126.96.36.199 The sources of technological change
New technology arises from a range of interacting drivers. The literature (for a review see, for example, Freeman, 1994, and Grubler, 1998) divides these drivers into three broad, overlapping categories: R&D, learning-by-doing, and spillovers. These drivers are distinctly different from other mechanisms that influence the costs of a given technology, such as. through economies of scale effects (see Box 2.3 below). Each of these entails different agents, investment needs, financial institutions and is affected by the policy environment. These are briefly discussed below, followed by a discussion of the empirical evidence supporting the importance of these sources and the linkages between them.
Research and Development (R&D): R&D encompasses a broad set of activities in which firms, governments, or other entities expend resources specifically to improve technology or gain new knowledge. While R&D covers a broad continuum, it is often parsed into two categories: applied R&D and fundamental research, and entails both science and engineering (and requires science and engineering education). Applied R&D focuses on improving specific, well-defined technologies (e.g. fuel cells). Fundamental research focuses on broader and more fundamental areas of understanding. Fundamental research may be mission-oriented (e.g. fundamental biological research intended to provide a long-term knowledge base to fight cancer or create fuels) or focus on new knowledge creation without explicit consideration of use (see Stokes (1997) regarding this distinction). Both applied R&D and fundamental research are interactive: fundamental research in a range of disciplines or research areas, from materials to high-speed computing, can create a pool of knowledge and ideas that might then be further developed through applied R&D. Obstacles in applied R&D can also feed research priorities back to fundamental research. As a rule of thumb, the private sector takes an increasingly prominent role in the R&D enterprise the further along the process toward commercial application. Similar terms found in the literature include: Research, Development, and Demonstration (RD&D), and Research, Development, Demonstration, and Deployment (RDD&D or RD3). These concepts highlight the importance of linking basic and applied research to initial applications of new technologies that are an important feedback and learning mechanism for R&D proper.
R&D from across the economic spectrum is important to climate change. Energy-focused R&D, basic or applied, as well as R&D in other climate-relevant sectors (e.g. agriculture) can directly influence the greenhouse gas emissions associated with these sectors (CO2, CH4). At the same time, R&D in seemingly unrelated sectors may also provide spillover benefits to climate-relevant sectors. For example, advances in computers over the last several decades have enhanced the performance of the majority of energy production and use technologies.
Learning-by-doing: Learning-by-doing refers to the technology-advancing benefits that arise through the use or production of technology, i.e. market deployment. The more that an individual or an organization repeats a task, the more adept or efficient that organization or individual becomes at that task. In early descriptions (for example, Wright, 1936), learning-by-doing referred to improvements in manufacturing labour productivity for a single product and production line. Over time, the application of learning-by-doing has been expanded to the level of larger-scale organizations, such as an entire firm producing a particular product. Improvements in coordination, scheduling, design, material inputs, and manufacturing technologies can increase labour productivity, and this broader definition of learning-by-doing therefore reflects experience gained at all levels in the organization, including engineering, management, and even sales and marketing (see, Hirsh, 1956; Baloff, 1966; Yelle, 1979; Montgomery and Day, 1985; Argote and Epple, 1990).
There are clearly important interactions between learning-by-doing and R&D. The production and use of technologies provides important feedbacks to the R&D process, identifying key areas for improvement or important roadblocks. In addition, the distinction between learning-by-doing and R&D is blurred at the edges: for example, everyday technology design improvements lie at the boundary of these two processes.
Spillovers: Spillovers refer to the transfer of knowledge or the economic benefits of innovation from one individual, firm, industry, or other entity to another. The gas turbine in electricity production, 3-D seismic imaging in oil exploration, oil platform technologies and wave energy, and computers are all spillovers in a range of energy technologies. For each of these obvious cases of spillovers there are also innumerable, more subtle instances. The ability to identify and exploit advances in unrelated fields is one of the prime drivers of innovation and improvement. Such advances draw from an enabling environment that supports education, research and industrial capacity.
