Framing flexibility: Theorising and data mining to develop a usefuldefinition of flexibility and related concepts

J. de Haan a,*, J.H. Kwakkel b,1, W.E. Walker b,2, J. Spirco c,3, W.A.H. Thissen b,4

a Centre for Water Sensitive Cities, Monash University, Building 11, Clayton Campus, Clayton (VIC), Melbourne, Australiab Delft University of Technology, Faculty of Technology, Policy, and Management, Policy Analysis Section, Jaffalaan 5, PO Box 5015, 2600 GA Delft, The Netherlandsc Delft University of Technology, Faculty of Technology, Policy, and Management, Section of Information & Communication Technology, Jaffalaan 5, PO Box 5015,

2600 GA Delft, The Netherlands

1. Introduction

The concept of flexibility applies very much to itself. The word is used in various contexts with varying meanings,conveniently meeting diverse needs. Moreover, there are many that have a similar ring to them and are used in similarcontexts. But, convenience can lead to confusion.

In this article, the discussion of the concept will be restricted to its use in relation with infrastructure systems. Therealisation that infrastructures are typically in place for decades or more, while demands and circumstances can change onmuch smaller timescales, makes the idea of a flexible infrastructure attractive. That there are many possible futuresconfronts society with large, and deep, uncertainties, and if infrastructures could be so flexible to accommodate thoseunforeseen changes in demand, function, or availability of resources, this would be highly beneficial.

Futures 43 (2011) 923–933

A R T I C L E I N F O

Article history:

Available online 17 June 2011

A B S T R A C T

Flexibility is a term used in various fields with widely differing interpretations. Moreover,

several related concepts, such as adaptability, exist that have an overlap in meaning or are

simply used synonymously. This article presents a framing of flexibility, and three

concepts with which it bears a close family resemblance, for the use in the context of

infrastructure constellations. The definitions proposed in this frame draw inspiration from

existing literature, though they are not based upon a classical literature review. Rather, a

usable set of definitions is proposed for the intended context. The definitions all have the

same structure to better appreciate how the concepts are related and how they differ. To

verify whether the definitions correspond to their practical use, a data-mining exercise is

performed on over 11,000 scientific articles that use the concepts of flexibility. After the

corpus of articles is identified that is close to the intended field of application

(infrastructure constellations), a co-occurrence analysis is carried out in order to clarify

the differences between the concepts and to give nuance to the meaning conveyed in the

definitions.

� 2011 Elsevier Ltd. All rights reserved.

* Corresponding author. Tel.: +61 3 9905 2616; fax: +61 3 9905 2948.

E-mail addresses: j.dehaan@monash.edu (J. de Haan), j.h.kwakkel@tudelft.nl (J.H. Kwakkel), w.e.walker@tudelft.nl (W.E. Walker), J.Spirco@tudelft.nl

(J. Spirco), w.a.h.thissen@tudelft.nl (W.A.H. Thissen).1 Tel.: +31 152788487; fax: +31 152786233.2 Tel.: +31 152785122; fax: +31 152786233.3 Tel.: +31 152786344; fax: +31 152783741.4 Tel.: +31 152786607; fax: +31 152786233.

Contents lists available at ScienceDirect

Futures

jou r nal h o mep ag e: w ww .e lsev ier . co m / loc ate / fu tu r es

0016-3287/$ – see front matter � 2011 Elsevier Ltd. All rights reserved.

doi:10.1016/j.futures.2011.06.002

In this manner, flexibility and related concepts have been proposed as ways to deal with the uncertainties that the futureholds. Just to give some examples from different viewpoints: Allen and Torrens [1] suggest, in the course of a theoreticaldiscussion introducing their special issue, that ‘‘adaptability and flexible response’’ are success factors in strategies, ratherthan ‘‘prediction, planning and control’’. In that same issue, Artigiani [2] argues from examples from naval history thatsystems need to be adaptive to survive under changing circumstances. Walker et al. [3] discuss a number of articles thatoutline actual adaptive approaches to deal with the deep uncertainties of the future.

Thus, having established that research on flexible infrastructures could benefit society, and acknowledging that it isresearched from various perspectives, what about research on the concept of flexibility in this context? To assess whatpossible solutions can be found to make infrastructures more flexible, it is important to make clear what exactly is meant bya flexible infrastructure. Only then can the investigation be systematic and the findings discussed in comparison. In otherwords, clarifying the concept can make research on what it refers to more scientific.

This article puts forward a conceptualisation of infrastructure systems, presenting a framework drawing from the field ofpolicy analysis [4], as well as the study of societal transitions [5] and socio-technical systems [6]. Using this, flexibility and therelated concepts are framed and an encompassing definition for each and all is constructed. Which is to say, each definition ofeach concept will have the same form. In this manner, the concepts will not only be defined by their definition, but also bycontrast – by how the definitions differ.

To demonstrate the validity and usefulness of this framework in a scientific manner, these differences will be comparedwith ‘empirical reality’. Because ‘‘meaning is use’’, as Wittgenstein [7] already knew, the empirical test of a definition wouldbe comparing it with the actual use of the concepts it refers to. In this article however, the empirical check will be whetherthe differences between the definitions reflect the differences in use of the concepts. Though this might seem an indirectapproach, it is actually more appropriate for the aim of this article, since here the definitions are proposed for a specificcontext – infrastructures – which might cause them to deviate from other formulations.

