Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

Systemic Analysis Approaches for Air Transportation Sheila Conway PhD by ericaburns


									 Systemic Analysis Approaches for Air Transportation
                                         Sheila Conway
                            PhD Candidate, Old Dominion University
                           Researcher, NASA Langley Research Center
                                    Hampton, Virginia USA

                  Abstract                          “attempt to represent the environment in such
                                                    a way as to maximally simplify problem-
    Air transportation system designers have        solving.” He continues that a model is not to
had only limited success using traditional          be judged on its correctness in terms of its
operations research and parametric modeling         ability to yield absolute or complete truth, but
approaches in their analyses of innovations.        rather on its ability to provide insight.
They need a systemic methodology for                     Simply stated, models are our windows to
modeling of safety-critical infrastructure that     understanding the world around us. Wilson
is comprehensive, objective, and sufficiently       (1998) concurs with this sentiment, saying, “A
concrete, yet simple enough to be used with         Model is the explicit interpretation of one’s
reasonable investment. The methodology must         understanding of a situation, or merely of
also be amenable to quantitative analysis so        one’s ideas about that situation. It can be
issues of system safety and stability can be        expressed in mathematics, symbols, or words,
rigorously addressed.            However, air       but it is essentially a description of entities
transportation has proven itself an extensive,      and the relationship between them. It may be
complex system whose behavior is difficult to       prescriptive or illustrative, but above all, it
describe, no less predict.                          must be useful.” Selecting an appropriate
    There is a wide range of system analysis        model is then highly dependent on the subject
techniques available, but some are more             and the addressed concern.
appropriate for certain applications than                Wilson’s statement raises the issue that
others. Specifically in the area of complex         modeling itself does not impose a particular
system analysis, the literature suggests that       implementation, method or tool, but rather is
both agent-based models and network analysis        the process of interpretation.        All models
techniques may be useful.                           share the same interpretive goal, but many
    This paper discusses the theoretical basis      different types of models, and their associated
for each approach in these applications, and        tools and techniques, may be able to achieve
explores their historic and potential further use   it. Selection should be based on the intended
for air transportation analysis.                    purpose of the model.
                                                         Modeling Systems. The real world is a
 Need for Complex System Modeling                   big place that can be difficult to understand.
                                                    To avoid being overwhelmed by its sheer size
    Why Model? Models are necessarily               or complexity, we are inclined to parse it into
simplified abstractions of reality for              manageable, and hopefully functional, parts
understanding specific properties of a subject.     through abstraction. This parsing can be at any
Heylighten (1993) likens models to                  granularity (pixilation, or scale), but tradeoffs
knowledge itself. As abstractions of reality,       are made between the size of the parts and the
he asserts they are vital simplifications and an    utility of the abstraction: too big, and it is still

unmanageable, too small and it may no longer       counterintuitive. In fact, there are numerous
represent the phenomena of interest. A             accounts of system changes that had
system is such a parsing. The context in           unanticipated and sometimes dramatic results,
which it is formed must be kept with it.           like the deregulation of the power industry in
    Kast and Rosenweig (1972) define a             California (which caused rolling power
system as an organized assemblage of smaller       outages). Often these ‘surprises’ are results of
units that form a unitary whole.          Others   oversimplification, ignoring system dynamics
identify the interdependence and complexity        or even misinterpreting the ‘system’ itself by
that their relationships inherently contribute.    not recognizing important elements, or
    Without models, we can’t do much with          including superfluous, confounding ones.
systems. In fact, Forrester (1987) held the            One obvious issue with complex system
belief that all “systems” are mental constructs    modeling is compliance with the principle of
of portions of the real world, and that we can’t   Occam’s Razor. How to select a model that is
even imagine a system without modeling.            simple enough to field, yet captures the
Such a bold statement brings to mind a riddle      nuances of a complex system? Heylighten’s
about a tree falling in the woods, but there is    statements imply that comprehensiveness is
no doubt that to understand or influence a         not necessary. On the contrary, he implores
system requires a model. Whether a mental          us to look for the simplest model that explains
model, a mathematical model, or a simulation,      observed behavior. As a guiding principle,
Forrester had a point: system models are a         controlling scope, or the inclusion of detail,
basis for action. We have expectations for         will require substantial domain knowledge and
our systems, and we would like to be able to       explicit recognition of the context. A complex
influence their behavior. However, a system’s      system cannot be trivialized with a simple
reaction to intervention is not always clear,      treatment.