Box 2.3 Economies of scale
Economies of scale refer to the decreases in the average cost of production that come with an increase in production levels, assuming a constant level of technology. Economies of scale may arise, for example, because of fixed production costs that can be spread over larger and larger quantities as production increases, thereby decreasing average costs. Economies of scale are not a source of technological advance, but rather a characteristic of production. However, the two concepts are often intertwined, as increased production levels can bring down costs both through learning-by-doing and economies of scale. It is for this reason that economies of scale have often been used as a justification for using experience curves or learning curves in integrated assessment models.
There are several dimensions to spillovers. Spillovers can occur between:
(1) Firms within an industry in and within countries (intra-industry spillovers).
(2) Industries (inter-industry spillovers).
(3) Countries (international spillovers).
The latter have received considerable attention in the climate literature (e.g. Grubb et al., 2002). Spillovers create a positive externality for the recipient industry, sector or country, but also limit (but not eliminate) the ability of those that create new knowledge to appropriate the economic returns from their efforts, which can reduce private incentives to invest in technological advance (see Arrow, 1962), and is cited as a primary justification for government intervention in markets for innovation.
Spillovers are not necessarily free. The benefits of spillovers may require effort on the part of the receiving firms, industries, or countries. Explicit effort is often required to exploit knowledge that spills over, whether that knowledge is an explicit industrial process or new knowledge from the foundations of science (see Cohen and Levinthal, 1989). The opportunities created by spillovers are one of the primary sources of knowledge that underlies innovation (see Klevorick, et al., 1995). There are different channels by which innovativions may spillover. For instance, the productivity achieved by a firm or an industry depends not only on its own R&D effort, but also on the pool of general knowledge to which it has access. There are also so-called ‘rent spillovers’, such as R&D leading to quality changes embodied in new and improved outputs which not necessarily yield higher prices. Finally, spillovers are frequent for products with high market rivalry effects (e.g. through reverse engineering or industrial espionage). However it is inherently difficult to distinguish clearly between these various channels of spillovers.
Over the last half century, a substantial empirical literature has developed, outside the climate or energy contexts, which explores the sources of technological advance. Because of the complexity of technological advance and the sizable range of forces and actors involved, this literature has proceeded largely through partial views, considering one or a small number of sources, or one or a small number of technologies. On the whole, the evidence strongly suggests that all three of the sources highlighted above – R&D, learning-by-doing, and spillovers – play important roles in technological advance and there is no compelling reason to believe that one is broadly more important than the others. The evidence also suggests that these sources are not simply substitutes, but may have highly complementary interactions. For example, learning from producing and using technologies provides important market and technical information that can guide both public and private R&D efforts.
Beginning with Griliches’s study of hybrid corn (see Griliches, 1992), economists have conducted econometric studies linking R&D to productivity (see Griliches, 1992, Nadiri, 1993, and the Australian Industry Commission, 1995 for reviews of this literature). These studies have used a wide range of methodologies and have explored both public and private R&D in several countries. As a body of work, the literature strongly suggests substantial returns from R&D, social rates well above private rates in the case of private R&D (implying that firms are unable to fully appropriate the benefits of their R&D), and large spillover benefits. Griliches (1992) writes that ‘… there have been a significant number of reasonably well done studies all pointing in the same direction: R&D spillovers are present, their magnitude may be quite large, and social rates of return remain significantly above private rates’.
Since at least the mid-1930s (see Wright, 1936), researchers have also conducted statistical analyses on ‘learning curves’ correlating increasing cumulative production volumes and technological advance. Early studies focused heavily on military applications, notably wartime ship and airframe manufacture (see Alchian, 1963 and Rapping, 1965). From 1970 through to the mid-1980s, use of experience curves was widely recommended for corporate strategy development. More recently, statistical analyses have been applied to emerging energy technologies such as wind and solar power. (Good summaries of the experience curve literature can be found in Yelle, 1979; Dutton and Thomas, 1984. Energy technology experience curves may be found in Zimmerman, 1982; Joskow and Rose, 1982; Christiansson, 1995; McDonald and Schrattenholzer, 2001).
Based on the strength of these correlations, large-scale energy and environmental models are increasingly using ‘experience curves’ or ‘learning curves’ to capture the response of technologies to increasing use (e.g. Messner, 1997; IEA, 2000; Rao et al., 2005; and the review by Clarke and Weyant, 2002). These curves correlate cumulative production volume to per-unit costs or other measures of technological advance.