The use of the concepts in scientific literature – the relevant empirical reality here – will be explored systematically usingdata-mining techniques. The ISI database will be queried for articles containing the word ‘flexibility’, its relatives, and theword ‘infrastructures’. The titles and abstracts retrieved will be analysed for their latent semantics in order to identifydiscourses. From these discourses, the one that is ‘closest to home’ in terms of research field will be selected, i.e. the discourseclosest to policy analysis for infrastructure systems. Subsequently, within this discourse, the difference in use of flexibilityand related concepts is studied by analysing their co-occurrence with certain disambiguating words – words that onetypically associates with, say, flexible, but not with adaptive.

2. Context and conceptualisation

2.1. Context

Flexibility and flexible, derive from the Latin verb flectere, meaning to bend. This suggests already a number of things. First,that flexibility is a property, an ability of something, something endowed with a capacity, be it metaphorically, to bend, to changeshape. Second, bending implies that some external force is applied; the something is flexible to some other thing or influence.Third, bending contrasts with breaking, suggesting that whatever is flexible can undergo change without changing itself.Fourth, bending is antonymic to stiffening, or becoming rigid, which are also ways to undergo change without changing.

The first meaning of flexible – and the one with the oldest uses – in the Oxford English Dictionary (second edition, 1989)states:

A. adj.

1. a. Capable of being bent, admitting of change in figure without breaking; yielding to pressure, pliable, pliant.1548 HALL Chron., Edw. IV, 212 Like a rede with every wind is agitable and flexible. 1562 W. BULLEIN Bk. Sicke Men

81a,Feele also the pacient..whither the partes be pained, or flexable, or haue loste their strength and are stiffe. 1606SHAKES. Tr. & Cr. I. iii. 50 When the splitting winde Makes flexible the knees of knotted Oakes. 1626 BACON Sylva §796And you shall finde. the Stalke harder and less Flexible, than it was. 1664 POWER Exp. Philos. I. 42 It hath a Cartilaginousflexible Tube or Channel. 1731 ARBUTHNOT Aliments ii. (1735) 40 An Animal, in order to be moveable, must be flexible.1802 BINGLEY Anim. Biog. (1813) II. 373 These parts, with the tail, are covered by a strong flexible skin. 1823 W.PHILLIPS Introd. Min. Introd. 9 A flexible granular quartz is found in Brazil. 1874 BOUTELL Arms & Arm. ii. 17 The long,flexible and pointless weapons that are described by the Roman historians.

Words with a meaning close to flexible, or having overlap with it, are abundant, even when one restricts oneself to thecontext of infrastructure systems. Think for instance of words like robustness, adaptivity, resilience, and such. All these termsseem to entail a form of dealing with changing or changed circumstances. The differences in meaning appear to lie either inhow this is accomplished or in the nature of the change in circumstances. For example, resilience seems to suggest thecapability to resume normal functioning after a shock, whereas adaptivity seems much more to imply a change in normalfunctioning in response to a more permanently changed situation.

Before the differences and commonalities can be elaborated in a more systematic way, it is necessary to clarify what thesubject is to which these properties apply. In other words, it is time to ponder somewhat on the topic of infrastructuresystems.

J. de Haan et al. / Futures 43 (2011) 923–933924

2.2. Conceptualisation of infrastructure systems and flexibility-like concepts

2.2.1. Infrastructure systems

Infrastructures are always part of a larger societal constellation. Such a constellation exists for a reason; it is there to meeta societal need. To do so, or at least attempt to do so, an institutional structure arranges how the infrastructure is used tomeet the societal need. This how is referred to as the functioning of the constellation of an infrastructure and its institutionalstructure.

That the constellation around an infrastructure functions to meet a societal need, does not imply it does so flawlessly.Moreover the relation between its functioning and the needs to be met need not be constant. The societal need itself can befurther specified as a demand for certain products or services, and requirements on the way these are delivered. For example,an energy supply infrastructure functions to meet a societal need for energy, as well as to do so in a green or safe way.

Constellations do not exist in isolation, they are surrounded by an environment. From the environment, resources areobtained and waste is deposited. The environment can, of course, contain other infrasystems with which exchange ofresources and waste is possible as well. Just as the relation between the functioning and the societal need could change, socan the relation with the environment. Similarly, the match between the constellation and its environment need not beperfect, resulting in shortages, pollution, or adverse interactions with other infrasystems.

This way of conceptualising infrasystems is much akin to the way societal systems are conceptualised in transitionstudies [8]. There, constellations are subsystems of larger societal systems, meeting societal needs in specific ways. Also theway a constellation here is enveloped in an environment resembles the way societal and socio-technical systems findthemselves in a landscape [6]. This environment is also close to the external forces distinguished in the system description inpolicy analysis [4]. There, also outcomes of interest are identified, which are much like the functioning described here,especially since they too need not be precisely the intended outcomes. The policymaking process [4] intends to close the gapbetween the actual functioning of the infrastructure constellation and the societal need. In policy analysis, the policymakers,or rather the entire policymaking process is viewed as external to the system. Although the framework proposed here doesnot explicitly include the policymaking process, nor policymakers for that matter, this viewpoint will be assumed herewhenever relevant.