nor easy to predict. As systems become more            Rather than assume that everything within
complex, with a greater number of interrelated     a system can be equivalently described by a
components, their response to intervention can     single type of abstraction, matching a model
become even more ambiguous. In fact, as            to reality begins with an honest assessment of
even linear systems become complex in the          the fidelity of what is 1) known and 2) what
structure of their interrelationships, their       knowledge is desired. This issue of scope for
behaviour can begin to appear chaotic.             each model element is paramount to
    Modeling Complex systems. Modeling             successful complex system modeling. One
complex systems adds some additional               might assume modelers are granted authority
limitations and assumptions to the already         to scope as they see fit, but unfortunately, that
challenging task of systems modeling. The          is not always the case.
scientific method and good engineering                 Another broad category of issues is
practice require us to observe, hypothesize,       expectations: 1) Qualities inherent to complex
predict, test, derive a conclusion, and repeat     systems may make understanding analytical
until a satisfactory solution is obtained.         results open to interpretation. 2) Complex
How is the safety of a new air traffic control     system analyses may disappoint those who are
technique to be tested without putting aircraft    concerned with output rather than outcomes.
and people at risk? Even if safe, what effect      3) Complex systems are often ill structured
will the new operation have on traffic flow        and changeable. Thus, modeling them can be
management, or an airline business?                difficult, costly, and never quite complete
    In a complex world, answers to such            compared to more simple systems.
questions are not always obvious, and can be

    Finally, though important in all systems       mechanical factors of the system and their
modeling, being able to gauge a system’s           interactions, or gross errors will occur. They
robustness     becomes       imperative    with    go on to recognize that this is a tall order
increasing system complexity.         Complex      indeed, and that a comprehensive ATS model
system models have limited optimizing              is a “grand challenge,” albeit necessary and
ability. The nature of complexity requires         attainable.
models to address fuzziness, non-determinism,           Actually, NASA has recognized the need
and multiple objectives by affording               for a more systemic study for some time.
exploration of sensitivities to assumptions and    They commissioned Krozel (2000) to review
environmental uncertainties.                       all the research related to distributed air traffic
                                                   management, a widely accepted development
Air Transportation System Analyses                 concept. He identified not only existing
                                                   research, but also research needs that were not
    “The changes that are coming are too big,      being met. In summary, he found that at the
too fundamental for incremental adaptation…        time, there were no tools capable of assessing
We need to modernize and transform our             both new and traditional ATS operations
global transportation system, starting right       simultaneously, or their interactions.
now.” (U.S. Transportation Secretary Norm               Carley      (1997)       claims      “social,
Minetta, 2004). Ultimately in the case of air      organizational and policy analysts have long
transportation, national governments are           recognized that groups, organizations,
largely responsible for both setting policy, and   institutions, and the societies in which they are
implementing infrastructure implied therein.       imbedded are complex systems.” When it
To do so necessitates consideration of both the    comes down to it, policy analysis is about
effectiveness and repercussions of actions         complex system design in light of uncertainty.
within the air transportation system (ATS).             Certainly, complexity and uncertainly will
    As Wieland et al (2002) point out,             abound in ATS transformation. Influencing
modeling the ATS, “with all its interrelated       ATS performance within itself is complicated
components – mechanics, human decision             enough, but ATS policy reaches outside this
making, and information flow, is a large effort    arbitrary system boundary. Sheate (1995)
involving multidisciplinary and ‘out-of-the-       complains that standard ATS policy decisions
box’ thinking. …The challenge is not only to       have lead to a business market that decides
represent physical NAS [National Airspace          “where capacity is needed and therefore fails
System] dynamics, but also to incorporate the      both to maximize the use of existing airport
behavioral and relational components of NAS        resources and to recognize the importance of
decision making that are an important part of      environmental capacity constraints.”            He
the system. …A comprehensive model is              argues for policy analyses that consider the
incomplete and subject to first order errors       interplay of system capacity, demand, and
unless all such interactions are incorporated to   aircraft capability.
