Learning Pathways for Energy Supply Technologies: bridging between Innovation Studies and Learning Rates

Mark Winskela,*, Nils Markussonb, Henry Jeffreya, Chiara Candelisec, Geoff Duttond, Paul Howarthe, Sophie Jablonskic, Christos Kalyvasf and David Wardg

aInstitute of Energy Systems, School of Engineering, University of Edinburgh, EH9 3JL, UK.

bSchool of Geosciences, University of Edinburgh, EH9 3JW, UK.

cImperialCollege Centre for Energy Policy and Technology, London, SW7 2AZ, UK.

d STFC Rutherford Appleton Laboratory, Didcot, Oxfordshire, OX11 0QX, UK.

e Dalton Nuclear Institute, University of Manchester, Manchester, M13 9PL, UK.

f Department of Earth Science and Engineering, ImperialCollege, LondonSW7 2AZ, UK.

g Culham Science Centre, Abingdon, Oxfordshire, OX14 3DB, UK.

* Corresponding author: Email: ; Address: Institute for Energy Systems, University of Edinburgh, KingsBuildings, Edinburgh EH9 3JL; Tel. +44 (0) 131 650 5594

Abstract

Supporting innovation and learning for different emerging low carbon energy supply technology fields is a key issue for policymakers, investors and researchers. A range of contrasting analytical approaches are used, often with little cross-over between them. Energy systems modelling using learning rates provides abstracted, quantitative and output oriented accounts, while innovation studies research offers contextualised, qualitative and process oriented accounts. Drawing on research evidence and expert consultation on learning for several different emerging energy supply technologies, this paper introduces a ‘learning pathways’ matrix to help bridge between the rich contextualisation of innovation studies and the systematic comparability of learning rates. The learning pathways matrix characterises technology fields by their relative orientation to radical or incremental innovation, and to concentrated or distributed organisation. A number of archetypal learning pathways are outlined to help learning rates analyses draw on innovation studies research,and so better acknowledge the different niche origins and learning dynamics of energy supply technologies. Finally, future research issues are outlined.

Keywords

Innovation; learning; niches;pathways; energy; electricity; technology

1. Introduction

National and international policy ambitions for greenhouse gas emissions reduction and low carbon technology deployment have focused attention on energy system transformation[1, 2]. One of the key dynamics – and uncertainties – associated with this envisaged transformation relates to the development and deployment of low carbon energy supply technologies. While technological innovation holds out considerable promise as an enabler of more affordable energy system change [3, 4], this promise is highly uncertain, particularly over decades-long timescales.

Technological innovation in the energy sector is driven by a complex mix of incentives and interests [5, 6], and there is now a large number of emerging low carbon energy supply technology fields, each supported by particular policy initiatives, investment programmes, developer firms and research institutions . Making sense of this activity – in terms of systematic ordering, comparing and assessing its effectiveness and potential– has become a major policy and research challenge in its own right [3, 7, 8].

A range of tools and frameworks are drawn on, including technology roadmaps [1, 9], energy system models [3, 10] and explorative scenario planning techniques [11, 12]. Each provides particular insights. Technology roadmaps specify the anticipated sequences involved in the progressive commercialisation of emerging technologies in considerable detail. System modelling provides ‘structured insights’ into the interactions and trade-offs between different parts of the whole energy system [13, 14]. Explorative scenario exercises consider alternative possible futures in the context of social and economic trends and potentially disruptive events [15], and may therefore capture more diverse combinations of envisaged social and technical futures than either system models or roadmaps.

These different techniques are not mutually exclusive, and indeed, they are sometimes used in combination[16]. At the same time, each approach has its limitations and blind-spots. Roadmaps may under represent the wider socio-technical context for innovation, including interactions between different technologies and the more socio-political aspects of innovation, factors which are often especially important for energy technologies. Different roadmaps may also articulate inconsistent levels of optimism or ambition across different technology communities. This is especially problematic in early stage technology assessment because of a lack of any empirical track record, and a tendency to ‘appraisal optimism’ [17]or even hype [18].

