streco-561; article in presseprints.soas.ac.uk/19216/1/2013 structural learning.pdf · 2014. 12....

GS

So

ACU

ARRAA

JDDLO

KPLS

1

peic

DtACRfatSs

0h

ARTICLE IN PRESS ModelTRECO-561; No. of Pages 17

Structural Change and Economic Dynamics xxx (2013) xxx– xxx

Contents lists available at ScienceDirect

Structural Change and Economic Dynamics

jou rn al h om epage : www.elsev ier .com/ locate /sced

tructural learning: Embedding discoveries and the dynamicsf production�

ntonio Andreoni ∗

entre for Science Technology and Innovation Policy, Institute for Manufacturing, Department of Engineering, University of Cambridge,nited Kingdom

a r t i c l e i n f o

rticle history:eceived January 2012eceived in revised form July 2013ccepted September 2013vailable online xxx

EL classification:2083

a b s t r a c t

Production and learning of productive knowledge are profoundly intertwined processesas the activation of either process triggers the other, very often implying interdependenttransformations. The paper aims to open the ‘production black box’ by proposing the ana-lytical map of production as a tool for disentangling the set of interdependent relationshipsamong capabilities, tasks and materials. The concept of structural learning is introduced toidentify the continuous process of structural adjustment triggered and oriented by existingproductive structures at each point in time. Structural learning trajectories allow for thetransformation of structural constraints such as bottlenecks and technical imbalances into

2333

eywords:roduction processearning

structural opportunities. Complementarities, similarities and indivisibilities are essentialfocusing devices for activating compulsive sequences of technological change as well asdiscovering structurally embedded opportunities. The paper then investigates the tensionbetween structure and agency present in structural learning trajectories, and examines theform it takes in different productive organisations.

tructural dynamics

. Introduction

Production and learning of productive knowledge arerofoundly intertwined processes as the activation of

Please cite this article in press as: Andreoni, A., Structural learnition. Struct. Change Econ. Dyn. (2013), http://dx.doi.org/10.101

ither process triggers the other, very often implyingnterdependent transformations. Production theory hasonventionally explained production processes as relation-

� This paper develops a line of research initially presented at theIME workshop “Production Theory-Process, Technology and Organisa-

ion: Towards a useful Theory of Production” (LEM, Scuola Superiore Sant’nna, Pisa 8–9 November 2010). I am grateful to Patrizio Bianchi, Ha-Joonhang, Mike Gregory, Prue Kerr, Michael Landesmann, Eoin O’ Sullivan andoberto Scazzieri for comments and discussions, and to Guido Buenstorf

or comments on an earlier draft of this paper. Finally, I am thankful to twononymous referees for their valuable comments to the manuscript andheir constructive suggestions. The usual caveat applies. The Centre forcience, Technology and Innovation Policy gratefully acknowledges theupport of the Gatsby Charitable Foundation.∗ Tel.: +44 1223 339738.

E-mail address: [email protected]

954-349X/$ – see front matter © 2013 Elsevier B.V. All rights reserved.ttp://dx.doi.org/10.1016/j.strueco.2013.09.003

© 2013 Elsevier B.V. All rights reserved.

ships between combinations of productive factors – i.e.input quantities – and certain quantities of outputs. Byassuming that producers ‘know how’ certain inputs maybe combined and transformed to obtain certain outputs,production functions do not make any explicit referenceto the capabilities needed to perform real production pro-cesses. Thus, in standard production theory, there is noproduction process strictly speaking (Loasby, 1999). Notonly is the production process treated as a black box, alsothe learning dynamics occurring in given production struc-tures are fundamentally ignored. Indeed, economists oftentreat learning as a costless and automatic process function-ally dependent on cumulative output, time, or investment,whose main effect is to reduce average production costs.

A very influential attempt to cope with the fundamen-tal limitations of more conventional production models

ng: Embedding discoveries and the dynamics of produc-6/j.strueco.2013.09.003

can be found in the capability theory of the firm, anapproach that emerged at the intersection of variousresearch fields, specifically organisational studies (Marchand Simon, 1993; Penrose, 1959; Richardson, 1960, 1972;

dx.doi.org/10.1016/j.strueco.2013.09.003


http://www.sciencedirect.com/science/journal/0954349X

http://www.elsevier.com/locate/sced

mailto:[email protected]


ING Model

d Econo

Learning dynamics of this second kind tend to be triggeredby a series of factors which are both internal and externalto the firm (Malerba, 1992). In terms of the former, not

1 The long tradition in learning curve studies is usually associated withthe empirical analysis of ‘learning by doing’ effects on productivity and

ARTICLESTRECO-561; No. of Pages 17

2 A. Andreoni / Structural Change an

Teece, 1980; Langlois, 1992; Morroni, 2006; Pitelis andTeece, 2009; Jacobides and Winter, 2012), and institutionaland evolutionary economics (Nelson and Winter, 1982;Cohen and Levinthal, 1990; Lundvall, 1992; Dosi et al.,2000), and empirical work in development economics (Bell,1982; Lall, 1992). With a particular focus on the trans-formation of cognitive contents and evolving capabilities,these contributions have shown how the knowledge ofproductive possibilities – i.e. input combinations – has tobe complemented by the availability of relevant capabil-ities for productive tasks being performed. Most notablyevolutionary economics has highlighted the complex cogni-tive dynamics underlying learning processes. It has drawnattention to the multifaceted nature of knowledge, its tacitcomponents as well as the complexities connected to itscreation, diffusion, adaptation, adoption and accumulationin organisational ‘routines’.

By integrating the above mentioned research streamswith structural theories dealing with the complex ‘archi-tecture of production’, this paper analyses productionstructures in transformation, examining the embeddedconstraints and opportunities which are responsible forlearning dynamics. From this perspective, learning isunderstood as a dynamic process triggered and constrainedby existing production structures. This means that produc-tion structures set the stage for learning dynamics, thatis, they prepare human minds for the intuitive discoveryof new productive possibilities. The paper also recognisesthat structures of cognition and structures of productionare linked by a bundle of bidirectional transformative rela-tionships.

The goal of the present paper is two-fold. Firstly, thepaper embeds different forms of learning such as ‘learningby doing’ and ‘learning by using’ in production struc-tures. The paper therefore proposes an ‘analytical mapof production’ as a stylised representation of the sys-tem of interrelated tasks through which transformationsof materials are performed according to different pat-terns of capacities/capabilities coordination, subject tocertain scale and time constraints (Section 2). Within thisnew analytical framework, the second contribution of thepaper is to introduce the concept of ‘structural learning’.In conventional approaches learning is simply describedas a cognitive/behavioural dynamic involving productionagents. In contrast, in our analytical framework, learningis understood as a process through which ‘structural con-straints’ in production such as bottlenecks and technicalimbalances are transformed into ‘structural opportunities’.In this context, static and dynamic complementarities,as well as similarities and indivisibilities, are essentialfocusing devices for triggering compulsive sequences oftechnological change which permit the discovery of new‘worlds of production’ (Section 3). Productive possibilitieshave to be ‘seen’, that is discovered and ‘actualised’ by pro-ductive organisations, for structural learning to be feasible.The concept of structural learning highlights a fundamen-tal analytical tension between structure and agency or,


more specifically, between productive structures and pro-ductive agents (the latter including both individuals andcollectivities). Given the same productive structures, struc-tural learning may follow different patterns according to

PRESSmic Dynamics xxx (2013) xxx– xxx

different forms of productive organisation (Section 4). Theanalytical account of specific historical cases is adoptedas main heuristic for disentangling structural learningdynamics.

2. Embedding learning in production dynamics

2.1. Learning in production: a taxonomy

In their critical review of learning curve studies,1 Adlerand Clark (1991, p. 270) proposed a fundamental distinc-tion between first-order and second-order learning.

First-order learning refers to those ‘learning by doing’processes directly experienced by workers via repetitionof productive tasks and the resulting incremental devel-opment of expertise. Here, learning is both an individualand collective process as interactions among workerswithin the firm are integral parts of their learning bydoing. The concept of ‘learning by doing’ expressed inKenneth Arrow’s (1962) seminal contribution captures theSmithian intuition that the accumulation of productionexperience increases workers’ productivity. In particular,Smith mentions three ‘different circumstances’ responsi-ble for this increase in labour productivity: ‘the increase ofdexterity in every particular workman’, ‘the saving of thetime which is commonly lost in passing from one speciesof work to another’, and ‘the invention of a great numberof machines which facilitate and abridge labour’ (Smith,1976[1776], p. 17).

Conventional learning models based on ‘learning bydoing’ and learning curves have been mainly used forexplaining productivity growth at the sectoral and macrolevel (Malerba, 1992, p. 846; Thompson, 2010). In thesemodels, production is treated as a timeless black boxand heroic assumptions are made concerning producers’knowledge of the entire spectrum of production possibil-ities as well as the availability of appropriate productivecapabilities.2 On the contrary, as the literature on localisedtechnical change (Atkinson and Stiglitz, 1969) has shown,given the local and cumulative character of knowledge,producers are only aware of a limited number of factorscomposition laws – i.e. proximate production possibilities.Moreover, as shown in the capability literature, production“has to be undertaken by human organisations embodyingspecifically appropriate experience and skills” (Richardson,1972, p. 888).3

Second-order learning refers to those managerial orengineering actions purposefully aimed at changing theinternal structure of production by introducing new tech-nologies, new equipments or investing in workers training.


was initiated by Wright (1936) and his work in the aircraft industry.2 The stochastic model by Jovanovic and Nyarko (1995) is an exception

in providing a microfoundation of Arrow’s. . .‘learning by doing’.3 The analytical and technical limitations of the production function

models are discussed in Georgescu-Roegen (1970), Scazzieri (1993).


ING Model

S

d Econo

oaiatAigoic(

acliuiee(emIaditdslboeifta

teirfiAceubpr

2

s

t

ARTICLETRECO-561; No. of Pages 17

A. Andreoni / Structural Change an

nly may the firm specifically invest in search activitiesnd production/technology research aimed at expandingts knowledge base (Nelson and Winter, 1982) but it maylso attempt to increase its learning and absorptive capaci-ies themselves (Stiglitz, 1988; Cohen and Levinthal, 1990).s for the latter, in some cases, triggers of learning dynam-

cs are external to the firm and may involve users of finaloods (Rosenberg, 1982; Rhee et al., 1984), other producersf intermediate or final goods in the same or in differentndustries (Lundvall, 1992) or possibly other actors, typi-ally those involved in scientific and technological researchKline and Rosenberg, 1986).