some degree.”                                           Unfortunately, policy analysts in the ATS
    Wieland et al stress the necessity for ATS     arena have continued to use methods more
modeling at three different time horizons for      suited to regularly-behaved systems to
various purposes: tactical (predictive),           develop strategy.       Apparently, this is a
strategic planning (investment and policy),        pervasive problem throughout the policy
and a posteriori analysis (also investment and     community. Bankes laments that there are
policy). Their claim is that a useful simulation   “few good examples of the classical policy
of the ATS intended for setting policy must        analysis tools being successfully used for a
model the economic, informational and

complete policy analysis of a problem where         that describe them.        Regardless of the
complexity and adaptation are central.” He          approach used, Andrews (2000) implores
continues to say that policy analysis in the        modelers to not loose touch with purpose of
face of “deep uncertainty” must focus on            their effort, and to build models that
robustness      rather     than      single-point   “appropriately and credibly” simplify reality
optimization. This reinforces the notion of         within specific context of the system at hand.
developing      many      different     plausible       Within the literature, there is consensus
environmental scenarios, and recommending           that: 1) Systems modeling is a useful a way of
policy that is viable across their range.           solving real world problems, particularly
Addressing this same concern, Iyer (2000)           when prototyping or experimenting with the
offered that the “basic contribution of             real system is expensive or impossible.
complexity theory [to planning] is its focus on     2) Different types of applications call for
systemic interactions at various scales…” that      different modeling techniques (figure 1).
can address uncertainty.
    Moss expresses the view that “Policy
analysis has to start with observation and the
specification of a problem to be solved.”
From there, appropriate analysis tools can be
defined. Moss, Iyer, and others suggest that
deterministic and even stochastic approaches
to complex policy development are
incompatible, though agent-based modeling
(ABM) may be workable.
    Though the ATS research community has            Figure 1. Methods and application within
attempted to model particular attributes of the       problem context (adapted from Daniel)
system, there hasn’t yet been a method                   Many different modeling approaches have
capable of answering questions regarding the        been offered in the literature. Generally
systemic response to substantive changes in         speaking, these different approaches are
operations. To date, agent-based, elemental         intended to address specific classes of
simulations have proven too expensive and           systems. A model’s capabilities have to
unwieldy to complete. Parametric simulations        match the system’s overall attributes. Many
have failed to provide the flexibility to be used   authors summarize these attributes in system
as design tools.                                    classifications, which, of course, vary within
    The dearth of appropriate analytical tools      the body of work. Authors find their own
is not due to a lack of demand, or trying. It       dimensions on which to split the space of
has simply proven to be very difficult. Calls       possible systems. In the realm of complex
for systemic simulation for operational design      systems, which nearly all classification
of the ATS continue to accrue, from the             schemes include, authors have deemed most
responsible government officials, to the            modeling approaches unsatisfactory, leaving
researchers in the trenches.                        precious few potential choices for those
                                                    interested in complex system modeling.
     A (Discretized) Continuum of                        Daniel (1990) described the possible space
    Potential Modeling Approaches                   of systems in two dimensions, along the
                                                    attributes of complexity and the number of
   Just as systems themselves differ in             objectives a system operates to control. He
objectives and complexity, so do the models         suggests that classic Operations Research

(OR) modeling techniques are best applied           as to say that soft approaches as well as
when the system can be described in great           traditional, “harder” ones will never support
detail and a single optimization function is the    effective policy analysis. How then to address
primary focus of study. He reports the static       complex systems in both a rigorous but
nature of OR models is an inherent limitation.      sufficiently realistic and tractable way? Moss
Not only does OR require detailed system            provides a suggestion, saying; “adaptive agent
knowledge, the system is implicitly expected        modeling [e.g. ABM] is an effective
to remain unchanged.                                substitute” for other analyses in the complex
     He offered cybernetics as an approach that     system realm.
