‘Linear’ developments

There was certainly theorizing about innovation before the Second World War. There were many sociologists, anthropologists, economists, business school professors, and industrialists proposing stage models of cultural and technological change from the 1920s to the 1940s (Godin, 2017) (this was before any anglophone had a chance to read Schumpeter). Indeed, Godin (2017, 2011) credits Maurice Holland – a director at the US National Research Council – with articulating the core ideas of the linear innovation model during the 1920s. But it was certainly the success of science during the Second World War – particularly the Manhattan Project – that solidified linear thinking within a ‘post-war paradigm’ of popular and academic theorizing (Nemet, 2009).

The most prominent voice for this paradigm was US presidential advisor Vannevar Bush. He had led the wartime US Office of Scientific Research and Development. And it was his report – Science, the Endless Frontier (Bush, 1945) – that most famously argued for postwar public investments in science to radically advance medicine, industry, and national defence (1945). There is no doubt that the Bush report was widely read and influential. It is frequently cited as either the source or turning point towards the linear model of innovation (for example, Irvine and Martin, 1984b; Freeman, 1996; Lundvall, 2013b). However, Bush did not elaborate an innovation model – sequential or otherwise (Godin, 2017). His report merely argued for a causal link between basic research and socioeconomic progress. Godin (2017) suggests that the closest Bush came to articulating a linear model was through his connection to Rupert Maclaurin.

Maclaurin, an economic historian at the Massachusetts Institute of Technology (MIT), had been secretary on one of the four committees that contributed to Bush’s report. He had also been a student of Schumpeter, whose ideas he later developed into a staged innovation process (Maclaurin, 1949, 1950) – a theoretical framework that would eventually become known as the linear model (Godin, 2008, 2017). Along the way, however, Maclaurin’s work would be forgotten; qualitative research was not respectable in economics at that time (Godin, 2008, 2017). We would also forget the many others, over many decades, who developed various stage and technology-push models. These would be subsumed under the ‘linear model’ label around the 1960s. So, despite the canon, Bush did not give us the linear model.

But Bush did influence the thought paradigm in which the linear model solidified: making science instrumental to technological change. In other words, science became a tool for producing technology. Pfotenhauer and Juhl go so far as to say that Bush’s report ‘castrates the government mandate by confining the state’s responsibility to the front end of the pipeline’ (Pfotenhauer and Juhl, 2017, p 72). This fundamentally reframed the role of government with respect to science, technology, and innovation. The idea that advancements in basic science drove technological progress then ‘held sway for 20 years or so’ (Martin, 2010, p 3). And, as we will see in Chapter 7, it continues to lurk in the background today through standardized innovation statistics (Godin, 2017).

Along the way, a debate emerged about the direction of the relationship between technoscientific progress and market demand. Those who ascribed to the 1940s postwar paradigm described an innovation process that was initiated and driven – or ‘pushed’ – by advancements in technoscientific knowledge. But in the 1960s, an alternative hypothesis emerged suggesting that market demand served to ‘pull’ – or determine the speed and direction of – innovation (Schmookler, 1966; Rosenberg, 1969). These two positions were entrenched by results from the US Department of Defense’s HINDSIGHT project in 1966 (arguing for ‘pull’) and the US National Science Foundation’s TRACES project in 1969 (arguing for ‘push’). In retrospect, we reject both positions as highly simplistic. Although he was a key proponent of demand-pull, John Langrish (Langrish et al, 1972; Langrish, 1974) now characterizes the whole debate as somewhat silly (Langrish, 2017). In describing it, he invokes the imaginary two-headed ‘pushmi-pullyu’ creature from the Dr Doolittle books (Lofting, 1920). Most stories about innovation theory gloss over the details of this silliness, especially the short-lived arguments for a demand-pull model (Godin, 2017). But Godin is right that this was a critical juncture in the shift towards a market bias for innovation research.

So, while the demand-pull model was quickly rejected, it had a lasting influence on the language of innovation studies. The push versus pull debate also triggered an influential series of studies on scientific instrument innovation (Utterback, 1971b, 1974; Freeman, 1974; von Hippel, 1976, 1988; Rosenberg, 1982).

‘Pushmi-pullyu’ devices

In the late 1960s and early 1970s, scientific instruments became a key empirical testing ground for the demand-pull model. As Chris Freeman and Luc Soete would later explain, ‘the increasingly intimate relationship between new materials, new process development and fundamental research is nowhere more apparent than in the field of instrumentation’ (1997, p 128). In other words, scientific instruments were as close as anyone could come to studying a real pushmi-pullyu – it was not immediately clear which end was the front. And, unsurprisingly, studying them eventually moved the debate beyond one-way linearity.