2.2.2. Flexibility and flexibility-like concepts

Now, with the terminology of, and the idea behind, infrastructures constellations introduced, all is set to frame flexibilityand its relatives. Flexibility and the concepts with which it shares a family resemblance apparently all denote ways tocontinue meeting a societal need under changing circumstances. Circumstances can change in:

� the relation of the infrastructure constellation with the environment,� the internal relations in the infrastructure constellation,� the relation of the infrastructure constellation with the societal need.

Changing circumstances can obviously be further distinguished into disturbances on long (trends) or short (shocks) timescales.Thus, flexibility, adaptivity, et cetera, are all qualities that an infrastructure can have to maintain functioning – although

perhaps differently – to meet a societal need, when circumstances change. In other words, to define what such a flexible-likequality is, it is necessary to describe what an infrastructure constellation does when dealing with changing circumstances in aflexible-like way. What does an infrastructure system have to do to maintain functioning? Does it need to alter itscomposition, be partially reconstructed? Is it prepared or can it recover? Does it need to be put to work differently? Does ithave enough surplus capacity, or otherwise margins, to cope with the change?

3. Defining, refining, and disambiguating

With the stage set for the infrastructure-constellation specific definitions, consider now some typical definitions of someflexible-like concepts that come from the literature. These four are the most frequently co-occurring flexible-like terms inthe literature considered, that are in the relevant discourse (see Section 4, for details). The definitions here are typical, tocatch the common connotations of each concept. But, to be more or less applicable to infrastructure systems, they are notnecessarily the standard or most-used ones.

� Flexibility

Strategic flexibility ‘‘(. . .) may be defined as the ability of the organization to adapt to substantial, uncertain, and fast-occurring (relative to required reaction time) environmental changes that have a meaningful impact on the organization’sperformance’’ [9].� Adaptability

Is the capacity for ‘‘adjustment in ecological, social, or economic systems in response to actual or expected climaticstimuli and their effects or impacts’’ [10].� Resilience

J. de Haan et al. / Futures 43 (2011) 923–933 925

1. ‘‘(. . .) refers to the tendency to rebound or recoil, showing buoyancy or recuperative power, and the capability towithstand shock without damage or rupture’’ [11].

2. For ecosystems, ‘‘(R)esilience is the capacity of a system to absorb disturbance and reorganize while undergoing changeso as to still retain essentially the same function, structure, identity, and feedbacks’’ [12].

� Robustness

1. (. . .) of a domain of interest is the ability to sustain a disruption [13].2. In complex systems is ‘‘(. . .) the maintenance of some desired system characteristics despite fluctuations in the behavior

of its component parts or its environment’’ [14].

3.1. Defining

The proposed definitions appear in Table 1.First some notes on these definitions. As one sees, they all have the structure of an infrastructure constellation able to do

something, referring to the flexible-like quality, to keep on or resume meeting a societal need, under changed circumstances.The structure of the definitions is the same on purpose to make the differences among them more apparent.

Also, there is mention of meeting a societal need under changed circumstances. However, one of the circumstances thatcould change is of course the societal need. Indeed, in the section on conceptualisation of infrastructure constellations thiswas mentioned as one the places where circumstances can change. Since the societal need is commonly described on a ratherhigh level of abstraction, like ‘energy supply’ or ‘public transportation’, speaking of, for instance, a similar societal need, couldintroduce unwanted connotations.

The definitions are as bare as reasonably possible for their intended application. Many refinements are obviously possible.Refinements on a set of definitions of related concepts can have two consequences: (1) increased precision, and thus morelimited applicability; (2) disambiguation, in which the differences between the concepts are amplified.

Using the examples from literature as a guide for reasoning, the next step is to refine the definitions and map out thevarious shades and hues in which the concepts can appear. This will lead to a number of disambiguating terms, with whichthe concepts can be delineated more clearly for certain applications, when desired, and with which the differences can beidentified sharper. These disambiguating terms are the stepping stones to the empirical, data-mining, approach where thescientific literature is probed for the common usage of the flexible-like concepts and their common connotations.

3.2. Refining

That flexibility and adaptability have a lot in common appears logical. Evans [11] in his literature review cites Stigler [15]to point out that adaptability ‘‘. . . differs from flexibility in that adaptability implies a singular and permanent adjustment toa newly transformed environment, whereas flexibility enables successive, but temporary approximations to this state ofaffairs.’’ This could also be thought of as flexibility being a capacity to change functioning that is already present in thesystem, whereas adaptability is the capability to change the structure of the system in order to change its functioning.

In examples from literature, the difference between robustness and flexibility appears to mainly lie in that robustnessdeals with change without changing the system. This, however, also is the ring that resilience has to it. It appears thatresilience is, so to say, a higher dimensional concept. Especially when it concerns ecological resilience [12], robustness can bethought of as one of the dimensions of resilience (dubbed ‘latitude’). Robustness and engineering resilience can then bethought of as roughly synonymous, with resilience more related to speed-of-return and robustness with, for example, safetymargins. A connotation that resilience seems to have retained from its dictionary meaning is that of the ability to ‘bounceback’ after a perturbation.