acknowledges the importance of both system              Borshev and Filippov (2004) interpret the
structure and the interaction of components         potential systems modeling space differently
that can cause dynamic behaviors. However,          than Daniel.         Borshev and Filippov’s
he implies its limitations lie in the singularity   orthogonal dimension to complexity is
of its optimization goals.         Others have      discreteness, that is, the level of abstraction or
enumerated additional challenges with               aggregation in model elements. Interestingly,
cybernetic deployment.                              they also make a distinction between system
     Daniel continues that of the many systems      types that necessitate simulation vs. those
modeling techniques described in the                better served by analytical models. They
literature, soft systems methods (SSM) are          prefer analytical solutions when a closed form
particularly well suited to complex systems:        solution is obtainable. Thus, they imply one
Complex systems are represented in the upper        ought at least to consider such a model first,
right of his systems space. They are context-       because simulation, they argue, is not trivial.
rich, non-linear problems that cannot be            However, they continue by saying that “for
expressed by a single set of objectives or          complex problems where time dynamics is
goals. SSM involve the development of a rich        important, simulation modeling is a better
picture of the problem, putting great emphasis      answer,” narrowing the field of potential
on framing the problem correctly within             modeling techniques.
context. However, these methods have been               Akin to Andrews, Borshev and Filippov
criticized for being unverifiable, non-             suggest matching modeling techniques to the
quantifiable, and lacking in rigor (Lane 1998).     “nature of the problem,” and that any one
     Additionally, if effecting systemic            technique will almost surely not be most
improvement is a modeler’s goal, the ability        appropriate for all systems. Rather they call
of the output/outcome to be influential has to      for modeling techniques that “would allow for
be considered. For a safety-critical system         integration and efficient cooperation between
with minimum performance criteria, mental           different modeling paradigms.” They discount
constructs (and the flexibility they provide as     other complex system simulation options, but
“controlling” qualities as in SSM) have not         conclude that there is a place for both system
proven influential in many circles. Many            dynamics (SD) and ABM. They found ABM
systems, air transport included, demand             well suited to systems where most knowledge
rigorous evaluation before change is even           is at the local level (e.g. agent-level) and little
considered.                                         or nothing is known about global
     Sterman (2002) warns that SSM often            interdependencies. They also concluded that
leads to “wildly erroneous inferences about         SD could be more efficient, particularly if
system behavior”, dramatic underestimation of       agents are uniform and/or have little true
the dynamics of systems, and incorrect              “active” or autonomous behavior, and discuss
conclusions. In fact, Moss (2002) goes so far       the use of both techniques in combination.

    While full-scale agent models can be as                                 and feedfoward loops causing attenuation and
complex and costly to develop as a large-scale                              amplification of system attributes respectively.
parametric model, there may be a means of                                   Often the metaphor of stocks and flows is
validating models and educing a number of                                   used to illustrate the approach. SD models are
higher-order effects without constructing and                               time-dependent linked mathematical models
running full-scale agent-based simulations.                                 exploiting differential calculus.
From the description above, it is clear that                                    Borshev and Filippov note that SD is
interaction among agents could be described                                 similar in nature to dynamic systems, or
by network structure: there are well-defined                                simply “dynamics,” taught in technical
nodes (agents) and links (interfaces,                                       engineering disciplines, but uses language and
interaction protocols).      Network analysis                               notation more familiar to systems analysts.
(NA), developed in the field of network                                     As with dynamics, rigorous treatment,
theory, could be applied to a network defined                               unavailable       with    cybernetic    models
by the agents’ communications demands.                                      equivalents, is possible. They comment that
These may provide a relatively simple and                                   dynamics are taught to mechanical, aero and
reliable means of evaluating the aggregate                                  electrical engineers “as a standard part of the
performance of a complex system, similar to                                 design process.” These members of the
SD, with less effort than an ABM (Figure 2).                                academic community acknowledge the
                                                                            necessity for systemic dynamic analysis for
                            ( re )F ra m in g
                                                                            design, at least for physical systems.