Daniel Shimshoni’s PhD thesis at Harvard (1966)¹ may have been the earliest study focused on innovation in scientific instruments. Trained as an engineer at Princeton and CalTech, Shimshoni helped build bombers during the war. He then led the development of the Israeli Air Force and became the first Director of the Israeli National Council for Research and Development. Returning briefly to the US, he completed his doctoral thesis on the interorganizational mobility of scientist-entrepreneurs in the instrument industry (Shimshoni, 1966, 1970). In his thesis (1966) and a subsequent paper in Minerva (1970), Shimshoni concluded that

an overwhelming majority of instrument innovations involved the movement of technical leaders to form their own companies or to join recently established firms. The scientific basis and the essential enabling technology of most of the innovations considered in the present study originated in university, government or large industrial laboratories, while new instrument products were largely the work of small firms. (Shimshoni, 1970, p 85)

And so, Shimshoni’s study came down on the side of a push model – albeit with people moving in all directions. He therefore argued for policy makers to emphasize public research and development (R&D) expenditures and to encourage employment mobility for all those involved. He then turned to a scholarly focus on Israeli politics (including science and technology policy). However, his scientific instrument study would have a direct, cited influence on some of the most famous innovation scholars, including James Utterback (1969, 1971a, 1971b, 1974), Eric von Hippel (1975, 1976, 1988), Nathan Rosenberg (Mowery and Rosenberg, 1979), and Chris Freeman (1974, 1982; Freeman and Soete, 1997).

Only a few years after Shimshoni, Utterback would complete his own PhD on innovation in scientific instruments (Utterback, 1969). Under the supervision of Donald Marquis at MIT, Utterback set out to understand idea generation and problem solving as the ‘first phases’ of innovation. He examined 32 cases of scientific instrument development in the Boston area. Like Shimshoni, he recognized that spin-off companies were the main mode of entry in the instrument industry. But, unsurprisingly, his focus was on an early version of the Myers and Marquis (1969) demand-pull model. Based on his sample, Utterback concluded that new product ideas were ‘predominantly (twenty-four of thirty-two cases) stimulated by information about a need’ (Utterback, 1969, p 2). His eight remaining cases were ‘stimulated by recognition of a technical possibility’ – yet, he quickly argued, even those ideas were ‘most often encountered in the course of work on a related problem’ (Utterback, 1969, p 2). Utterback would go on to publish this work in IEEE Transactions (Utterback, 1971a), the Academy of Management Journal (Utterback, 1971b), and Science (Utterback, 1974). Hedging his bets, he would argue that ‘most often’ (Utterback, 1971b, p 83), given ‘the weight of evidence’ (Utterback, 1971a, p 131), and ‘in most cases’ (Utterback, 1974, p 183), ‘market forces appear to be the primary influence on innovation’ (Utterback, 1974, p 621). He also noted that the source of information about new instrument needs (framed as economic opportunities) was often outside the firm. Nonetheless, his study kept innovation neatly within the boundaries of the firm, ‘or divisions of firms in the Boston area’ (Utterback, 1969, p 2).

Two years after Utterback’s journal publications, Eric von Hippel took this debate beyond the firm in his own PhD thesis (1975) and in a subsequent article in Research Policy (von Hippel, 1976). Here, von Hippel examined the development of 111 scientific instrument innovations in the US, including four broad classes of scientific instruments: gas chromatographs, nuclear magnetic resonance spectrometers, ultraviolet spectrophotometers, and transmission electron microscopes. He used the Myers and Marquis (1969) demand-pull model and mapped the steps undertaken by scientists – which he referred to as ‘users’ – versus the steps undertaken by scientific instrument manufacturing firms, which he referred to as ‘producers’. He found that in nearly all cases, scientist-users performed all steps of the innovation process up to and including the development and use of precommercial prototypes. Only after this precommercial testing did producers in the private sector acquire and begin to both commercialize and incrementally improve these instruments.

Frank Spital, one of the research assistants on von Hippel’s project (von Hippel, 1988, 1975), later extended the research (Spital, 1979). He added some nuance to the original observations, determining that scientist-users were responsible for major innovations and many minor improvement innovations except those that were initiated by manufacturers in response to their competitors. Nearly 20 years later, William Riggs worked with von Hippel to revisit user innovation in the context of scientific instruments. Riggs and von Hippel (1994) reconfirmed the dominant role of users in scientific instrument innovation. In their new dataset, they found ‘user innovators almost never gained direct financial benefit from their instrument innovations when those were commercialized by instrument firms’ (Riggs and von Hippel, 1994, p 465). They further suggested that user-driven scientific instrument innovations are more radical than producer-driven innovations, which are more incremental.

