loss of functionality and the time to recover, or F i ¼ g i T; qm τi 1  ∑nj¼ 1 aij
By capturing the overall resilience through the F i metric in
terms of the other metrics we are trying to infer system behavior The values on the contour isolines, which denote F i values, indicate
in terms of its maximum damage and speed to recovery. Eqs. (12) to the decision maker the outcome of different choices of resilience
through (15) are complex matrix functions that can become depending upon the initial inoperability (qi ð0Þ), recovery time (τi ) and
difficult to solve as they require measures for recovery times and measure for amount of recovery (Li ). The observations in Fig. 7 can be
maximum inoperabilities for all the sectors. A particular form of used to infer the following decision making insights:
the dynamic inoperability model is explored here to obtain a


functional relationship. As given in Eq. (23), F ¼ g T; qm ; τ ; K is 1. For the same maximum impact the sector average level of
derived under the assumptions that (i) dynamic response in Eq. (5) operability decreases as its recovery time increases, which is a
is due to initial inoperability only and there are no further demand result of slower recovery rates, and is indicative of less resource
perturbations (c ðt Þ ¼ 0; 8 t), and (ii) the initial inoperability cor- and redundancy measures.
responds to the maximum loss of functionality (qm ¼ q ð0Þ; 8 t). 2. Observations closer to the axes suggest better preparedness for
1
 resilient recovery, which progressively becomes worse as we
F ¼ 1  ½1  e  KðI  A Þτ ½K I  An   1 qm
n
ð23Þ move away from the axes.
T
The scalar form of Eq. (23), which represents the sector-specific
relationship among the resilience metrics, follows the form of
4.2.1. Numerical example
Eq. (24), where αi denotes
some
 measure of interdependence
To illustrate the usefulness of the dynamic resilience metrics in
suggested through the K I  A matrix.
deriving a decision-space we again use the two-sector case study data
1
Fi ¼ 1  ½1  e  αi τi qm ð24Þ
αi T i

Oliva et al. [25] suggest that the term ∑nj¼ 1 aij can be treated as
a quick global means to evaluate sector resilience, calling it a
dependency index as it reflects the capability of an infrastructure or
industry sector to maintain functionality under maximum inter-
dependent inoperabilities. For diagonal K, the term αi therefore
can take on the value given by Eq. (25).
!
n
αi ¼ kii 1  ∑ aij ð25Þ
j¼1

For a fixed T, different values of the F i metric in Eq. (24) can be

calculated over different combinations of αi ; τi ; and qm i . To gen-
erate any meaningful insights, we are interested in a decision
space among F i ; τi and qm i only. τ i is also a function of αi , and in
fact αi is analogous to a recovery rate parameter, which influences
the time to recovery and subsequently the overall time averaged
operability. If Eq. (24) is combined with a constant αi τi ¼ Li , then
the functional relationship between the three resilience metrics is
expressed as Eq. (26).
1 
Fi ¼ 1  1  e  Li τ i q m
i ð26Þ
Li T
Fig. 7. Contour plot depicting the relationship among the resilience metrics
This describes a rectangular hyperbola for fixed F i and different derived for the condition that the product of the recovery rate and recovery speed
combinations of τi and qmi . As shown in Fig. 7, the expression in is constant.

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
10 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎

from Table 2. In this example, disruptions are occurring on the supply

side of the sector economic outputs. In real world scenarios, physical
damages to the transportation infrastructures result in destruction of
major roads, railways, or ports, causing supply losses. For the
Manufacturing sector, closure of a manufacturing facility leads to
supply losses, resulting in inoperable machinery or damaged inven-
tory. The sector-wise relationships between the resilience metrics
calculated from Eq. (24) are given in Eq. (28).
 
F 1 ¼ 1  0:25k
1
11 T
1  e  0:25k11 τ1 qm
1

1
F2 ¼ 1  ½1  e  0:85k22 τ2 qm
2 ð28Þ
0:85k22 T
Argued previously, Li is a measure of amount of recovery from
disruption. Suppose r i A ð0; 1 denotes the fraction of recovery
from the observed maximum impact in sector i, then ð1  r i Þqm i
denotes the remaining inoperability still remaining in the sector.
Using an interpretation similar to the Eq. (7) notion of recovery,
we define the constant value of Li to be lnðqm i =ð1  r i Þqi Þ. A value
m