                            th e p ro ble m
                                                                                Unfortunately, SD requires extensive
                                                                            system knowledge a priori, including all
                                                                            system        elements        and      potential
 Co n ce pt M o de ls:       A n alysis o f           R ecom m e n d
                                                                            communications between them. This makes
 d escribin g m o de l         Ne two rk
     e le m e n ts o r
    “a ge n ts” a n d
                            D e scrib e d by
                         A g en t d efinitio n s
                                                   S ystem tran sfo rm s
                                                   b a se d on fe asib le   building a comprehensive model of a complex
                                                      m od ification s
     in te ractio ns
                                                                            system an enormous effort. Systems that
                                                                            change frequently or have a high degree of
                           E xe cu tio n a n d
                                                                            uncertainty may not be amenable to SD at all.
                             An a lysis of
                            A gen t-B a se d
                                                                                Despite this major drawback, many
                             Sim ula tio n
                                                                            authors have used SD to model complex
                                                                            systems. In its favor, analysis and control
      Figure 2. ABM and NA Relationship
                                                                            techniques for the resultant mathematical
    By using dynamic or adaptive modelling                                  models are well established and have proven
methods when dealing with complex systems,                                  to be highly serviceable.
the possible modeling space is reduced                                          Agent-Based Modeling (ABM). Agent-
dramatically. The three methods identified in                               based modeling (ABM) techniques have been
the literature as applicable for capturing                                  proposed as an alternative to traditional
dynamic behavior are worthy of further                                      parametric models because they can exhibit
consideration:                                                              higher-order behaviors based on a relatively
    System Dynamics (SD). Forrester, the                                    simple rule set. ABM uses agents to execute
father of SD, describes system dynamics as                                  model functions.         They are the active
the discipline of interpreting real life systems                            components of an agent-based simulation.
as simulation models. These models highlight                                    Agents are ‘autonomous’ in that they have
the structure and decision-making processes                                 interfaces to the general simulation, but carry
within a system that give rise to its behavior.                             within them their own ability to perform their
As the name implies, SD is the study of the                                 assigned tasks without a centralized controller.
interactions of system elements via feedback

Agents are interactive entities that capture            While the latter two arguments are similar
salient but generally localized behavior of         to those of Jennings, Bankes claims
system elements. Using simple rules to              dissatisfaction with the restrictions imposed
determine each agent’s actions, higher-order        by alternative modeling formalisms is driving
systemic behaviors can emerge. Jennings             modelers to agent-based solutions. In his
(2000) offers further detail, saying agents:        opinion, the most widely used alternatives,
  1) have defined boundaries & interfaces.          systems of differential equations and statistical
  2) are situated in a particular environment.      modeling, are viewed as imposing restrictive
  3) strive for specific objectives.                or unrealistic assumptions that limit many
  4) are both reactive and proactive, and           applications. He says “The list of assumptions
  5) are autonomous (distinct from objects).        that have been objected to is lengthy, but it
    Jennings would most likely agree that           includes linearity, homogeneity, normality,
ABM is not well suited to all systems.              and stationarity.”
However, he outlines his argument in favor of           What Bankes fails to mention is that these
ABM of complex systems, saying complex              shortcomings are not necessarily avoided just
system development requirements and ABM             by deploying ABM approaches, and certainly
are highly compatible. He argues that ABM is        not by agent implementations of standard
particularly well suited to complex systems         methods. A model still has to be appropriately
because it:                                         defined to describe significant features for the
  1) partitions a complex problem space.            system served.       Additionally, addressing
  2) naturally abstracts complex systems, and       issues such as homogeneity requires not only
  3) captures dependencies and interactions.        more effort in model specificity, but also more
    However, he also admits that these same         information related to distributions of
properties     can     lead    to    issues    of   variables or behaviors. These data may not be
unpredictability     and     apparent     chaotic   available. A homogeneous population model
behavior. Unpredictability is a problem in the      might be of sufficient fidelity for describing
simulation world because it makes internal          some systems, while an assumed (but
validation very difficult when exact results        erroneous) normal distribution, for example,
cannot be repeated. The lack of deterministic       might yield misleading results. A more
behavior is also a problem for validation.          complex or detailed model (e.g. at the agent
Jennings and others claim that these                rather than the aggregate level) is not
difficulties can be circumvented by formally        necessarily more accurate.