Especially resilience appears to differ from flexibility in the sense that the change it responds to is more abrupt. Alsoflexibility and adaptability seem to be more anticipatory qualities, allowing a system to change along with the change that itencounters.

Then there is the distinction between anticipated and unanticipated. Although anticipated might have the connotation ofplanning to it, this is not necessary. The ability of a system to respond to unanticipated changes could be coincidental. Also,anticipated suggests that there is knowledge of the future, and this is also somewhat misleading. For example, in the context

Table 1

Proposed definitions of flexible-like concepts.

� A flexible infrastructure constellation. . .

. . .can be employed differently to keep on meeting a societal need under changed circumstances

�An adaptive infrastructure constellation. . .

. . .can be altered to keep on meeting a societal need under changed circumstances

� A resilient infrastructure constellation. . .

. . .can resume meeting a societal need under changed circumstances

�A robust infrastructure constellation. . .

. . .can keep on meeting a societal need under changed circumstances

J. de Haan et al. / Futures 43 (2011) 923–933926

of policymaking, robustness is a desired quality, which means that the policy performs acceptably well for a range of possiblefutures [16]. Here, anticipation acknowledges a lack of knowledge of the future. However, in recent literature onpolicymaking under deep uncertainty [17] the additional demand is made that a policy is also robust under unanticipatedchanges. For this, the approach to policymaking is made adaptive and the quality of the resulting policy is consequently morethan robust, namely adaptively robust.

The kind of change the system responds to in a flexible-like manner can also bring some refinement to the concepts. Thedynamics of the change, for example. Is what is happening to the system, happening to it abruptly or gradually? Is it a shockor disturbance after which the situation returns to normal and does the system consequently only need to withstand it orrecover from it? Or, is it a trend-like change, a development leading to a permanently altered state of affairs that needs to befaced from the moment of its onset?

Another aspect to the concepts, reflected in the proposed definitions already, is that of whether the system is alteredwhen it responds in flexible-like manner. In other words, can the system accommodate the changes or does it needs toadjust? Are its margins broad enough to warrant satisfactory functioning under the changed circumstances or can it alter theway it is functioning to deal with them? And, after the fact, does the system do more or less the same what it did before, orsomething qualitatively different?

3.3. Disambiguating

To be able to use data-mining techniques, the distinction, or distinctions, need to be made in terms of disambiguating

words – words that clearly point in the direction of one of the flexible-like words and away from another. This idea ofdirection will be exploited more later on. Thus, one needs words that one associates with a certain flexible-like term – thatcarry similar connotations – and not with another.

Differences, and even nuances in meaning, such as those pointed out above, can be characterised with dichotomies. Thatis, a certain difference can be described by a certain word pair that is different in the same way. In other words, the differencein definition of a tortoise and a hare will entail the dichotomy slow – fast. Thus, in line with the small discussion of thedifferences of the flexible-like concepts, the following three main dichotomies appear relevant:

� anticipation: recovery (or anticipated – unanticipated).� disturbance: change.� accommodation: adjustment.

Anticipate implies planning, intent, or, in any case, actions or initiatives before the fact, as opposed to recover from, whichentails recuperation or restoring to normal functioning after the fact.

Disturbance is to be understood as a shock or disruption from the normal of a temporary character, a sudden peakdemand or shortage of resources, after which the normal situation is restored. Change, however, is understood asaltering the situation permanently and taking place more slowly. Disturbance vis-a-vis change is the conflation of twocriteria of distinction, namely temporary–permanent and fast–slow. Since a very slow shock feels as a trend-likedevelopment and two rapidly evolving trends, moving away from, and back to, normal, appear as one disturbance, thisconflation is justified.

Adjustment implies that, to keep on functioning, something be altered, as opposed to accommodation, which suggests thatthere is enough margin allowing normal functioning to be sustained in the changed situation.

Obviously other dichotomies can be used as well to distinguish among the concepts, like normal – new, referring to thestate of the system after the facts. Or, swiftly – gradually, referring to the speed of that process. In the following some of thesedichotomies will be studied empirically with data-mining techniques.

4. Data mining

4.1. Outline of the approach

The empirical approach consists essentially of two steps:

1. Identifying the discourses

There are many fields of research, and consequently many discourses. It is possible that the usage of a flexible-likeconcept in a certain discourse that is remote from that on infrastructures does not match the proposed definitions at all.Moreover, the proposed definitions would be useless for those research fields. Consequently, it is necessary to find outwhich discourses are close enough for the proposed definitions to be of use.

2. Probing the concepts with the disambiguating words

In addition to the core definitions in Table 1, it is interesting to know whether a certain flexible-like concept has, when itcomes to its meaning, an inclination towards or away from a certain disambiguating term. For example, whether ‘adaptivity’is an ‘anticipation’ or a ‘recovery’ inclined word. That is, does it have the connotation of anticipation or recovery? For this, the

J. de Haan et al. / Futures 43 (2011) 923–933 927

idea of direction needs to be implemented for words, so one can speak of the ‘angle’ between, and thus inclination to or from,words.