of r i ¼ 0:5 thus suggests 50% recovery from that maximum impact,

or, in other words, if the maximum impact were 30% loss of
functionality, then the system would recover to 85% operability.
The parameter r i is employed to indicate a preference for the
level of desired recovery for which we want to model resilience. If Fig. 8. Evolution of sector resilience based on amount of investments made into sector
preparedness in the two-sector case study. A and B correspond to planning strategies
we fix r i ¼ 0:95 for both sectors in this example (indicating 95%
for which the curve is used to measure recovery times from the y-axis and the contour
recovery from the experienced loss is sufficient), it results in values.
Li ¼ 2:9957. Both sectors have the same contour representation
on the resilience metrics as shown in Fig. 8.
As mentioned previously the contour isolines provide the decision-
maker with the outcome of resilience based on the maximum
inoperability, recovery time, and measure for amount of recovery.
The useful insight is added into the decision making process by
including the curves specifying the relationships between the recov-
ery rate and time to recovery. Fig. 9 depicts the plots for the recovery
rates versus recovery times for the two sectors obtained from the
equations 0:25k11 τ1 ¼ 2:9957 and 0:85k22 τ2 ¼ 2:9957, respectively.
A combined analysis of Figs. 7 and 8 indicates that for the maximum
impact, recovery time, and overall operability level, the Manufacturing
sector must have a higher recovery rate parameter than the Trans-
portation sector. This suggests that the Manufacturing sector should
invest more in recovery planning in order to maintain similar
functionality compared to the Transportation sector. For decision
making purposes, this also implies that the Transportation sector is
more resilient than the Manufacturing sector. A comparison between
two planning strategies highlights these insights. For resilience plan-
ning, we want to know the different K matrix coefficients when the Fig. 9. Plots depicting values for the sector recovery rates as functions of recovery
sector resilience is governed by the (F i ; qm i ) metrics. For both times. The outcomes of strategies A and B indicate the recovery rates corresponding
strategies A and B represented in Table 3, we are interested in the to recovery time estimates.
effectiveness of different recovery outcomes from the same levels of
maximum inoperability (in this case, initial damage). As such, the pair
dynamic representations. We integrate a mathematical construct
ðF i ; qm
i Þ indicates the desired level of overall operability during of resilience with a data-driven risk-based interdependency
recovery. Using Fig. 8 the recovery time τi for a sector can be inferred
model, the inoperability input–output model (IIM) [36,18] and
for the given resilience scenario, and interpolated to Fig. 9 relationship
derive resilience metrics derived from static and dynamic repre-
to estimate the kii for each sector. As the number below show to
sentations of the interdependency model. Further, we represent
maintain a high level of functionality (F i ¼ 0:968), the Manufacturing
graphical depictions of resilience for decision support.
sector requires a 12.2 day recovery achieved by a large k11 (¼0.982),
The static risk-based IIM quantifies the ability of the system of
which might be an unmanageable resilience option. On the other
interdependent infrastructures to move from one equilibrium
hand the Strategy B outcomes shows a relatively high resilience at
state to another after disruption. This means that the model is
lower planning investments. The Transportation sector can achieve
able to capture the system's capability to rearrange itself, suggest-
the desired resilience values for both strategies with lower k22 , which
ing a resilience quality. Much of this quality depends on the ability
suggests it is able to cope better and is thus more resilient Table 4.
of the system to resist the immediate impact of the external
perturbation to which it is subjected. We generate a static
5. Concluding remarks resilience decision space by considering the relationship between
resource allocation and initial impact from a disruptive event.
The primary focus of this research is to improve our under- For a dynamic system, resilience is indicative of its ability to
standing of interdependent economic resilience through static and resist initial impact effects (like static resilience) but then also

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i
 R. Pant et al. / Reliability Engineering and System Safety ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11