analyzed interaction protocols, limiting the            Arthur (1994) suggests agents are a natural
nature of agent interaction, and adopting rigid     way to deal with ill-defined or complicated
organizational structure among the agents.          “reasoning” within a system, oft induced by
    Much hope is laid at the feet of ABM,           inclusion of humans. He argues, “beyond a
particularly in the social science realm where      certain level of complexity, human logical
complexity and uncertainty are paramount.           capacity ceases to cope – human rationality is
From recent literature, Bankes (2002)               bounded.” Agents can be designed to mimic
summarizes three reasons why ABM is                 the inductive behavior of people when placed
potentially important: 1) the unsuitability of      in unfamiliar or complicated environments.
competing modeling formalisms to address the        However, the example he provides, a problem
problems of social science, (2) the ability to      of deciding whether or not to frequent a bar
use agents as a natural ontology for many           based on the expected crowd, exemplifies a
social problems, and (3) the ability to capture     prime concern with assuming agent
emergent behavior.                                  “intelligence” (which has to be present to

differentiate the agent from a mere object in           All networks can be analyzed by some
Jennings terms). In his example, the agents        basic, quantifiable measures including their
select from a pre-determined set of schemata       degree distribution and their average
based on some outcome metric (actual number        clustering coefficient (Wuchty et al, 2003).
of bar patrons). Can this be considered true       Stemming from these basic metrics, networks
inductive behavior? The “induction” was            often exhibit higher-order dynamic functions,
accomplished [by the modeler] in the               thought to be associated with their unique
generation of the options, not by the agent in     structures. These include robustness, fragility,
their selection later on.                          percolation and searchability.
     If appropriate strategies were not included        The ability of NA to differentiate
in the agent’s definition, Arthur’s agents         operationally unique airline route strategies
would have never succeeded. Recognizing            and their resultant distinctive structures is yet
this, he does acknowledge that people’s            to be shown. Due to the relatively small
inductive ability [emulated by agents using        number of nodes in air traffic networks, nodal
lists, genetic algorithms, etc.] is a “deep        separation distance and searchability tend to
question in psychology” and thus can only be       be straightforward to determine and not too
marginally imitated.        Generally speaking,    instructive.       However, because of the
agent “intelligence” at best will be limited by    criticality of the application, resilience to
the degrees of freedom their internal models       cascading failure, percolation, and congestion
are allowed to explore, and may be further         robustness are of utmost interest in the ATS.
limited by the methods of exploration.             It is not clear if NA will be able to reveal these
     Bonabeau (2002) claims that ABM is “by        qualities sufficiently. Braha and Bar-Yam
its very nature the canonical approach to          (2004) suggest that the approach is worthy of
modeling emergent phenomena” of complex            pursuit, as functional classes of networks
systems, necessary for analysis of non-linear      might be expected to have differences in their
behaviors,     localized     phenomena,      and   topologies, such as directedness. These in
heterogeneous populations. However, like           turn could be expected to lead to particular
Jennings, he acknowledges difficulties in          dynamic potentials.