A preliminary step was to identify which flexible-like words were to be taken into account in the further analysis. To doso, a thesaurus [18] was used to create a list of 23 such words. The analysis was done on scientific articles, or rather their titlesand abstracts. The data were gathered from the ISI Web of Science database, querying journal records from 1991 onwards (ISIstarted indexing abstracts then, in addition to titles and such). An initial data set was created using the query (flexibl* ORadapt* OR resil* OR robust* OR agil*) AND infrastr*, with the star indicating that the stem of the word was used. The setcontained 11,029 articles. Stop words (words considered to carry little meaning) were removed on the basis of a list of 328stop words. In this set, the 100 most frequently occurring words, excluding the 23 flexible-like words, were identified, andthe frequency of the 23 flexible-like words was counted.5 Results are tabulated in Tables 2 and 3.

That words like infrastructure and system occur frequently is hardly surprising, and the presence of words such as network

and data suggests much ICT-related content. As regards the flexible-like words, it is remarkable that after robust* thefrequency of these words drops quite abruptly. A reason for this might be that adapt*, flexibl*, and robust* are used genericallythroughout scientific literature, whereas words like resil*, agil*, and others, have much more specific usage, perhaps specificto certain research areas.

4.2. Step 1. Identifying the discourses

The first step consisted of mapping whether the word flexible and its relatives, in the context of infrastructures, was usedin distinct ways. That is, whether discourses could be identified, for example pertaining to specific research areas. This entailssome way of identifying the similarity in meanings of a flexible-like word over various articles. Taking Wittgenstein’s‘‘meaning is use’’ seriously implies here that patterns of words are to be used as the semantic indicator. To this end, co-wordanalysis can be applied. Its use for science mapping, to which the analysis here is much akin, goes back to Bauin [19]. It hasbeen widely used, and especially the work of Peters and van Raan is exemplary [20,21].

Co-word analyses traditionally rely on factor-analytic techniques [22]. These techniques assume that there is a hiddenstructure to the data, a latent semantics that can be revealed by a factor analysis. A well-known approach using this line ofthinking, from the field of document retrieval, is latent semantic indexing [23,24] or LSI, in short. In this article an extensionof this approach is employed, the MFA, or mixtures of factors analysers [25,26]. The MFA uses a generalisation of the factoranalysis allowing multiple centroids of the data. What this means is that, in contrast to a standard factor analysis, multiple‘regions’ with differing patterns of words, and thus different semantic structures, can be identified. These regions are thediscourses mentioned earlier. Kwakkel and Cunningham [27] show that the MFA is a useful approach especially whendealing with polysemy and synonymy. They also point out a number of other reasons why this approach is attractive for thecase at hand. For example, that it has the ability to reduce high-dimensional data, as well as to simultaneously revealdiscourses and latent semantics.

Some parameter choices need to be made when using the MFA, for example, the number of discourses and factors towhich to fit the data has to be supplied. The choice of parameters is not trivial, and over fitting needs to be avoided, sincemore parameters might produce a ‘better’ fit but less insight or an invalid model when extra data become available. To thisend, various techniques are available, among which are metrics of model quality, for example Akaike’s [28] InformationCriteria (AIC) and the Bayesian Information Criterion (BIC) [29]. Using the latter criterion, the MFA was probed with the

Table 2

Frequently occurring words.

Most frequently occurring words Frequency

Infrastructure 9066

System 7320

Network 6193

Data 6142

Systems 5209

Table 3

Frequently occurring flexible-like words.

Most frequently occurring flexible-like words Frequency

Adapt* 6112

Flexibl* 4616

Robust* 2577

Resili* 601

Agil* 473

5 The analysis was carried out using Python and Matlab scripts. These are available from the authors upon request.

J. de Haan et al. / Futures 43 (2011) 923–933928

number of discourses and factors both ranging from 1 to 30. This revealed that 4 discourses and 20 factors was the bestparameterisation.

The discourses and their leading journals are:

� Discourse 1

World Development, American Journal Of Physical Anthropology, Journal of the American Planning Association,Climate Research, Environmental Science and Policy.� Discourse 2

Fiber and Integrated Optics, IEEE Communications Magazine.� Discourse 3

International Journal of Software Engineering and Knowledge Engineering, Future Generation Computer Systems – TheInternational Journal of Grid Computing, IEEE Conference on Web Services.� Discourse 4

Wireless Networks, Bell Labs Technical Journal, IEEE-ACM Transactions on Networking.

The conclusion was that Discourse 1 was the relevant one in the context of infrastructure constellations. All analysespresented in the remainder of the paper were performed on this subset of articles. It turned out that the concept ‘‘agility’’ washardly, if at all, used in this discourse. Consequently the focus was on the four most frequently occurring flexible-like words:flexibl*, adapt*, resil*, and robust*.

4.3. Step 2. Probing the concepts with the disambiguating words

The proposed definitions in Table 1 can be thought of as core definitions. Obviously, each flexible-like concept carriesmore connotations and subtle nuances of meaning than can be precipitated in the kind of bare sentences of these proposeddefinitions. What can be done, however, is to probe empirically what the connotations of the concepts are. As was alreadydiscussed in Section 3.2 Refining, this can be done using disambiguating terms. By means of a dichotomy, such as anticipation

vs. recovery, one can find the side of the dichotomy with which the concept has the most affinity.