Rosenberg’s (1982, p. 122) concept of ‘learning by using’rises from the recognition that “in an economy withomplex new technologies, there are essential aspects ofearning that are a function not of the experience involvedn producing the product but of its utilisation by finalsers. . .the performance characteristics of a durable cap-

tal good often cannot be understood until after prolongedxperience with it”.4 The related ideas of ‘learning byxporting’ (Rhee et al., 1984) and ‘learning by interacting’Lundvall, 1992) with upstream and downstream produc-rs develop Rosenberg’s fundamental intuition and, thus,ay be considered as sub-categories of ‘learning by using’.

n Lundvall’s (1992) framework learning by interacting is critical feature of societies. Capabilities are collectivelyeveloped through social interactions mainly by observ-

ng and imitating others’ actions as well as by mirroringheir attitudes. This is why the organisational design of pro-uction processes as well as firms’ underlying relationaltructures can affect people’s disposition towards mutualearning and knowledge discovery. Historically, learningy interacting has taken various forms from more co-perative to more competitive ones such as learning byxporting and, thus, though upgrading products character-stics, but also importing and copying machines, recruitingoreign skilled workers and technician exchange, pooling ofechnology, organisation of expos (industrial exhibitions)nd industrial espionage (Chang, 2002; Poni, 2009).

In all these cases, learning dynamics are initiallyriggered by factors external or internal to the firm thatventually result in the reconfiguration of the firm’snternal production structure. Of course this reconfigu-ation may or may not happen depending on how therm in question reacts to the internal or external stimuli.ll of the above suggests that in order to analyse theseompulsive sequences of transformation it is necessary tombed learning dynamics in production structures and tonderstand in which ways these dynamics are constrained,ut also triggered, by existing production structures. Fig. 1rovides a taxonomy of the different forms of learningeviewed above.

.2. The analytical map of production


In mainstream economics, production functions repre-ent complete sets of feasible input combinations for a

4 Mukoyama (2006) develops a stochastic model of learning based onhis idea.

PRESSmic Dynamics xxx (2013) xxx– xxx 3

given output; in an isomorphic way, utility functions estab-lish a relationship between combinations of consumptiongoods and the satisfaction that they provide – i.e. util-ity. Both production and utility functions are designed toshow in the universe of rational choice and equilibriumallocations how the combinations chosen (respectivelyof inputs and consumption goods) reflect relative prices.Given that conventional production theory does notprovide any analytical representation of the internal struc-ture of production processes, qualitative transformationsgenerated by innovations and changes in the technologyand structures of production remain completely unex-plored. In other words, conventional economics adoptsan ‘outside the production machine perspective’ and, asa result, production and, thus, learning dynamics remainblack boxes.

In contrast, the analysis of the internal structure ofproduction combined with a strong emphasis on the repre-sentation of the complex system of interrelated productionprocesses in different sectors was at the centre of the classi-cal theories of production. Classical economists focused onthe limited availability of non-producible goods, the uti-lisation problem and the various constraints determinedby the production scale and its time structure. There arefour main components of the Classical theoretical frame-work. Francois Quesnay’s early formulation of the conceptof productive interdependencies called attention to the ‘cir-cular flow’ of wealth production and reproduction (seealso Leontief, 1928). Adam Smith’s analysis of the inter-nal structure of the pin factory revealed the microeconomicadvantages of the division of labour and the macroeconomicconditions on which it is based–i.e. stock of circulatingcapital flows. Charles Babbage’s focus ‘on the causes andconsequences of large factories’ led to the formulation ofthe law of multiples and, thus, to the discovery of differ-ent patterns of proportional utilisation and maintenance ofindivisible inputs. Finally, Karl Marx’s analysis of differentarrangements of production processes highlighted the mainfeatures of the modern factory system and thus the work-ing of the so called ‘collective machine’ (Landesmann, 1986,1988; Scazzieri, 1993).

In this line, more recently, Wassily Leontief’s (1947)input–output analysis and Nicholas Georgescu-Roegen’s(1970, 1971, 1990) fund-flow model developed the buildingblocks for a series of structural approaches to produc-tion (Landesmann, 1986; Scazzieri, 1981, 1993; Bianchi,1984; Morroni, 1992; Landesmann and Scazzieri, 1996;Buenstorf, 2004, 2007). These contributions view a givenproduction process Pr (r = 1,. . ., k) as a particular systemof interrelated tasks through which a sequence of transfor-mations of materials are performed according to differentcombinations of flow inputs (such as productive agents andmechanical artefacts) and fund inputs (such as fuel, chem-ical catalysts and electricity), subject to certain scale andtime constraints.

Approaching production from the point of view of struc-tural economics implies an analytical focus on the follow-


ing set of both quantitative and qualitative coordinationproblems: (i) how to synchronise and arrange the systemof interrelated tasks in time; (ii) how to arrange the pro-duction process given the specific properties of materials in


ARTICLE IN PRESSG Model

STRECO-561; No. of Pages 17

4 A. Andreoni / Structural Change and Economic Dynamics xxx (2013) xxx– xxx

Trigg ers

Learning dy namic

Internal to th e fir m External to the fir m

First-order learning Produ cer – producti on Learn ing by d oing

Second-order learning Produc er – re sear ch Learn ing by search ing

Learning to learn

Produc ers – science Learn ing from sc ienc e

Produc ers – users Produc ers – producer s

Learn ing by u sing

- Learn ing b y int eracting

- Lea rning b y exporting

dynami
Fig. 1. Learning
transformation; (iii) how to organise and activate the pro-duction process by combining different fund inputs eachof them endowed with certain capabilities or capacities.5

Interdependencies among these coordination problems arepervasive in the sense that for example tasks arrangementdepend on both the properties of materials in transforma-tion and firm’s availability of capabilities/capacities.6

Tasks refer to those production operations that are pur-posefully performed in a given production process. Eachtask Tj (with j = 1, 2,. . ., j,. . ., J) can be decomposed into ele-mentary operations or clustered in groups of tasks. Theycan be arranged simultaneously or sequentially in variousstages of fabrication j (with j = 1, 2,. . ., j,. . ., J), sometimesin a discrete way but sometimes in a continuous way,that is, with or without interruptions.7 This last distinctionproves to be very relevant as soon as we consider how dif-ferent forms of production organisation have historicallydeveloped different techniques for inventory and storagecapacities management (Rosenberg, 1994; Landesmannand Scazzieri, 1996, chap. 8).

Materials refer to what is transformed in the fabricationstages of a production process.8 The relationship betweenmaterials and stages of fabrication can be represented bya descriptive matrix M = [mij] in which any element refersto the material i (with i = 1, 2,. . ., n) that has been trans-


formed in the fabrication stage j through the execution ofthe task Tj. At each fabrication stage, given a certain stockof materials available to the productive organisation, only

5 Mechanical artefacts present certain production capacities, while eachproductive agent is characterised by a certain set of production capabilities.

6 Richardson (1972, p. 885) stresses how “the habit of working withmodels which assume a fixed list of goods may have the unfortunate resultof causing us to think of coordination merely in terms of the balancing ofquantities of inputs and outputs and thus leave the need for qualitativecoordination out of account”.

7 As for the time structure, the material transformation processes can bevisualised as a system of pipelines (Landesmann, 1986; see also Morroni,1992).

8 In the case of ‘immaterial production’ – e.g. service activities – ‘mate-rials in process cannot be identified, at least in the usual sense, and theproduction process generally takes the form of a close interaction amongfund agents, in the course of which some of the characteristics of suchagents (and sometimes their capabilities as well) may get transformed(Landesmann and Scazzieri, 1996, pp. 252–253).

cs: a taxonomy.

some materials will be utilised and thus transformed. Thisimplies that for each production process we will observea certain ‘realised’ matrix M* = [mij], whose internal struc-ture represents all the materials in use in the stages of thatproduction process.

In order to perform a certain system of interrelatedtasks through which materials are transformed into finalcommodities, the production process has to be ‘activated’by a series of inputs such as fuel, chemical catalysts, butalso machines and productive agents, that is, workers.Flow inputs such as fuel, chemical catalysts, electricity andfertilisers are utilised in certain stages of material trans-formation but they do not materially constitute the finaloutput of the process as the materials in use do. Flow inputsused in a certain production process can be describedthrough a descriptive matrix F = [fij] in which any elementrefers to the flow input i that has been consumed in thefabrication stage j through the execution of the task Tj. Foreach production process we will observe a certain realisedmatrix F* = [fij].

M =

⎛⎜⎜⎜⎝

m11 m1J

· · ·mij

· · ·mn1 mnj

⎞⎟⎟⎟⎠ F =

⎛⎜⎜⎜⎝

f11 f1J

· · ·fij

· · ·fn1 fnj

⎞⎟⎟⎟⎠

In contrast, fund inputs are both mechanical artefactssuch as machines, tools and equipment and productiveagents (i.e. workers, supervisors, engineers and managers).Fund inputs maintain their characteristics substantiallyunaltered during the production process, provided thatcertain tolerance thresholds are not violated (Georgescu-Roegen, 1970; Landesmann, 1986). Mechanical artefactspresent a certain production capacity, while each productiveagent is characterised by a set of complementary productivecapabilities. By activating some of these capabilities, eachproductive agent is able to perform a single task or a set ofsimilar tasks (i.e. those tasks which require the utilisation ofthe same set of complementary capabilities). Although pro-


ductive agents may learn to perform different tasks, theircapabilities are limited so they cannot switch between allproductive tasks, especially when complex products areconsidered.


ING Model

S

d Econo

decaaTltTbtfapetdiFtmio

C

cPmiwpiopmiclocab

oetSfiot

g“tmp

The network of relationships and interdependenciesamong the spaces C, T, M has to be synchronised over



Just as we did with materials and flow inputs, we canistinguish between the bundle of capacities/capabilitiesmbodied in a certain set of fund inputs (i.e. potentialapacities/capabilities) and the capacities/capabilities actu-lly utilised by the productive organisation in performing

certain set of tasks (i.e. capacities/capabilities in use).9

he former is described by the matrix C = [cij] while theatter by the matrix C* = [cij], where cij denotes the rela-ionship between the capacity/capability i and the taskj performed at the stage of fabrication j. The distinctionetween the matrix C = [cij] and C* = [cij] illustrates howhe same production process Pr (r = 1,. . ., k) can be per-ormed by using different bundles of mechanical artefactsnd productive agents, that is, different combinations ofroduction capacity and productive capabilities. However,ven when two different productive organisations performhe same process Pr by combining the same bundle of pro-uctive capacities/capabilities, the latter can be employed

n different proportions. For example, two firms Firm1 andirm2 can perform the same production process by usinghe same two fund agents – i.e. workers w and machines

– but in different combinations – for example, one fund-nput combination may be more labour intensive than thether.

1∗ =(

5w 7w

1m 1m

)C2∗ =

(1w 2w

3m 3m

)

Thus, by comparing the two matrices C1* and C2* wean discover specific features of the production processr performed by Firm1 and Firm2. In particular, the twoatrices express different relationships of complementar-

ty among the two fund inputs considered (machines andorkers respectively). In our case, the first stage T1 of theroduction process Pr can be performed either by combin-

ng one machine with five workers or three machines andne worker (see above). Given these relationships of com-lementarity between fund inputs and also the fact thatachines tend to be tasks-specific and only partially flex-

ble, the kind of combinations of fund inputs that firmsan select from the space C = [cij] for performing Pr areimited. Moreover, scaling up the production process notnly requires the consideration of these relationships ofomplementarity but also that a law of proportionalitymong all the components of the process is satisfied (seeelow).