Von Hippel explained that he chose this empirical focus because Shimshoni and Utterback had already ‘ascertained that innovation in response to user need was prominent in scientific instruments’ (von Hippel, 1976, p 215). And so, this class of technology was ideal – not only for testing the demand-pull model, but also for building a stellar career trajectory around innovation outside the firm. Yet there was something about scientific instruments that forced von Hippel to quickly move on. He wrote that ‘to explore this matter, I decided to conduct a second study in other, more “normal” fields, before suggesting that users-as-innovators might be a generally significant phenomenon’ (1988, p 20). Indeed, he is now known for his work on open, distributed, and free innovation. His work is more readily associated with empirical domain of open-source software.

This response gives us a rare insight into the processes that focus our scholarly attention one way or another. Not only do we tend to dismiss nonconformist research, but we also tend to dismiss interesting qualitative cases in favour of large statistical datasets. I will return to this latter point in Chapter 7.

For now, let me emphasize that von Hippel’s (1976) paper was a turning point for innovation research. Bogers et al argue that this work was the first to notice that users can be innovators, and that it ‘set off a substantial amount of research investigating users as the sources of innovation’ (2010, p 859). The findings were reprinted as a key part of von Hippel’s book The Sources of Innovation (1988), which Fagerberg et al (2012a) ranked as thirteenth on their list of top contributions to innovation studies. Over time, von Hippel’s research programme helped to dispel the myth that the locus of innovation activity (von Hippel, 1976) rests within manufacturing firms. He established that the ‘locus’ of this activity can also rest in ‘users’.

Von Hippel and his colleagues foreshadowed this decline of linear models in two ways. First, they directly observed two-way – bidirectional – interaction at the point where precommercial instruments created by users were transformed into commercial instruments by producers (von Hippel, 1976). Second, they indirectly foreshadowed the fall of this model by selecting a context where scientists were operating at both ends of the linear flow: exerting both science-push and demand-pull. Von Hippel and his colleagues positioned their language on the ‘demand-pull’ side of the debate. However, the ‘locus’ of this demand rested in those individuals who were pushing back the frontiers of science. There would have been no language available to reconcile this complexity from inside the pushmi-pullyu debate.

There would also have been no discursive space to consider the public nature of these innovations in the US in the 1960s and 1970s (which was hungover from McCarthyism). But there is no doubt that von Hippel’s ‘users’ must have been employed in public or quasi-public organizations (for example, public or land grant universities, public research laboratories). There is no mention of public organizations in von Hippel’s scientific instrument studies. There is only a passing reference in later work to ‘universities’ writ large (with no discussion of public/private distinctions) and ‘government laboratories’ (Riggs and von Hippel, 1994, pp 461–2). And yet, von Hippel’s users were most definitely ‘pulling’ scientific instrument innovation from within the same institutions that were ‘pushing’ technological change. And so, let’s move on from these pushmi-pullyu studies with the sense that they were unintentionally and unconsciously observing public sector innovation in goods.

‘Chain-linked’ processes

In the 1950s and 1960s, the linear nature of innovation was mostly taken for granted. Debate focused on directionality: technology-push and demand-pull were seen as mutually exclusive hypotheses (Chidamber and Kon, 1994; Nemet, 2009). But by the 1980s, it was widely accepted that ‘innovation is neither smooth nor linear, nor often well-behaved’ (Kline and Rosenberg, 1986, p 285). Research on science, technology, and innovation became loaded with words like coupling, interaction, and symbiosis (Godin, 2017). Linear models persisted, but became buried beneath layers of feedback loops (Godin, 2017).

The most well known of these layered models was Stephen Kline and Nathan Rosenberg’s ‘chain-linked model’ of innovation (Kline, 1985; Kline and Rosenberg, 1986). It may not have been all that novel (Godin, 2017). It also had ‘the problem that if you start trying to explain [the chain-linked model] to policy makers their eyes start glazing over!’ (Martin, 2010, p 4). But it is often cited as the turning point away from linear models and towards a systems approach (for example, Martin, 2013). This is because the chain-linked model does not assume that innovation begins with research or with market demand. Instead, it highlights the ongoing interactions between R&D activities.

Based on his 30 years of consulting to industry, Kline proposed a ‘linked-chain’ model of innovation as an improvement to the ‘oversimple and inadequate’ linear model (Kline, 1985, p 36). In his model, Kline separated research activities (which he defined as the processes that produce knowledge) from the product development process (which he labelled as ‘the chain-of-innovation’) (Kline, 1985, p 36). He then argued that innovation involved not one sequential process, but five flows or pathways. In the first paper, he illustrated this across seven different figures (the paper was only ten pages long). The figures included 29 arrows indicating the flows between research, development, and stocks of knowledge. These are the details in the model that tend to make eyes gloss over. They are discussed extensively in the original papers. But for my purposes, one line is more interesting than all the others. This is the line Kline used to denote the critical role of scientific instrument innovation.