Table 4 [16] Klein RJT, Nicholls RJ, Thomalla F. Resilience to natural hazards: how useful is
Outcomes of two different investment strategies for dynamic resilience planning. this concept? Global Environmental Change Part B: Environmental Hazards
2003;5(1-2):35–45.
Planning strategy Sector resilience (F i ; qm
i ) Sector recovery (τ i ; k11 ; k22 ) [17] Leontief WW. Input–output economics. Oxford, UK: Oxford University Press;
1966.
A (0.968, 0.40) (12.2, 0.982,0.288) [18] Lian C, Haimes YY. Managing the risk of terrorism to interdependent infra-
B (0.952, 0.40) (18.1, 0.662,0.195) structure systems through the dynamic inoperability input–output model.
Systems Engineering 2006;9(3):241–58.
[19] MacKenzie CA, Barker K, Grant FH. Evaluating the consequences of an inland
waterway port closure with a dynamic multiregional interdependency model.
return to adequate functionality within a desirable amount of time. IEEE Transactions on Systems, Man, and Cybernetics Part A 2012;42
The dynamic inoperability input–output model (DIIM) quantifies the (2):359–70.
[20] MacKenzie CA, Santos JR, Barker K. Measuring international production losses
trajectory of inoperability from the time of the disruptive event up from a disruption: case study of the Japanese earthquake and tsunami.
until the system has made recovery. We explain the dynamic risk- International Journal of Production Economics 2012;138(2):293–302.
based interdependency model as a resilience estimation construct. [21] MacKenzie CA, Barker K. Empirical data and regression analysis for estimation
Resilience is quantified through appropriate metrics, which provide of infrastructure resilience, with application to electric power outages. Journal
of Infrastructure Systems 2013;19(1):25–35.
an operational framework for understanding interdependent system [22] McDaniels T, Chang S, Cole D, Mikawoz J, Longstaff H. Fostering resilience to
recovery and planning. These metrics, called the time averaged level extreme events within infrastructure systems: characterizing decision con-
of operability, maximum loss of functionality, time to recovery, texts for mitigation and adaptation. Global Environmental Change 2008;18
(2):310–8.
respectively denote the systems' ability to maintain functionality
[23] Miller R, Blair P. Input–output analysis. Cambridge University Press; 2009.
throughout the post-disruption response, withstand the maximum [24] Okuyama Y, Hewings GJD, Sonis M. Measuring the economic impacts of
disruptive impact and still recover, and progress towards recovery disasters: interregional input–output analysis using the sequential interin-
with some speed. The time-averaged level of operability metric can dustry model. In: Okuyama Y, Chang S, editors. Modeling Spatial and
Economic Impacts of Disasters. Berlin: Springer; 2004.
be expressed as a function of the other two metrics and improving [25] Oliva G, Panzieri S, Setola R. Agent-based input–output interdependency
system speed to recovery can come at the cost of incurring more model. International Journal of Critical Infrastructure Protection 2010;3
maximum impact. These complex behaviors are better understood if (2):76–82.
the metrics can be used to generate a decision space that allows a [26] Olsson S, editor. Springer; 2009.
[27] Ouyang M, Dueñas-Osorio L, Min X. A three-stage resilience analysis frame-
user to choose appropriate levels of values and thus affect sector and work for urban infrastructure systems. Structural Safety 2012;36(1):23–31.
interdependent resilience. [29] Pant, R, K Barker. 2012. Building dynamic resilience estimation metrics for
Future work will put these metrics to use in describing interdependent infrastructures. In: Proceedings of the European safety and
reliability conference, Helsinki, Finland.
infrastructure recovery following a disruption, as well as with
[28] Pant R, Barker K, Grant FH, Landers TL. Interdependent impacts of inoper-
the development of a preparedness decision making framework ability at multi-modal transportation container terminals. Transportation
with which to promote interdependent economic resilience. Research Part E: Logistics and Transportation 2011;47(5):722–37.
[30] Reed DA, Kapur KC, Christie RD. Methodology for assessing the resilience of
networked infrastructure. IEEE Systems Journal 2009;3(2):174–80.
References [31] Rose A. Defining and measuring economic resilience to disasters. Disaster
Prevention and Management 2004;13(4):307–14.
[1] Aven T. On some recent definitions and analysis frameworks for risk, [32] Rose A. Economic resilience to natural and man-made disasters: multidisci-
vulnerability, and resilience. Risk Analysis 2011;31(4):515–22. plinary origins and contextual dimensions. Environmental Hazards 2007;7