Based on Cartwright (1989), it has been noted how veryften the capabilities (and capacities) of fund inputs can bexpressed in a quantitative form, so that we can assumehey are comparable in cardinal space (Landesmann andcazzieri, 1996, p. 197). One possible quantitative speci-


cation of the matrix C = [cij] relies on the considerationf the time structure of the production process in relationo the capacity/capabilities in use. The matrix C* = [cij] can

9 As has been stressed, this distinction leads us to interpret the emer-ence of new productive structures within the space of virtual practices asthe outcome of a clustering process that brings about a rearrangement ofhe primitive elements of productive activity; [thus] structural change

ay be considered as a case of variation within a spectrum of virtualossibilities” (Scazzieri, 1999, p. 230).


be transformed into a matrix of capacities/capabilities use –times �* = [�ij] where the generic �ij represents the use-time of the capacity/capability cij in the production processPr.

C =

⎛⎜⎜⎜⎝

c11 c1J

· · ·cij

· · ·cn1 cnj

⎞⎟⎟⎟⎠ �∗ =

⎛⎜⎜⎜⎝

�11 �1J

· · ·�ij

· · ·�n1 �nj

⎞⎟⎟⎟⎠

Taking the case in which two different productiveorganisations perform the same process Pr with the samecombination of fund inputs described by C* = [cij], wecan compare the matrices �* to discover if one timearrangement of Pr is more time wasting than another. Forexample, the reconfiguration of the time structure of Prfrom one in line to one in parallel can reduce the amountof idle time of fund inputs across fabrication stages (seebelow). Given appropriate transformations such as theone proposed above, as soon as the productive capabili-ties and capacities become comparable in cardinal space,the capacity–capability ratios can be calculated for eachproductive task (or groups of tasks) and organised in amatrix. This will elucidate the interdependencies betweendifferent kinds of fund inputs. However, the set of inter-dependencies characterising each production process doesnot simply involve one subset of its components (here, fundinputs). Instead, each production process requires the coor-dination of all its components (namely tasks, materials andflow inputs as well as fund inputs).

Interdependencies among components can be visu-alised by mapping the relationships between capac-ity/capabilities, tasks and materials.10 The entire spectrumof possible combinatorics is represented through the ana-lytical map of production relationships (see Fig. 2).11 Themapping from the capacity/capability space C to the taskspace T (i.e. job specification programme) can be determinedfollowing different criteria (Landesmann and Scazzieri,1996). For example different combinations of fund inputsmay be relatively more or less adequate for the execu-tion of one task (or cluster of tasks) than another. Also,a reconfiguration of the job specification programme (i.e.different mapping from the space C to the space T) mayallow the activation of previously unused fund inputs orthe achievement of higher efficiency in the utilisation ofthe capacity/capabilities in use.


time and according to specific scale requirements deter-mined by the existence of process indivisibilities as well

10 For clarity the flow agents are taken out of the picture. The decision toprivilege the other three dimensions matrices C, T and M is due to the factthat commonly there are higher degrees of freedom in their combinatoricsand the use of flow agents is strictly dependent on the utilisation of fundagents.

11 The concept of ‘analytical map of the true [interpersonal] relations’ isproposed in Georgescu-Roegen (1976, p. 205) as one possible realisationof the ‘entire spectrum of peasant institutions’.





ap of pr
Fig. 2. The analytical m
as indivisible fund inputs.12 As for the time structure, syn-chronisation has to be pursued at three different levels(coordination in the utilisation of fund inputs, arrangementof interdependent tasks over time, and transformation ofmaterials over time). The difficulty of matching the ‘timesequencing requirements’ of these three dimensions makesperfect synchronisation across the three above mentionedlevels impossible and explains the co-existence of pat-terns of simultaneity or sequentiality. Time gaps and idletime in production processes are thus largely structurallydetermined and, within the given structure, only partiallyreducible through various forms of learning (Landesmannand Scazzieri, 1996).

Moving on to indivisibilities, processes are indivisiblewhen they are not ‘indifferent to size’. As in the biologicalworld, all individual production processes “follow exactlythe same pattern: beyond a certain scale some collapse,others explode, or melt, or freeze. In a word, they ceaseto work at all. Below another scale, they do not evenexist” (Georgescu Roegen, 1976, p. 288). The fact that pro-cesses are ‘scale-specific’ (in other words that they arecharacterised by upper and lower bounds) implies that con-ducting a process on a smaller or a larger scale can onlybe done if a law of proportionality among the componentsof the process is satisfied. This idea was originally formu-lated by Charles Babbage’s law of multiple. In Babbage’sview: ‘[w]hen the number of processes into which it is mostadvantageous to divide [the production process], and thenumber of individuals to be employed in it, are ascertained,then all factories which do not employ a direct multiple ofthis latter number, will produce the article at a greater cost’(Babbage, 1835, p. 211).

At the level of the components, limitations in thebundling and unbundling of fund inputs are extremely


stringent, while flow inputs as well as materials in trans-formation are more often divisible. As far as fund inputsare concerned, the existence of indivisible funds of capaci-

12 For a comprehensive discussion of the role of time and scale in pro-duction see Landesmann (1986), Bianchi (1984), Morroni (1992), Scazzieri(1993), Landesmann and Scazzieri (1996).

oduction relationships.

ties (such as a machine tool with certain scale-determinedtechnical characteristics and specifications) as well as indi-visible funds of capabilities (that is, productive agents suchas workers and engineers at the shop floor) mean that, fora fund input to be fully utilised, a specific scale of produc-tion has to be achieved. For small scales of production, fundinputs would inevitably be underutilised. However, if fundinputs are not too specialised in the execution of some pro-ductive tasks, productive organisations can overcome scaleconstraints by utilising the same indivisible fund inputs forthe production of other commodities. Nevertheless thesenew commodities generally possess a certain degree ofsimilarity as fund inputs are endowed with only a limitedset of complementary capacities or capabilities.

2.3. Embedding learning dynamics

The structural representation of production providedabove now allows us to see some of the many limitationsarising from the understanding of learning dynamics as adisembedded process, as is the case with today’s main-stream economics (see Section 2.1). To give one exampleof the main form of first-order learning, Arrow’s concept of‘learning by doing’ refers to a process involving one sub-set of the space C (i.e. capabilities of fund inputs such asworkers, engineers and managers). In this case, ‘learningby doing’ is nothing more than an increase in productivecapabilities, which generally result in a reduction of capa-bilities use-times. In other words the execution of the sameproductive task will require less time due to accumulatedexperience and as a result the overall productivity of theproductive organisation will be increased. However, as weshall see below, our analytical map of production shows


how ‘learning by doing’ does not always imply such pro-ductivity increases and it might even lead to the emergenceof bottlenecks and imbalances.13

13 In fact in his seminal work on ‘learning by doing’, Kenneth Arrowrecognises how “learning associated with repetition of essentially thesame problem is subject to sharply decreasing returns” and, thus, thatlearning mainly consists of finding new solutions to emerging ‘stimulus


ING Model

S

d Econo

amiialcfomgiet‘

scfctnhtic(kitwihthp1

i‘fnispdode

sil

oo

lal

and allow them to discover structurally embedded oppor-



The reason for this becomes clear as soon as we visu-lise interdependencies among components, i.e. tasks,aterials, flow inputs and fund inputs (and their capac-

ty/capability). The development of increasing capabilitiesn the execution of a certain set of productive tasks gener-lly implies that certain stages of fabrication will requireess time, while other stages remain invariant as a result ofonstraining factors such as fixed times for material trans-ormation (e.g. time needed for certain chemical reactions)r the scale of other existing fund inputs, in particularachines and equipment. These latter stages of fabrication,

iven their invariant properties, will appear as bottlenecksn the production process and may end up affecting thentire job specification programme, potentially even neu-ralising or counteracting the productivity increases oflearning by doing’.

To give a second example of the importance of under-tanding learning as embedded we can look at Rosenberg’soncept of ‘learning by using’, the latter being the mainorm of second-order learning discussed above. The con-ept of learning by using was developed with referenceo ‘products involving complex interdependent compo-ents or materials’. As a result of the particular industrye focused on, that is, aircraft,14 Rosenberg underlinedhe fact that ‘learning by using’ implies a “feedback loopn the development stage which, in turn, increases effi-iency and/or requires changes in productive techniques”1982, p. 123).15 Rosenberg distinguishes between twoinds of useful knowledge arising from ‘learning by using’n both products and processes. Embodied knowledge ishat which requires ‘appropriate design modifications’,hile disembodied knowledge “leads to certain alterations

n use that require no (or only trivial) modifications inardware design”, although even the latter still “leadso new practices that increase the productivity of theardware” – for example modification in maintenanceractices in the aerospace industry (Rosenberg, 1982, p.24).

Of course, these two forms of ‘learning by using’ arentertwined. By generating new embodied knowledge,learning by using’ in fact facilitates the discovery of neworms of disembodied knowledge and even makes themecessary. What is implicitly suggested here is that ‘learn-

ng by using’ may trigger the rearrangement of the jobpecification programme. This occurs because of the newroductive tasks and fund inputs required to cope withesign modifications (embodied knowledge), or as a result


f alterations in productive practices whose performancesepend on the rearrangement of fund inputs available (dis-mbodied knowledge).

ituations’ (Arrow, 1962, p. 155). However, the effects of ‘evolving stimuli’n the transformation of productive structures are not analysed given theack of an analytical map of production.14 Even today aircraft are among the most complex products, composedf almost 6 million parts (by way of comparison a car is typically composedf just 6 thousands parts).15 In this respect, see also Hippel von (1988) whose contribution links theearning by using dynamics to product diversification patterns. Also, Klinend Rosenberg (1986) presents a ‘chain-linked model’ where feedbackoops in the innovation process are recognised as key factors.


The first case is more clearly detectable as it requiresdefinite design modifications (i.e. technological improve-ments) while the second set of transformations tend tobe under-estimated as they do not call for the intro-duction of any new fund inputs. The analytical map ofproduction allows us to understand how a productionprocess may be qualitatively transformed even withoutequipping the productive organisation with new fundinputs or without transforming the existing ones. Instead,the production process may be transformed just by re-arranging fund inputs among the system of tasks whichhave to be performed or by synchronising tasks in adifferent way over time. In fact there are various waysof combining elementary operations into new tasks orclustering existing tasks in new ways. Once again theextent to which this can be done depends on the capac-ities/capabilities embedded in funds inputs and theirdegree of utilisation, as well as on the properties ofthe materials in transformation and on time arrange-ments.