Energy system models, in elaborating a broad system-level view, may over over-simplify, either by under-representing the uncertainties, contingencies and non-linearities of system change, or by only allowing for highly aggregated, crude representations of key system drivers such as technological innovation. Even relatively detailed bottom-up energy system models tend to characterise supply technologies by a small set of parameters, such as capital and operating cost, resource availability and conversion efficiency, with innovation dynamics often represented by a single parameter: the experience (or learning) rate[19] (the term ‘learning rate’ is used in this paper because of its resonance with the concern here for learning effects). Reducing down innovation processes to a single aggregated parameter means that many their important properties go unrepresented, such as the qualitative difference between early stage and later stage innovation dynamics[20], or the often key role of market diversity, including niche markets, in early stage innovation [21].

Finally, explorative scenarios often provide only rather ‘broad-brush’ characterisations of socio-technical trends or possible reconfigurations, and tend to lack any detailed account of the causal mechanisms (agents, institutions and policies) by which their envisaged outcomes may be realised [22, 23].

Alongside these widely used tools and methods is a body of mainly qualitative social science research, innovation studies, which also analyses the dynamics of emerging technology systems. Low carbon energy innovation studies has become a highly active research field over recent years, developing and applying frameworks such as Technological Innovation Systems (TIS) [24-26]and the multi-level perspective (MLP) on system transition [27-30]. Although TIS and MLP have distinctive strengths and weaknesses [31] both provide richly contextualised and contingent accounts of innovation, in terms of interwoven and co-evolving social and technical elements; case study research based on them has provided many detailed accounts of the evolution of energy technology systems[20, 32, 33].

However, despite their common concern to take into account the role of innovation in socio-technical system change, there is strikingly little cross-over between the abstracted representations of learning rates and system modelling, and the contextualised, contingent accounts of innovation studies. The premise for the learning pathways framing outlined here is that there are missed opportunities for cross-disciplinary interaction here, and in particular, that innovation studies offer a valuable research resource to enrich learning rates analysis.For example, learning rates and system modelling tend to see techno-economic performance, measured as unit cost, as the determining factor on technological change, while innovation studies highlights a broader set of socio-cultural forces (such as the role of knowledge flows and information sharing in early stage innovation, political legitimacy and societal acceptability, and the enabling or blocking role of incumbent organisational interests) [25, 32, 34].

The aim of this paper is to help bridge between system modelling and innovation studies by developing a simple analytical framework that allows for systematic comparison of energy technologies, while retaining some of the contextual richness of innovation studies. The formidable task of developing a full synthesis of system modelling and innovation studies is beyond our scope. Instead, we take technology-specific innovation studies as a starting point, develop a 2x2 matrix to improve its comparability without losing too much specificity, and then consider the cross-overs or implications for learning rates and system modelling.

Our efforts are inspired and informed by earlier attempts within innovation and organisational studies at developing comparative frameworks for technology comparison [35-41]. Within the ‘comparison-oriented’ stream of innovation and organisational studies, the learning pathways approach pays particular attention to the differing niche origins of emerging energy technologies and their different development pathways. By describing a set of archetypical or generic learning pathways, the prospect is opened up oflearning narratives that are part-contextualised but which also allow technologies to be compared.

The paper proceeds as follows. Section 2 summarises and contrasts system modelling using learning rates and innovation studies; Section 3 introduces the learning pathways matrix, drawing on detailed analyses of a few energy supply technologies; Section 4 applies the learning pathways matrix to compare the niche origins and learning paths of selected technologies. Section 5 presents a set of archetypal learning pathways for use in technology forecasting for energy system change. Section 6 summarises and concludes the paper, and identifies some areas for future research.