4.3.1. Meaning as direction

The method to go about determining a concept’s affinity is fairly straightforward. For each dichotomy that one wouldlike to probe, two lists of words are made. One list is made for each side of the dichotomy, which contains words that areclose in meaning, have the same ring to it, have the same connotations, or however one wants to put it. For example, inthe anticipation vs. recovery case, one makes a small list of words that come to mind for the term anticipation and a smalllist of words for recovery. The resulting lists are then made more consistent using a thesaurus [18, in the case of thisarticle]. The lists used in this paper can be found in the Appendix. Special attention needs to be paid to thedisambiguating character of the words in the list. That is, when disambiguating between accommodate and adjust, theword ‘change’ can readily be found as being related to both terms. It, therefore, does not help in the disambiguation; infact it obscures the distinction.

Next is to probe whether a certain flexible-like concept points more in the direction of one end of a dichotomy or the other;that is, whether the concept is more aligned with the words in one list than the other. This gives insight into whether, forexample, resilience is more closely related to ‘recovery’ than ‘anticipation’. The idea of direction and alignment can be takenquite literally. For this, the space of words in the articles used in Cluster 1, needs to be considered a vector space of sorts. Inthe approach taken here, this implies considering each distinct word a dimension of this ‘meaning space’. The point a certainword occupies in this space is then defined by how often it co-occurs with each other word. So each word is represented by avector that has as components the number of times it co-occurs with the other words. However, that some words are morecommonly used than others, might introduce a bias here. If a word occurs more often, it will probably co-occur more often aswell. To correct for this, the total number of occurrences per word is divided out.

For each dichotomy, two specific vectors need to be defined – those pointing to the two sides of the dichotomy. Theseare simply vectors containing ones for the words that are in their list of connotation words and zeroes for the rest. Theangles between a certain flexible-like word and both ends of the dichotomy can now be calculated. An angle of zerodegrees with one end would imply complete alignment, that is, exactly matching the connotations as suggested in thelists corresponding to that end of the dichotomy. Note that the angle with one end of the dichotomy says relatively littleabout the angle with the other end. That is, a word having a large angle (i.e. not very similar) with, say, anticipation doesnot imply that it is very similar to the other end of the dichotomy, in this case recovery. This is because the space is highlydimensional.

How can the angle between two words be calculated? Since the flexible-like words and the dichotomies are allrepresented by vectors, the straightforward way seems to be to use the angle defined by their (Euclidian) inner product.6

6 To obtain the angle between two vectors in this manner, one takes the inner product of the two, and divides this by the product of their lengths. The

angle is then the arccosine of this result.

J. de Haan et al. / Futures 43 (2011) 923–933 929

Coincidently, this also much akin to a cosine similarity measure [see e.g. 30]. For clear presentation, in this paper the choice ismade to scale the retrieved angles between zero and 100%, with zero meaning that one vector points in the opposite directionof the other and that the words are as unrelated as can be measured in this fashion. 100%, of course, means that the wordsexactly match in terms of connotations.

4.3.2. Data-mining results

The four flexible-like concepts (flexibility, adaptivity, resilience, and robustness) were compared with four dichotomies:anticipation vs. recovery; disturbance vs. change; accommodate vs. adjust; and swift vs. gradual. The word lists used aretabulated in Appendix. The resulting calculated angles for all pairs of words are summarised in Tables 4–7.

The first thing one notices is that there are hardly any ‘extreme’ angles, and, for instance in Table 5, the angles sometimesare rather close together. It is therefore even more important to interpret these results with care and modesty. That resilience

in Table 4 was an 11.4% anticipation word and a 41.2% recovery word, does not give much insight. However, noticing thatflexibility is more than five times as much on the anticipation side than resilience, which is some two times as much on therecovery side, does hold clues for the interpretation of the differences between these concepts. This is the way the results willbe interpreted – by exploring how they make the contrasts between the concepts more clear. The results are discussed in thenext section, where also the connection will be made with the intuitive connotations, which were the basis for thedichotomies used, and with the definitions proposed in this paper.

5. Conclusions and thoughts

5.1. Discussing the results

It is interesting to see that some intuitive connotations of flexible-like concepts are in line with their apparent use in theliterature and others not at all.

Table 4 was reassuring. One of the intuitions that was quite strongly confirmed was that flexibility differs from adaptivityin that the former is about anticipating change and the latter about recovering from it. Resilience was considered very much a

Table 5

Angles for disturbance vs. change (%).

Disturbance Change

Flexible 37.3 45.4

Adaptive 37.2 54.2

Robust 30.8 41.5

Resilience 35.3 10.4

Table 4

Angles for anticipation vs. recovery (%).

Anticipation Recovery

Flexible 59.0 20.7

Adaptive 25.9 36.9

Robust 31.8 37.7

Resilience 11.4 41.2

Table 6

Angles for accommodate vs. adjust (%).

Accommodate Adjust

Flexible 23.7 37.3

Adaptive 33.6 21.1

Robust 34.3 21.0

Resilience 24.3 6.5

Table 7

Angles for swift vs. gradual (%).

Swift Gradual

Flexible 49.5 28.0

Adaptive 47.6 17.3

Robust 40.3 31.8

Resilience 25.3 17.8

J. de Haan et al. / Futures 43 (2011) 923–933930

quality of recovery rather than anticipation, which also shows from the results. That robustness is generally indifferent to thedichotomy between anticipation and recovery also raised no eyebrows.