Learning processes are intrinsically heterogeneous andoccur through time at several nested levels of produc-tion, the latter being structurally determined by productiveinterdependencies.16 As soon as we attempt a restructu-ring of ‘learning by doing’ or ‘learning by using’, it becomesobvious that the majority of existing studies focus theirattention on what triggers the learning process or whatits output is. The process per se not discussed. In otherwords, the conventional analysis of learning ends exactlywhere the learning process starts. Even when there is amore detailed investigation of learning dynamics in pro-duction, as in the work of economic historians (Rosenberg,1969, 1976, 1979, 1982; Noble, 1986; Piore and Sabel, 1984;Mowery and Rosenberg, 1998; Mokyr, 2002; Landes, 2000;Poni, 2009), production structures are generally seen onlyas constraints that productive agents overcome throughproblem-solving activities and changes in productive tech-niques. For example, Mokyr (1990, p. 9 italics added) arguesthat ‘[t]echnological change involves an attack by an indi-vidual on a constraint that everyone else has taken asgiven’.

However, as we shall see below, an analytical account ofa number of historical cases allows us to understand howexisting and evolving production structures are not justconstraints. Instead, existing production structures orien-tate productive agents towards certain learning trajectories


tunities. As shown by Hicks (1969), the adoption of ananalytical approach to economic history can be a vehicle

16 Another aspect that conventional approaches tend to forget is that‘learning in time’ can proceed at different speeds according to the timerequired for reconfiguring the production structure or according to thetime knowledge requires to flow (i.e. be disseminated and absorbed)throughout the production organisation or at the inter-firm level. Inother words the problem is not only ‘what to learn’ or ‘how to learn tolearn’ but also ‘how to learn faster’. As shown by Dodgson (1991), thedifferential ability in learning quickly about technological opportunitiesis a crucial determinant especially in those sectors (e.g. biotechnol-ogy) characterised by an uncertain and generally rapid process oftransformation.


ING Model

d Econo

erable extension of the useful life of a wide range ofcapital equipment’ as well as to other ‘cumulative improve-ments’.



for developing a ‘quasi-theory’, that is to say a stylised rep-resentation of economic facts through which theories canbe developed.

3. Structural learning: an analytical framework

3.1. Learning in a structured space: an analytical accountof historical cases

The analytical map of production relationship eluci-dates firstly the ‘architecture of complexity’ in production(Simon, 1962; Buenstorf, 2005) and secondly, the factthat learning dynamics are realised in a ‘structured space’over time. This means that learning in production isnot simply a process occurring as a result of cognitivedynamics; rather it is also a process triggered and ori-entated by structural dynamics. The latter open up thepossibility of transforming ‘structural constraints’ suchas bottlenecks and technical imbalances into ‘structuralopportunities’. As highlighted in structural analyses of pro-duction (see above), coordination problems in the spaceof capacity/capabilities, materials and tasks may be solvedin multiple (albeit interdependent) ways. In other words,there are ‘worlds of production’, that is, a variety of produc-tion arrangements (‘world of possibilities’) that are feasibleeven under same sets of contextual conditions (Salais andStorper, 1997).

‘Worlds of possibilities’ permit the transformation of pro-duction processes and their outcomes – i.e. process andproduct innovations (Sabel and Zeitlin, 1997). Of coursesaying that there are multiple possibilities should notblind us to the fact that bottlenecks, materials proper-ties or technical imbalances are all pervasive constraints.In fact, discovering these possibilities, given certain struc-tural constraints, is the very essence of what I call thestructural process of learning. The concept of structurallearning is introduced here to identify the continuousprocess of structural adjustment and transformation ofproduction ‘triggered’ and ‘orientated by’ existing andevolving production structures. Static and dynamic com-plementarities, as well as similarities and indivisibilities,are essentially focusing devices for activating compul-sive sequences of technological change and discoveringnew production possibilities at the firm and inter-firmlevel.

We will now move to an analytical account of his-torical cases in which “[c]omplex technologies createinternal compulsions and pressures which, in turn, initiateexploratory activity in particular directions” (Rosenberg,1969, p. 4). This historical analysis is the first step towardsdisentangling those structural dynamics that prepare thesetting for learning and those specific factors trigger-ing learning processes in production. The second stepis to identify a number of structural learning trajectoriesand illustrate them with an analytical map of productionrelationships. The third step will be to re-link dynam-ics occurring at the level of production structures with


those occurring at the level of the structures of cogni-tion in productive organisations. As we shall see, this thirdstep will allow us to show the analytical tension betweenstructure and agency in learning dynamics and also


elaborate how structure and agency are linked by a bun-dle of bidirectional transformative relationships (Bourdieu,1972).17

Rosenberg (1969) identifies three main ‘induce-ment mechanisms’ of learning, namely technical imbal-ances or bottlenecks, labour-saving/uncertainty-reducingmachines and substitutes or alternative sources of supplyof fund and flow inputs or materials. A number of histori-cal examples will help to illustrate this point. In 1900 themachine tool industry was revolutionised by the introduc-tion of high-speed steel which allowed an increase in thehardness of cutting tools. However “it was impossible totake advantage of higher cutting speeds with machine toolsdesigned for the older carbon steel cutting tools becausethey could not withstand the stresses and strains or providesufficiently high speeds in the other components of themachine tool” (Rosenberg, 1969, p. 7; see also Andreoniand Gregory, 2013). As a consequence, transmissions, con-trol elements and other machine tool components had tobe redesigned and this change “in turn, enlarged consider-ably the scope of their practical operations and facilitatedtheir introduction into new uses” (Rosenberg, 1969, p. 8).This is a typical example of a technical imbalance leadingto changes in complementary processes as well as compo-nents, that is, tasks, materials and capabilities/capacities.It highlights how a technical constraint can actually acti-vate a process of exploration and searching in which “thesize of the discovery need bear no systematic relation-ship to the size of the initial stimulus” (Rosenberg, 1969,p. 9).

Indeed, the initial technical imbalance in a certainindustry may trigger structural learning processes in otherindustries and sectors. In the early nineteenth century USagricultural sector, before tractors were introduced, JohnDeere revolutionised agricultural production by inventingthe steel plow (Andreoni, 2011a,b). A biological constrainttriggered the introduction of steel plows, but also of othercomplementary tools made with the same or differentmaterials according to specific task requirements. Tradi-tional wood plows could not plow the rich soil of theMiddle-West without breaking. At that time given thescarcity of steel and the need to import it from Great Britain,John Deere made his first plow out of an old blade saw.After a series of tests on different types of soil, the newsteel plow was ready to be absorbed into the ‘crop-growingtechnique’ adopted at that time. In turn, the introductionof the steel plow triggered new complementary discover-ies as well as the application of the same new material toother equipments requiring the same hardness. In fact, asrecognised by Rosenberg (1979, p. 37), ‘the substitution ofnew materials (e.g. aluminium and rust-resistant steels)for old ones, improved techniques of friction reduction(lubrication and roller bearings) have led to a consid-


17 At the centre of Bourdieu’s analysis there is the dialectic between‘externalising the internal’ and ‘internalising the external’ which attemptsto go beyond precisely the same tension.


ING Model

S

d Econo

swfttlliwtIowoub

piw“nttsiii

moiofi(hdcitpd

aotcacoedIsofis2bc

Principles of similarity and complementarity also operateat the firm level and are responsible for distinct structurallearning trajectories.



Thus, the process of structural learning in a givenector may develop an intersectoral character. In otherords, complementarities (as well as innovations suitable

or similar tasks performed in other productive activi-ies) may spread from one sector of an economic systemo another sector triggering a specific form of structuralearning which we label here intersectoral learning. Theatter expression identifies a dynamic process of interlock-ng and mutual reinforcing technological developments

hich links the innovative patterns of two or more sec-ors in a relationship of complementarity and/or similarity.n sum, technical complementarities among fund inputsr the application of an innovation – e.g. a new materialith certain properties – in the execution of a broad set

f similar productive tasks are the fundamental dynamicsnderlying the learning processes considered by Rosen-erg.

It is not just constrained posed by existing technicalrocesses that can work as triggers for structural learn-

ng dynamics: social process can function in this way asell. In the Poverty of Philosophy Karl Marx observed how

after each new strike of any importance there appeared aew machine” (n.d.: 134; first source Rosenberg, 1969). Thehreat of strikes introduces a critical element of uncertaintyo the supply of labour and strongly affects the delicate timetructure of a production process, thereby promptly thenvention of new labour-saving machines. Social changesnduce the invention or discovery of new machines and thisn turn sets off a further train of changes.

Robert’s self-acting mule, the Jacquard punchingachine and the introduction by the British Government

f the ‘American System of Manufacturing’ in the gun mak-ng industry in 1854, are all cases in which the inventionr acquisition of a new more powerful machine is just therst step in a subsequent process of structural learningRosenberg, 1969; see also Chang, 2002). All these casesighlight how when a new machine becomes available pro-uction that was technically feasible but not economicallyonvenient becomes possible. This possibility depends onncreasing the scale of complementary machines or inhe re-arrangement of workers in the production unit,rovided that they can perform a certain set of similar pro-uctive tasks.

Together with the above mentioned inducement mech-nisms identified by Rosenberg, the need/opportunityf increasing the scale of production is another factorriggering processes of structural learning. For exampleomplementary innovations such as refrigerators, railwaysnd steamships affected the reduction of transportationosts, increased the degree of regional specialisation andpened the opportunity to benefit from scale technologyxpansions and from specialisation in a limited set of pro-uctive tasks performed at high productivity standards.

ndeed, as soon as the scale of production increases ‘ahifting succession of bottlenecks’ will emerge. Focusingn them, engineers will start exploring new possible con-gurations of the production process, which may lead to


erendipitously discovering ‘singleton techniques’ (Mokyr,002). Problems related to scale constraints, which ariseoth from indivisible fund inputs and indivisible pro-esses, may trigger the discovery of innovative (structural


as well as organisational) configurations of productionprocesses (Andreoni and Scazzieri, 2013). A good exam-ple of this is the typical problems faced by small farmersand small firms when trying to gain access to indivisi-ble fund inputs such as machines and other equipment.Historically fund inputs indivisibilities as well as scaleinvariant processes have triggered institutional innova-tions such as ‘renting/sharing’ solutions implemented byproducer cooperatives (Lissoni, 2005) as well as forcingproductive agents to rearrange job specification pro-grammes.

3.2. Structural learning trajectories

The historical cases document how inducement mecha-nisms of learning dynamics, and the resulting ‘compulsivesequences’ of transformations, are embedded in and trig-gered by existing production structures at each pointin time. Specifically complementarities and similaritiesamong tasks or materials, as well as fund inputs indivis-ibilities, have been crucial focusing devices in structurallearning dynamics. The analytical account of these histor-ical cases leads to the identification of three fundamentalstructural learning trajectories. Given a certain bottleneckor technical imbalance in production, the first two struc-tural learning trajectories are triggered by the existence ofsimilarities and of complementarity among materials, tasksand fund inputs. The third structural learning’s trajectory istriggered by the existence of indivisibilities in production.