building agent models of large systems                   Latora and Marchiori (2001) call for the
because of the myriad low-level details and        measurement of average path length,
the “extremely computation intensive and           clustering coefficient, average degree, and
therefore time consuming” model that results.      degree distribution as do Strogatz, Watts, and
     Network Models. Network theory is an          others, but also suggest the use of efficiency
extension of graph theory. By definition,          and cost. They define efficiency at both the
nodes that constitute a network are                local and global level as “the measure of how
interconnected in some way or another by           efficiently it [the network] exchanges
links.      The resultant network can be           information.” They suggest that efficiency is
categorized by its structure. In turn, this        really a more general measure for path length
structure imparts peculiar characteristics to      and clustering, useful because other measures
both the system as a whole and to the              can only be defined for certain network sub-
individual nodes.           Following specific     classes. Efficiency can be applied to any
connectivity rules, some networks have some        network, but it can be difficult to calculate.
nodes that are highly connected while others            Latora and Marchiori argue that the
have only a few connections. In other              Watts/Strogatz measures are only effective in
networks, links are randomly formed but still      quantifying a network in the “topological
obey statistically generalizable patterns.         abstraction, where the only information

retained is about the existence or absence of a    paradigm. For example, a good modeler
link.” Following the above arguments that          would be hard pressed to generate any “hard”
quality/cost of the links are paramount to         model (e.g. ABM or SD) without some effort
describing operational functionality, it appears   to capture significant system context or a clear
unlikely then that topological metrics alone       understanding of the problem at hand. Using
will be useful abstractions for describing air     SSM adds formalism to this step that in turn
transport networks. Using the Boston Subway        may improve the product.
as an example, they suggest that substituting          What Mingers and others offer is balance
efficiency measurements resolves difficulties      to the process. Their claim, based on a
in general application of network topologic        number of examples, is that modelers will
analyses to weighted and directed systems.         tend to focus on the data at hand, and not on
    Once measured, Latora and Marchiori            modeling the primary driving functions of a
show that efficiency metrics can be used as        system. Models tend to be concentrated on
indicators of potential cascading failure, and     directly measurable quantities, and ignore or
can be used as a “measure of performance” of       de-emphasize less well-behaved system
the network.          They showed marked           components (such as people). Freeing the
differences in the non-linear behavior (onset      modeler to use all available and suitable
of cascading failure) of two different, well-      techniques rather than a single model for the
documented network topologies.                     entire system should produce a better, more
    Multimethod Approaches. The use of             tractable product with less effort.
more than one method in a single modeling              Regardless of the declaration of multi-
effort may be the most promising approach to       methodology or not, the concept is well
complex system analysis. Multimethodology          established in practice.
may enable modeling of inhomogeneous
elements of a complex system, each element                          Summary
matched to an appropriate modeling method.
This strategy may be important where a lack            The majority of ATS researchers have
of complete system knowledge inhibits the use      joined Wieland et al. in suggesting that
of a single model type. It might also be useful    specific classes of tools represented by ABM
when the scope of the system represented in a      and/or network analysis are perhaps the only
single scale would cause the model to become       modeling solutions currently available that
too cumbersome. Multiple methods could be          offer systemic utility. Holmes and Scott
applied in successive phases of an effort, or in   (2004) say, “Proposed ideas for changing the
parallel, representing different levels of         NAS should not be contemplated lightly, due
fidelity for various subsystem models.             to the sheer size and complexity of the system.
     Mingers (2000) proclaims, "Multimethod        Instead it will require a fundamental
is not the name of a single method, or a           reconsideration of how such complex systems
specific way of combining methods. Rather it       are analyzed and designed if the system to
refers in general to utilizing a plurality of      evolve remains productive and viable.
methods or techniques, both quantitative and       Traditional methods for analyzing changes to
qualitative, within a real-world intervention."    complex systems fail when applied to highly
Declaring multimethodology as a distinct           dynamic and interconnected system such as
analytical approach or even as “new” may be        the Internet or the NAS.” They outline a case
a disservice to best practices within systems      for using agents operating on networks as a
science. Perhaps multimethodology is more a        viable analytical alternative.