2. Review: Perspectives on Technology Learning

2.1 Learning Rates

Learning rates first emerged from historical evidence of cost reduction with cumulative production in manufacturing industries[42, 43]. The learning rate is the percentage reduction in technology ‘unit costs’ associated with each doubling of installed cumulative capacity [44]. Over recent years, in the context of policy targets for energy system change, learning rates have been used in many energy system modelling exercises [45-47], either formulated endogenously within the model, or factored-in exogenously using off-model calculations [48].

The learning rate is a powerful analytical construct, given its apparent ability to capture and quantify innovation, and project it forwards as a key part of wider socio-technical system change. In practice, technological innovation is more complex and less predictable than this suggests, and comparing different technologies on the basis of learning rates disguises important differences. As a number of observers have recognised, using learning rates for long-term energy system projections raises particular concerns [43, 49-56]; these include:

  • the assumed correlation between deployment and cost reduction is not always observed: history shows examples of some energy technologies, such as nuclear power, coal-fired steam turbines and offshore wind energy failing to lower costs despite significant deployment [57-60].
  • even when a correlation is observed, the direction of causality is often unclear:unit cost reductions may result from market growth, or be a driver of market growth [61].
  • over time, apparently small differences in learning rate estimates have dramatic impacts on suggested investments needed for commercial breakthrough of individual technologies, and, at the system level, optimal energy mixes [51, 56, 57].
  • cumulative deployment is a poor measure of learning for research-intensive technologies. ‘Two-factor’ learning rates [62, 63] explicitly allow for learning by research, alongside learning by doing, but they exacerbate the problem of input data uncertainties. Junginger et al. [18] suggested that time rather than deployment may be a better indicator of cost reduction potential for R&D intensive technologies.
  • using a single learning rate for a technology field is likely to disguise significant contextual diversity over place, time and content:
  • energy innovation policies are still determined, to a significant degree, at the national and sub-national level, and innovation dynamics differ across regions, nations and organisations
  • discontinuities and step-changes in learning are often seen over time, through different phases of development. Colpier and Cornland [64] distinguished between price umbrella, shakeout and stability phases;Grubler [65] and Wilson [66] identified four phases of development, including early phase experimentation, unit scaling, industry scaling and global diffusion.
  • learning effects often differ considerably between the component parts of a technology system [67].

While these differencesare ironed-out over long-rungloballearning rate studies, they are significant for those operating at any level of detail, including policymakers, business strategists and research programme managers.

The need for improved representations of innovation in learning rates and energy system modelling has been recognised by modelling researchers. Gielen et al. [68]outlined lines of enquiry for a ‘technology learning research agenda’, including analysis of clustering of interrelated technologies, identification of global and regional learning patterns, and attention to underlying ‘autonomous trends’ such as increased computer capacity and advanced materials. Berglund and Söderholm [69]suggested linking exogenous assumptions about learning rates to different assumptions about policy, cumulative capacity and R&D investment. Clarke et al.[61](p593) called forstudy of ‘any distortions of policy conclusions from models with limited representations of technological change’.

2.2 Innovation Studies

Learning rates are a widely adopted, influential tool, and their refinement has become a highly active research area in recent years. At the same time, their well-documented flaws suggest a need to draw on accounts of innovationable to retain greater complexity and contextualisation. Reflecting its conceptual groundings in evolutionary economics and sociology, innovation studies – especially in its more technology-rich strands – emphasises the contingent nature of technological development, and the need to analyse technology systems in their socio-historical context, rather than by reference to technical or economic imperatives [70, 71]. A central insight offered by this body of research is that the various elements that make up a technology system: technical artefacts and knowledge, and also practices, institutions and expectations, interact together in co-evolutionary and path-dependent ways over time [31]. Foxon [72] identified five key domains for energy system co-evolution: technologies, institutions, business strategies, natural ecosystems and social practices.