The fundamental intuition behind the first two struc-tural learning trajectories may be found in Richardson’s(1960, 1972, 2003) observation that different forms ofinter-firm cooperation we see arise from different patternsof similarity and complementarity among productiveactivities.18 Richardson breaks down the production ofeach final commodity into various stages or activities, eachof them executable by different types of firms. “Activitieswhich require the same capability for their undertaking”are called similar activities (Richardson, 1972, p. 888). Onthe other hand, activities are complementary “when theyrepresent different phases of a process of production andrequire in some way or another to be coordinated [. . .]both quantitatively and qualitatively” (Richardson, 1972,pp. 889–890). Building on this dichotomy, Richardsonexplains how the complex and interlocking clusters,groups and alliances of firms we observe are in realitydifferent responses to the same problem: the need to coor-dinate “closely complementary but dissimilar activities”.19

As firms cannot accumulate all the capabilities requiredfor performing a broad set of dissimilar activities, theywill specialise in a few activities and cooperate with thosefirms specialised in closely complementary activities.


18 See also Ménard (2004), Gibbons and Roberts (2011), Garnsey andMcGlade (2006).

19 This analytical point is developed in Section 4 with respect to thedifferent forms of production organisation.


ING Model

d Econo



3.2.1. Structural learning trajectory triggered bysimilarities

Overcoming a productive constraint by introducing anew set of tasks, capabilities or materials may induce thesame or other firms to adopt the same set of tasks, capa-bilities or materials for overcoming a similar constraint, inthe same or other kind of productive processes. As docu-mented by Rosenberg (1963, pp. 422–423, italics added)“industrialisation was characterised by the introduction ofa relatively small number of broadly similar productive pro-cesses to a large number of industries. This follows fromthe familiar fact that industrialisation in the nineteenthcentury involved the growing adoption of a metal-usingtechnology employing decentralised sources of power”.Furthermore, discovering a new way of performing a certaintask affects all those productive processes in which simi-lar tasks are performed. This explains why “many of thebenefits of increased productivity flowing from an innova-tion are captured in industries other than the one in whichthe innovation was made” (Rosenberg, 1979, p. 41; see alsoUsher, 1954).20

Many examples might be provided which highlightthe existence of technological linkages among apparentlyuncorrelated products such as guns, sewing machines,bicycles, motorcycles, and automobiles. Among the manyhistorical examples ‘the development of the universalmilling machine by Brown and Sharpe is, perhaps, themost outstanding example of a machine which was ini-tially developed as a solution to a narrow and specificrange of problems and which eventually had enormousunintended ramifications as the technique was appliedto similar productive processes over a wide range ofmetal-using industries’ (Rosenberg, 1963, p. 432, italicsadded).

In the specific case of firms whose production processconsists of a system of similar tasks, the discovery of a newway of performing a certain task or the introduction of anew material implies a complete reconfiguration of theentire process. However, as in this specific case produc-tive agents would already be endowed with similar kindsand amounts of capabilities, they will be substitutable andcan be arranged in many different ways across time. Theproduction process of more complex products (or compo-nents) tends to assume the form of a system of dissimilartasks. Indeed, complex products are defined as those “com-posed of many subsystems that interact in complex ways”(Rosenberg, 1982, p. 136). In the case of complex productsrequiring the performance of closely interdependent dis-similar tasks, intra and inter-firm complementarities willbe pervasive.

3.2.2. Structural learning trajectory triggered bycomplementarities

“[I]nnovations hardly ever function in isolation”


(Rosenberg, 1979, p. 26). The theoretical framework wehave constructed allows us to analytically specify andexplain this intuitive insight of Rosenberg’s. Innovation

20 This analytical point will form the basis of our discussion in the ThirdEssay of the concept of intersectoral learning.


occurs in this bunched fashion because of the utilisa-tion and the productivity of fund inputs (i.e. machineswith certain capacities or productive agents with certaincapabilities) both critically depend on the simultaneousavailability of complementary fund inputs. Complemen-tarities among fund inputs may trigger direct learningdynamics, or learning dynamics over time. Direct learningdynamics occur when one fund input makes the func-tioning of another fund input possible or more efficient.Learning dynamics over time occur when one fund inputmakes the functioning or introduction of other fund inputspossible over time.

In the specific case of a production process constitutedby a system of dissimilar tasks, fund inputs performing aspecific task in one stage of fabrication are combined withothers performing other tasks in other stages of fabricationin a relationship of complementarity rather than of substi-tutability. Now if tasks are very dissimilar and complex,productive agents (or even entire productive organisa-tions) have to specialise in the execution of only one task,or even in performing elementary operations of more com-plex tasks. In this case, a number of processes of the sametype can be organised in series (also called in sequence) sothat specialised productive agents (or organisations) canperform the task in which they are specialised without longperiods of inactivity. Discovering this possibility and apply-ing it to the production process allows firms to reduce timewastage as productive agents will shift over time from oneprocess to another.

Additionally, according to the degree of decomposabil-ity of a given production process, firms may decide toadopt a modularisation strategy (Langlois, 2002; Buenstorf,2005). Interestingly, in the case of productive processescomposed of closely complementary but dissimilar tasks,modularisation may guarantee static efficiency at thecost of dynamic efficiency. This problem occurs becausemodularisation tends to reduce the number of learning tra-jectories triggered by complementarities.

3.2.3. Structural learning trajectory triggered byindivisibilities

Indivisible fund inputs and materials as well as scale-invariant tasks (or processes) impose a proportionalitypath on all transformations of the internal structure ofproduction (see above the reference to Babbage’s law ofmultiples). This means, for example, that if a certain indi-visible fund input (e.g. a new machine) is adopted, then, thefirm has to reconfigure the job specification programme insuch a way that scale economies generated by the use ofthe new machine are exploited and potential bottlenecksand time or material wastes are avoided.

The existence of indivisibility might also triggerincremental innovations both at the technological andorganisational levels. For example adopters of the newindivisible input (or scale-invariant task) “could inventaround the new machine and remove those technolog-ical constraints that limit their ex ante or ex post size.


[. . .] Alternatively they could attack it directly by findingthe way to split the different functions that the originalinnovation performs jointly, thus decomposing the latterinto a few (possibly compatible) modules, each of them


ING Model

S

d Econo

b3aauaopv

ic‘sateiTgimwaaee1

3a

steicntbppimfgpnd

tttepf4obp

lighted by Herbert Simon (1957) in his analysis of thebounded rationality problem. Simon introduced the ideathat individuals’ learning is socially constructed, in otherwords, that ‘[w]hat an individual learns in an organisation



eing cheaper than the original item” (Lissoni, 2005, p.64, italics added). Indivisibility-led innovations may alsoffect the way in which indivisible fund inputs are acquirednd adopted and, thus, production is organised. As doc-mented in Paul David’s (1966) analysis of smaller firmsnd their ways to deal with indivisibilities, another formf innovation triggered by indivisibilities is the creation ofroducers’ consortia and cooperatives or pro-renting inno-ations.

Structural learning trajectories triggered by indivisibil-ties also interact with those triggered by similarities andomplementarities (see above). Specifically, indivisibilitiesshape the form’ of those learning trajectories triggered byimilarities and complementarities. This means that thepplication of new indivisible fund inputs and materialso similar production activities (in the same or differ-nt sectors) will introduce in the new production contextn which they have been applied a new indivisibility.he latter will reshape the overall job specification pro-ramme and the scale of the production process (when itsntroduction is not too costly). As for the case of comple-

entarities, at the firm level they mainly arise in threeays: (i) among indivisible fund inputs or materials, (ii)

s a result of scale-invariant processes, and finally, (iii)mong different production processes when the combinedxecution of scale-invariant tasks reduced the costs ofach production process (i.e. economies of scope, Morroni,992).

.2.4. Analytical map of structural learning trajectories:n illustration

One of the possible ways to visualise how these threetructural learning trajectories interact is to make use ofhe analytical map of production relationships we developedarlier. Different examples can be inserted into the analyt-cal map of production relationships (see figure 3). Let usonsider the following illustrative case. The acquisition of aew fund input c53 (e.g. a new machine introduced throughechnology transfers or a traditional machine transformedy small improvements) can trigger a cascade process ofroduction reconfiguration. The task T2 has to be decom-osed into two tasks (T′

2 and T′3) while the obsolete fund

nput c23 can be dismissed. The discovery of a new material54 requires the execution of a new task T4 to be trans-

ormed. The scale of fund input c41 has to be changed,iven the introduction of a new indivisible fund input c53. Areviously unutilised capabilities fund c64 is activated so aew complementarity with the new material m54 has beeniscovered.

These dynamics may be also analysed by adoptinghe virtual matrix C = [cij] and C* = [cij] (see above). Here,hrough the process of structural learning, some rela-ionships of complementarity among funds inputs willnd, others will change (although maintaining the sameroportions) while others still will be completely trans-ormed. A similar idea is presented by Simon (1962, p.


75) when he suggests that, given a hierarchical systemf interdependencies, the architecture of complexity maye disentangled by adopting what he calls a nearly decom-osable matrix.


3.3. Structures of production and structures of cognition

The cascade process of reconfiguration triggered bystructural learning dynamics cannot be thought of as anautomatic one. In order to react to structural stimuli andfeedback loops, they have to be ‘seen’ (that is, discovered) byproductive agents and ‘used’ by productive organisationsfor reconfiguring their internal processes.21 Poni’s compar-ative study of the silk industry in Lyon and Bologna clearlyillustrates this point when he highlights how feedbackmust “end in the hands of the right persons [as feedbackmanagement] require capabilities and knowledge of tech-niques which are not necessarily available in the rightmoment, in the right sector, in the right hands” (Poni, 2009,p. 297; my translation).

The likelihood that opportunities embedded in pro-ductive structures are seen by productive agents andused by productive organisations depends on three set ofissues:

(i) the individual and collective cognitive dynamicsthrough which opportunities are discovered;

(ii) the collective capabilities of productive organisationsto transform production structures (provided that cer-tain new opportunities have been discovered);

(iii) the specific form assumed by the productive organi-sation. We will discuss the first issue in the presentsection and then the last two conditions in the follow-ing section.

To return to the issue of individual and collectivecognition dynamics, psychologists (e.g. Kellogg, 1995)and experts in behavioural and organisational studies(Simon, 1986) have done a great deal of research on themechanisms responsible for agents’ memorisation. Mostinterestingly they have looked at agents’ embodiment ofperceived stimuli and past experiences such as variousforms of ‘analogical thinking’. Moreover the same stud-ies have elucidated how the positions held in certainstructures affect agents’ understanding and representationof stimuli and experiences (for a review see Buenstorf,2004).22 This set of results have been synthesised by Marchand Simon (1993, p. 335 italics added) when they stresshow “[p]roblems in learning from experience stem partlyfrom inadequacies of human cognition habits, partly fromfeatures of organisation, partly from characteristics of thestructure of experience”. In this passage there is a clearemphasis on the collective and structural dimensions ofcognition and learning.

The collective character of learning was originally high-


21 The mapping of the reconfiguration problem, as done in Fig. 3, is anheuristic for tracking complex evolving interdependencies.

22 Buenstorf (2007) is one of the few contributions addressing theseanalytical conjunctures by posing the building blocks of an evolutionarytheory of production.





al learni

organisations.24

23 See also Simon (1962) on techniques for decomposing the architectureof complexity and Scazzieri (1993, pp. 11–13) on the distinction betweensocial and technical division of labour.