The results in Table 5 on the dichotomy disturbance (shock, short term) vs. change (gradual, trend like), also confirmedintuition. Resilience proved to be used most in the context of disturbances and far less in the context of more gradual change.Flexible and adaptive are, as expected, qualities related to change on the longer term, as is robustness.

The results in Table 6 are more subtle. Flexibility is more related to adjust than to accommodate, which was not expected.The hypothesis was that flexibility is an accommodate word, because a flexible system is not expected to change itself inresponse to changed circumstances. This ‘changing itself’ was considered to be captured with ‘adjust’ and similar words.Perhaps the subtlety of changes made in the system vs. change in the way it deals with changed circumstances is difficult tocapture in this manner.

That adaptation is an accommodation word, rather than an adjust one, was only surprising at first glance. To adapt is toadjust, right? However, recalling that Discourse 1 contained many climate-change related journals clarified matters. In thisdiscourse, the two ways of dealing with climate change are mitigation and adaptation, where the first refers to stoppingclimate change and the latter refers to accommodating to climate change – accommodation by adaptation. Resilience androbustness, by being both more accommodation than adjust words, conform more to intuition here.

Table 7 shows how difficult to interpret the results become when there is ambiguity concerning the disambiguatingterms. The dichotomy was intended to differentiate between a swift or gradual response to the changed circumstances. Theseand similar terms, however, are of course also used to refer to those changes themselves. Thus that resilience is a swift wordstill makes sense, since it often refers to the bouncing back of a system after a shock. Adaptation was thought to have stronggradual connotations, for example, because of its relation to the evolutionary discourse. However, its usage is apparentlymore in swift terms.

5.2. Revisiting the definitions

All in all, the conclusion is that the data-mining approach as taken in this article is very useful in exploring theconnotations of concepts. It provides an empirical tool for conceptual analysis. What it does not do for the conceptualanalyst, though, is produce the definitions. These needed to be constructed on the basis of a tentative scan of usage in theliterature and the demands of the field of application. Moreover, as mentioned before, the definitions are to be regarded ascore definitions, they are as basic as possible. The nuances, and more importantly, the defining differences, can then comefrom the data mining. Thus, also concluding this paper, the contrasts can be brought to light in following way:

Flexibility and adaptivity

. . .are both concepts referring to changes on the longer term. The former, however, carries the connotation that thesystem is already prepared for the change and only needs to be employed differently, whereas the latter is associated with anability to change along with the circumstances.

Flexibility and resilience

. . .differ in the type of response they are used to describe, flexibility associated with being prepared and longer timescales,and resilience being associated with bouncing back, recovering from shocks and disturbances.

Flexibility and robustness

. . .are both concepts used in the context of longer timescales, differing in their use in that flexibility anticipates ratherthan recovers, whereas robustness is more or less indifferent to that distinction.

Adaptivity and resilience

. . .are both associated with recovering. Resilience, however, is more about recovering from change on short timescales,whereas adaptivity is about changing along with the circumstances on longer timescales.

Adaptivity and robustness

. . .are both concepts used in the context of longer timescales, used differently in that adaptivity favours to recover, afterthe fact, rather than to anticipate, before the fact, whereas robustness apparently has no preference in this.

Resilience and robustness

. . .differ in that robustness is indifferent with respect to recovery or anticipation, whereas resilience has a preference forrecovery. Moreover, resilience is associated with short timescales, shocks and disturbances, in contrast with robustness,which has an inclination towards longer timescales.

Appendix: Lists of words used for disambiguation

Dichotomy Anticipation Recovery Dichotomy Disturbance Change

Connotations Anticipate Recuperate Connotations Disaster Development

Advance Recover Disasters Developments

Blueprint Recovery Shock Scenario

Plan Disaster Shocks Scenarios

J. de Haan et al. / Futures 43 (2011) 923–933 931

Appendix: Lists of words used for disambiguation (Continued )

Dichotomy Anticipation Recovery Dichotomy Disturbance Change

Planning Shock Disturbance Trend

Prior Aftermath Disturbances Trends

Prepare Catastrophe Aftermath Extrapolation

Margins Upheaval Sudden Longterm

Preliminary Disruption Quick Gradual

Prognosis Disturbance Fast Stretch

Earlier Event Disruption Slow

Forecast Incident Disruptions Period

Forecasting Disaster Continuing

Model Disasters Long

Modelling Prompt Prognosis

Modelling Short Foresight

Foresight Radical Forecast

Policy Disruptive Forecasting

Future

Dichotomy Accommodate Adjust Dichotomy Swiftly Gradually

Connotations Accommodate Adjust Connotations Swift Gradually

Margin Reform Fast Gradual

Buffer Restructure Rapid Piecewise

Shield Restructured Immediate Slow

Safeguard Restructuring Direct Late

Margins Transformation Quick Incremental

Buffers Modify Instantaneous Stepwise

Shields Alter Responsive Inert

Safeguards Mutation Shortterm Impervious

Boundary Mutations Prompt Creeping

Scope Reorganisation Speedy Lagging

Allow Rearrange Abrupt Reluctant

Tolerate Reorder Nimble Tardy

Protect Reordered Bounce Prolonged

Maintain Reordering Prompt Long

Maintaining Reorient Brief Developing

Defend Reoriented Light Continuous

Bound Remodel Steady

Cope Remodeled

Coping Remodeling

Withstand

Withstanding

Endure

Repel

Repulse

References

[1] P.M. Allen, P.M. Torrens, Knowledge and complexity, Futures 37 (2005) 581–584.[2] R. Artigiani, Leadership and uncertainty: complexity and the lessons of history, Futures 37 (2005) 585–603.[3] W.E. Walker, V.A.W.J. Marchau, D. Swanson, Addressing deep uncertainty using adaptive policies: introduction to section 2, Technological Forecasting and