Two prominent conceptual frameworks have developed within innovation studies over the last two decades: the Technological Innovation Systems (TIS) approach[20, 24-26, 32] and the Multi-Level Perspective (MLP) on system transitions [27-30]. TIS studies emphasise multiple agency and distributed learning in innovation processes. Rather than all-powerful technologists, or linear knowledge flows, the focus tends to be on interaction and feedbacks across different system elements: actors, networks and institutions [73]. Two broad phases of technology development are often identified: an initial, formative phase, characterised by the generation of technological variety, the testing of different designs, interactive learning, niche markets, and efforts to establish organisational and political legitimacy; and a subsequent market expansion phase, in which positive feedbacks between market growth, learning by doing and scale economies enable diffusion of a dominant technology into mass markets, allowing for the run down of public policy support mechanisms [20, 25, 32].

The MLP conceives technological innovation as an outcome of the interplay of social and technical elements over three distinct levels of aggregation: micro-level niches, meso-level regimes and macro-level societal landscapes [74, 75]Within this, ‘system innovations’ (innovations with the greatest impact on the design and character of systems, including their environmental performance) are seen to originate mainly in radical niches, which over time become stabilised in dominant designs, and which may subsequently break through to reconfigure regimes[76]. Case research using the MLP has often involved long-term historical studies, such as the transition from coal- to gas-based energy supply in The Netherlands [77]. In the context of the ‘managed transitions’ of energy systems to meet decarbonisation targets, Shackley and Green [78] suggested that the approach could usefully supplement modelling and scenario-based analysis. Verbong and Geels [79], for example,described three different prospective ‘transition pathways’ for the electricity system and grid infrastructures, based on different niche-regime-landscape interactions. Foxon et al. have elaborated different transition pathways for UK electricity system transition [16].

The rising prominence of TIS and the MLPin innovation studiesprovoked some criticism. For TIS, and innovation systems studies more broadly, criticisms included inconsistencies across different studies in terms of system delineation and measures of system performance, and reliance on mainly ex-post qualitative analysis[80, 81]. Criticisms of the MLP included inconsistent conceptual framing, a neglect of agency, an over-emphasis on niches as drivers of system change [38, 82]. Later contributions to both theories have sought to respond to these criticisms. In TIS, this involved the development of a more standard, prescriptive analytical framework, based on a set of system ‘functions’ (including, for example, knowledge development, market formation and resource mobilization) [24, 25]. In MLP studies, in an effort to overcome ‘niche-driven bias’, Geels and Schot [28]introduced a small set of archetypal transition pathways, based on particular niche-regime-landscape relationships.

These later contributions have directed innovation studies towards greater comparability between different cases. In offering these more standardised accounts, there is a danger of under-emphasising significant differences of content or context between different technologies. For example, TIS’s framing bysystem functions and inter-functional patterns may fail to capture important differences in niche origin and learning paths (such as different levels of importance of learning by doing, interacting, and researching, scale economies, or learning by transfer from other industries [18, 61, 83-85]. Bergek et al.[25] identified a need for a taxonomy of archetypal development pathways for emerging innovation systems; Section 5 of this paper offers such a taxonomy.

Case research using the MLP is deeply concerned with niche context, and a particular strand of transitions research based on ‘innovation journeys’ has paid particular attention to the shifting dynamics of technology learning over time [33, 86]. Even so, the MLP tends to a niche-led account of system innovation, with more significant system change arising from radical and disruptive niches. By doing so, like TIS, it may under-represent the different origins and learning pathsof different technologies, or the powerful role, over time, of incremental innovation. Smith et al. [87]called for greater attention to be paid to the plurality of niches and ‘niche-regime’ interactions. The learning pathways matrix can be seen as a response here, and also, to the prospect of regime-led system innovation under urgent change imperatives[88]. Foxon [72] has noted the neglect of cost and economic factors in much innovation studies based on the MLP; this strengthens the case for seeking bridges between innovation studies and cost-based methods such as learning rates.