24 Of course this essay recognises “the importance of the immaterial sideof production, that is, of the complex network of cognitive rules and prac-tices, customs and social norms from which production is made possible”(Landesmann and Scazzieri, 1996, p. 4). However here we have focused ourattention on the often-overlooked role that production structures play in

Fig. 3. Structur

is very much dependent on what is already known to (orbelieve by) other members of the organisation and whatkinds of information are present in the organisational envi-ronment’ (Simon, 1991, p. 125). However, as the collectiveand thus organisational dimension affect human cogni-tion habits, also the structure of experience does. Nowa fundamental analytic tension arises here. Structures ofproduction and structures of cognition are linked by a bun-dle of bidirectional transformative relationships. On theone hand, agent’s cognitive structures are continuouslyshaped by evolving productive structures (given that theformer are embedded in the latter). On the other hand, pro-ductive agents may take different decisions and reshapeproductive structures in a unpredictable way (based ofcourse on certain stimuli coming from productive struc-tures).

It will now become possible to see how the three struc-tural learning trajectories discussed above provide agentswith focusing devices to decompose the complex archi-tecture of production and select from amongst the set ofpossible learning trajectories the one they want to fol-low. In order to explain this process we make use ofthe work of Simon (1962, p. 468) and conceptualise theproduction process as a complex system “composed ofinterrelated subsystems, each of the latter being, in turn,hierarchic in structure until we reach some lowest levelof elementary subsystem”. When faced with the ‘architec-ture of complexity’, Simon (1962, pp. 472–473) suggeststhat “[P]roblem solving requires selective trial and error”and adds that “[t]he selectivity derives from various rulesof thumb, or heuristics, that suggest which paths should betried first and which leads are promising”. Thus seen, thecomplementarities, similarities and indivisibilities embed-ded in production structures can be seen to trigger andorientate cognitive dynamics in Simon’s sense, giving rise


to what we have called here structural learning dynam-ics.

Now from a methodological standpoint, in order todecompose the complex architecture of production and

ng trajectories.

investigate further these structural learning dynamics, thepaper has maintained a separation between two fun-damental levels of analysis (Simon, 1962; Rosenberg,1963; Pasinetti, 2007; Andreoni and Scazzieri, 2013). Assuggested by Luigi Pasinetti’s (2007, p. 255) separationtheorem “it is possible to disengage those investigationsthat concern the foundational bases of economic relations– to be detected at a strictly essential level of basic eco-nomic analysis – from those investigations that must becarried out at the level of the actual economic institu-tions, which at any time any economic system is landedwith, or has chosen to adopt, or is trying to achieve”.23

A similar methodological approach was envisioned byRosenberg (1963, p. 440) when he noted: “an analyti-cal explanation of many of the technological changes inthe manufacturing sector of the economy may be fruit-fully approached at the purely technological level. Thisis not to deny, of course, that the ultimate incentivesare economic in nature; rather, the point is that complextechnologies create internal compulsions and pressureswhich, in turn, initiate exploratory activity in particu-lar directions”. The next section investigates structurallearning dynamics from the level of the actual production


triggering and orientating learning dynamics of the cognitive kind. In theconcluding section the paper addresses the issue of how structural learn-ing trajectories, given the same set of structural stimuli, may be framedin different ways by different organisations according to their collectivecapabilities and the specific organisational form assumed.


ING Model

S

d Econo

4l

dbatea(asiihsfipodeMgoddctc

psSboaopthaTarosbsioitoodpt

p

agents cannot make such products. The same problem mayarise when firms attempting to satisfy increasing levels ofdemand have to introduce specialised fund inputs in order



. The organisation of production and structuralearning

As we have already discussed, opportunities embed-ed in the productive structure not only have to be ‘seen’y productive agents but also have to be captured andctualised by production organisations. The latter mayake various forms in different historical contexts and arendowed with different organisational capabilities to oper-te as collective entities. As stressed by Luigi Pasinetti2007, p. 271) “[The production paradigm cannot] abstract,s the models of exchange usually do, from historicalpecificities, since the kind of institutions that shape anndustrial society, besides being far more complex, arenherently subject to changes induced by the evolvingistorical events, much more extensively than those thathaped the era of trade”. In this respect, the notion oforms of production organisation captures the different waysn which “coordination problems have been resolved inarticular circumstances, taking into account the statef technological knowledge, the evolution of patterns ofemand, natural resources and environmental constraints,tc.” (Landesmann and Scazzieri, 1996, p. 218; see also,orroni, 2006; Jacobides and Winter, 2012). The emer-

ence and disappearance of different forms of productionrganisation testify that the coordination of tasks, pro-uctive agents and materials in transformation can followifferent patterns according to specific objectives andonstraints (see above). Thus, the ‘virtual coordination pat-erns’ actualise as ‘real responses’ to specific historical andontextual circumstances.

In evolutionary economics the collective capabilities ofroduction organisations have been referred to as ‘organi-ational capabilities’ (Nelson and Winter, 1982; Amit andchoemaker, 1993; Dosi et al., 2000). Organisational capa-ilities are a particular form of know-how which enablerganisations to perform their “basic characteristic outputctions – particularly, the creation of a tangible productr the provision of a service, and the development of newroducts and services” (Dosi et al., 2000, p. 1). In this con-ext, for an organisation ‘to be capable of something [itas] to have a generally reliable capacity to bring that thingbout as a result of intended action’ (Dosi et al., 2000, p. 2).o the opposite of organisational routines, which are char-cterised by a high degree of tacitness, automaticity andepetitiveness, capabilities are developed and deployed byrganisations as a result of intentional and conscious deci-ions. However, as routines constitute one of the buildinglocks of organisational capabilities as well as individualkills contribute to the emergence of organisational rout-nes, these two functional features of organisations – i.e.rganisational capabilities and routines – remain stronglyntertwined. Here, the central point is to understand con-extually to what extent a capability became routinised andr a routine emerge as a distinct capability. In this respectrganisational routines that are characterised by a highegree of tacitness, automaticity and repetitiveness are


roblematic since structural learning dynamics will tendo destroy old routines and introduce new ones.

By definition structural learning is not an individualrocess, since productive agents (and/or productive units


understood as organised sets of production agents) areintrinsically interdependent. In other words, to differentdegrees, structural learning involves a number of inter-dependent tasks, as well as fund inputs and materials intransformation. Thus structural learning is a systemic pro-cess, which means that firms have to be endowed withorganisational capabilities to manage all the transforma-tions entailed by structural learning trajectories. To theextent that learning trajectories are variously triggered bysimilarities, complementarities and indivisibilities, firmswill require increasing amounts of organisational capabili-ties in order to reconfigure the analytical map of productionrelationships. However, the organisational capabilitiesrequired will be different according to the specific orga-nisational form adopted by the firm and this will affectpossibility of following certain structural learning trajec-tories.

4.1. The job-shop, the putting-out system and the factorymodel

Not all forms of production organisation – such as thejob-shop, the putting-out system or the traditional factorymodel – are suitable for transforming certain structural‘constraints’ into structural ‘opportunities’ along certainstructural learning trajectories. Thus it may happen thatcertain organisational forms have to be abandoned for newones. Otherwise we can face situations in which struc-tural learning trajectories that are feasible will never beenrealised in historical time.25

For example, the job-shop model, adopted in the craftsystem, is a form of production organisation characterisedby (i) multi-task productive agents performing similartasks and (ii) a ‘stop and go’ process of material trans-formations. These two features provide the craft systemwith high flexibility and adaptability in solving unexpectedproblems, although low capacity in satisfying increasinglevels of demand. With the exception of a few productiveagents who coordinate the entire production process, in thejob-shop model productive agents tend to be highly substi-tutable and each of them is only capable of performing thesame limited set of tasks. Given this organisational form,structural learning dynamics will tend to follow patterns ofdiversification in similar activities. These do not require anyinvestment in the acquisition of new fund inputs endowedwith different capacities/capabilities.

Clearly, this organisational form is limited by a numberof quantitative (e.g. scale) and qualitative (e.g. special-isation) constraints. Thus in the case of increasinglycomplex products whose production requires the perfor-mance of dissimilar but closely complementary activities,firms using the job-shop model may have to change theirorganisational forms because non-specialised multi-tasks


25 Similarly, Henderson and Clark (1990) focus on the interplay betweenstructural features of production and organisational capabilities in thecontext of architectural innovations.


ING Model

d Econo



to scale up processes (according to specific laws of pro-portionality). Production processes operating at differentscales may then require different organisational forms.

The putting-out and the traditional factory model areresponses to some of the above-mentioned structural con-straints faced by the job shop model (Landesmann andScazzieri, 1996). The putting-out model (also known asVerlagssystem) is structured as a network of separate ‘spe-cialised workshops’, each of them performing a limitednumber of tasks related to a specific stage of fabrication.Very often the workshop (or the merchant) executing thefinal stage of production is responsible for the coordina-tion of the different production processes performed in thedifferent workshops. Sometimes they are also involved inprevious stages of fabrication, for example by assuring theprovision of raw materials (Hicks, 1969). Here, given thehigh degree of interdependence among productive tasksperformed by each member of the network as well as thefact that each workshop is highly specialised, we observeoverlapping structural learning trajectories triggered byindivisibilities and new complementarities.

In contrast, the traditional factory model was developedas a concentrated form of production in which complexproductive tasks were subdivided in an increasing numberof elementary operations performed by highly specialisedproductive agents inside the same production organisa-tion. Historically the traditional factory model was adoptedin the automotive industry at the time in which Ford andGeneral Motors were dominating the global car produc-tion. Here, both workers and machines and, thus, theircapabilities and capacities, are coordinated in a way thatguarantee their full and continuous utilisation in executingnetworks of dissimilar tasks (Landesmann, 1986, p. 294).By increasing the scale of production, in the factory con-text indivisible funds can be more efficiently utilised and,thus, both economies of scale and scope achieved. In thetraditional factory context, structural learning dynamicstriggered by indivisibilities tend to be pervasive. This lastpoint was outlined in the ‘Maxcy-Silberston curve’ in thespecific context of Western vehicle manufacturers.

4.2. The lean production system

Just as the factory model developed as a response toa series of structural constraints characterising previousforms of production organisation (see above), the ‘lean pro-duction system’ was introduced as a response to structurallimitations of the traditional factory model. Firstly, the tra-ditional factory model is too rigid for responding to firms’increasing need to accommodate consumer preferences forproduct diversity and, as a result, to produce large varietiesin small volumes. Secondly, given the scale and organisa-tion in time of production stages, the traditional factorymodel is handicapped by large inventories and a relativelyhigh number of defects. Thirdly, diversification in closelycomplementary but dissimilar activities, such as producing


“a particular car with a particular brake and a particularbrake lining” in the same factory (Richardson, 1972, pp.891–892), requires increasing investment for building (oracquiring) new bundles of very specialised and diversified


capabilities and, very often, a complete reconfiguration ofthe job specification programme.