Kline labelled this the ‘initiation of science link’ (1985, p 41). He grounded it in Derek de Solla Price’s (1984) notion of scientific instrumentalities, explaining that ‘the production of new instruments, tools, and processes has in many instances made possible new forms of research’ (Kline 1985, p 41). Kline’s version of the paper listed the telescope, the microscope, and radiometric dating as historical examples, and this discussion was expanded in Kline and Rosenberg (1986). Kline and Rosenberg also discussed the CAT scan, the electroencephalogram, and the ‘digital computer’ as examples of the ongoing ‘feedback from innovation, or more precisely from the products of innovations, to science’ (1986, p 293, emphasis added). And so, the chain-link model recognized the important flow of new instruments and techniques into science. This flow was seen as critical to the model, but it is easily overlooked in the muddle of boxes, lines, and arrows.

However, it should be noted that scientific instruments were represented by a one-way arrow in the chain-link model. This was consistent with Nathan Rosenberg’s work at the time. Kline’s (1985) pathways between research and development were based in Rosenberg’s earlier assertion that ‘science is not entirely exogenous’ (Rosenberg, 1982, p 142). In other words, the chain-link model did not consider science to be disconnected from the market; instead, it considered scientific research and technological development to be directly and indirectly linked. Rosenberg developed this sense of the links between science and markets in a 1981 conference paper. The paper appears in his book Inside the Black Box (1982) – another of the top twenty contributions to innovation studies (Fagerberg et al, 2012a). There, Rosenberg argued that ‘improvements in instrumentation, through their differential effects upon the possibilities of observation and measurement in specific subfields of science, have long been a major determinant of scientific progress’ (1982, p 158). In other words, technology pushes science. Rosenberg’s understanding of scientific instrument innovation at that time appears to have been influenced by the technology-push discourse. Although the chain-link model was attempting to overcome linear flows, scientific instruments were illustrated and described as flowing in one direction. However, Rosenberg knew that this analysis was ‘only the first small step on a long intellectual journey’ (Rosenberg, 1982, p 142).

Here Rosenberg echoes the recurring sentiment that scientific instrumentalities are a highly important innovation context due to their wide diffusion through society. This diffusion is at least partly thanks to the work of private industry. Like others (von Hippel, 1976; Spital, 1979; von Hippel 1988; Riggs and von Hippel, 1994), Rosenberg (1992) notes that private sector manufacturers make incremental improvements to scientific instruments. These improvements in performance, versatility, price, and usability for those with less training in the original applications of the technology help to facilitate diffusion of the innovations. But further to his collaboration with Kline (1986), Rosenberg reminds his readers that innovation is not linear. A new scientific instrumentality can stimulate follow-on research with respect to performance, materials, or ancillary technologies, as well as open new fields of research, be adapted to other fields of research, and be adapted to commercial applications (Rosenberg, 1992). He concludes that, in the context of scientific instruments, the ‘scientific research community undertook radical innovative initiatives that led, in many cases, to the eventual supplying of its own internal demand and, in the process, provided large external benefits as well’ (Rosenberg, 1992, p 389).

By 1992, Rosenberg agreed with de Solla Price (1984) on the widespread importance of scientific process innovations as well as the nature of the relationships between scientists and the scientific instrumentality industry. They both rejected the idea that knowledge flows one-way from science to industry via instruments or any other means. It is appropriate to think of scientific instruments as the inputs or ‘capital goods of the scientific research industry’ (Rosenberg 1992, p 381), yet it is also important to recognize that ‘scientific instrument firms are quite often spin-offs from great national facilities in experimental science … [and] the mechanism for the entrepreneuring and expansion of such crucial high technology laboratories has been government procurement’ (de Solla Price, 1984, p 18). And so, the relationship between research organizations and instrumentality companies can be described as ‘interactive’ (Rosenberg, 1982, p 158), ‘complementary’ (Rosenberg, 1992, p 386), or ‘symbiotic’ (Rosenberg, 1992, p 386). The research organizations are primarily but not exclusively populated by scientists, while the private instrument companies are primarily but not exclusively populated by engineers/technicians. According to Rosenberg, ‘the migration of scientific instruments to industry has been matched by a reverse flow of fabrication and design skills that have vastly expanded the capacity of university scientists to conduct research’ (1992, p 386). In other words, Rosenberg (1992) understood scientific instrument innovation to be a critical bidirectional link between ‘research’ and technological ‘development’. It was a key point of linkage, coupling, or interaction in the broad process of innovation. This complexity might be reduced to a single one-way line in a transitory model, but there is no doubt that scientific instrument innovation was important to Kline and Rosenberg’s theorizing.