matter of emphasis than a totally new                  The literature suggests that both NA and
                                                   ABM are well suited to study emergent,

complex behavior within the context of air            Tpolgies”, 4th Intg CNS Conf, Fairfax, VA
transportation. From the outset, differences      Iyer (2000)“Can Complexity Theory Enter the
in both scope of effort to establish these two        World of Planning” Critl Planning, v7 p25
models and expectations for their results         Jennings, (2000) “On Agent-based software
should be acknowledged. NA is focused at              engineering” Artfcl Intlgnce 117, 277-296
systemic-level solutions, much like system        Kast & Rosenweig (1972) “General systems
dynamics, while ABM revolves around the               theory” Acad Mgmt Jrnl, v.4 p. 47
“unit” of the system.         Is the additional   Krozel, (2000) “Free Flight Research Issues
information (at the agent level) necessary or         and Lit. Search”, NASA, NAS2-980005
even useful for a transport system study? Is      Lane and Oliva, “The greater whole: towards
the system so sensitive to assumptions of             a synthesis of SD and SSM” Euro Jrnl of
individual behaviors that ABM predictions are         Opra’l Rsrch, v. 107 n.1 p.214-235, 1998.
no better, or in fact worse, than more            Latora & Marchiori(2001) “Efficient Behavior
generalized network analyses? On the other            of Small-World Networks”, Physl Rvw
hand, are NA so aggregated that system                Lters, Amer. Physical Society, v.87 n.19
dynamics are poorly described?                    Mingers (2000), “Variety is the spice of life:
    Either approach, or perhaps both in               Combining soft and hard OR/MS” Intern’l
combination as Mingers might suggest, may             Transactions in OR 7, 6, pp. 673-691
provide clues for uncovering problems,            Moss (2002) “Policy analysis from first
provide      confidence      about    systemic        principles”, PNAS v.99 sup.3, p.7267
performance, and contribute to developing         Sheate, “Transport policy: a critical role for
mitigation strategies for systemic ATS issues.      strategic environmental assessment” World
                                                    Trnsprt Polcy and Prctice, v1 n4, p.17, 1995
                References                        Sterman, “All models are wrong” Systems
                                                      Dynamics Review, v. 18 n. 4 2002
Andrews, “Restoring Legitimacy to the             U.S. Transportation Secretary Minetta, speech
   Systems Approach” IEEE Tech & Society              to Aero Club of Wash DC, Jan 27, 2004
   Mag, v. 19 n.4, p.38-44, winter 2000/2001.     Wieland, Wanke, Niedringhaus and Wojcik,
Arthur (1994), “Complexity in Economic                “Modeling the NAS: A grand challenge
   Theory” Amer Econ Revw, v.84, n.4, p.406           for the simulation community” Soc. for
Bankes(2002) “Agent based modeling:                   Computer Sim, San Antonio, TX, 1/2002
   Revolution?” PNAS, v99 n3 p7199-7200           Wilson, E. (1998) Consilience: The unity of
Bonabeau (2002) “ABM: Mthds & technqs for             knowledge. Knopf, New York, NY
   sim human sys”, PNAS 99/3 p 7280-7287          Wuchty, Ravasz and Barabasi (2003) "The
Borshchev & Filippov (2004) “From Systems             Architecture of Biological Networks",
   Dynamics and Discrete Event to Practical           Complex Systems Science in Biomedicine,
   ABM”, Sys Dymcs Soc Conf, Oxford, UK               Kluwer Academic, New York
Braha and Bar-Yam (2004), “The Topology of
   Large-Scale Engineering Problem-Solving                        Biography
   Networks”, Physical ReviewE, v.69
Carley (1997) “Extracting team mental models          Sheila Conway is a researcher at NASA
 thru textual analysis” Jrl Org Bhvr, 18, p533    Langley Research Center in Hampton,
Forrester, J. “Fourteen ‘Obvious Truths’” Sys     Virginia and a Doctoral Candidate at Old
   Dynamics Revw, v.3 n.2 p.156-159. 1987         Dominion University.          Her experience
Heylighen (1993): "Epistemology", Principia       includes engineering design and systems
   Cybernetica, URL: //          development. She is a commercial pilot and
Holmes and Scott (2004), “Trnsprtn Ntwk           an instrument flight instructor.


To top