The lean production system was pioneered by Toyotaand resulted from the visionary ideas of its mechani-cal engineer Taiichi Ohno (Cusumano, 1985; Ohno, 1988;Fujimoto, 1999). The Toyota Production System inventedby Ohno was based on two fundamental pillars: ‘autono-mation’ and ‘just in time’ (JIT). The former, the introductionof ‘autonomous machines’ in production, opened up thepossibility of reducing costs by eliminating waste of mate-rials and machines’ idle times. The JIT developed the ideaaccording to which “in a comprehensive industry such asautomobile manufacturing, the best way to work would beto have all the parts of the assembly at the side of the linejust in time for their user” (Ohno, 1988, p. 75).

The application of these two principles resulted in the‘small-lot’ production technique, i.e. a combination of theflexibility and high quality standards of craft productionwith the low cost of mass production techniques (Womacket al., 1990). This form of production organisation is char-acterised by higher levels of flexibility for two reasons: (i)the costs of switching from one product line to anotherare minimised and (ii) multi-task workers organised inteams are equipped with less highly specialised machinesand tools than those used in the mass production factorymodel. The high quality standards of production are alsomade possible by the fact that every worker is allowed tostop production every time a fault is discovered (insteadof assigning this decision to the senior line manager) andby the fact that product’s components are supplied to thework station just in time (instead of keeping large stocks ofeach component beside the work station).

Although this organisational arrangement implies thatinitially stoppages in the production line are frequent, inthe medium run workers are increasingly able to discoverthe sources of problems (and their interdependencies)in the space of capabilities, ‘materials in use’ and ‘taskexecution’. These discoveries trigger a sequence of struc-tural learning dynamics according to which solutions toa certain production problem or bottleneck are appliedto similar problems. Additionally solutions to a particu-lar production problem or bottleneck make the solution ofcomplementary problems necessary. As the complemen-tary production problems are identified and solved thenumber of stoppages diminishes to the point that theybecome much less frequent than in the typical mass pro-duction assembly line.

In the lean production system structural learning trajec-tories are also triggered by the fact that design teams workclosely with production engineers and producers of prod-uct components. As a result the specification of productdesign proceeds hand in hand with the design, calibrationand adaptation of tools and equipment that are used in pro-duction. In this way not only does the overall productionsystem achieve high quality standards, but also in the prod-uct’s design process a stream of diversified products rapidlydevelop. Thus the overall production system experiences


reductions in the unit costs of production. The successfulapplication in the Japanese car industry of this form of pro-duction organisation was at the centre of the InternationalMotor Vehicle Programme started in 1979 whose results


ING Model

S

d Econo

wo

oinesdii5aaieimaptdtaa2

tcattt(smi

5

oehmcitHg(eas

st

a



ere collected five years later in the MIT book The Futuref Automobile (1984).26

As is amply documented in Fujimoto’s (1999) reviewf the evolution of the Toyota Production System andts transformation into the lean production system, theseew forms of production organisation allowed firms tonter structural learning trajectories which were unfea-ible within the traditional factory model. Indeed theiriscovery was the result of a structural learning dynamic

n itself which involved precisely what we have called herentersectoral learning. In the words of Fujimoto (1999, p.0 italics added) ‘Toyota’s production organisation [. . .]dopted various elements of the Ford system selectivelynd in unbundled forms, and hybridised them with theirngenious system and original ideas. It also learnt fromxperiences with other industries’. In particular, as reportedn Cusumano (1985), Taiichi Ohno declared that “the auto-

otive loom was a text book in front of [his] eyes”. Thepplication of the same solutions to similar productionroblems arising in different sectors was at the very root ofhe Toyota Production System and its evolution in lean pro-uction system or lean production technique. Throughouthe 1990s lean production techniques were increasinglypplied from the automotive to other industries suchs aerospace, producing highly-complex products (Roos,003).

To recap, in all the above-mentioned cases, even if cer-ain structural stimuli and feedback loops were to makeertain structural learning trajectories feasible, only firmsdopting a certain organisational form will be capableo follow these trajectories. Specific organisational fea-ures of production organisation may enable (or block) thehree fundamentally alternative routes described abovestructural learning trajectories triggered by discoveringimilarities, those triggered by discovering new comple-entarities and, finally, those triggered by overcoming

ndivisibilities).

. Concluding remarks

Learning dynamics are the main transformative forcesf economic systems. Economists have always been inter-sted in this fundamental reality. Some recent studiesave attempted to link micro-learning dynamics andacro-transformative effects by tracking countries’ spe-

ialisation/diversification patterns driven by similaritiesn the ‘product space’, a network-type representation ofhe international market architecture (e.g. Hidalgo andausmann, 2009). However, these studies do not disentan-le the different forms of learning realised at the firm levelas we have done in this paper) and thus are not able toxplain how learning dynamics trigger structural changend economic growth of economic systems at differenttages of development.


As learning processes are embedded in productivetructures, any attempt to understand how economic sys-ems change over time through learning dynamics cannot

26 Among others see Bianchi’s contribution on ‘Flexible Manufacturingnd Product Differentiation in the automobile industry’.


avoid looking at the reality of production processes. In thisrespect, the contribution of this paper is twofold. Firstly,building on structural theories dealing with the com-plex architecture of production, the paper has proposeda new heuristic for analysing interdependencies amongcomponents of production processes. The ‘analytical mapof production relationships’ provides a stylised represen-tation of the system of interrelated tasks through whichtransformations of materials are performed according todifferent patterns of capacities/capabilities coordination,subject to certain scale and time constraints. On this basisthe two main forms of learning (‘learning by doing’ and‘learning by using’) have been re-formulated in a way thatsees them as being affected by and affecting the productionstructure.

Learning in production takes many forms and is realisedat several interconnected (nested) levels, the pattern ofnesting being structurally determined. Thus the conceptof structural learning has been introduced to identify thecontinuous process of structural adjustment triggered andorientated by existing and evolving productive structures.The paper identifies three main structural learning trajec-tories. Static and dynamic complementarities, similaritiesand indivisibilities are essential triggers for activating com-pulsive sequences of technological change as well as fordiscovering new productive possibilities at the firm and inter-firm level.

These structural learning dynamics have to be ‘seen’ byproductive agents and ‘used’ by productive organisations.At this point, we have identifies a fundamental tensionunderlying structural learning dynamics. Namely, the factthat structures of production and structures of cognitionare linked by a bundle of bidirectional transformative rela-tionships. This tension is partially solved by the adoptionof different forms of productive organisations, whose spe-cific features may enable (or block) a number of structurallearning trajectories.

From a methodological standpoint, in order to decom-pose the ‘architecture of complexity’ in production as wellas investigate structural learning dynamics, the paper hasmaintained a ‘separation’ between two fundamental lev-els of analysis. At one level, the analysis focused on thosedynamics inherent in productive structures independentlyof specific organisational/institutional configurations. At adeeper level, the analysis considered how, given certainpossibilities embedded in productive structures, differentorganisations may follow different structural learning tra-jectories. The analytical account of specific historical caseshas been adopted as the main heuristic for disentanglingstructural learning dynamics. At this point in the analy-sis the historical emergence and reappearance of differentforms of productive organisation have been stressed.

Looking at the production process and its transfor-mations from a structural perspective has allowed us tore-link production and learning dynamics. This goal hasprofound implications for policy design. Structural learningtrajectories are transformative processes operating within


the black box of production and, as such tend to remain‘invisible’ to policymakers. The ‘political economy of struc-tural learning’ suggests a number of unconventional policyoptions, such as the possibility of policy intervention in


ING Model

d Econo



sectors, tasks and capabilities that are similar or comple-mentary to those that were initially taken as the objectof policy intervention. In this respect Silver (1984) arguedthat “in developing countries, or in developed economieswhen innovation renders the market’s existing capabilitiesobsolete, a firm may have to integrate into many dissimilaractivities in order to generate all the complementary activ-ities it needs” (Langlois, 1992, p. 108; see also O’Sullivanet al., 2013). In economies in their catch-up phase, whereconstraints in production structures appear pervasive,the structural learning perspective also suggests possiblestrategies for overcoming indivisibilities or scale-invariantprocess constraints. Finally it points to the possibility ofdiscovering unexploited opportunities embedded in exist-ing productive structures at the sectoral and intersectorallevels, following patterns of diversification in similar activ-ities, and building/exploiting technological linkages withthose dissimilar activities towards the selective creation ofnew productive opportunities. The final aim of these struc-tural learning policies is to facilitate the discovery of new‘worlds of possibilities’ and, thus, the emergence of ‘new‘worlds of production’.

References

Adler, P., Clark, K., 1991. Behind the learning curve: a sketch of the learningprocess. Manage. Sci. 37, 267–281.

Amit, R., Schoemaker, P., 1993. Strategic assets and organisational rent.Strat. Manage. J. 14, 33–46.

Andreoni, A., 2011a. Manufacturing agrarian change: intersectoral learn-ing, technological capabilities and transformative policies, DRUIDWorking paper, 11–13.

Andreoni, A., 2011b. Productive Capabilities Indicators for Industrial Pol-icy Design, UNIDO Working Paper, 17–11.

Andreoni, A., Scazzieri, R., 2013. Triggers of change: produc-tion structures and economic dynamics. Camb. J. Econ.,http://dx.doi.org/10.1093/cje/bet034.

Andreoni, A., Gregory, M., 2013. Why and How Does Manufacturing stillMatter: Old Rationales, New Realities. Rev. d’Écon. Indus. (forthcom-ing).

Altshuler, A., Anderson, M., Jones, D.T., Roos, D., Womack, J., 1984. TheFuture of the Automobile. The MIT Press, Cambridge, MA.

Atkinson, A., Stiglitz, J., 1969. A new view of technological change. Econ. J.79, 573–578.

Arrow, K., 1962. The Economic Implications of Learning by Doing. Rev.Econ. Stud. 29 (3), 155–173.

Babbage, C., 1835. [1832] On the Economy of Machinery and Manufactur-ing. Charles Knight, London.

Bell, M., 1982. Technical Change in Infant Industries: A Review of theEmpirical Evidence. SPRU, University of Sussex (mimeo).

Bianchi, P., 1984. Divisione del Lavoro e Ristrutturazione Industriale. IlMulino, Bologna.

Bourdieu, P., 1972. Outline of a Theory of Practice. Cambridge UniversityPress, Cambridge.

Buenstorf, G., 2004. The Economics of Energy and the Production Process:An Evolutionary Approach. Edward Elgar, Cheltenham.

Buenstorf, G., 2005. Sequential production, modularity and technologicalchange. Struct. Change Econ. Dyn. 16, 221–241.

Buenstorf, G., 2007. Complementarity, cognition and capabilities: towardsan evolutionary theory of production. In: Ioannides, S., Nielsen, K.(Eds.), Economics and the Social Sciences. Edward Elgar, Cheltenham,UK.

Cartwright, N., 1989. Nature’s Capacities and their Measurement.Clarendon Press, Oxford.