Changing lenses

The chain-linked model of innovation is often described as the intermediary step that led from old linear process models to new systems approaches. And thus far I have followed that storyline: push models led to pull models, which led to the chain-linked model and ultimately the systems approach. However, I will now argue that the systems approach and the chain-linked model have overlapping origins: both were responses to the push/pull debate, and both were shaped by an understanding of scientific instrument innovation.

This means that process models and system models are different ways of framing innovation. The systems approach does not merely add dimensions (or linkages, interactions, etc.) atop unidirectional models to produce a more complex chain. Instead, it shifts focus away from the work of individual actors (Martin, 2010). This is a qualitatively different perspective on innovation. There are certainly strong arguments for using a systems approach rather than a (linear) process model. But regardless of what these models might give us, the point is that ‘process’ and ‘system’ are different lenses for observing the world of innovation. The systems approach is not merely a better process lens; it is a different lens entirely. And in the 1960s, both lenses were being honed through research on scientific instrument innovation.

Focusing on ‘systems’

The innovation systems approach is often traced back to the work of Christopher Freeman (1987), Bengt-Åke Lundvall (1988), and Richard Nelson (1993) in the late 1980s and early 1990s. But Lundvall himself goes back even further. He argues that the first articulation of innovation systems theory was Freeman’s (1974) analysis of results from Project SAPPHO (Scientific Activity Predictor from Patterns with Heuristic Origins). Beginning in 1967, SAPPHO (Curnow and Moring, 1968; Rothwell et al, 1974) was the first major undertaking of the newly formed Science Policy Research Unit (SPRU) at the University of Sussex. SPRU is perhaps the most famous centre for research on science, technology, and innovation policy (Fagerberg et al, 2012b; Soete, 2019). Christopher Freeman was its founding director and would come to be known as a ‘founding father’ of innovation studies (for example, Lundvall, 2013a; Martin, 2013; Soete, 2019). This was in no small part due to that first major research project. The SAPPHO results ‘attracted much attention, particularly in industry’ for both SPRU and Freeman (Fagerberg et al, 2011, p 901). Lundvall argues that Freeman’s analysis of SAPPHO was the first recognition of ‘the importance of interaction between individuals and departments within firms as well as the important interaction with suppliers, customers, and science institutes’ (Lundvall, 2013b, p 41). In other words, SAPPHO provided much of the theoretical foundation for the systems of innovation approach.

Empirically, SAPPHO was a study of science-intensive industrial innovation. The first phase (Curnow and Moring, 1968) was an examination of 58 innovations in chemicals and scientific instruments (see Table 5.1 in Freeman, 1982; and Table 8.1 in Freeman and Soete, 1997). Early on, Freeman was asked to explain this empirical focus. He noted that the work had been influenced by ‘the capabilities of the people that were engaged on the study’ and by prior studies: work by Enos on petrochemicals and Shimsonhi on scientific instruments (Williams, 1973, p 252). However, he explained that ‘there had been no special significance about this [choice of industries] and that they would like in the future to investigate other branches’ (Williams, 1973, p 252). Indeed, Freeman initially framed scientific instruments as representing a subset of the broader electronics industry (Freeman, 1973, 1974, 1982). Decades later, Freeman and Soete (1997) would retrospectively explain that scientific instrument innovations were key to the SAPPHO study because they were found at the intersection of fundamental research and new technology development. This seems to be why SAPPHO data were so useful to Freeman in moving beyond linear push and pull logic: much of the data came from observations of pushmi-pullyu devices. However, this is only evident in retrospect. The qualities of scientific instrument innovation were not an explicit part of the SAPPHO design.

The resulting analysis identified 27 firm-level factors that differentiated between successful and unsuccessful innovations. Most of the success factors identified by the SAPPHO team were related to marketing practices, and some were related to organizational structure (Freeman, 1974). But according to Freeman, ‘the single measure which differentiated most clearly between success and failure was “user-needs understood”’ (1974, p 188). He explained that successful innovations were the result of a close ‘match’ between technology and user needs. He also noted that ‘better external communications were associated with success, but the strongest difference emerged with respect to communication with that specialized part of the outside scientific community which had knowledge of the work closely related to the innovation’ (Freeman, 1974, p 189, emphasis in original). The researchers had asked: ‘What was the degree of coupling with the outside scientific and technological community in the specialized field involved?’ (Freeman, 1974, p 179, emphasis added). And it is in these ideas of matching and coupling between producers, users, and researchers that we can most clearly see the beginnings of the systems of innovation approach. From these results we begin to get the concept of ‘interactive learning’, which has been described as the ‘theoretical core’ of the systems of innovation approach (Lundvall, 2013b, p 32).