[2] Barker K, Haimes YY. Assessing uncertainty in extreme events: applications to (4):383–98.
risk-based decision making in interdependent infrastructure sectors. Relia- [33] Rose, A. 2009. Economic resilience to disasters, CARRI research report 8.
bility Engineering and System Safety 2009;94(4):819–29. Technical report, center for risk and economic analysis of terrorism events,
[3] Barker K, Santos JR. Measuring the efficacy of inventory with a dynamic input– University of Southern California.
output model. International Journal of Production Economics 2010;126 [34] Rose A, Liao S. Modeling regional economic resilience to disasters: a compu-
(1):130–43. table general equilibrium analysis of water service disruptions. Journal of
[4] Barker K, Santos JR. A risk-based approach for identifying key economic and Regional Science 2005;45(1):75–112.
infrastructure sectors. Risk Analysis 2010;30(6):962–74. [36] Santos JR, Haimes YY. Modeling the demand reduction input–output (I–O)
[5] Barker K, Ramirez-Marquez JE, Rocco S. CM. Resilience-based network inoperability due to terrorism of interconnected infrastructures. Risk Analysis
component importance measures. Reliability Engineering and System Safety 2004;24(6):1437–51.
2013;117(1):89–97. [35] Santos JR, Barker K, Zelinke PJ. Sequential decision-making in interdependent
[6] Baroud, H, JE Ramirez-Marquez, K Barker, CM Rocco. 2013. Measuring and sectors with multiobjective inoperability decision trees. Economic Systems
planning for stochastic network resilience: application to waterway commod- Research 2008;20(1):29–56.
ity flows in revision. Risk Analysis. [37] The Infrastructure Security Partnership. 2011. Regional disaster resilience
[7] Bruneau M, Chang SE, Eguchi RT, Lee GC, O'Rourke TD, Reinhorn AM, guide for developing an action plan. Technical report, American Society of
Shinozuka M, Tierney K, Wallace WA, von Winterfeldt D. A framework to Civil Engineers.
quantitatively assess and enhance the seismic resilience of communities. [38] Tierney KJ. Business impacts of the Northridge earthquake. Journal of Con-
Earthquake Spectra 2003;19(4):733–52.
tingencies and Crisis Management 1997;5(2):87–97.
[8] Chang SE, Shinozuka M. Measuring improvements in the disaster resilience of
[39] Ulieru, M. 2007. Design for resilience of networked critical infrastructures. In:
communities. Earthquake Spectra 2004;20(3):739–55.
2007 inaugural IEEE international conference on digital ecosystems and
[9] Department of Homeland Security. 2009. National infrastructure protection
technologies, 540-545.
plan: partnering to enhance protection and resiliency. Technical report,
[40] Vugrin ED, Warren DE, Ehlen MA, Camphouse RC. A framework for assessing
Department of Homeland Security.
the resilience of infrastructure and economic systems. In: Gopalakrishnan K,
[10] Haimes YY. On the definition of resilience in systems. Risk Analysis 2009;29
(4):498–501. Peeta S, editors. Sustainable and resilient critical infrastructure systems:
[11] Henry D, Ramirez-Marquez JE. Generic metrics and quantitative approaches simulation, modeling, and intelligent engineering. Berlin, Germany: Springer;
for system resilience as a function of time. Reliability Engineering and System 2010. p. 77–116.
Safety 2012;99(1):114–22. [41] Whitson JC, Ramirez-Marquez JE. Resiliency as a component importance
[12] Homeland Security Advisory Council. 2006. Report of the critical infrastruc- measure in network reliability. Reliability Engineering and System Safety
ture task force. Technical report, Department of Homeland Security. 2009;94(10):1685–93.
[13] Horowitz, K, M Planting. 2006. Concepts and methods of the U.S. input–output [42] Wixted B, Yamano N, Webb C. Input–output analysis in an increasingly
accounts. Bureau of Economic Analysis Papers. globalised world: applications of OECD's harmonised international tables
[14] Johansson J, Hassel H. An approach for modelling interdependent infrastruc- (No. 2006/7). OECD Publishing; 2006.
tures in the context of vulnerability analysis. Reliability Engineering and [43] Zobel, CW. 2010. Comparative visualization of predicted disaster resilience. In:
System Safety 2010;95(12):1335–44. Proceedings of the 7th international ISCRAM conference (p. 1–5).
[15] Joint Committee on Structural Safety. 2008. Risk assessment in engineering: [44] Zobel CW. Representing perceived tradeoffs in defining disaster resilience.
principles, system representation, and risk criteria. Edited by MH Faber. Decision Support Systems 2011;50(2):394–403.

Please cite this article as: Pant R, et al. Static and dynamic metrics of economic resilience for interdependent infrastructure and
industry sectors. Reliability Engineering and System Safety (2013), http://dx.doi.org/10.1016/j.ress.2013.09.007i