Social Change 77 (2010) 917–923.[4] W.E. Walker, Policy analysis: a systematic approach to supporting policy making in the public sector, Journal of Multi-criteria Decision Analysis (2000) 11–

27.[5] J. Rotmans, R. Kemp, M. van Asselt, More evolution than revolution: transition management in public policy, Foresight (2001) 17.[6] A. Rip, R. Kemp, Technological change, in: S. Rayner, E.L. Malone (Eds.), Human Choice and Climate Change, Battelle Press, Columbus, OH, 1998, pp. 327–399.[7] L. Wittgenstein, Philosophical Investigations, Revised 4th ed., Wiley-Blackwell, 2009 (1953).[8] J. de Haan, J. Rotmans, Patterns in transitions: understanding complex chains of change, Technological Forecasting and Social Change 78 (1) (2011) 90–102.[9] D.A. Aaker, B. Mascarenhas, The need for strategic flexibility, The Journal of Business Strategy 5 (1984) 74–82.

[10] B. Smit, O. Pilifosova, Adaptation to climate change in the context of sustainable development and equity, in: Climate Change 2001: Working Group II:Impacts, Adaptation and Vulnerability, IPCC, 2001, pp. 877–912.

[11] J.S. Evans, Strategic flexibility for high technology manoeuvres: a conceptual framework, Journal of Management Studies 28 (1991) 69–89.[12] B. Walker, C.S. Holling, S.R. Carpenter, A. Kinzig, Resilience, adaptability and transformability in social–ecological systems, Ecology and Society 9 (2004) 5.[13] M.D. Quirk, S.J. Fernandez, Infrastructure robustness for multiscale critical missions, Journal of Homeland Security 2 (2005) 2.[14] J.M. Carlson, J. Doyle, Complexity and robustness, Proceedings of the National Academy of Sciences (PNAS) 99 (2002) 2538.[15] G. Stigler, Production and distribution in the short run, Journal of Political Economy 47 (1939) 305–327.[16] W.E. Walker, S.A. Rahman, J. Cave, Adaptive policies, policy analysis, and policy-making, European Journal of Operational Research 128 (2001) 282–289.[17] W.E. Walker, R.J. Lempert, J.H. Kwakkel, Deep Uncertainty, in: S.I. Gass, M. Fu (Eds.), Encyclopedia of Operations Research and Management Science, Springer,

forthcoming.[18] Roget’s, Roget’s International Thesaurus, sixth ed., HarperCollins Publishers, New York, 2001.[19] S. Bauin, Aquaculture: a field by bureaucratic fiat, in: M. Callon, J. Law, A. Rip (Eds.), Mapping the Dynamics of Science and Technology, MacMillan Press Ltd.,

London, 1986.[20] H.P.F. Peters, A.F.J. van Raan, Co-word-based science maps of chemical engineering. Part I: representations by direct multidimensional scaling, Research

Policy 22 (1993) 23–45.

J. de Haan et al. / Futures 43 (2011) 923–933932

[21] H.P.F. Peters, A.F.J. van Raan, Co-word-based science maps of chemical engineering. Part II: representations by combined clustering and multidimensionalscaling, Research Policy 22 (1993) 47–71.

[22] M.J. Greenacre, Theory and Application of Correspondence Analysis, Academic Press, London, 1984.[23] S. Deerwester, S.T. Dumais, G.W. Furnas, T.K. Landauer, D. Harshman, Indexing by latent semantic analysis, Journal of the American Society for Information

Science 41 (1990) 391–407.[24] K. Borner, C. Chen, K.W. Boyack, Visualizing knowledge domains, Annual Review of Information Science and Technology 37 (2003) 179–255.[25] Z. Ghahramani, G.E. Hinton, The EM Algorithm for Mixtures of Factor Analyzers, Department of Computer Science, University of Toronto, 1997.[26] G.J. McLachlan, D. Peel, Finite Mixture Models, John Wiley & Sons, 2000.[27] J.H. Kwakkel, S. Cunningham, Managing polysemy and synonymy in science mapping using the mixtures of factor analyzers model, Journal of the American

Society for Information Science and Technology 60 (2009) 2064–2078.[28] H. Akaike, A new look at the statistical model identification, IEEE Transaction on Automatic Control 19 (1974) 716–723.[29] G. Schwarz, Estimating the dimension of a model, The Annals of Statistics 6 (1978) 461–464.[30] W.P. Jones, G.W. Furnas, Pictures of relevance: a geometric analysis of similarity measures, Journal of the American Society for Information Science 38 (1987)

420–442.

J. de Haan et al. / Futures 43 (2011) 923–933 933