Of course, the systems approach is not normally referred to as a theory or model. However, it is certainly of that ilk (Godin, 2017). It has been described as a ‘focusing device’ – a kind of social scientific theory (Lundvall, 1992). It focuses our attention on processes of interactive learning that unfold among actors and within an institutional environment (for example, rules and norms). This framing was more fully developed in Freeman’s later book on Japan’s national innovation system and in decades of subsequent research on national (for example, Lundvall, 1988; Nelson, 1993), regional (Asheim and Isaksen, 2002; Asheim et al, 2011), and ‘sectoral’ systems (see Malerba, 2005). In Chapter 6, I will engage with the problem of defining the boundaries for an innovation system and the ‘danger of getting “lost in the woods” while searching for the institutional component’ (Doloreux and Parto, 2005, p 146). What is important now, in this chapter, is the assertion that ‘Freeman’s experiences from project SAPPHO provided the ground for the innovation systems perspective’ (Lundvall, 2013b, p 41).

Prior to Freeman’s analysis, the project seemed poised to take a side in the pushmi-pullyu debate. After all, SAPPHO had been designed in the aftermath of the HINDSIGHT and TRACES research projects – those two large studies that entrenched the push and pull perspectives in the US. It was also conducted in the wake of a high-profile British study (from Manchester Business School) that examined winners of the Queen’s Award for Innovation. That study had concluded in support of the demand-pull argument (Langrish et al, 1972; Langrish, 1974). So, when Curnow and Moring (1968) presented the plans for SAPPHO in Futures, it was not surprising that they articulated a linear model on the first page. Theirs was a glossy technology push model consisting of three stages: ‘technical, industrial and commercial steps, and then the commercial acceptance’ (Curnow and Moring, 1968, p 82). In this, SAPPHO appeared like it might become SPRU’s counterpoint to the Manchester study, much like HINDSGHT and TRACES were a contrasting set in the US. However, the demand-pull framing suggested by Curnow and Moring (1968) was soon replaced by Freeman’s ideas about complexity and systems.

By the time SAPPHO data collection was under way, Freeman had already written some systems language into the first SPRU annual report (Fagerberg et al, 2011; 2012a; Lundvall, 2013b). And once the first phase of SAPPHO was completed in 1970, Freeman began to clearly articulate his perspective in both the SPRU mission statement (see Fagerberg et al, 2011) and in his analysis of the SAPPHO results (Freeman, 1973). His first written analysis of SAPPHO was presented at a conference of the International Economic Association in 1971 and published in the proceedings two years later (Freeman, 1973). That publication was accompanied by ‘minutes’ of the conference discussion (Freeman, 1973). There we can see that the lead discussant wasted no time applauding Freeman’s approach to the push-pull debate. He said that Freeman’s SAPPHO paper ‘clearly indicated that single factor explanations were not sufficient to explain success. It was clear that both elements of demand and of research had to be taken into account in explaining the difference between success and failure’ (J.J. Paunio, paraphrased in Williams, 1973, p 246). Here, in 1971, Freeman was already moving his peers towards a ‘systems’ view.

Although the SAPPHO methodology was hotly debated at the IEA conference (Williams, 1973), Freeman would use the results to close debate on the push and pull models (see also Godin, 2017, pp 114–15). His conference paper became the core of his tremendously influential book The Economics of Industrial Innovation (Freeman, 1974, 1982; Freeman and Soete, 1997). That book ‘for a long time held a virtual monopoly in presenting the “state of the art” of knowledge in the field’ (Fagerberg et al, 2012a, p 1136) and had a substantial influence on other major works, notably Nelson and Winter (1982). Starting in the first edition, Freeman asserted that ‘innovation is essentially a two-sided or coupling activity’ (1974, p 165). He wrote off linear models, saying: ‘Whilst there are instances in which one or the other may appear to predominate, the evidence of the innovations considered here points to the conclusion that any satisfactory theory must simultaneously take into account both elements’ (Freeman, 1974, p 166, emphasis in original). In later editions, Freeman would go further in weighing the SAPPHO evidence against Schmookler (1966) (pull) and ‘counter-Schmookler’ (push) positions (Freeman, 1982, p 128; Freeman and Soete, 1997, pp 219–20). He would suggest that the push-pull debate was driven by a difference in focus, with one side (push) emphasizing radical innovations and the other (pull) measuring more incremental ones. In his view, the push and pull models simply ‘measure something rather different’ from each other (Freeman, 1982, p 128).