An important methodological issue arising in the use of these curves is that the statistical correlations on which they are based do not address the causal relationships underlying the correlations between cumulative production and declining costs, and few studies address the uncertainties inherent in any learning phenomenon (including negative learning). Because these curves often consider technologies over long time frames and many stages of technology evolution, they must incorporate the full range of sources that might affect technological advance or costs and performance more generally, including economies of scale, changes in industry structure, own-industry R&D, and spillovers from other industries and from government R&D. Together, these sources of advance reduce costs, open up larger markets, and result in increasing cumulative volume (see Ghemawat, 1985; Day and Montgomery, 1983; Alberts, 1989; Soderholm and Sundqvist, 2003). Hence, the causal relationships necessarily operate both from cumulative volume to technological advance and from technological advance to cumulative volume.
A number of studies have attempted to probe more deeply into the sources of advance underlying these correlations (see, for example, Rapping, 1965; Lieberman, 1984; Hirsh, 1956; Zimmerman, 1982; Joskow and Rose, 1985; Soderholm and Sundqvist, 2003, and Nemet, 2005). On the whole, these studies continue to support the presence of learning-by-doing effects, but also make clear that other sources can also be important and can influence the learning rate. This conclusion is also confirmed by recent studies following a so-called ‘two-factor-learning-curve’ hypothesis that incorporates both R&D and cumulative production volume as drivers of technological advance within a production function framework (see, for example, Kouvaritakis et al., 2000). However, Soderholm and Sundqvist (2003) conclude that ‘the problem of omitted variable bias needs to be taken seriously’ in this type of approach, in addition to empirical difficulties that arise, because of the absence of public and private sector technology-specific R&D statistics and due to significant co-linearity and auto-correlation of parameters (e.g. Miketa and Schrattenholzer, 2004).
More broadly, these studies, along with related theoretical work, suggest the need for further exploration of the drivers behind technological advance and the need to develop more explicit models of the interactions between sources. For example, while the two-factor-learning-curves include both R&D and cumulative volume as drivers, they often assume a substitutability of the two forms of knowledge generation that is at odds with the (by now widely accepted) importance of feedback effects between ‘supply push’ and ‘demand pull’ drivers of technological change (compare Freeman, 1994). Hence, while modelling paradigms such as two-factor-learning-curves might be valuable methodological steps on the modelling front, they remain largely exploratory. For a (critical) discussion and suggestion for an alternative approach see, for example, Otto et al., 2005.
A range of additional lines of research has explored the sources of technological advance. Authors have pursued the impacts of ‘general-purpose technologies’, such as rotary motion (Bresnahan and Trajtenberg, 1992), electricity and electric motors (Rosenberg, 1982), chemical engineering (Rosenberg, 1998), and binary logic and computers (Bresnahan and Trajtenberg, 1992). Klevorick et al. (1995) explored the sources of technological opportunity that firms exploit in advancing technology, finding important roles for a range of knowledge sources, depending on the industry and the application. A number of authors (see, for example, Jaffe and Palmer 1996; Lanjouw and Mody 1996; Taylor et al., 2003; Brunnermier and Cohen, 2003; Newell et al. 1998) have explored the empirical link between environmental regulation and technological advance in environmental technologies. This body of literature indicates an important relationship between environmental regulation and innovative activity on environmental technologies. On the other hand, this literature also indicates that not all technological advance can be attributed to the response to environmental regulation. Finally, there has been a long line of empirical research exploring whether technological advance is induced primarily through the appearance of new technological opportunities (technology-push) or through the response to perceived market demand (market pull). (See, for example, Schmookler, 1962; Langrish et al., 1972; Myers and Marquis, 1969; Mowery and Rosenberg, 1979; Rosenberg 1982; Mowery and Rosenberg, 1989; Utterback, 1996; Rycroft and Kash 1999). Over time, a consensus has emerged that ‘the old debate about the relative relevance of “technology push” versus “market pull” in delivering new products and processes has become an anachronism. In many cases one cannot say with confidence that either breakthroughs in research “cause” commercial success or that the generation of successful products or processes was a predictable “effect” of having the capability to read user demands or other market signals accurately’ (Rycroft and Kash, 1999).