Chang, H.-J., 2002. Kicking Away the Ladder. Anthem Press, London.Cohen, W.M., Levinthal, D., 1990. Absorptive capacity: a new perspective

on learning and innovation. Adm. Sci. Q. 35 (1), 128–152.


Cusumano, M.A., 1985. The Japanese Automobile Industry: Technologyand Management at Nissan and Toyota. Harvard University Press,Boston.

David, P.A., 1966. The mechanisation of reaping in ante bellum Midwest.In: Rosovsky H. (Eds.), Industrialization in Two Systems: Essays in


Honour of Alexander Gerschenkron. Wiley, New York. Republished inDavid, P.A., 1975. Technical Choice, Innovation and Economic Growth.Cambridge University Press, Cambridge.

Dodgson, M., 1991. The Management of Technological Learning. DeGruyter, Berlin.

Dosi, G., Nelson, R.R., Winter, S., 2000. The Nature and Dynamics of Orga-nizational Capabilities. Oxford University Press, Oxford.

Fujimoto, T., 1999. The Evolution of a Manufacturing System at Toyota.Oxford University Press, Oxford.

Garnsey, E., McGlade, J., 2006. Complexity and Co-Evolution: Continuityand Change in Socio-Economic Systems. Edward Elgar, Cheltenham,UK/Northampton, USA.

Georgescu-Roegen, N., 1970. The economics of production. Am. Econ. Rev.LX (2), 1–9.

Georgescu-Roegen, N., 1971. The Entropy Law and the Economic Process.Harvard University Press, Cambridge, MA.

Georgescu-Roegen, N., 1976. Energy and Economic Myths. PergamonPress, New York.

Georgescu-Roegen, N., 1990. Production process and dynamic eco-nomics’. In: Baranzini, M., Scazzieri, R. (Eds.), The Economic Theoryof Structure and Change. Cambridge University Press, Cambridge,pp. 198–226.

Gibbons, R., Roberts, J. (Eds.), 2011. Handbook of Organisational Eco-nomics. Princeton University Press, Princeton.

Henderson, R., Clark, K., 1990. Architectural innovation: the reconfigura-tion of existing product technologies and failure of established firms.Adm. Sci. Q. 35 (1), 9–30.

Hicks, J., 1969. A Theory of Economic History. Clarendon Press, Oxford.Hidalgo, C., Hausmann, R., 2009. The building blocks of economic com-

plexity. Proc. Natl. Acad. Sci. 106 (26), 10570–10575.Hippel von, E., 1988. The Sources of Innovation. Oxford University Press,

Oxford.Jacobides, M.G., Winter, S.G., 2012. Capabilities: structure, agency and

evolution. Organ. Sci. 23 (5), 1365–1381.Jovanovic, B., Nyarko, Y., 1995. A Bayesian learning model fitted to a variety

of empirical learning curves. Brook. Pap. Microecon., 247–305.Kline, S., Rosenberg, N., 1986. An overview of innovation. In: Landau, R.,

Rosenberg, N. (Eds.), The Positive Sum Strategy. National AcademyPress.

Kellogg, R.T., 1995. Cognitive Psychology. SAGE, Thousand Oaks.Lall, S., 1992. Technological capabilities and industrialization. World

Develop. 20 (2), 165–186.Landesmann, M., 1986. Conceptions of technology and the production pro-

cess. In: Baranzini, M., Scazzieri, R. (Eds.), Foundations of Economics.Blackwell, Oxford, pp. 281–310.

Landesmann, M., 1988. Where do we go from here? Technical change anddynamics. Camb. J. Econ. 12, 169–177.

Landesmann, M., Scazzieri, R., 1996. Production and Economic Dynamics.Cambridge University Press, Cambridge.

Langlois, R.N., 1992. Transaction-cost economics in real time. Indus. Corp.Change 1 (1), 99–127.

Langlois, R.N., 2002. Modularity in technology and organization. J. Econ.Behav. Organ. 49, 19–37.

Landes, D., 2000. Revolution in Time. Clocks and the Making of the ModernWorld. Belknap Press of Harvard University Press, Cambridge, MA.

Leontief, W., 1928. The economy as a circular flow. Struct. Change Econ.Dyn. 2, 177–212.

Leontief, W., 1947. Introduction to a theory of the internal structure offunctional relationships. Econometrica 15 (4), 361–373.

Lissoni, F., 2005. The reaper and the scanner: indivisibility-led incrementalinnovations and the adoption of new technologies. Camb. J. Econ. 29,359–379.

Loasby, B.J., 1999. Knowledge, Institutions, and Evolution in Economics.Routledge, London New York.

Lundvall, B.A., 1992. National Systems of Innovation: Towards a Theory ofInnovation and Interactive Learning. Pinter, London.

Malerba, F., 1992. Learning by firms and incremental technical change.Econ. J. 102 (413), 845–859.

March, J., Simon, H., 1993. [1958] Organizations, 2nd ed. Wiley and Sons.Ménard, C., 2004. The economics of hybrid organizations. J. Inst. Theor.

Econ. 160 (3), 345–376.Mokyr, J., 1990. The Lever of the Riches: Technology Creativity and Eco-

nomic Progress. Oxford University Press, Oxford.Mokyr, J., 2002. The Gifts of Athena: Historical Origins of the Knowledge


Economy. Princeton University Press, Princeton.Morroni, M., 1992. Production Process and Technical Change. Cambridge

University Press, Cambridge.Morroni, M., 2006. Knowledge, Scale and Transactions in the Theory of the

Firm. Cambridge University Press, Cambridge.


http://refhub.elsevier.com/S0954-349X(13)00065-9/sbref0005



























































dx.doi.org/10.1093/cje/bet034
















































































































































































































































































































































































































































































































































































































































































































ING Model

S

d Econo

M

M

N

N

O

O

P

P

P

P

PR

RR

R

R

R

R

R

R

R

Press, Cambridge, MA.Womack, J.P., Jones, D.T., Roos, D., 1990. The Machine that changed the



owery, D., Rosenberg, N., 1998. Paths of Innovation: Technologi-cal Change in 20th-Century America. Cambridge University Press,Cambridge.

ukoyama, T., 2006. Rosenberg’s “learning by using” and technology dif-fusion. J. Econ. Behav. Organ. 61, 123–144.

elson, R.R., Winter, S., 1982. An Evolutionary Theory of Economic Change.Belknap Press, Cambridge, MA.

oble, D., 1986. Forces of production: a social history of industrial automa-tion. Oxford University Press, Oxford.

hno, T., 1988. The Toyota Production System: Beyond Large-Scale Pro-duction. Productivity Press, Portland.

’Sullivan, E., Andreoni, A., Lopez-Gomez, C., Gregory, M., 2013. Whatis new in the new industrial policies? Oxf. Rev. Econ. Policy,http://dx.doi.org/10.1093/oxrep/grt027.

asinetti, L.L., 2007. Keynes and the Cambridge Keynesians. CambridgeUniversity Press, Cambridge.

enrose, E.T., 1959. The Theory of the Growth of the Firm. Oxford Univer-sity Press, Oxford.

iore, M., Sabel, C., 1984. The Second Industrial Divide: Possibilities forProsperity. Macmillan, New York.

itelis, C., Teece, D., 2009. The (new) nature and essence of the firm. Eur.Manage. Rev. 6, 5–15.

oni, C., 2009. La seta in Italia. Il Mulino, Bologna.ichardson, G.B., 1960. Information and Investment. Oxford University

Press, Oxford.ichardson, G.B., 1972. The organisation of industry. Econ. J. 82, 883–896.ichardson, G.B., 2003. The organisation of industry re-visited. In: DRUID

Summer Conference.hee, Y.W., Ross-Larson, B., Pursell, G., 1984. Korea’s Competitive Edge.

John Hopkins University Press, Baltimore.oos, D., 2003. The Development of Engineering Systems at MIT

, Engineering Systems Workshop, MIT, October, Cambridge,MA.

osenberg, N., 1963. Technological Change in the Machine Tool Industry,1840–1910. J. Econ. Hist. 23 (4), 414–443.

osenberg, N., 1969. The direction of technological change: inducementmechanisms and focusing devices. Econ. Develop. Cult. Change 18,1–24.

osenberg, N., 1976. Perspectives on Technology. Cambridge University


Press, Cambridge.osenberg, N., 1979. Technological Interdependence in the American

Economy. Technol. Cult. 20 (1), 25–50.osenberg, N., 1982. Inside the Black Box: Technology and Economics.

Cambridge University Press.


Rosenberg, N., 1994. Exploring the Black Box: Technology, Economics, andHistory. Cambridge University Press, Cambridge.

Salais, R., Storper, M., 1997. Worlds of Production: The Action Frameworksof the Economy. Harvard University Press, Cambridge, MA.

Sabel, C.F., Zeitlin, J. (Eds.), 1997. World of Possibilities: Flexibility andMass Production in Western Industrialization. Cambridge UniversityPress, New York.

Scazzieri, R., 1981. Efficienza produttiva e livelli di attività: un contributodi teoria economica. Il Mulino, Bologna.

Scazzieri, R., 1993. A Theory of Production. Tasks, Processes and TechnicalPractices. Clarendon Press, Oxford.

Scazzieri, R., 1999. A Theory of resilient flow-fund linkages. In: Mayumi,K., McGowdy, J. (Eds.), Bioeconomics and Sustainability. Edward Elgar,North. USA.

Silver, M., 1984. Enterprise and the Scope of the Firm. Martin Robertson,London.

Simon, H., 1957. Administrative Behaviour. A Study of Decision-MakingProcess in Administrative Organisation. Macmillan, New York.

Simon, H., 1962. The Architecture of Complexity. Proc. Am. Philos. Soc. 106(6), 467–482.

Simon, H., 1986. Rationality in psychology and economics. Journal of Busi-ness 59, S209–S224.

Simon, H., 1991. Bounded rationality and organizational learning. Organ.Sci. 2 (1), 125–134.

Smith, A., 1976[1776]. An Inquiry into the Nature and Causes of the Wealthof Nations, General Editors R.H. Campbell and A.S. Skinner; TextualEditor W. B. Todd. Clarendon Press, Oxford.

Stiglitz, J., 1988. Learning to learn, localized learning and tech-nological progress. In: Dasgupta, P., Stiglitz, J. (Eds.), EconomicPolicy and Technological Performance. Cambridge University Press,Cambridge.

Teece, D., 1980. Economics of scope and the scope of the enterprise. J. Econ.Behav. Organ. 1, 223–247.

Thompson, P., 2010. Learning by doing. In: Hall, B.H., Rosenberg, N.(Eds.), Handbook of the Economics of Innovation. Elsevier, NorthHolland.

Usher, A.P., 1954. A History of Mechanical Inventions. Harvard University


World. Harper-Collins, New York.Wright, T.P., 1936. Factors affecting the cost of airplanes. Journal of Aero-

nautical Sciences 3 (4), 122–128.


































































dx.doi.org/10.1093/oxrep/grt027



























































































































































































































































































































































































































streco-561; article in presseprints.soas.ac.uk/19216/1/2013 structural learning.pdf · 2014. 12....

Documents