Interestingly, Freeman’s book barely mentions the limits of his own perspective. Those had been raised by the audience at the 1971 conference. And there, Freeman had admitted that the SAPPHO data did not fully account for government–firm relations or the role of public sector organizations as key users of scientific instrument innovations. Nonetheless, ‘he admitted that this did have an effect’ (Williams, 1973, p 253). He knew that public organizations were important to the successful scientific instrument innovations that had been studied, and that this was not captured in the SAPPHO data or results. Indeed, Freeman’s book positioned government only as R&D financier and policy maker. It said ‘in capitalist societies most industrial R and D is performed by enterprises and innovations are made by firms’ (Freeman, 1974, p 287). There was no space in Freeman’s systems perspective for public employees to become user-innovators. And although the SAPPHO results would be cited by von Hippel (1976) in building the case for user innovation, he did not acknowledge the public sector role either.

Models as instruments

Across several high-profile studies of scientific instrument innovation, we have seen that similar empirical material can be observed using very different epistemic lenses. Observations of scientific instrument innovation were used to support claims about the importance of pushes (Shimshoni, 1966, 1970), pulls (von Hippel, 1976; Utterback, 1971b, 1974), links (Kline, 1985; Kline and Rosenberg, 1986), and coupling/interaction (Freeman, 1974, 1982). This is not to say that any one author’s observations were wrong; rather, the mistake is in assuming that theoretical models flow from empirical matters. Models are not merely generalizations or abstractions from material reality. They are also tools for observation and sensemaking. They are instrumental technologies that shape – and are shaped by – our efforts to understand the world around us. The relation between theory and empirical reality is not linear (or even chain-linked).

This is a foreign and potentially invasive concept in innovation studies, where the positivist paradigm still reigns. But it is a rather old idea elsewhere. As I suggested in Chapter 1, STS has a rather more robust understanding of how scientific instrumentalities are woven into science, technology, and innovation. Instruments are highlighted as actors (or actants) in the laboratory studies of Latour and Woolgar (1986) and ‘after’ (Law and Hassard, 1999). And scientific instruments have been a focal point for many studies in the history (for example, Hughes, 1976; de Solla Price, 1984; Taub, 2011), sociology (for example, Joerges and Shinn, 2001; Shinn, 2005), and philosophy (for example, Marcacci, 2019) of science and technology. These different lines of inquiry share an interest in the epistemology of scientific instruments: a sense that these artefacts defy old Greek ideas about differentiated kinds of knowledge. In other words, scientific instruments are at once episteme and techne (and, as we will see in the next section, they carry ethical concerns as well).

David Baird’s book Thing Knowledge is the most expansive of these investigations. He includes models as one form of scientific instrument (Baird, 2004). However, his focus is on physical models and physical instruments. This is because his goal is to outline a materialist epistemology – a sense that ‘the material products of science and technology constitute knowledge … in a manner different from theory, and not simply “instrumental to” theory’ (Baird, 2004, p 18). And so, Baird establishes a fundamental separation between physical and conceptual models. For Baird, models-as-instruments require maintenance effort, whereas noncorporeal models, like mathematical equations, do not. He argues that conceptual models ‘exist in the unchanging, self-sufficient world of ideas’ (Baird, 2004, p 35). This is unlike his former colleague Richard Hughes, who was interested in how the diversity of conceptual and physical models ‘provide representations of parts of the world, or of the world as we describe it’ (Hughes, 1997, p S325). While I appreciate Baird’s work, my perspective is closer to that of Hughes (1997): I am approaching models as instrumentalities.

Yes, some theories are primarily tacit. However, this does not make them less powerful actors in the construction of knowledge. Indeed, this chapter suggests that several ideas about innovation – pushes, pulls, chains, and couplings – were powerfully shaping the ways in which prominent innovation scholars understood their empirical observations. These ideas were what my colleague and conference co-adventurer Chris Hartt has called ‘noncorporeal actants’ (Hartt, 2013; Hartt, 2019). His theory of the noncorporeal actant uses Actor–Network Theory (ANT), historiography, and sensemaking to consider how ‘ideas, values, concepts and beliefs are acting upon the decision maker to choose their actions’ (Hartt, 2019, p 1). Although his work is focused on noncorporeal actants in past managerial practices, Chris’ work helps us position conceptual models alongside other scientific instruments in the production of knowledge. It establishes symmetry between corporeal and noncorporeal actants.

Meanwhile, let’s notice that most theoretical models leave physical traces. Some become things (for example, ball-and-stick molecular models) and others are written/illustrated on paper – or mostly on magnetic drives in large server farms. In this way, theories – archived as artefacts, words, and diagrams – are traces of past places and times. Some traces persist and others do not. This is a historicized view of ‘extant’ theory – the theory available to us today. It recognizes that, over time, knowledge ‘which fits with conventional wisdom (not, significantly, with the empirical evidence) is preserved while the rest is truncated’ (Jacques, 2006, p 34). The bits and pieces that remain become tools – scientific instrumentalities – for understanding the present. They enable and constrain our present-day thinking.

Instrumental neoliberalism

We get a new angle on the constraints imposed by these innovation models when we look at them as scientific instrumentalities being deployed in the study of scientific instrument innovation. Again, we have the benefit that different models relate to scientific instrument innovation in different ways. We can certainly see how chain-linked and systems models are improvements over ‘simplistic’ linear thinking. But we can also see how, despite their differences, all the models in this chapter focus on firms and markets. Multiple times, prominent innovation scholars came close to noticing that physical technologies were being developed within universities and public research laboratories. But each time, they turned away. Their models obscured any direct observation of public innovation.

Again, this is not to say that any of these scholars was wrong. Rather, this points to the way in which neoliberalism is ‘scripted’ (Akrich, 1994) into the shared toolkit of innovation studies. Pfotenhauer and Juhl (2017) have already argued that the push, pull, chain-linked, system, and triple helix³ models all presume a market ontology. They write that ‘the history of innovation models has remained captive to an instrumental dyadic logic that seeks to connect technologies with markets, and that sees the state as both external and subservient to these two poles’ (Pfotenhauer and Juhl, 2017, p 87). Godin (2017) agreed, pointing out the irony that it was the success of scientific research during the Second World War that led to the valorization of technological innovation and ‘marginalization’ of research starting in the 1950s. Later, he expanded on this analysis, arguing that the idea of ‘research as a source of progress’ now ‘competes, for better or worse, with that of the market: that technological innovation is the commercialization of new goods, and the principal agents in the process are firms, not scientists’ (Godin, 2020, p 145). But while his historical analysis is more than sound, I think Godin was too soft in suggesting ‘competition’ between public science and private markets. I prefer the stronger conclusion drawn by Pfotenhauer and Juhl: ‘The history of innovation models can be interpreted as one of systematic exclusion of the political state and the constituency it serves in favour of a purified, technocratic understanding of what innovation is, what it is for, and who needs to be involved’ (Pfotenhauer and Juhl, 2017, p 79, emphasis added).

I share the view that the state is systematically sidelined in innovation studies. This was evident in the way that Freeman (1973, 1974), von Hippel (1976), and Kline and Rosenberg (1986) all asymptotically approached the place of public sector organizations in scientific instrument innovation. But this does not imply any ‘dark’ intent on their parts. It only confirms that, whether consciously or not, neoliberal politics were embedded in the innovation models that were at play. Models-as-instruments are therefore not only technical and epistemic, but also ethical. However quietly, theoretical models carry values. Although this point might be commonplace for those trained in STS, it is crucial for advancing (critical) innovation studies. I present it here to spur sensemaking about innovation theory – to argue, as Karl Weick (1996) did in organization studies, that these tools are weighing us down.

Instrumentality innovation

In assuming the editorship of Administrative Sciences Quarterly, Weick (1996) famously presented the Mann Gulch and South Canyon wildfire disasters as an allegory for the future of organization studies. He described how 27 firefighters died, within sight of safe zones, because they failed to drop the heavy tools that were slowing down their escape, despite direct orders to do so. He then reviewed several slightly less existential threats to organization studies and discussed the ‘heavy’ scholarly tools impeding progress. His editorial called for a return to ‘the lightness associated with “the play of ideas”, improvisation, and experimentation’, but warned that this would be impeded ‘when dropping ideas or keeping them becomes confused with dropping or keeping group ties’ (Weick, 1996, p 312). Throughout this book I will argue that his advice and warning apply more acutely to innovation studies.

In this chapter we have seen that innovation models are heavy and potentially problematic tools. We have also seen that they are strongly connected to the identity of innovation studies as a field of scholarship. Indeed, my argument goes further than Weick. I have suggested that innovation models reflect more than shared ideas; they also reflect shared (neoliberal) values. I worry that some might see these values as core to any community of innovation scholarship. They might therefore misconstrue my critique as a personal or community attack. To a certain extent, Godin was right when he said ‘the persistence of the market-first perspective speaks more about the values of the scholars promoting it than to its contribution to understanding technological innovation’ (2017, p 125). But I take solace in the playfulness of the pushmi-pullyu debate. Yes, this debate was academically fierce and politically charged. However, as we have seen in this chapter, it spurred tremendous scholarly innovation. Studying the ‘unusual’ field of scientific instruments, from many different perspectives, produced a range of novel ideas about innovation. Models were being retooled left, right, and centre. And so, while the field might now be experiencing ‘disciplinary sclerosis’ (Martin, 2013, p 179), it also has a history of examining phenomena that do not quite fit and thereby developing new social-scientific instrumentalities. We can draw on that history to address the ‘heaviness’ of the models explored here and the ‘heaviness’ of other instrumentalities in the chapters to come.